WeightWatcher.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# WeightWatcher\n",
    "\n",
    "https://calculationconsulting.com"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:51:16.544327Z",
     "start_time": "2019-03-05T23:51:16.531457Z"
    },
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# Suppress the powerlaw package warnings\n",
    "# \"powerlaw.py:700: RuntimeWarning: divide by zero encountered in true_divide\"\n",
    "# \"powerlaw.py:700: RuntimeWarning: invalid value encountered in true_divide\"\n",
    "import warnings\n",
    "warnings.simplefilter(action='ignore', category=RuntimeWarning)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1. Quick start example"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1.1 Import your model (Keras or PyTorch)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:51:48.865394Z",
     "start_time": "2019-03-05T23:51:16.555209Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:From /Users/charleshmartin/anaconda3/lib/python3.7/site-packages/tensorflow/python/framework/op_def_library.py:263: colocate_with (from tensorflow.python.framework.ops) is deprecated and will be removed in a future version.\n",
      "Instructions for updating:\n",
      "Colocations handled automatically by placer.\n"
     ]
    }
   ],
   "source": [
    "from keras.models import load_model\n",
    "from keras.applications import vgg16\n",
    "\n",
    "kmodel = vgg16.VGG16\n",
    "model = kmodel(weights='imagenet')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 1.2 Run WeightWatcher"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:51:54.297463Z",
     "start_time": "2019-03-05T23:51:48.878699Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:00:47,407 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:00:47,408 INFO Analyzing model 'vgg16' with 23 layers\n",
      "2019-12-24 14:00:48,324 INFO ### Printing results ###\n",
      "2019-12-24 14:00:49,763 INFO Norm: min: 2.4488985538482666, max: 23.428979873657227, avg: 4.01861572265625\n",
      "2019-12-24 14:00:49,763 INFO Norm compound: min: 2.728064775466919, max: 23.428979873657227, avg: 6.7535719871521\n",
      "2019-12-24 14:00:49,764 INFO LogNorm: min: 0.3889707922935486, max: 1.369753360748291, avg: 0.5674788951873779\n",
      "2019-12-24 14:00:49,765 INFO LogNorm compound: min: 0.43449220061302185, max: 1.369753360748291, avg: 0.6947276592254639\n"
     ]
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "results = watcher.analyze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:51:54.412503Z",
     "start_time": "2019-03-05T23:51:54.309238Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{0: {'id': 0,\n",
       "  'type': <keras.engine.input_layer.InputLayer at 0x106968cf8>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 1: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 64,\n",
       "   'M': 3,\n",
       "   'Q': 21.333333333333332,\n",
       "   'summary': 'Weight matrix 1/9 (3,64): Skipping: too small (<50)'},\n",
       "  1: {'N': 64,\n",
       "   'M': 3,\n",
       "   'Q': 21.333333333333332,\n",
       "   'summary': 'Weight matrix 2/9 (3,64): Skipping: too small (<50)'},\n",
       "  2: {'N': 64,\n",
       "   'M': 3,\n",
       "   'Q': 21.333333333333332,\n",
       "   'summary': 'Weight matrix 3/9 (3,64): Skipping: too small (<50)'},\n",
       "  3: {'N': 64,\n",
       "   'M': 3,\n",
       "   'Q': 21.333333333333332,\n",
       "   'summary': 'Weight matrix 4/9 (3,64): Skipping: too small (<50)'},\n",
       "  4: {'N': 64,\n",
       "   'M': 3,\n",
       "   'Q': 21.333333333333332,\n",
       "   'summary': 'Weight matrix 5/9 (3,64): Skipping: too small (<50)'},\n",
       "  5: {'N': 64,\n",
       "   'M': 3,\n",
       "   'Q': 21.333333333333332,\n",
       "   'summary': 'Weight matrix 6/9 (3,64): Skipping: too small (<50)'},\n",
       "  6: {'N': 64,\n",
       "   'M': 3,\n",
       "   'Q': 21.333333333333332,\n",
       "   'summary': 'Weight matrix 7/9 (3,64): Skipping: too small (<50)'},\n",
       "  7: {'N': 64,\n",
       "   'M': 3,\n",
       "   'Q': 21.333333333333332,\n",
       "   'summary': 'Weight matrix 8/9 (3,64): Skipping: too small (<50)'},\n",
       "  8: {'N': 64,\n",
       "   'M': 3,\n",
       "   'Q': 21.333333333333332,\n",
       "   'summary': 'Weight matrix 9/9 (3,64): Skipping: too small (<50)'}},\n",
       " 2: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 64,\n",
       "   'M': 64,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.499732,\n",
       "   'lognorm': 0.39789346,\n",
       "   'summary': 'Weight matrix 1/9 (64,64): Lognorm: 0.3978934586048126'},\n",
       "  1: {'N': 64,\n",
       "   'M': 64,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.8417575,\n",
       "   'lognorm': 0.45358703,\n",
       "   'summary': 'Weight matrix 2/9 (64,64): Lognorm: 0.45358702540397644'},\n",
       "  2: {'N': 64,\n",
       "   'M': 64,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.545549,\n",
       "   'lognorm': 0.40578145,\n",
       "   'summary': 'Weight matrix 3/9 (64,64): Lognorm: 0.40578144788742065'},\n",
       "  3: {'N': 64,\n",
       "   'M': 64,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.8463395,\n",
       "   'lognorm': 0.4542867,\n",
       "   'summary': 'Weight matrix 4/9 (64,64): Lognorm: 0.45428669452667236'},\n",
       "  4: {'N': 64,\n",
       "   'M': 64,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.1401582,\n",
       "   'lognorm': 0.49695152,\n",
       "   'summary': 'Weight matrix 5/9 (64,64): Lognorm: 0.49695152044296265'},\n",
       "  5: {'N': 64,\n",
       "   'M': 64,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.8666604,\n",
       "   'lognorm': 0.45737624,\n",
       "   'summary': 'Weight matrix 6/9 (64,64): Lognorm: 0.45737624168395996'},\n",
       "  6: {'N': 64,\n",
       "   'M': 64,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.5378573,\n",
       "   'lognorm': 0.4044672,\n",
       "   'summary': 'Weight matrix 7/9 (64,64): Lognorm: 0.4044671952724457'},\n",
       "  7: {'N': 64,\n",
       "   'M': 64,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.8256307,\n",
       "   'lognorm': 0.4511154,\n",
       "   'summary': 'Weight matrix 8/9 (64,64): Lognorm: 0.4511153995990753'},\n",
       "  8: {'N': 64,\n",
       "   'M': 64,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.4488986,\n",
       "   'lognorm': 0.3889708,\n",
       "   'summary': 'Weight matrix 9/9 (64,64): Lognorm: 0.3889707922935486'}},\n",
       " 3: {'id': 3,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb2853b1d0>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 4: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 128,\n",
       "   'M': 64,\n",
       "   'Q': 2.0,\n",
       "   'norm': 2.7612133,\n",
       "   'lognorm': 0.44109997,\n",
       "   'summary': 'Weight matrix 1/9 (64,128): Lognorm: 0.44109997153282166'},\n",
       "  1: {'N': 128,\n",
       "   'M': 64,\n",
       "   'Q': 2.0,\n",
       "   'norm': 2.8930957,\n",
       "   'lognorm': 0.4613628,\n",
       "   'summary': 'Weight matrix 2/9 (64,128): Lognorm: 0.4613628089427948'},\n",
       "  2: {'N': 128,\n",
       "   'M': 64,\n",
       "   'Q': 2.0,\n",
       "   'norm': 2.7353733,\n",
       "   'lognorm': 0.4370166,\n",
       "   'summary': 'Weight matrix 3/9 (64,128): Lognorm: 0.4370166063308716'},\n",
       "  3: {'N': 128,\n",
       "   'M': 64,\n",
       "   'Q': 2.0,\n",
       "   'norm': 2.9293149,\n",
       "   'lognorm': 0.46676606,\n",
       "   'summary': 'Weight matrix 4/9 (64,128): Lognorm: 0.4667660593986511'},\n",
       "  4: {'N': 128,\n",
       "   'M': 64,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.3123863,\n",
       "   'lognorm': 0.520141,\n",
       "   'summary': 'Weight matrix 5/9 (64,128): Lognorm: 0.5201410055160522'},\n",
       "  5: {'N': 128,\n",
       "   'M': 64,\n",
       "   'Q': 2.0,\n",
       "   'norm': 2.9468317,\n",
       "   'lognorm': 0.46935534,\n",
       "   'summary': 'Weight matrix 6/9 (64,128): Lognorm: 0.46935534477233887'},\n",
       "  6: {'N': 128,\n",
       "   'M': 64,\n",
       "   'Q': 2.0,\n",
       "   'norm': 2.7955456,\n",
       "   'lognorm': 0.44646657,\n",
       "   'summary': 'Weight matrix 7/9 (64,128): Lognorm: 0.4464665651321411'},\n",
       "  7: {'N': 128,\n",
       "   'M': 64,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.0312111,\n",
       "   'lognorm': 0.4816162,\n",
       "   'summary': 'Weight matrix 8/9 (64,128): Lognorm: 0.48161619901657104'},\n",
       "  8: {'N': 128,\n",
       "   'M': 64,\n",
       "   'Q': 2.0,\n",
       "   'norm': 2.8000274,\n",
       "   'lognorm': 0.44716227,\n",
       "   'summary': 'Weight matrix 9/9 (64,128): Lognorm: 0.4471622705459595'}},\n",
       " 5: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 128,\n",
       "   'M': 128,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.851535,\n",
       "   'lognorm': 0.45507872,\n",
       "   'summary': 'Weight matrix 1/9 (128,128): Lognorm: 0.45507872104644775'},\n",
       "  1: {'N': 128,\n",
       "   'M': 128,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.0472515,\n",
       "   'lognorm': 0.4839083,\n",
       "   'summary': 'Weight matrix 2/9 (128,128): Lognorm: 0.4839082956314087'},\n",
       "  2: {'N': 128,\n",
       "   'M': 128,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.8800385,\n",
       "   'lognorm': 0.4593983,\n",
       "   'summary': 'Weight matrix 3/9 (128,128): Lognorm: 0.4593982994556427'},\n",
       "  3: {'N': 128,\n",
       "   'M': 128,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.067396,\n",
       "   'lognorm': 0.48676983,\n",
       "   'summary': 'Weight matrix 4/9 (128,128): Lognorm: 0.48676982522010803'},\n",
       "  4: {'N': 128,\n",
       "   'M': 128,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.4112387,\n",
       "   'lognorm': 0.53291214,\n",
       "   'summary': 'Weight matrix 5/9 (128,128): Lognorm: 0.5329121351242065'},\n",
       "  5: {'N': 128,\n",
       "   'M': 128,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.0651598,\n",
       "   'lognorm': 0.48645312,\n",
       "   'summary': 'Weight matrix 6/9 (128,128): Lognorm: 0.486453115940094'},\n",
       "  6: {'N': 128,\n",
       "   'M': 128,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.854957,\n",
       "   'lognorm': 0.45559958,\n",
       "   'summary': 'Weight matrix 7/9 (128,128): Lognorm: 0.4555995762348175'},\n",
       "  7: {'N': 128,\n",
       "   'M': 128,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.0803354,\n",
       "   'lognorm': 0.488598,\n",
       "   'summary': 'Weight matrix 8/9 (128,128): Lognorm: 0.4885979890823364'},\n",
       "  8: {'N': 128,\n",
       "   'M': 128,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.8168051,\n",
       "   'lognorm': 0.4497568,\n",
       "   'summary': 'Weight matrix 9/9 (128,128): Lognorm: 0.44975680112838745'}},\n",
       " 6: {'id': 6,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb28592048>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 7: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 256,\n",
       "   'M': 128,\n",
       "   'Q': 2.0,\n",
       "   'norm': 2.7108333,\n",
       "   'lognorm': 0.43310282,\n",
       "   'summary': 'Weight matrix 1/9 (128,256): Lognorm: 0.4331028163433075'},\n",
       "  1: {'N': 256,\n",
       "   'M': 128,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.176145,\n",
       "   'lognorm': 0.5019003,\n",
       "   'summary': 'Weight matrix 2/9 (128,256): Lognorm: 0.501900315284729'},\n",
       "  2: {'N': 256,\n",
       "   'M': 128,\n",
       "   'Q': 2.0,\n",
       "   'norm': 2.7279358,\n",
       "   'lognorm': 0.43583414,\n",
       "   'summary': 'Weight matrix 3/9 (128,256): Lognorm: 0.435834139585495'},\n",
       "  3: {'N': 256,\n",
       "   'M': 128,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.2911222,\n",
       "   'lognorm': 0.517344,\n",
       "   'summary': 'Weight matrix 4/9 (128,256): Lognorm: 0.5173439979553223'},\n",
       "  4: {'N': 256,\n",
       "   'M': 128,\n",
       "   'Q': 2.0,\n",
       "   'norm': 4.274396,\n",
       "   'lognorm': 0.63087475,\n",
       "   'summary': 'Weight matrix 5/9 (128,256): Lognorm: 0.630874752998352'},\n",
       "  5: {'N': 256,\n",
       "   'M': 128,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.26429,\n",
       "   'lognorm': 0.51378876,\n",
       "   'summary': 'Weight matrix 6/9 (128,256): Lognorm: 0.5137887597084045'},\n",
       "  6: {'N': 256,\n",
       "   'M': 128,\n",
       "   'Q': 2.0,\n",
       "   'norm': 2.6995878,\n",
       "   'lognorm': 0.43129745,\n",
       "   'summary': 'Weight matrix 7/9 (128,256): Lognorm: 0.4312974512577057'},\n",
       "  7: {'N': 256,\n",
       "   'M': 128,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.1601713,\n",
       "   'lognorm': 0.49971062,\n",
       "   'summary': 'Weight matrix 8/9 (128,256): Lognorm: 0.4997106194496155'},\n",
       "  8: {'N': 256,\n",
       "   'M': 128,\n",
       "   'Q': 2.0,\n",
       "   'norm': 2.6716125,\n",
       "   'lognorm': 0.42677346,\n",
       "   'summary': 'Weight matrix 9/9 (128,256): Lognorm: 0.42677345871925354'}},\n",
       " 8: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.8918858,\n",
       "   'lognorm': 0.46118113,\n",
       "   'summary': 'Weight matrix 1/9 (256,256): Lognorm: 0.4611811339855194'},\n",
       "  1: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.238382,\n",
       "   'lognorm': 0.5103281,\n",
       "   'summary': 'Weight matrix 2/9 (256,256): Lognorm: 0.5103281140327454'},\n",
       "  2: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.907241,\n",
       "   'lognorm': 0.46348104,\n",
       "   'summary': 'Weight matrix 3/9 (256,256): Lognorm: 0.46348103880882263'},\n",
       "  3: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.1938913,\n",
       "   'lognorm': 0.50432014,\n",
       "   'summary': 'Weight matrix 4/9 (256,256): Lognorm: 0.5043201446533203'},\n",
       "  4: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.92556,\n",
       "   'lognorm': 0.59390163,\n",
       "   'summary': 'Weight matrix 5/9 (256,256): Lognorm: 0.5939016342163086'},\n",
       "  5: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.2178304,\n",
       "   'lognorm': 0.5075632,\n",
       "   'summary': 'Weight matrix 6/9 (256,256): Lognorm: 0.5075631737709045'},\n",
       "  6: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.8660247,\n",
       "   'lognorm': 0.45727992,\n",
       "   'summary': 'Weight matrix 7/9 (256,256): Lognorm: 0.45727992057800293'},\n",
       "  7: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.1748025,\n",
       "   'lognorm': 0.50171673,\n",
       "   'summary': 'Weight matrix 8/9 (256,256): Lognorm: 0.5017167329788208'},\n",
       "  8: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.8920782,\n",
       "   'lognorm': 0.46121,\n",
       "   'summary': 'Weight matrix 9/9 (256,256): Lognorm: 0.4612100124359131'}},\n",
       " 9: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.9971552,\n",
       "   'lognorm': 0.47670922,\n",
       "   'summary': 'Weight matrix 1/9 (256,256): Lognorm: 0.4767092168331146'},\n",
       "  1: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.3329842,\n",
       "   'lognorm': 0.5228332,\n",
       "   'summary': 'Weight matrix 2/9 (256,256): Lognorm: 0.5228332281112671'},\n",
       "  2: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.0220184,\n",
       "   'lognorm': 0.48029712,\n",
       "   'summary': 'Weight matrix 3/9 (256,256): Lognorm: 0.48029711842536926'},\n",
       "  3: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.258065,\n",
       "   'lognorm': 0.5129597,\n",
       "   'summary': 'Weight matrix 4/9 (256,256): Lognorm: 0.5129597187042236'},\n",
       "  4: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.892659,\n",
       "   'lognorm': 0.5902464,\n",
       "   'summary': 'Weight matrix 5/9 (256,256): Lognorm: 0.5902463793754578'},\n",
       "  5: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.3484614,\n",
       "   'lognorm': 0.5248453,\n",
       "   'summary': 'Weight matrix 6/9 (256,256): Lognorm: 0.52484530210495'},\n",
       "  6: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 2.9902184,\n",
       "   'lognorm': 0.4757029,\n",
       "   'summary': 'Weight matrix 7/9 (256,256): Lognorm: 0.4757029116153717'},\n",
       "  7: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.2758641,\n",
       "   'lognorm': 0.5153259,\n",
       "   'summary': 'Weight matrix 8/9 (256,256): Lognorm: 0.5153259038925171'},\n",
       "  8: {'N': 256,\n",
       "   'M': 256,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.008209,\n",
       "   'lognorm': 0.478308,\n",
       "   'summary': 'Weight matrix 9/9 (256,256): Lognorm: 0.4783079922199249'}},\n",
       " 10: {'id': 10,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb28601e10>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 11: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 512,\n",
       "   'M': 256,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.4128516,\n",
       "   'lognorm': 0.5331174,\n",
       "   'summary': 'Weight matrix 1/9 (256,512): Lognorm: 0.533117413520813'},\n",
       "  1: {'N': 512,\n",
       "   'M': 256,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.6959422,\n",
       "   'lognorm': 0.5677252,\n",
       "   'summary': 'Weight matrix 2/9 (256,512): Lognorm: 0.5677251815795898'},\n",
       "  2: {'N': 512,\n",
       "   'M': 256,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.4338121,\n",
       "   'lognorm': 0.53577656,\n",
       "   'summary': 'Weight matrix 3/9 (256,512): Lognorm: 0.5357765555381775'},\n",
       "  3: {'N': 512,\n",
       "   'M': 256,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.6989415,\n",
       "   'lognorm': 0.56807745,\n",
       "   'summary': 'Weight matrix 4/9 (256,512): Lognorm: 0.5680774450302124'},\n",
       "  4: {'N': 512,\n",
       "   'M': 256,\n",
       "   'Q': 2.0,\n",
       "   'norm': 4.368356,\n",
       "   'lognorm': 0.64031804,\n",
       "   'summary': 'Weight matrix 5/9 (256,512): Lognorm: 0.6403180360794067'},\n",
       "  5: {'N': 512,\n",
       "   'M': 256,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.674093,\n",
       "   'lognorm': 0.56515014,\n",
       "   'summary': 'Weight matrix 6/9 (256,512): Lognorm: 0.5651501417160034'},\n",
       "  6: {'N': 512,\n",
       "   'M': 256,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.379677,\n",
       "   'lognorm': 0.52887523,\n",
       "   'summary': 'Weight matrix 7/9 (256,512): Lognorm: 0.5288752317428589'},\n",
       "  7: {'N': 512,\n",
       "   'M': 256,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.6538746,\n",
       "   'lognorm': 0.5627536,\n",
       "   'summary': 'Weight matrix 8/9 (256,512): Lognorm: 0.5627536177635193'},\n",
       "  8: {'N': 512,\n",
       "   'M': 256,\n",
       "   'Q': 2.0,\n",
       "   'norm': 3.359951,\n",
       "   'lognorm': 0.526333,\n",
       "   'summary': 'Weight matrix 9/9 (256,512): Lognorm: 0.5263329744338989'}},\n",
       " 12: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.8165078,\n",
       "   'lognorm': 0.5816662,\n",
       "   'summary': 'Weight matrix 1/9 (512,512): Lognorm: 0.5816661715507507'},\n",
       "  1: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.9761157,\n",
       "   'lognorm': 0.599459,\n",
       "   'summary': 'Weight matrix 2/9 (512,512): Lognorm: 0.5994589924812317'},\n",
       "  2: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.805959,\n",
       "   'lognorm': 0.5804641,\n",
       "   'summary': 'Weight matrix 3/9 (512,512): Lognorm: 0.5804641246795654'},\n",
       "  3: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.8709376,\n",
       "   'lognorm': 0.5878162,\n",
       "   'summary': 'Weight matrix 4/9 (512,512): Lognorm: 0.5878161787986755'},\n",
       "  4: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.3149986,\n",
       "   'lognorm': 0.6349807,\n",
       "   'summary': 'Weight matrix 5/9 (512,512): Lognorm: 0.6349806785583496'},\n",
       "  5: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.8804407,\n",
       "   'lognorm': 0.5888811,\n",
       "   'summary': 'Weight matrix 6/9 (512,512): Lognorm: 0.5888810753822327'},\n",
       "  6: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.7826579,\n",
       "   'lognorm': 0.57779706,\n",
       "   'summary': 'Weight matrix 7/9 (512,512): Lognorm: 0.5777970552444458'},\n",
       "  7: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.932986,\n",
       "   'lognorm': 0.5947224,\n",
       "   'summary': 'Weight matrix 8/9 (512,512): Lognorm: 0.5947223901748657'},\n",
       "  8: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.7902145,\n",
       "   'lognorm': 0.57866377,\n",
       "   'summary': 'Weight matrix 9/9 (512,512): Lognorm: 0.5786637663841248'}},\n",
       " 13: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.9701836,\n",
       "   'lognorm': 0.5988106,\n",
       "   'summary': 'Weight matrix 1/9 (512,512): Lognorm: 0.598810613155365'},\n",
       "  1: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.2078204,\n",
       "   'lognorm': 0.6240572,\n",
       "   'summary': 'Weight matrix 2/9 (512,512): Lognorm: 0.6240571737289429'},\n",
       "  2: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.9831755,\n",
       "   'lognorm': 0.60022944,\n",
       "   'summary': 'Weight matrix 3/9 (512,512): Lognorm: 0.6002294421195984'},\n",
       "  3: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.0450754,\n",
       "   'lognorm': 0.6069266,\n",
       "   'summary': 'Weight matrix 4/9 (512,512): Lognorm: 0.6069266200065613'},\n",
       "  4: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.5052695,\n",
       "   'lognorm': 0.6537208,\n",
       "   'summary': 'Weight matrix 5/9 (512,512): Lognorm: 0.6537207961082458'},\n",
       "  5: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.028195,\n",
       "   'lognorm': 0.60511047,\n",
       "   'summary': 'Weight matrix 6/9 (512,512): Lognorm: 0.6051104664802551'},\n",
       "  6: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.950565,\n",
       "   'lognorm': 0.59665924,\n",
       "   'summary': 'Weight matrix 7/9 (512,512): Lognorm: 0.596659243106842'},\n",
       "  7: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.1731806,\n",
       "   'lognorm': 0.6204672,\n",
       "   'summary': 'Weight matrix 8/9 (512,512): Lognorm: 0.6204671859741211'},\n",
       "  8: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 3.9536626,\n",
       "   'lognorm': 0.5969996,\n",
       "   'summary': 'Weight matrix 9/9 (512,512): Lognorm: 0.5969995856285095'}},\n",
       " 14: {'id': 14,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb28686cf8>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 15: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.379643,\n",
       "   'lognorm': 0.6414387,\n",
       "   'summary': 'Weight matrix 1/9 (512,512): Lognorm: 0.6414387226104736'},\n",
       "  1: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.554939,\n",
       "   'lognorm': 0.65848255,\n",
       "   'summary': 'Weight matrix 2/9 (512,512): Lognorm: 0.658482551574707'},\n",
       "  2: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.3807607,\n",
       "   'lognorm': 0.6415495,\n",
       "   'summary': 'Weight matrix 3/9 (512,512): Lognorm: 0.6415495276451111'},\n",
       "  3: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.3441844,\n",
       "   'lognorm': 0.6379083,\n",
       "   'summary': 'Weight matrix 4/9 (512,512): Lognorm: 0.6379082798957825'},\n",
       "  4: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.932443,\n",
       "   'lognorm': 0.69306207,\n",
       "   'summary': 'Weight matrix 5/9 (512,512): Lognorm: 0.6930620670318604'},\n",
       "  5: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.3207736,\n",
       "   'lognorm': 0.6355615,\n",
       "   'summary': 'Weight matrix 6/9 (512,512): Lognorm: 0.6355615258216858'},\n",
       "  6: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.344898,\n",
       "   'lognorm': 0.6379796,\n",
       "   'summary': 'Weight matrix 7/9 (512,512): Lognorm: 0.6379796266555786'},\n",
       "  7: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.515352,\n",
       "   'lognorm': 0.6546916,\n",
       "   'summary': 'Weight matrix 8/9 (512,512): Lognorm: 0.6546915769577026'},\n",
       "  8: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.344318,\n",
       "   'lognorm': 0.6379216,\n",
       "   'summary': 'Weight matrix 9/9 (512,512): Lognorm: 0.6379215717315674'}},\n",
       " 16: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.449406,\n",
       "   'lognorm': 0.6483021,\n",
       "   'summary': 'Weight matrix 1/9 (512,512): Lognorm: 0.6483020782470703'},\n",
       "  1: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.6948195,\n",
       "   'lognorm': 0.6716189,\n",
       "   'summary': 'Weight matrix 2/9 (512,512): Lognorm: 0.6716188788414001'},\n",
       "  2: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.4539843,\n",
       "   'lognorm': 0.6487487,\n",
       "   'summary': 'Weight matrix 3/9 (512,512): Lognorm: 0.6487486958503723'},\n",
       "  3: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.2787437,\n",
       "   'lognorm': 0.6313163,\n",
       "   'summary': 'Weight matrix 4/9 (512,512): Lognorm: 0.6313163042068481'},\n",
       "  4: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.8996086,\n",
       "   'lognorm': 0.6901614,\n",
       "   'summary': 'Weight matrix 5/9 (512,512): Lognorm: 0.690161406993866'},\n",
       "  5: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.289923,\n",
       "   'lognorm': 0.6324495,\n",
       "   'summary': 'Weight matrix 6/9 (512,512): Lognorm: 0.6324495077133179'},\n",
       "  6: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.401268,\n",
       "   'lognorm': 0.6435778,\n",
       "   'summary': 'Weight matrix 7/9 (512,512): Lognorm: 0.6435778141021729'},\n",
       "  7: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.616433,\n",
       "   'lognorm': 0.6643065,\n",
       "   'summary': 'Weight matrix 8/9 (512,512): Lognorm: 0.6643065214157104'},\n",
       "  8: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.391785,\n",
       "   'lognorm': 0.64264107,\n",
       "   'summary': 'Weight matrix 9/9 (512,512): Lognorm: 0.6426410675048828'}},\n",
       " 17: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  0: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.2785525,\n",
       "   'lognorm': 0.6312969,\n",
       "   'summary': 'Weight matrix 1/9 (512,512): Lognorm: 0.6312968730926514'},\n",
       "  1: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.525857,\n",
       "   'lognorm': 0.6557008,\n",
       "   'summary': 'Weight matrix 2/9 (512,512): Lognorm: 0.6557008028030396'},\n",
       "  2: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.268781,\n",
       "   'lognorm': 0.6303039,\n",
       "   'summary': 'Weight matrix 3/9 (512,512): Lognorm: 0.6303039193153381'},\n",
       "  3: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.2651267,\n",
       "   'lognorm': 0.6299319,\n",
       "   'summary': 'Weight matrix 4/9 (512,512): Lognorm: 0.6299319267272949'},\n",
       "  4: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.7191625,\n",
       "   'lognorm': 0.6738649,\n",
       "   'summary': 'Weight matrix 5/9 (512,512): Lognorm: 0.6738649010658264'},\n",
       "  5: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.2333026,\n",
       "   'lognorm': 0.6266793,\n",
       "   'summary': 'Weight matrix 6/9 (512,512): Lognorm: 0.6266793012619019'},\n",
       "  6: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.2681537,\n",
       "   'lognorm': 0.63024,\n",
       "   'summary': 'Weight matrix 7/9 (512,512): Lognorm: 0.6302400231361389'},\n",
       "  7: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.5327334,\n",
       "   'lognorm': 0.6563602,\n",
       "   'summary': 'Weight matrix 8/9 (512,512): Lognorm: 0.6563602089881897'},\n",
       "  8: {'N': 512,\n",
       "   'M': 512,\n",
       "   'Q': 1.0,\n",
       "   'norm': 4.2659492,\n",
       "   'lognorm': 0.6300157,\n",
       "   'summary': 'Weight matrix 9/9 (512,512): Lognorm: 0.6300156712532043'}},\n",
       " 18: {'id': 18,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb2870c128>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 19: {'id': 19,\n",
       "  'type': <keras.layers.core.Flatten at 0xb2870c748>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 20: {'id': 20,\n",
       "  'type': <keras.layers.core.Dense at 0xb2872eb00>,\n",
       "  'layer_type': <LAYER_TYPE.DENSE: 1>,\n",
       "  0: {'N': 25088,\n",
       "   'M': 4096,\n",
       "   'Q': 6.125,\n",
       "   'norm': 23.42898,\n",
       "   'lognorm': 1.3697534,\n",
       "   'summary': 'Weight matrix 1/1 (4096,25088): Lognorm: 1.369753360748291'}},\n",
       " 21: {'id': 21,\n",
       "  'type': <keras.layers.core.Dense at 0xb28746be0>,\n",
       "  'layer_type': <LAYER_TYPE.DENSE: 1>,\n",
       "  0: {'N': 4096,\n",
       "   'M': 4096,\n",
       "   'Q': 1.0,\n",
       "   'norm': 18.021769,\n",
       "   'lognorm': 1.2557974,\n",
       "   'summary': 'Weight matrix 1/1 (4096,4096): Lognorm: 1.2557973861694336'}},\n",
       " 22: {'id': 22,\n",
       "  'type': <keras.layers.core.Dense at 0xb287647f0>,\n",
       "  'layer_type': <LAYER_TYPE.DENSE: 1>,\n",
       "  0: {'N': 4096,\n",
       "   'M': 1000,\n",
       "   'Q': 4.096,\n",
       "   'norm': 16.757492,\n",
       "   'lognorm': 1.2242091,\n",
       "   'summary': 'Weight matrix 1/1 (1000,4096): Lognorm: 1.2242090702056885'}}}"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:51:54.441885Z",
     "start_time": "2019-03-05T23:51:54.418688Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'norm': 4.0186157,\n",
       " 'norm_compound': 6.753572,\n",
       " 'lognorm': 0.5674789,\n",
       " 'lognorm_compound': 0.69472766}"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "watcher.get_summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:51:54.487413Z",
     "start_time": "2019-03-05T23:51:54.454110Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:00:49,835 INFO ### Printing results ###\n",
      "2019-12-24 14:00:51,212 INFO Norm: min: 2.4488985538482666, max: 23.428979873657227, avg: 4.01861572265625\n",
      "2019-12-24 14:00:51,212 INFO Norm compound: min: 2.728064775466919, max: 23.428979873657227, avg: 6.7535719871521\n",
      "2019-12-24 14:00:51,213 INFO LogNorm: min: 0.3889707922935486, max: 1.369753360748291, avg: 0.5674788951873779\n",
      "2019-12-24 14:00:51,214 INFO LogNorm compound: min: 0.43449220061302185, max: 1.369753360748291, avg: 0.6947276592254639\n"
     ]
    }
   ],
   "source": [
    "watcher.print_results()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2. Advanced examples"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2.1 Filter by layer type (CONV1D, CONV2D, DENSE)\n",
    "\n",
    "In this example we are interested in the DENSE layers only"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:52:19.042816Z",
     "start_time": "2019-03-05T23:51:54.498749Z"
    },
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:00:54,204 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:00:54,205 INFO Analyzing model 'vgg16' with 23 layers\n",
      "2019-12-24 14:00:54,606 INFO ### Printing results ###\n",
      "2019-12-24 14:00:54,878 INFO Norm: min: 16.757492065429688, max: 23.428979873657227, avg: 19.402746200561523\n",
      "2019-12-24 14:00:54,879 INFO Norm compound: min: 16.757492065429688, max: 23.428979873657227, avg: 19.402746200561523\n",
      "2019-12-24 14:00:54,879 INFO LogNorm: min: 1.2242090702056885, max: 1.369753360748291, avg: 1.2832533121109009\n",
      "2019-12-24 14:00:54,880 INFO LogNorm compound: min: 1.2242090702056885, max: 1.369753360748291, avg: 1.2832533121109009\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{0: {'id': 0,\n",
       "  'type': <keras.engine.input_layer.InputLayer at 0xb38ad7390>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 1: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 2: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 3: {'id': 3,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38acde80>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 4: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 5: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 6: {'id': 6,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38b304e0>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 7: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 8: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 9: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 10: {'id': 10,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38b9bf60>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 11: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 12: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 13: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 14: {'id': 14,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38c2cbe0>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 15: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 16: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 17: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 18: {'id': 18,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38ada7f0>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 19: {'id': 19,\n",
       "  'type': <keras.layers.core.Flatten at 0xb38adacf8>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 20: {'id': 20,\n",
       "  'type': <keras.layers.core.Dense at 0xb38acda58>,\n",
       "  'layer_type': <LAYER_TYPE.DENSE: 1>,\n",
       "  0: {'N': 25088,\n",
       "   'M': 4096,\n",
       "   'Q': 6.125,\n",
       "   'norm': 23.42898,\n",
       "   'lognorm': 1.3697534,\n",
       "   'summary': 'Weight matrix 1/1 (4096,25088): Lognorm: 1.369753360748291'}},\n",
       " 21: {'id': 21,\n",
       "  'type': <keras.layers.core.Dense at 0xb38cd6cc0>,\n",
       "  'layer_type': <LAYER_TYPE.DENSE: 1>,\n",
       "  0: {'N': 4096,\n",
       "   'M': 4096,\n",
       "   'Q': 1.0,\n",
       "   'norm': 18.021769,\n",
       "   'lognorm': 1.2557974,\n",
       "   'summary': 'Weight matrix 1/1 (4096,4096): Lognorm: 1.2557973861694336'}},\n",
       " 22: {'id': 22,\n",
       "  'type': <keras.layers.core.Dense at 0xb38cf3f60>,\n",
       "  'layer_type': <LAYER_TYPE.DENSE: 1>,\n",
       "  0: {'N': 4096,\n",
       "   'M': 1000,\n",
       "   'Q': 4.096,\n",
       "   'norm': 16.757492,\n",
       "   'lognorm': 1.2242091,\n",
       "   'summary': 'Weight matrix 1/1 (1000,4096): Lognorm: 1.2242090702056885'}}}"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from keras.applications import vgg16\n",
    "\n",
    "kmodel = vgg16.VGG16\n",
    "model = kmodel(weights='imagenet')\n",
    "\n",
    "import weightwatcher as ww\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "watcher.analyze(layers=ww.LAYER_TYPE.DENSE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:52:19.078746Z",
     "start_time": "2019-03-05T23:52:19.050673Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:00:54,899 INFO ### Printing results ###\n",
      "2019-12-24 14:00:55,144 INFO Norm: min: 16.757492065429688, max: 23.428979873657227, avg: 19.402746200561523\n",
      "2019-12-24 14:00:55,145 INFO Norm compound: min: 16.757492065429688, max: 23.428979873657227, avg: 19.402746200561523\n",
      "2019-12-24 14:00:55,146 INFO LogNorm: min: 1.2242090702056885, max: 1.369753360748291, avg: 1.2832533121109009\n",
      "2019-12-24 14:00:55,146 INFO LogNorm compound: min: 1.2242090702056885, max: 1.369753360748291, avg: 1.2832533121109009\n"
     ]
    }
   ],
   "source": [
    "watcher.print_results()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2.2 Filter by multiple layer types\n",
    "\n",
    "In this example we are interested in the CONV1D and DENSE layers.\n",
    "\n",
    "Filter the layers using a bitmask."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:52:20.812400Z",
     "start_time": "2019-03-05T23:52:19.101208Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:00:55,162 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:00:55,163 INFO Analyzing model 'vgg16' with 23 layers\n",
      "2019-12-24 14:00:55,455 INFO ### Printing results ###\n",
      "2019-12-24 14:00:55,696 INFO Norm: min: 16.757492065429688, max: 23.428979873657227, avg: 19.402746200561523\n",
      "2019-12-24 14:00:55,696 INFO Norm compound: min: 16.757492065429688, max: 23.428979873657227, avg: 19.402746200561523\n",
      "2019-12-24 14:00:55,697 INFO LogNorm: min: 1.2242090702056885, max: 1.369753360748291, avg: 1.2832533121109009\n",
      "2019-12-24 14:00:55,697 INFO LogNorm compound: min: 1.2242090702056885, max: 1.369753360748291, avg: 1.2832533121109009\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{0: {'id': 0,\n",
       "  'type': <keras.engine.input_layer.InputLayer at 0xb38ad7390>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 1: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 2: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 3: {'id': 3,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38acde80>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 4: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 5: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 6: {'id': 6,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38b304e0>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 7: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 8: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 9: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 10: {'id': 10,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38b9bf60>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 11: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 12: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 13: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 14: {'id': 14,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38c2cbe0>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 15: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 16: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 17: {'layer_type': <LAYER_TYPE.CONV2D: 4>,\n",
       "  'message': 'Skipping (Layer type not requested to analyze)'},\n",
       " 18: {'id': 18,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38ada7f0>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 19: {'id': 19,\n",
       "  'type': <keras.layers.core.Flatten at 0xb38adacf8>,\n",
       "  'message': 'Skipping (Layer not supported)'},\n",
       " 20: {'id': 20,\n",
       "  'type': <keras.layers.core.Dense at 0xb38acda58>,\n",
       "  'layer_type': <LAYER_TYPE.DENSE: 1>,\n",
       "  0: {'N': 25088,\n",
       "   'M': 4096,\n",
       "   'Q': 6.125,\n",
       "   'norm': 23.42898,\n",
       "   'lognorm': 1.3697534,\n",
       "   'summary': 'Weight matrix 1/1 (4096,25088): Lognorm: 1.369753360748291'}},\n",
       " 21: {'id': 21,\n",
       "  'type': <keras.layers.core.Dense at 0xb38cd6cc0>,\n",
       "  'layer_type': <LAYER_TYPE.DENSE: 1>,\n",
       "  0: {'N': 4096,\n",
       "   'M': 4096,\n",
       "   'Q': 1.0,\n",
       "   'norm': 18.021769,\n",
       "   'lognorm': 1.2557974,\n",
       "   'summary': 'Weight matrix 1/1 (4096,4096): Lognorm: 1.2557973861694336'}},\n",
       " 22: {'id': 22,\n",
       "  'type': <keras.layers.core.Dense at 0xb38cf3f60>,\n",
       "  'layer_type': <LAYER_TYPE.DENSE: 1>,\n",
       "  0: {'N': 4096,\n",
       "   'M': 1000,\n",
       "   'Q': 4.096,\n",
       "   'norm': 16.757492,\n",
       "   'lognorm': 1.2242091,\n",
       "   'summary': 'Weight matrix 1/1 (1000,4096): Lognorm: 1.2242090702056885'}}}"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "watcher.analyze(layers=ww.LAYER_TYPE.CONV1D|ww.LAYER_TYPE.DENSE)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 2.3 Filter by layer Ids"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:52:22.428909Z",
     "start_time": "2019-03-05T23:52:20.830391Z"
    },
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:00:55,717 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:00:55,718 INFO Analyzing model 'vgg16' with 23 layers\n",
      "2019-12-24 14:00:55,976 INFO ### Printing results ###\n",
      "2019-12-24 14:00:56,195 INFO Norm: min: 23.428979873657227, max: 23.428979873657227, avg: 23.428979873657227\n",
      "2019-12-24 14:00:56,197 INFO Norm compound: min: 23.428979873657227, max: 23.428979873657227, avg: 23.428979873657227\n",
      "2019-12-24 14:00:56,197 INFO LogNorm: min: 1.369753360748291, max: 1.369753360748291, avg: 1.369753360748291\n",
      "2019-12-24 14:00:56,198 INFO LogNorm compound: min: 1.369753360748291, max: 1.369753360748291, avg: 1.369753360748291\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "{0: {'id': 0,\n",
       "  'type': <keras.engine.input_layer.InputLayer at 0xb38ad7390>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 1: {'id': 1,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38ad7a58>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 2: {'id': 2,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38ad7ac8>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 3: {'id': 3,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38acde80>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 4: {'id': 4,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38ad52e8>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 5: {'id': 5,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38b165c0>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 6: {'id': 6,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38b304e0>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 7: {'id': 7,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38b30be0>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 8: {'id': 8,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38b68400>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 9: {'id': 9,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38ad8630>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 10: {'id': 10,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38b9bf60>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 11: {'id': 11,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38bb65f8>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 12: {'id': 12,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38bf1240>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 13: {'id': 13,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38c0c7f0>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 14: {'id': 14,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38c2cbe0>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 15: {'id': 15,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38c2cf60>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 16: {'id': 16,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38c62a58>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 17: {'id': 17,\n",
       "  'type': <keras.layers.convolutional.Conv2D at 0xb38c7db38>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 18: {'id': 18,\n",
       "  'type': <keras.layers.pooling.MaxPooling2D at 0xb38ada7f0>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 19: {'id': 19,\n",
       "  'type': <keras.layers.core.Flatten at 0xb38adacf8>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 20: {'id': 20,\n",
       "  'type': <keras.layers.core.Dense at 0xb38acda58>,\n",
       "  'layer_type': <LAYER_TYPE.DENSE: 1>,\n",
       "  0: {'N': 25088,\n",
       "   'M': 4096,\n",
       "   'Q': 6.125,\n",
       "   'norm': 23.42898,\n",
       "   'lognorm': 1.3697534,\n",
       "   'summary': 'Weight matrix 1/1 (4096,25088): Lognorm: 1.369753360748291'}},\n",
       " 21: {'id': 21,\n",
       "  'type': <keras.layers.core.Dense at 0xb38cd6cc0>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'},\n",
       " 22: {'id': 22,\n",
       "  'type': <keras.layers.core.Dense at 0xb38cf3f60>,\n",
       "  'message': 'Skipping (Layer id not requested to analyze)'}}"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "watcher.analyze(layers=[20])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-01T22:51:12.554859Z",
     "start_time": "2018-10-01T22:51:12.518859Z"
    }
   },
   "source": [
    "## 2.4 Get the return values per layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:52:24.982563Z",
     "start_time": "2019-03-05T23:52:22.440026Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:00:56,216 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:00:56,216 INFO Analyzing model 'vgg16' with 23 layers\n",
      "2019-12-24 14:00:56,635 INFO ### Printing results ###\n",
      "2019-12-24 14:00:58,026 INFO Norm: min: 2.4488985538482666, max: 23.428979873657227, avg: 4.01861572265625\n",
      "2019-12-24 14:00:58,026 INFO Norm compound: min: 2.728064775466919, max: 23.428979873657227, avg: 6.7535719871521\n",
      "2019-12-24 14:00:58,027 INFO LogNorm: min: 0.3889707922935486, max: 1.369753360748291, avg: 0.5674788951873779\n",
      "2019-12-24 14:00:58,028 INFO LogNorm compound: min: 0.43449220061302185, max: 1.369753360748291, avg: 0.6947276592254639\n"
     ]
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "results = watcher.analyze()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-05T23:52:25.022930Z",
     "start_time": "2019-03-05T23:52:24.990119Z"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Layer 2, Slice 0: Lognorm: 0.3978934586048126\n",
      "Layer 2, Slice 1: Lognorm: 0.45358702540397644\n",
      "Layer 2, Slice 2: Lognorm: 0.40578144788742065\n",
      "Layer 2, Slice 3: Lognorm: 0.45428669452667236\n",
      "Layer 2, Slice 4: Lognorm: 0.49695152044296265\n",
      "Layer 2, Slice 5: Lognorm: 0.45737624168395996\n",
      "Layer 2, Slice 6: Lognorm: 0.4044671952724457\n",
      "Layer 2, Slice 7: Lognorm: 0.4511153995990753\n",
      "Layer 2, Slice 8: Lognorm: 0.3889707922935486\n",
      "Layer 4, Slice 0: Lognorm: 0.44109997153282166\n",
      "Layer 4, Slice 1: Lognorm: 0.4613628089427948\n",
      "Layer 4, Slice 2: Lognorm: 0.4370166063308716\n",
      "Layer 4, Slice 3: Lognorm: 0.4667660593986511\n",
      "Layer 4, Slice 4: Lognorm: 0.5201410055160522\n",
      "Layer 4, Slice 5: Lognorm: 0.46935534477233887\n",
      "Layer 4, Slice 6: Lognorm: 0.4464665651321411\n",
      "Layer 4, Slice 7: Lognorm: 0.48161619901657104\n",
      "Layer 4, Slice 8: Lognorm: 0.4471622705459595\n",
      "Layer 5, Slice 0: Lognorm: 0.45507872104644775\n",
      "Layer 5, Slice 1: Lognorm: 0.4839082956314087\n",
      "Layer 5, Slice 2: Lognorm: 0.4593982994556427\n",
      "Layer 5, Slice 3: Lognorm: 0.48676982522010803\n",
      "Layer 5, Slice 4: Lognorm: 0.5329121351242065\n",
      "Layer 5, Slice 5: Lognorm: 0.486453115940094\n",
      "Layer 5, Slice 6: Lognorm: 0.4555995762348175\n",
      "Layer 5, Slice 7: Lognorm: 0.4885979890823364\n",
      "Layer 5, Slice 8: Lognorm: 0.44975680112838745\n",
      "Layer 7, Slice 0: Lognorm: 0.4331028163433075\n",
      "Layer 7, Slice 1: Lognorm: 0.501900315284729\n",
      "Layer 7, Slice 2: Lognorm: 0.435834139585495\n",
      "Layer 7, Slice 3: Lognorm: 0.5173439979553223\n",
      "Layer 7, Slice 4: Lognorm: 0.630874752998352\n",
      "Layer 7, Slice 5: Lognorm: 0.5137887597084045\n",
      "Layer 7, Slice 6: Lognorm: 0.4312974512577057\n",
      "Layer 7, Slice 7: Lognorm: 0.4997106194496155\n",
      "Layer 7, Slice 8: Lognorm: 0.42677345871925354\n",
      "Layer 8, Slice 0: Lognorm: 0.4611811339855194\n",
      "Layer 8, Slice 1: Lognorm: 0.5103281140327454\n",
      "Layer 8, Slice 2: Lognorm: 0.46348103880882263\n",
      "Layer 8, Slice 3: Lognorm: 0.5043201446533203\n",
      "Layer 8, Slice 4: Lognorm: 0.5939016342163086\n",
      "Layer 8, Slice 5: Lognorm: 0.5075631737709045\n",
      "Layer 8, Slice 6: Lognorm: 0.45727992057800293\n",
      "Layer 8, Slice 7: Lognorm: 0.5017167329788208\n",
      "Layer 8, Slice 8: Lognorm: 0.4612100124359131\n",
      "Layer 9, Slice 0: Lognorm: 0.4767092168331146\n",
      "Layer 9, Slice 1: Lognorm: 0.5228332281112671\n",
      "Layer 9, Slice 2: Lognorm: 0.48029711842536926\n",
      "Layer 9, Slice 3: Lognorm: 0.5129597187042236\n",
      "Layer 9, Slice 4: Lognorm: 0.5902463793754578\n",
      "Layer 9, Slice 5: Lognorm: 0.52484530210495\n",
      "Layer 9, Slice 6: Lognorm: 0.4757029116153717\n",
      "Layer 9, Slice 7: Lognorm: 0.5153259038925171\n",
      "Layer 9, Slice 8: Lognorm: 0.4783079922199249\n",
      "Layer 11, Slice 0: Lognorm: 0.533117413520813\n",
      "Layer 11, Slice 1: Lognorm: 0.5677251815795898\n",
      "Layer 11, Slice 2: Lognorm: 0.5357765555381775\n",
      "Layer 11, Slice 3: Lognorm: 0.5680774450302124\n",
      "Layer 11, Slice 4: Lognorm: 0.6403180360794067\n",
      "Layer 11, Slice 5: Lognorm: 0.5651501417160034\n",
      "Layer 11, Slice 6: Lognorm: 0.5288752317428589\n",
      "Layer 11, Slice 7: Lognorm: 0.5627536177635193\n",
      "Layer 11, Slice 8: Lognorm: 0.5263329744338989\n",
      "Layer 12, Slice 0: Lognorm: 0.5816661715507507\n",
      "Layer 12, Slice 1: Lognorm: 0.5994589924812317\n",
      "Layer 12, Slice 2: Lognorm: 0.5804641246795654\n",
      "Layer 12, Slice 3: Lognorm: 0.5878161787986755\n",
      "Layer 12, Slice 4: Lognorm: 0.6349806785583496\n",
      "Layer 12, Slice 5: Lognorm: 0.5888810753822327\n",
      "Layer 12, Slice 6: Lognorm: 0.5777970552444458\n",
      "Layer 12, Slice 7: Lognorm: 0.5947223901748657\n",
      "Layer 12, Slice 8: Lognorm: 0.5786637663841248\n",
      "Layer 13, Slice 0: Lognorm: 0.598810613155365\n",
      "Layer 13, Slice 1: Lognorm: 0.6240571737289429\n",
      "Layer 13, Slice 2: Lognorm: 0.6002294421195984\n",
      "Layer 13, Slice 3: Lognorm: 0.6069266200065613\n",
      "Layer 13, Slice 4: Lognorm: 0.6537207961082458\n",
      "Layer 13, Slice 5: Lognorm: 0.6051104664802551\n",
      "Layer 13, Slice 6: Lognorm: 0.596659243106842\n",
      "Layer 13, Slice 7: Lognorm: 0.6204671859741211\n",
      "Layer 13, Slice 8: Lognorm: 0.5969995856285095\n",
      "Layer 15, Slice 0: Lognorm: 0.6414387226104736\n",
      "Layer 15, Slice 1: Lognorm: 0.658482551574707\n",
      "Layer 15, Slice 2: Lognorm: 0.6415495276451111\n",
      "Layer 15, Slice 3: Lognorm: 0.6379082798957825\n",
      "Layer 15, Slice 4: Lognorm: 0.6930620670318604\n",
      "Layer 15, Slice 5: Lognorm: 0.6355615258216858\n",
      "Layer 15, Slice 6: Lognorm: 0.6379796266555786\n",
      "Layer 15, Slice 7: Lognorm: 0.6546915769577026\n",
      "Layer 15, Slice 8: Lognorm: 0.6379215717315674\n",
      "Layer 16, Slice 0: Lognorm: 0.6483020782470703\n",
      "Layer 16, Slice 1: Lognorm: 0.6716188788414001\n",
      "Layer 16, Slice 2: Lognorm: 0.6487486958503723\n",
      "Layer 16, Slice 3: Lognorm: 0.6313163042068481\n",
      "Layer 16, Slice 4: Lognorm: 0.690161406993866\n",
      "Layer 16, Slice 5: Lognorm: 0.6324495077133179\n",
      "Layer 16, Slice 6: Lognorm: 0.6435778141021729\n",
      "Layer 16, Slice 7: Lognorm: 0.6643065214157104\n",
      "Layer 16, Slice 8: Lognorm: 0.6426410675048828\n",
      "Layer 17, Slice 0: Lognorm: 0.6312968730926514\n",
      "Layer 17, Slice 1: Lognorm: 0.6557008028030396\n",
      "Layer 17, Slice 2: Lognorm: 0.6303039193153381\n",
      "Layer 17, Slice 3: Lognorm: 0.6299319267272949\n",
      "Layer 17, Slice 4: Lognorm: 0.6738649010658264\n",
      "Layer 17, Slice 5: Lognorm: 0.6266793012619019\n",
      "Layer 17, Slice 6: Lognorm: 0.6302400231361389\n",
      "Layer 17, Slice 7: Lognorm: 0.6563602089881897\n",
      "Layer 17, Slice 8: Lognorm: 0.6300156712532043\n",
      "Layer 20, Slice 0: Lognorm: 1.369753360748291\n",
      "Layer 21, Slice 0: Lognorm: 1.2557973861694336\n",
      "Layer 22, Slice 0: Lognorm: 1.2242090702056885\n"
     ]
    }
   ],
   "source": [
    "for layer_id, result in results.items():\n",
    "    for slice_id, summary in result.items():\n",
    "        if not str(slice_id).isdigit() or \"lognorm\" not in summary:\n",
    "            continue\n",
    "        lognorm = summary[\"lognorm\"]\n",
    "        print(\"Layer {}, Slice {}: Lognorm: {}\".format(layer_id, slice_id, lognorm))    "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-01T22:51:12.554859Z",
     "start_time": "2018-10-01T22:51:12.518859Z"
    }
   },
   "source": [
    "## 2.5 Power Law Fit"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T00:16:38.794143Z",
     "start_time": "2019-03-05T23:52:25.030000Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:00:58,060 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:00:58,061 INFO Analyzing model 'vgg16' with 23 layers\n",
      "2019-12-24 14:05:27,511 INFO ### Printing results ###\n",
      "2019-12-24 14:05:28,882 INFO Norm: min: 2.4488985538482666, max: 23.428979873657227, avg: 4.01861572265625\n",
      "2019-12-24 14:05:28,883 INFO Norm compound: min: 2.728064775466919, max: 23.428979873657227, avg: 6.7535719871521\n",
      "2019-12-24 14:05:28,883 INFO LogNorm: min: 0.3889707922935486, max: 1.369753360748291, avg: 0.5674788951873779\n",
      "2019-12-24 14:05:28,884 INFO LogNorm compound: min: 0.43449220061302185, max: 1.369753360748291, avg: 0.6947276592254639\n",
      "2019-12-24 14:05:28,885 INFO Alpha: min: 1.2069766877846264, max: 4.558140877329403, avg: 2.7189644206475023\n",
      "2019-12-24 14:05:28,885 INFO Alpha compound: min: 1.2176340849793623, max: 3.9603022474454272, avg: 2.6610036230330296\n",
      "2019-12-24 14:05:28,886 INFO Alpha Weighted: min: -1.0876175288204757, max: 2.868902178916181, avg: -0.1764910229031938\n",
      "2019-12-24 14:05:28,886 INFO Alpha Weighted compound: min: -0.7291792298817236, max: 2.868902178916181, avg: 0.23297558826130024\n"
     ]
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "results = watcher.analyze(alphas=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-01T22:51:12.554859Z",
     "start_time": "2018-10-01T22:51:12.518859Z"
    }
   },
   "source": [
    "## 2.6 Debug and Custom Logging"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-01T22:51:12.554859Z",
     "start_time": "2018-10-01T22:51:12.518859Z"
    }
   },
   "source": [
    "### Custom Logging at Debug Level"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T00:16:41.769719Z",
     "start_time": "2019-03-06T00:16:38.803181Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:__main__:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:__main__:Analyzing model 'vgg16' with 23 layers\n",
      "DEBUG:__main__:Layer 1: <keras.engine.input_layer.InputLayer object at 0xb38ad7390>\n",
      "DEBUG:__main__:Layer 1: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 2: <keras.layers.convolutional.Conv2D object at 0xb38ad7a58>\n",
      "DEBUG:__main__:Keras tensor shape detected: 3x64 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 2: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 2/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 3/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 4/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 5/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 6/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 7/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 8/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 9/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:Layer 3: <keras.layers.convolutional.Conv2D object at 0xb38ad7ac8>\n",
      "DEBUG:__main__:Keras tensor shape detected: 64x64 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 3: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,64): Lognorm: 0.3978934586048126\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,64): Lognorm: 0.45358702540397644\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,64): Lognorm: 0.40578144788742065\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,64): Lognorm: 0.45428669452667236\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,64): Lognorm: 0.49695152044296265\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,64): Lognorm: 0.45737624168395996\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,64): Lognorm: 0.4044671952724457\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,64): Lognorm: 0.4511153995990753\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,64): Lognorm: 0.3889707922935486\n",
      "DEBUG:__main__:Layer 4: <keras.layers.pooling.MaxPooling2D object at 0xb38acde80>\n",
      "DEBUG:__main__:Layer 4: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 5: <keras.layers.convolutional.Conv2D object at 0xb38ad52e8>\n",
      "DEBUG:__main__:Keras tensor shape detected: 64x128 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 5: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,128): Lognorm: 0.44109997153282166\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,128): Lognorm: 0.4613628089427948\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,128): Lognorm: 0.4370166063308716\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,128): Lognorm: 0.4667660593986511\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,128): Lognorm: 0.5201410055160522\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,128): Lognorm: 0.46935534477233887\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,128): Lognorm: 0.4464665651321411\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,128): Lognorm: 0.48161619901657104\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,128): Lognorm: 0.4471622705459595\n",
      "DEBUG:__main__:Layer 6: <keras.layers.convolutional.Conv2D object at 0xb38b165c0>\n",
      "DEBUG:__main__:Keras tensor shape detected: 128x128 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 6: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,128): Lognorm: 0.45507872104644775\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,128): Lognorm: 0.4839082956314087\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,128): Lognorm: 0.4593982994556427\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,128): Lognorm: 0.48676982522010803\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,128): Lognorm: 0.5329121351242065\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,128): Lognorm: 0.486453115940094\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,128): Lognorm: 0.4555995762348175\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,128): Lognorm: 0.4885979890823364\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,128): Lognorm: 0.44975680112838745\n",
      "DEBUG:__main__:Layer 7: <keras.layers.pooling.MaxPooling2D object at 0xb38b304e0>\n",
      "DEBUG:__main__:Layer 7: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 8: <keras.layers.convolutional.Conv2D object at 0xb38b30be0>\n",
      "DEBUG:__main__:Keras tensor shape detected: 128x256 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 8: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,256): Lognorm: 0.4331028163433075\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,256): Lognorm: 0.501900315284729\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,256): Lognorm: 0.435834139585495\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,256): Lognorm: 0.5173439979553223\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,256): Lognorm: 0.630874752998352\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,256): Lognorm: 0.5137887597084045\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,256): Lognorm: 0.4312974512577057\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,256): Lognorm: 0.4997106194496155\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,256): Lognorm: 0.42677345871925354\n",
      "DEBUG:__main__:Layer 9: <keras.layers.convolutional.Conv2D object at 0xb38b68400>\n",
      "DEBUG:__main__:Keras tensor shape detected: 256x256 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 9: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,256): Lognorm: 0.4611811339855194\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,256): Lognorm: 0.5103281140327454\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,256): Lognorm: 0.46348103880882263\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,256): Lognorm: 0.5043201446533203\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,256): Lognorm: 0.5939016342163086\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,256): Analyzing ...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "DEBUG:__main__:    Weight matrix 6/9 (256,256): Lognorm: 0.5075631737709045\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,256): Lognorm: 0.45727992057800293\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,256): Lognorm: 0.5017167329788208\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,256): Lognorm: 0.4612100124359131\n",
      "DEBUG:__main__:Layer 10: <keras.layers.convolutional.Conv2D object at 0xb38ad8630>\n",
      "DEBUG:__main__:Keras tensor shape detected: 256x256 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 10: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,256): Lognorm: 0.4767092168331146\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,256): Lognorm: 0.5228332281112671\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,256): Lognorm: 0.48029711842536926\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,256): Lognorm: 0.5129597187042236\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,256): Lognorm: 0.5902463793754578\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,256): Lognorm: 0.52484530210495\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,256): Lognorm: 0.4757029116153717\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,256): Lognorm: 0.5153259038925171\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,256): Lognorm: 0.4783079922199249\n",
      "DEBUG:__main__:Layer 11: <keras.layers.pooling.MaxPooling2D object at 0xb38b9bf60>\n",
      "DEBUG:__main__:Layer 11: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 12: <keras.layers.convolutional.Conv2D object at 0xb38bb65f8>\n",
      "DEBUG:__main__:Keras tensor shape detected: 256x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 12: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,512): Lognorm: 0.533117413520813\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,512): Lognorm: 0.5677251815795898\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,512): Lognorm: 0.5357765555381775\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,512): Lognorm: 0.5680774450302124\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,512): Lognorm: 0.6403180360794067\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,512): Lognorm: 0.5651501417160034\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,512): Lognorm: 0.5288752317428589\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,512): Lognorm: 0.5627536177635193\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,512): Lognorm: 0.5263329744338989\n",
      "DEBUG:__main__:Layer 13: <keras.layers.convolutional.Conv2D object at 0xb38bf1240>\n",
      "DEBUG:__main__:Keras tensor shape detected: 512x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 13: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Lognorm: 0.5816661715507507\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Lognorm: 0.5994589924812317\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Lognorm: 0.5804641246795654\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Lognorm: 0.5878161787986755\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Lognorm: 0.6349806785583496\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Lognorm: 0.5888810753822327\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Lognorm: 0.5777970552444458\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Lognorm: 0.5947223901748657\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Lognorm: 0.5786637663841248\n",
      "DEBUG:__main__:Layer 14: <keras.layers.convolutional.Conv2D object at 0xb38c0c7f0>\n",
      "DEBUG:__main__:Keras tensor shape detected: 512x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 14: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Lognorm: 0.598810613155365\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Lognorm: 0.6240571737289429\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Lognorm: 0.6002294421195984\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Lognorm: 0.6069266200065613\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Lognorm: 0.6537207961082458\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Lognorm: 0.6051104664802551\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Lognorm: 0.596659243106842\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Lognorm: 0.6204671859741211\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Lognorm: 0.5969995856285095\n",
      "DEBUG:__main__:Layer 15: <keras.layers.pooling.MaxPooling2D object at 0xb38c2cbe0>\n",
      "DEBUG:__main__:Layer 15: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 16: <keras.layers.convolutional.Conv2D object at 0xb38c2cf60>\n",
      "DEBUG:__main__:Keras tensor shape detected: 512x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 16: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Lognorm: 0.6414387226104736\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Lognorm: 0.658482551574707\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Lognorm: 0.6415495276451111\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Lognorm: 0.6379082798957825\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Lognorm: 0.6930620670318604\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Lognorm: 0.6355615258216858\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Lognorm: 0.6379796266555786\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Lognorm: 0.6546915769577026\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Lognorm: 0.6379215717315674\n",
      "DEBUG:__main__:Layer 17: <keras.layers.convolutional.Conv2D object at 0xb38c62a58>\n",
      "DEBUG:__main__:Keras tensor shape detected: 512x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 17: Analyzing 9 weight matrices...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Lognorm: 0.6483020782470703\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Lognorm: 0.6716188788414001\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Lognorm: 0.6487486958503723\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Lognorm: 0.6313163042068481\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Lognorm: 0.690161406993866\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Lognorm: 0.6324495077133179\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Lognorm: 0.6435778141021729\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Lognorm: 0.6643065214157104\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Lognorm: 0.6426410675048828\n",
      "DEBUG:__main__:Layer 18: <keras.layers.convolutional.Conv2D object at 0xb38c7db38>\n",
      "DEBUG:__main__:Keras tensor shape detected: 512x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 18: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Lognorm: 0.6312968730926514\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Lognorm: 0.6557008028030396\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Lognorm: 0.6303039193153381\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Lognorm: 0.6299319267272949\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Lognorm: 0.6738649010658264\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Lognorm: 0.6266793012619019\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Lognorm: 0.6302400231361389\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Lognorm: 0.6563602089881897\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Lognorm: 0.6300156712532043\n",
      "DEBUG:__main__:Layer 19: <keras.layers.pooling.MaxPooling2D object at 0xb38ada7f0>\n",
      "DEBUG:__main__:Layer 19: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 20: <keras.layers.core.Flatten object at 0xb38adacf8>\n",
      "DEBUG:__main__:Layer 20: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 21: <keras.layers.core.Dense object at 0xb38acda58>\n",
      "DEBUG:__main__:Layer 21: Analyzing 1 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,25088): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,25088): Lognorm: 1.369753360748291\n",
      "DEBUG:__main__:Layer 22: <keras.layers.core.Dense object at 0xb38cd6cc0>\n",
      "DEBUG:__main__:Layer 22: Analyzing 1 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,4096): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,4096): Lognorm: 1.2557973861694336\n",
      "DEBUG:__main__:Layer 23: <keras.layers.core.Dense object at 0xb38cf3f60>\n",
      "DEBUG:__main__:Layer 23: Analyzing 1 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (1000,4096): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (1000,4096): Lognorm: 1.2242090702056885\n",
      "INFO:__main__:### Printing results ###\n",
      "DEBUG:__main__:Layer 2: Norm compound: 2.728064775466919\n",
      "DEBUG:__main__:Layer 2: LogNorm compound: 0.43449220061302185\n",
      "DEBUG:__main__:Layer 4: Norm compound: 2.9116666316986084\n",
      "DEBUG:__main__:Layer 4: LogNorm compound: 0.46344295144081116\n",
      "DEBUG:__main__:Layer 5: Norm compound: 3.0083017349243164\n",
      "DEBUG:__main__:Layer 5: LogNorm compound: 0.477608323097229\n",
      "DEBUG:__main__:Layer 7: Norm compound: 3.108454704284668\n",
      "DEBUG:__main__:Layer 7: LogNorm compound: 0.48784738779067993\n",
      "DEBUG:__main__:Layer 8: Norm compound: 3.1452996730804443\n",
      "DEBUG:__main__:Layer 8: LogNorm compound: 0.49566471576690674\n",
      "DEBUG:__main__:Layer 9: Norm compound: 3.2361817359924316\n",
      "DEBUG:__main__:Layer 9: LogNorm compound: 0.5085809230804443\n",
      "DEBUG:__main__:Layer 11: Norm compound: 3.630833148956299\n",
      "DEBUG:__main__:Layer 11: LogNorm compound: 0.5586807727813721\n",
      "DEBUG:__main__:Layer 12: Norm compound: 3.9078686237335205\n",
      "DEBUG:__main__:Layer 12: LogNorm compound: 0.5916056036949158\n",
      "DEBUG:__main__:Layer 13: Norm compound: 4.090791702270508\n",
      "DEBUG:__main__:Layer 13: LogNorm compound: 0.6114423274993896\n",
      "DEBUG:__main__:Layer 15: Norm compound: 4.457479476928711\n",
      "DEBUG:__main__:Layer 15: LogNorm compound: 0.6487328410148621\n",
      "DEBUG:__main__:Layer 16: Norm compound: 4.497330188751221\n",
      "DEBUG:__main__:Layer 16: LogNorm compound: 0.6525691151618958\n",
      "DEBUG:__main__:Layer 17: Norm compound: 4.373068809509277\n",
      "DEBUG:__main__:Layer 17: LogNorm compound: 0.6404882073402405\n",
      "DEBUG:__main__:Layer 20: Norm: 23.428979873657227\n",
      "DEBUG:__main__:Layer 20: LogNorm: 1.369753360748291\n",
      "DEBUG:__main__:Layer 21: Norm: 18.02176856994629\n",
      "DEBUG:__main__:Layer 21: LogNorm: 1.2557973861694336\n",
      "DEBUG:__main__:Layer 22: Norm: 16.757492065429688\n",
      "DEBUG:__main__:Layer 22: LogNorm: 1.2242090702056885\n",
      "INFO:__main__:Norm: min: 2.4488985538482666, max: 23.428979873657227, avg: 4.01861572265625\n",
      "INFO:__main__:Norm compound: min: 2.728064775466919, max: 23.428979873657227, avg: 6.7535719871521\n",
      "INFO:__main__:LogNorm: min: 0.3889707922935486, max: 1.369753360748291, avg: 0.5674788951873779\n",
      "INFO:__main__:LogNorm compound: min: 0.43449220061302185, max: 1.369753360748291, avg: 0.6947276592254639\n"
     ]
    }
   ],
   "source": [
    "import logging\n",
    "logging.basicConfig(level=logging.INFO)\n",
    "logger = logging.getLogger(__name__)\n",
    "logger.setLevel(logging.DEBUG)\n",
    "\n",
    "import weightwatcher as ww\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model, logger=logger)\n",
    "\n",
    "results = watcher.analyze()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "ExecuteTime": {
     "end_time": "2018-10-01T22:51:12.554859Z",
     "start_time": "2018-10-01T22:51:12.518859Z"
    }
   },
   "source": [
    "### Disable Logging"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T00:16:43.234403Z",
     "start_time": "2019-03-06T00:16:41.779045Z"
    }
   },
   "outputs": [],
   "source": [
    "import weightwatcher as ww\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model, log=False)\n",
    "\n",
    "results = watcher.analyze()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 3. pyTorch Models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T00:16:43.249725Z",
     "start_time": "2019-03-06T00:16:43.240611Z"
    }
   },
   "outputs": [],
   "source": [
    "data = []"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T00:40:17.723085Z",
     "start_time": "2019-03-06T00:16:43.259483Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:05:34,227 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:weightwatcher.weightwatcher:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:05:34,229 INFO Analyzing model\n",
      "INFO:weightwatcher.weightwatcher:Analyzing model\n",
      "2019-12-24 14:09:42,933 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 14:09:44,455 INFO Norm: min: 3.203004837036133, max: 51.2126579284668, avg: 7.801441669464111\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 3.203004837036133, max: 51.2126579284668, avg: 7.801441669464111\n",
      "2019-12-24 14:09:44,456 INFO Norm compound: min: 3.56429123878479, max: 51.2126579284668, avg: 14.09792709350586\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 3.56429123878479, max: 51.2126579284668, avg: 14.09792709350586\n",
      "2019-12-24 14:09:44,457 INFO LogNorm: min: 0.5055575966835022, max: 1.7093772888183594, avg: 0.8381870985031128\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.5055575966835022, max: 1.7093772888183594, avg: 0.8381870985031128\n",
      "2019-12-24 14:09:44,458 INFO LogNorm compound: min: 0.5506355166435242, max: 1.7093772888183594, avg: 0.9792784452438354\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.5506355166435242, max: 1.7093772888183594, avg: 0.9792784452438354\n",
      "2019-12-24 14:09:44,459 INFO Alpha: min: 1.525053187957775, max: 4.343339437091599, avg: 2.89233801062374\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.525053187957775, max: 4.343339437091599, avg: 2.89233801062374\n",
      "2019-12-24 14:09:44,460 INFO Alpha compound: min: 1.7535254182659035, max: 3.8419123407107967, avg: 2.767717801865902\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.7535254182659035, max: 3.8419123407107967, avg: 2.767717801865902\n",
      "2019-12-24 14:09:44,461 INFO Alpha Weighted: min: 0.35369982858727583, max: 3.7137959804872724, avg: 1.405072984118814\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: 0.35369982858727583, max: 3.7137959804872724, avg: 1.405072984118814\n",
      "2019-12-24 14:09:44,462 INFO Alpha Weighted compound: min: 0.4626615096713753, max: 3.7137959804872724, avg: 1.7930749068678844\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: 0.4626615096713753, max: 3.7137959804872724, avg: 1.7930749068678844\n"
     ]
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "import torchvision.models as models\n",
    "\n",
    "model = models.vgg16(pretrained=True)\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "results = watcher.analyze(alphas=True)\n",
    "\n",
    "data.append({\"name\": \"vgg16torch\", \"summary\": watcher.get_summary()})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T01:03:30.884210Z",
     "start_time": "2019-03-06T00:40:17.729503Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:09:45,911 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:weightwatcher.weightwatcher:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:09:45,913 INFO Analyzing model\n",
      "INFO:weightwatcher.weightwatcher:Analyzing model\n",
      "2019-12-24 14:13:50,826 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 14:13:52,463 INFO Norm: min: 1.9416159391403198, max: 47.83592224121094, avg: 6.997468948364258\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 1.9416159391403198, max: 47.83592224121094, avg: 6.997468948364258\n",
      "2019-12-24 14:13:52,464 INFO Norm compound: min: 2.2552099227905273, max: 47.83592224121094, avg: 13.280407905578613\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 2.2552099227905273, max: 47.83592224121094, avg: 13.280407905578613\n",
      "2019-12-24 14:13:52,465 INFO LogNorm: min: 0.28816333413124084, max: 1.6797541379928589, avg: 0.7700976729393005\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.28816333413124084, max: 1.6797541379928589, avg: 0.7700976729393005\n",
      "2019-12-24 14:13:52,466 INFO LogNorm compound: min: 0.3503240942955017, max: 1.6797541379928589, avg: 0.9220765233039856\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.3503240942955017, max: 1.6797541379928589, avg: 0.9220765233039856\n",
      "2019-12-24 14:13:52,467 INFO Alpha: min: 1.5416896884676328, max: 5.67734879325649, avg: 3.099085407095711\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.5416896884676328, max: 5.67734879325649, avg: 3.099085407095711\n",
      "2019-12-24 14:13:52,468 INFO Alpha compound: min: 1.6648925749775423, max: 4.570109325217148, avg: 2.966024122982959\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.6648925749775423, max: 4.570109325217148, avg: 2.966024122982959\n",
      "2019-12-24 14:13:52,470 INFO Alpha Weighted: min: -0.14159670551181727, max: 5.226979652357834, avg: 1.0820403179586902\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: -0.14159670551181727, max: 5.226979652357834, avg: 1.0820403179586902\n",
      "2019-12-24 14:13:52,470 INFO Alpha Weighted compound: min: 0.1388687968794394, max: 5.226979652357834, avg: 1.5490586351060547\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: 0.1388687968794394, max: 5.226979652357834, avg: 1.5490586351060547\n"
     ]
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "import torchvision.models as models\n",
    "\n",
    "model = models.vgg16_bn(pretrained=True)\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "results = watcher.analyze(alphas=True)\n",
    "\n",
    "data.append({\"name\": \"vgg16bntorch\", \"summary\": watcher.get_summary()})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T01:24:27.165312Z",
     "start_time": "2019-03-06T01:03:30.908016Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:13:53,950 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:weightwatcher.weightwatcher:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:13:53,951 INFO Analyzing model\n",
      "INFO:weightwatcher.weightwatcher:Analyzing model\n",
      "2019-12-24 14:17:37,639 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 14:17:38,592 INFO Norm: min: 5.2481255531311035, max: 57.39590072631836, avg: 10.107156753540039\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 5.2481255531311035, max: 57.39590072631836, avg: 10.107156753540039\n",
      "2019-12-24 14:17:38,593 INFO Norm compound: min: 5.895910263061523, max: 57.39590072631836, avg: 19.885963439941406\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 5.895910263061523, max: 57.39590072631836, avg: 19.885963439941406\n",
      "2019-12-24 14:17:38,594 INFO LogNorm: min: 0.7200042009353638, max: 1.758880853652954, avg: 0.9474983215332031\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.7200042009353638, max: 1.758880853652954, avg: 0.9474983215332031\n",
      "2019-12-24 14:17:38,595 INFO LogNorm compound: min: 0.7691774368286133, max: 1.758880853652954, avg: 1.1386475563049316\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.7691774368286133, max: 1.758880853652954, avg: 1.1386475563049316\n",
      "2019-12-24 14:17:38,596 INFO Alpha: min: 1.3854856636971848, max: 3.875876983334032, avg: 2.5724933311485363\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.3854856636971848, max: 3.875876983334032, avg: 2.5724933311485363\n",
      "2019-12-24 14:17:38,597 INFO Alpha compound: min: 1.522580930995347, max: 3.5210144189935715, avg: 2.5370047984035393\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.522580930995347, max: 3.5210144189935715, avg: 2.5370047984035393\n",
      "2019-12-24 14:17:38,598 INFO Alpha Weighted: min: 0.6363944562361387, max: 4.999376215087901, avg: 1.8439950665352716\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: 0.6363944562361387, max: 4.999376215087901, avg: 1.8439950665352716\n",
      "2019-12-24 14:17:38,599 INFO Alpha Weighted compound: min: 1.0371089622425371, max: 4.999376215087901, avg: 2.4581932538546676\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: 1.0371089622425371, max: 4.999376215087901, avg: 2.4581932538546676\n"
     ]
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "import torchvision.models as models\n",
    "\n",
    "model = models.vgg11(pretrained=True)\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "results = watcher.analyze(alphas=True)\n",
    "\n",
    "data.append({\"name\": \"vgg11torch\", \"summary\": watcher.get_summary()})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T01:45:08.643722Z",
     "start_time": "2019-03-06T01:24:27.176445Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:17:40,033 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:weightwatcher.weightwatcher:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:17:40,034 INFO Analyzing model\n",
      "INFO:weightwatcher.weightwatcher:Analyzing model\n",
      "2019-12-24 14:21:21,319 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 14:21:22,343 INFO Norm: min: 3.9749603271484375, max: 53.424800872802734, avg: 9.151532173156738\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 3.9749603271484375, max: 53.424800872802734, avg: 9.151532173156738\n",
      "2019-12-24 14:21:22,344 INFO Norm compound: min: 4.532721519470215, max: 53.424800872802734, avg: 19.01651382446289\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 4.532721519470215, max: 53.424800872802734, avg: 19.01651382446289\n",
      "2019-12-24 14:21:22,345 INFO LogNorm: min: 0.5993328094482422, max: 1.7277429103851318, avg: 0.8902592658996582\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.5993328094482422, max: 1.7277429103851318, avg: 0.8902592658996582\n",
      "2019-12-24 14:21:22,346 INFO LogNorm compound: min: 0.6529925465583801, max: 1.7277429103851318, avg: 1.0959337949752808\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.6529925465583801, max: 1.7277429103851318, avg: 1.0959337949752808\n",
      "2019-12-24 14:21:22,347 INFO Alpha: min: 1.3312675711718847, max: 5.229885854374285, avg: 2.7469664048443074\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.3312675711718847, max: 5.229885854374285, avg: 2.7469664048443074\n",
      "2019-12-24 14:21:22,348 INFO Alpha compound: min: 1.66424651984324, max: 3.9272856564161924, avg: 2.638103821130217\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.66424651984324, max: 3.9272856564161924, avg: 2.638103821130217\n",
      "2019-12-24 14:21:22,349 INFO Alpha Weighted: min: 0.30176628869093886, max: 5.329173539336359, avg: 1.5976468201672023\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: 0.30176628869093886, max: 5.329173539336359, avg: 1.5976468201672023\n",
      "2019-12-24 14:21:22,350 INFO Alpha Weighted compound: min: 0.6444258908863996, max: 5.329173539336359, avg: 2.201919500846972\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: 0.6444258908863996, max: 5.329173539336359, avg: 2.201919500846972\n"
     ]
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "import torchvision.models as models\n",
    "\n",
    "model = models.vgg11_bn(pretrained=True)\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "results = watcher.analyze(alphas=True)\n",
    "\n",
    "data.append({\"name\": \"vgg11bntorch\", \"summary\": watcher.get_summary()})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T02:07:08.066376Z",
     "start_time": "2019-03-06T01:45:08.651095Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:21:23,777 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:weightwatcher.weightwatcher:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:21:23,778 INFO Analyzing model\n",
      "INFO:weightwatcher.weightwatcher:Analyzing model\n",
      "2019-12-24 14:25:09,536 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 14:25:10,705 INFO Norm: min: 3.4243783950805664, max: 57.03801345825195, avg: 8.586669921875\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 3.4243783950805664, max: 57.03801345825195, avg: 8.586669921875\n",
      "2019-12-24 14:25:10,706 INFO Norm compound: min: 3.7364306449890137, max: 57.03801345825195, avg: 16.972938537597656\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 3.7364306449890137, max: 57.03801345825195, avg: 16.972938537597656\n",
      "2019-12-24 14:25:10,707 INFO LogNorm: min: 0.53458172082901, max: 1.7561644315719604, avg: 0.8640468120574951\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.53458172082901, max: 1.7561644315719604, avg: 0.8640468120574951\n",
      "2019-12-24 14:25:10,708 INFO LogNorm compound: min: 0.5716044902801514, max: 1.7561644315719604, avg: 1.0408902168273926\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.5716044902801514, max: 1.7561644315719604, avg: 1.0408902168273926\n",
      "2019-12-24 14:25:10,709 INFO Alpha: min: 1.4229916456933371, max: 4.882618427826814, avg: 2.6896965121426315\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.4229916456933371, max: 4.882618427826814, avg: 2.6896965121426315\n",
      "2019-12-24 14:25:10,710 INFO Alpha compound: min: 1.6193138952563046, max: 3.654240384837193, avg: 2.6287548746902396\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.6193138952563046, max: 3.654240384837193, avg: 2.6287548746902396\n",
      "2019-12-24 14:25:10,711 INFO Alpha Weighted: min: 0.37454269723312644, max: 4.799173592745528, avg: 1.6453829975237424\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: 0.37454269723312644, max: 4.799173592745528, avg: 1.6453829975237424\n",
      "2019-12-24 14:25:10,712 INFO Alpha Weighted compound: min: 0.6126543377111549, max: 4.799173592745528, avg: 2.1923071329174553\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: 0.6126543377111549, max: 4.799173592745528, avg: 2.1923071329174553\n"
     ]
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "import torchvision.models as models\n",
    "\n",
    "model = models.vgg13(pretrained=True)\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "results = watcher.analyze(alphas=True)\n",
    "\n",
    "data.append({\"name\": \"vgg13torch\", \"summary\": watcher.get_summary()})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T02:28:16.756345Z",
     "start_time": "2019-03-06T02:07:08.077271Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:25:12,122 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:weightwatcher.weightwatcher:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:25:12,124 INFO Analyzing model\n",
      "INFO:weightwatcher.weightwatcher:Analyzing model\n",
      "2019-12-24 14:28:52,518 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 14:28:53,764 INFO Norm: min: 2.1068990230560303, max: 52.42221450805664, avg: 7.682347774505615\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 2.1068990230560303, max: 52.42221450805664, avg: 7.682347774505615\n",
      "2019-12-24 14:28:53,765 INFO Norm compound: min: 2.452697515487671, max: 52.42221450805664, avg: 16.066328048706055\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 2.452697515487671, max: 52.42221450805664, avg: 16.066328048706055\n",
      "2019-12-24 14:28:53,766 INFO LogNorm: min: 0.3236437141895294, max: 1.719515323638916, avg: 0.7937618494033813\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.3236437141895294, max: 1.719515323638916, avg: 0.7937618494033813\n",
      "2019-12-24 14:28:53,767 INFO LogNorm compound: min: 0.3866419494152069, max: 1.719515323638916, avg: 0.9850755333900452\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.3866419494152069, max: 1.719515323638916, avg: 0.9850755333900452\n",
      "2019-12-24 14:28:53,768 INFO Alpha: min: 1.5395371164868283, max: 7.42747746192009, avg: 2.9530773615514185\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.5395371164868283, max: 7.42747746192009, avg: 2.9530773615514185\n",
      "2019-12-24 14:28:53,769 INFO Alpha compound: min: 1.853543286417002, max: 4.015678609398574, avg: 2.816647560031195\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.853543286417002, max: 4.015678609398574, avg: 2.816647560031195\n",
      "2019-12-24 14:28:53,770 INFO Alpha Weighted: min: -0.05021955395170501, max: 5.28496241458978, avg: 1.3551243085548361\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: -0.05021955395170501, max: 5.28496241458978, avg: 1.3551243085548361\n",
      "2019-12-24 14:28:53,771 INFO Alpha Weighted compound: min: 0.45352451514183456, max: 5.28496241458978, avg: 1.9049098069460684\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: 0.45352451514183456, max: 5.28496241458978, avg: 1.9049098069460684\n"
     ]
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "import torchvision.models as models\n",
    "\n",
    "model = models.vgg13_bn(pretrained=True)\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "results = watcher.analyze(alphas=True)\n",
    "\n",
    "data.append({\"name\": \"vgg13bntorch\", \"summary\": watcher.get_summary()})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T02:52:14.088253Z",
     "start_time": "2019-03-06T02:28:16.767312Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:28:55,277 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:weightwatcher.weightwatcher:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:28:55,278 INFO Analyzing model\n",
      "INFO:weightwatcher.weightwatcher:Analyzing model\n",
      "2019-12-24 14:33:28,339 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 14:33:30,180 INFO Norm: min: 3.2178001403808594, max: 47.974002838134766, avg: 7.2684478759765625\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 3.2178001403808594, max: 47.974002838134766, avg: 7.2684478759765625\n",
      "2019-12-24 14:33:30,182 INFO Norm compound: min: 3.559237241744995, max: 47.974002838134766, avg: 12.337833404541016\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 3.559237241744995, max: 47.974002838134766, avg: 12.337833404541016\n",
      "2019-12-24 14:33:30,183 INFO LogNorm: min: 0.507559061050415, max: 1.6810059547424316, avg: 0.8185057640075684\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.507559061050415, max: 1.6810059547424316, avg: 0.8185057640075684\n",
      "2019-12-24 14:33:30,184 INFO LogNorm compound: min: 0.5500112771987915, max: 1.6810059547424316, avg: 0.9365273714065552\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.5500112771987915, max: 1.6810059547424316, avg: 0.9365273714065552\n",
      "2019-12-24 14:33:30,186 INFO Alpha: min: 1.4433522167552777, max: 5.027317242480103, avg: 2.964672695949022\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.4433522167552777, max: 5.027317242480103, avg: 2.964672695949022\n",
      "2019-12-24 14:33:30,187 INFO Alpha compound: min: 1.5696912807219614, max: 4.10398537172971, avg: 2.8479756155799745\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.5696912807219614, max: 4.10398537172971, avg: 2.8479756155799745\n",
      "2019-12-24 14:33:30,188 INFO Alpha Weighted: min: 0.2145263550533989, max: 3.591568654707832, avg: 1.158888231628616\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: 0.2145263550533989, max: 3.591568654707832, avg: 1.158888231628616\n",
      "2019-12-24 14:33:30,189 INFO Alpha Weighted compound: min: 0.3858703649968457, max: 3.591568654707832, avg: 1.5002341935342836\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: 0.3858703649968457, max: 3.591568654707832, avg: 1.5002341935342836\n"
     ]
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "import torchvision.models as models\n",
    "\n",
    "model = models.vgg19(pretrained=True)\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "results = watcher.analyze(alphas=True)\n",
    "\n",
    "data.append({\"name\": \"vgg19torch\", \"summary\": watcher.get_summary()})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T03:16:03.620082Z",
     "start_time": "2019-03-06T02:52:14.095969Z"
    }
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 14:33:31,677 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:weightwatcher.weightwatcher:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 14:33:31,678 INFO Analyzing model\n",
      "INFO:weightwatcher.weightwatcher:Analyzing model\n",
      "2019-12-24 14:37:55,892 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 14:37:57,878 INFO Norm: min: 1.6498754024505615, max: 45.86455154418945, avg: 6.413882255554199\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 1.6498754024505615, max: 45.86455154418945, avg: 6.413882255554199\n",
      "2019-12-24 14:37:57,880 INFO Norm compound: min: 1.9919495582580566, max: 45.86455154418945, avg: 11.5429048538208\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 1.9919495582580566, max: 45.86455154418945, avg: 11.5429048538208\n",
      "2019-12-24 14:37:57,881 INFO LogNorm: min: 0.21745115518569946, max: 1.6614772081375122, avg: 0.7364940643310547\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.21745115518569946, max: 1.6614772081375122, avg: 0.7364940643310547\n",
      "2019-12-24 14:37:57,881 INFO LogNorm compound: min: 0.293550968170166, max: 1.6614772081375122, avg: 0.8661542534828186\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.293550968170166, max: 1.6614772081375122, avg: 0.8661542534828186\n",
      "2019-12-24 14:37:57,882 INFO Alpha: min: 1.0748766362053024, max: 5.64448912713187, avg: 3.0578078027714213\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.0748766362053024, max: 5.64448912713187, avg: 3.0578078027714213\n",
      "2019-12-24 14:37:57,883 INFO Alpha compound: min: 1.664717727538821, max: 4.799351170421366, avg: 2.9531824880317767\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.664717727538821, max: 4.799351170421366, avg: 2.9531824880317767\n",
      "2019-12-24 14:37:57,884 INFO Alpha Weighted: min: -0.3748439788066682, max: 5.183230373691309, avg: 0.813032970569857\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: -0.3748439788066682, max: 5.183230373691309, avg: 0.813032970569857\n",
      "2019-12-24 14:37:57,885 INFO Alpha Weighted compound: min: -0.043707289421253376, max: 5.183230373691309, avg: 1.231837513595454\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: -0.043707289421253376, max: 5.183230373691309, avg: 1.231837513595454\n"
     ]
    }
   ],
   "source": [
    "import weightwatcher as ww\n",
    "import torchvision.models as models\n",
    "\n",
    "model = models.vgg19_bn(pretrained=True)\n",
    "\n",
    "watcher = ww.WeightWatcher(model=model)\n",
    "\n",
    "results = watcher.analyze(alphas=True)\n",
    "\n",
    "data.append({\"name\": \"vgg19bntorch\", \"summary\": watcher.get_summary()})"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T03:16:03.656840Z",
     "start_time": "2019-03-06T03:16:03.629371Z"
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[{'name': 'vgg16torch',\n",
       "  'summary': {'norm': 7.8014417,\n",
       "   'norm_compound': 14.097927,\n",
       "   'lognorm': 0.8381871,\n",
       "   'lognorm_compound': 0.97927845,\n",
       "   'alpha': 2.89233801062374,\n",
       "   'alpha_compound': 2.767717801865902,\n",
       "   'alpha_weighted': 1.405072984118814,\n",
       "   'alpha_weighted_compound': 1.7930749068678844}},\n",
       " {'name': 'vgg16bntorch',\n",
       "  'summary': {'norm': 6.997469,\n",
       "   'norm_compound': 13.280408,\n",
       "   'lognorm': 0.7700977,\n",
       "   'lognorm_compound': 0.9220765,\n",
       "   'alpha': 3.099085407095711,\n",
       "   'alpha_compound': 2.966024122982959,\n",
       "   'alpha_weighted': 1.0820403179586902,\n",
       "   'alpha_weighted_compound': 1.5490586351060547}},\n",
       " {'name': 'vgg11torch',\n",
       "  'summary': {'norm': 10.107157,\n",
       "   'norm_compound': 19.885963,\n",
       "   'lognorm': 0.9474983,\n",
       "   'lognorm_compound': 1.1386476,\n",
       "   'alpha': 2.5724933311485363,\n",
       "   'alpha_compound': 2.5370047984035393,\n",
       "   'alpha_weighted': 1.8439950665352716,\n",
       "   'alpha_weighted_compound': 2.4581932538546676}},\n",
       " {'name': 'vgg11bntorch',\n",
       "  'summary': {'norm': 9.151532,\n",
       "   'norm_compound': 19.016514,\n",
       "   'lognorm': 0.89025927,\n",
       "   'lognorm_compound': 1.0959338,\n",
       "   'alpha': 2.7469664048443074,\n",
       "   'alpha_compound': 2.638103821130217,\n",
       "   'alpha_weighted': 1.5976468201672023,\n",
       "   'alpha_weighted_compound': 2.201919500846972}},\n",
       " {'name': 'vgg13torch',\n",
       "  'summary': {'norm': 8.58667,\n",
       "   'norm_compound': 16.972939,\n",
       "   'lognorm': 0.8640468,\n",
       "   'lognorm_compound': 1.0408902,\n",
       "   'alpha': 2.6896965121426315,\n",
       "   'alpha_compound': 2.6287548746902396,\n",
       "   'alpha_weighted': 1.6453829975237424,\n",
       "   'alpha_weighted_compound': 2.1923071329174553}},\n",
       " {'name': 'vgg13bntorch',\n",
       "  'summary': {'norm': 7.682348,\n",
       "   'norm_compound': 16.066328,\n",
       "   'lognorm': 0.79376185,\n",
       "   'lognorm_compound': 0.98507553,\n",
       "   'alpha': 2.9530773615514185,\n",
       "   'alpha_compound': 2.816647560031195,\n",
       "   'alpha_weighted': 1.3551243085548361,\n",
       "   'alpha_weighted_compound': 1.9049098069460684}},\n",
       " {'name': 'vgg19torch',\n",
       "  'summary': {'norm': 7.268448,\n",
       "   'norm_compound': 12.337833,\n",
       "   'lognorm': 0.81850576,\n",
       "   'lognorm_compound': 0.9365274,\n",
       "   'alpha': 2.964672695949022,\n",
       "   'alpha_compound': 2.8479756155799745,\n",
       "   'alpha_weighted': 1.158888231628616,\n",
       "   'alpha_weighted_compound': 1.5002341935342836}},\n",
       " {'name': 'vgg19bntorch',\n",
       "  'summary': {'norm': 6.4138823,\n",
       "   'norm_compound': 11.542905,\n",
       "   'lognorm': 0.73649406,\n",
       "   'lognorm_compound': 0.86615425,\n",
       "   'alpha': 3.0578078027714213,\n",
       "   'alpha_compound': 2.9531824880317767,\n",
       "   'alpha_weighted': 0.813032970569857,\n",
       "   'alpha_weighted_compound': 1.231837513595454}}]"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T03:16:03.721719Z",
     "start_time": "2019-03-06T03:16:03.665193Z"
    }
   },
   "outputs": [],
   "source": [
    "# pytorch Model accuracies \n",
    "# https://github.com/Cadene/pretrained-models.pytorch\n",
    "\n",
    "accuracies = {\n",
    "    \"vgg11torch\": 68.970,\n",
    "    \"vgg11bntorch\": 70.452,\n",
    "    \"vgg13torch\": 69.662,\n",
    "    \"vgg13bntorch\": 71.508,\n",
    "    \"vgg16torch\": 71.636,\n",
    "    \"vgg16bntorch\": 73.518,\n",
    "    \"vgg19torch\": 72.080,\n",
    "    \"vgg19bntorch\": 74.266,\n",
    "}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T03:16:03.740667Z",
     "start_time": "2019-03-06T03:16:03.728138Z"
    }
   },
   "outputs": [],
   "source": [
    "# pytorch Model accuracies \n",
    "# https://github.com/Cadene/pretrained-models.pytorch\n",
    "\n",
    "accuracies5 = {\n",
    "    \"vgg11torch\": 88.746,\n",
    "    \"vgg11bntorch\": 89.818,\n",
    "    \"vgg13torch\": 89.264,\n",
    "    \"vgg13bntorch\": 90.494,\n",
    "    \"vgg16torch\": 90.354,\n",
    "    \"vgg16bntorch\": 91.608,\n",
    "    \"vgg19torch\": 90.822,\n",
    "    \"vgg19bntorch\": 92.066,\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3.1 Log Norm of Weight Matrices vs Accuracies of models"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following graph demonstrates the linear relationship between the average Log Norm of Weight matrices and the test accuracies of the models (notice we didnt't need the test data):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T03:16:06.458114Z",
     "start_time": "2019-03-06T03:16:03.747317Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "plt.rcParams['figure.figsize'] = [8,8]\n",
    "\n",
    "for modelname, accuracy in accuracies5.items():\n",
    "    x = accuracy\n",
    "    summary = [d[\"summary\"] for d in data if d[\"name\"] == modelname]\n",
    "    y = summary[0][\"lognorm\"]\n",
    "    label = modelname\n",
    "    plt.scatter(x,y,label=label)\n",
    "\n",
    "plt.legend()\n",
    "plt.title(r\"Test Accuracy vs Average Log Norm $\\langle\\log\\Vert W\\Vert\\rangle$\"+\"\\nPretrained VGG and VGG_BN Models\")\n",
    "plt.xlabel(r\"Test Accuracy\")\n",
    "plt.ylabel(r\"$\\langle\\log\\Vert W\\Vert\\rangle$\");"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's compare the average Log Norm with the average Log Norm compound:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T03:16:07.301228Z",
     "start_time": "2019-03-06T03:16:06.465048Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "plt.rcParams['figure.figsize'] = [8,8]\n",
    "\n",
    "x = []\n",
    "y1, y2 = [], []\n",
    "for modelname, accuracy in accuracies5.items():\n",
    "    x.append(accuracy)\n",
    "    summary = [d[\"summary\"] for d in data if d[\"name\"] == modelname]\n",
    "    y1.append(summary[0][\"lognorm\"])\n",
    "    y2.append(summary[0][\"lognorm_compound\"])\n",
    "    label = modelname\n",
    "plt.scatter(x,y1,label=\"Log Norm\", color='r')\n",
    "plt.scatter(x,y2,label=\"Log Norm Compound\", color='b')\n",
    "\n",
    "plt.legend()\n",
    "plt.title(r\"Test Accuracy vs (Average Log Norm $\\langle\\log\\Vert W\\Vert\\rangle$ and Log Norm Compound)\"+\"\\nPretrained VGG and VGG_BN Models\")\n",
    "plt.xlabel(r\"Test Accuracy\")\n",
    "plt.ylabel(r\"$\\langle\\log\\Vert W\\Vert\\rangle$\");"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 3.2 Power law fitting (Alpha) of Weight Matrices vs Accuracies of models"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The linear relationship between the Power law fitting (Alpha) of the weight matrices and the accuracies of the models is demonstrated in the following graph:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T03:16:08.268963Z",
     "start_time": "2019-03-06T03:16:07.307689Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "plt.rcParams['figure.figsize'] = [8,8]\n",
    "\n",
    "for modelname, accuracy in accuracies5.items():\n",
    "    x = accuracy\n",
    "    summary = [d[\"summary\"] for d in data if d[\"name\"] == modelname]\n",
    "    y = summary[0][\"alpha_weighted\"]\n",
    "    label = modelname\n",
    "    plt.scatter(x,y,label=label)\n",
    "\n",
    "plt.legend()\n",
    "plt.title(r\"Test Accuracy vs Weighted Alpha\"+\"\\nPretrained VGG and VGG_BN Models\")\n",
    "plt.xlabel(r\"Test Accuracy\")\n",
    "plt.ylabel(r\"Weighted Alpha\");"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The more accurate the model, the lower the exponent of the power law fit of the weight matrices is."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "ExecuteTime": {
     "end_time": "2019-03-06T03:16:09.090898Z",
     "start_time": "2019-03-06T03:16:08.275006Z"
    }
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAfgAAAH/CAYAAAC7Lc/HAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjAsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy+17YcXAAAgAElEQVR4nOzdeZxcVZ3//9c7IRACYU1EISQdEdmXYAhbxKAOICCgwAzQQtTBgA4q6HwVjQqi8ecObjNMqwhiG3RYFBVkGY1hh4QtQEC2JIQ1hC0QkSWf3x/ndOd2p6q6Ot3Vy837+XjUo6rO3c69das+95x76hxFBGZmZlYuQ/o7A2ZmZtb7HODNzMxKyAHezMyshBzgzczMSsgB3szMrIQc4M3MzErIAd5KQ9JZkk5qwHrXkfSSpM3rmHe4pJA0prfz0Rvrl/RNST/r7Xn7kqRtJb3eR9s6T9Ln6pz3QklfamBeVnv93TlmfXl8y0bSgZIezK8l6XZJW/dXfhzga8g/6m2PFZL+UXjf3IP13iTpQ3XMt1He5iWru601haQtgCOBc/P7hZIOLUx/Tw6MndOelaRa646If0bE+hHxeC/k8yRJ1/R0PXVs50JJr0oa1ehtDSSSvtr5+yLp0Spph3e1voj4cER8uxfy1dALv8J2Dszb+VQjt2Ndi9TJzFnAGf2VBwf4GvKP+voRsT6wCHh/Ia21D7Lwb8By4GBJm/bB9tpJWqsvt9cLPgr8LiJeze9nA+8qTN8XuK9C2nVRst6eJG0IHAYsA47p5+z0tdnAO9su2iSNB/4J7NkpbQvg2n7LZeNMBZ7Nz9b/LqEffr/bOMD3gKShkr4s6WFJz0hqlbRRnrZeLkU9K+l5STdL2ljS94DdgZ/lmoDv1djEVOBs4CE6/VBLapL0+7zdZ4rrkfQJSfdJWiZpnqSdKpUgilV+bVVLeX+eAv5b0mhJV0hakvfj95LeUlh+lKRfSnpS0nOSfpPTH5T0L4X5hkt6QdJ2FY7hw5LeW3i/Tp53+2rHsMqxeh/wt8L72aQA3uadwLcqpM0ubPtESffn7f0p1wqsUvqS9KZ8XF7MtTHfrFAqf5+kh/JxOSsvN4H0eU7Jn/2TOX1dSWfnUuWTkn4kaZ1CvqZLekrSYqDLmh/SheFjeX+r/tArV8XmWoUnJD0u6ZOdZltX0sx8Lt0ladfC8l+R9Eiedrekg2tsa5/8+b2Qt3NW20Vk4fh+rPMxy9PXkvQDSUuVqj//pdp2gBuBDYHt8/t3AlcDj3ZKuzcilub17yjpL3m781Uo2atTtbikL7V9FpKmdf5OAaMkXZmPyfWSxuX0tvPs/vzZH57X94F8XJ+XdK2k7QvbmiTpzryuXwFr19jv4oXdScAuknasMe9Nkr4maW7+TC7Oyxfn+UjezyWS/l8hvepnWWVbU/L2XpC0SNKxOX0TSb/O639E0uek9ouwk/Jn8uO83AOSJuZj/lj+DI4ubOPC/L35az5e/6f8/c3T3yXptryumyTtXpj2pKTJhfftt6YK35Fqx2I9pd/95yXNAyYU9z0iXgLmAe+lP0SEH3U8gAXAezulnUYqBWwODAfOA36Rp30auAhYF1iLFNTXy9NuAj7Uxfa2BlYAWwHTgVsK04YB84FvAiPyNvbO044DFpJONAHbAGNy/gIYU1jPhcCX8usDgdeBM0k/JOsCm5F+MNYl/Wj+HriwsPz/ARcAG+Vl9s3pXwHOL8z3b8CtVfbzG8DPC++PAO7o6hhWWM8yYKfC+23y/qyfj9eSnMcnCmkvA5Py/EfnY/r2PO3rwF/ztA7HDvgd8Mucr53zOq/pNO8lwAbAeOB5YEqeflLbvIW8npP3c6N8nK8ETs/TDicF621zvi/u/DlWOBbX589xy3wO7VCY9k3gZ/n1tnld5+d9mUAq/U0uzLucFFCHkqobZ3X6XN9CKigclz+DUVXyNCl/fkNJ5/SDwEl1HrNTSD+SmwOjgeuA12vs/43Ax/PrnwHHAt/rlPZf+fUG+fNrznnbPR+Dt1X4jhwOLCadW+sBv+10XlwIPA3sls+hi4DzKp1DOW3PvO135G1PA/5OOteHA48Dn8jraiadz1+qsd8fI333Rbqo+XZh2rbFY0b6DVpYOK/+UOG8+EnOx+7Aq8Bbu/osK+TpbcBLpO/1Wvnz2yVP+y3wv3n7bwMeAZoL35PX8me3FvCdnN+zSN/jQ4HngOGFY/88sFfO8zms/E6+CXgR+Ne8rg+Tfg82zNOfJJ/zNb4j1Y7F2aTfwY1I5+39wIOdjkEL8I2exJ/VffR74BwsDyoH+EeAfQrvx5N+EJW/mH8DdqywrnoC/NeBmwrrXQFsl9/vR/rRH1Jhub8BJ1ZIryfAvwwMq5GnPYEnCnl6FRhZYb4m4AVgRH7/R+BTVda5Q/6irp3fXwx8Lr+uegw7rWNo3remTulPAAfkL/3/5bTfFdKWAWvl9L+Sf1zy+2GkH5jNiscuv14BjCvM+11WDfATC9MvA07JrzsEeNIPzqvAFoW0/YD5+fWvgTMK03bu/Dl22ue35enbFs6HbxWmV/rxaipM/yHwk8K8fyxM2w14vsbncB9wQJ3fp9OAmXUesxuADxemHUrtAP/NwrrvJ13oHN4p7d/y66nA1Z2WPx/4fIXvyK/JF175/Y6sGuB/XJj+QVZerFb6/v0CmN5p2wuBPYD9gUc6TbuN2gH+OuCb+fVHSBcIQwufdecAXzyvdgNe7nRejCpMvws4vKvPssK0r1aaBqwDvEEOlDnt08CfC9+TeYVpu+c8bVhIe5mV5/mF5Iup/H6TPP9o0oXP7E7bvx04Or+uJ8BXPBb5GE8pTPsUqwb475EvKPv64Sr61ZSrkrYELs/VM8+TTpohwKbAz0k/rhflqp1vSBrajXUfB7QCRMQjpC/k1DzLlqQv/4oKi29JqtJfHU9GxGuFfIyUdG6uVnsRuApoa7S1JfB0RCzrvJKIWEA6FodJGg28m/QFXEVE3EOqPn2fpA1IVe0z8+S6jmFEvEEK1iM7TbqWVCW/Lyvvt15XSLshItpaC48Dzil8lktIJabOjaLeTLqAW1xIe7TCrj1ZeL2cVEqpZHPSxcQ9hW3/jlTqaJteXP/CKutpMxW4LSLuy+9bgQ9JqvVd77z+4r8Fqu6HpH8vVC8/T7q4qNioT+mWyxW5avVFUi1P53mrbau7x6DtPvxmpAvHR0mfe1va21lZZT4O2LdtH/J+HEGqmeiscz568rm3bfuLnbY9mtQ+YHM6nmNQY78lbQXsQ/7NIF0ob0zt2xmdj+mIQjX9GxHxTKV9qfOzbFPt9+jNpN/KRZ3ysEXh/VOF1/8A/hkRL3RKKx7f9v2JiGdJNQeb50fnY9d5W7VUPBb5d3ozuj43R5JqF/qcA/xqinRp9hjw7ojYqPAYHhHPRGp5/ZWI2JYUTI4iVQNDuiKsZT9gLHBGvj/0JLALK3+oHwWaqvxoP0qqNuvsVVKJdEQh7c2dd6vT+9NIAW73iNiAVKpoa3H+KPAmSdV+wM4n3S8+GvhLRDxdZT5IAf0Y0g/rrfkHmS6OYWd3kX64i9ruw7+TlQH+2kLa7MK8j5JKicXPct2ImNtpnU+SjlPxx2HLGvvWWedj/ATpQmKrwnY3jIhNC9OL6x9bbcWFC8PtCufNN0g/cLXuAXZef5f/FpD0duBHpGrlTSJiI1JVbbV/JPyUVALdKp9LZ9aYt7O6j0F2HSlAfzi/Jv9AL8tpD0bEE3neR4GrOn3u60fEKVXyUbzg68nn3rbtr3Ta9oiIuKTCtqD2frdd/F+dP/e2qv7jayzT+Zgu7xRAq+nOZ1nt9+hJUk1YcZ/Gkn5TV1f7/kjahBT8nyCdz+M6zVvc1svU/l2sKMeAp+n63NwOuLOedfY2B/ieOQf4pqQtob3x1fvz6/fmK90hpPs/r5OqpCBdmb61xnqnkqq1dwB2zY9dSNVO7yH9aC0DviZphFIjrb3zsj8DTpO0i5K3SxqTS/vzgGalxoHvJ1VT1zKSdLX6vNLfrdobG+VahdnAjyVtKGltScUGbBcBk4GPk+5X1zITOAQ4gVQNCnR5DDu7nI4t5Mn5251U5XlTTruNdFwn0zHAnwN8SdI2edsbSzqi80Yi4hXS/cqvKjUO25F0n7BeTwFbShqW1/ca6a99P1BqtChJW2plI8XfAifkz3F9UmmpmimkwLYbK8+bHUmluanVF+P0fA7tQrpA+E0d+7E+6Qd6CTBEqf+Bt9WYfyTwQkS8JGkHUrVpvX4LnCrpLfk8rPm/9Ih4kfSD+hk6tpS/LqcVP/ffARMk/ZukYfk83jNfwFTKxwmStpa0HoXvQ1ci4p+k21bF730L8EmlxmOStL6kQyWNyHkcrtTYbC1Jx5Buz6yicGH3RVZ+7ruSzsvDc81YJR8unFdnUN/nDt37LH8JHKLUmHCoUsPdnfPxuBT4hlJDta1IVfS/qjMPlRwmaQ+lBqptbWieJt3umSDpyHwsjycF4j/n5e4AjsnT9iS1O6rXb4Hp+TdwHOm2Yrt8nuxEuk/f5xzge+bbwDXAXyQtI90r3C1P24LUKG0ZcDcpAP02TzsLOF6p1W6H/9jmL9sRwA8j4snC40FSNffUHBQOIgX9xaRqrg8CRMQFwPdJAfZFVjbeAjiZ1DDqOeADpIuIWr5LqnpbSvpxvLzT9GNI1csPkK7IP942IVfd/yEfh8tqbSRX6d9JCsb/W5hU6xh2dh7px6zY0nge6er8kYh4OW/rNdIXeh3glkIeZgI/Bi7J1Y53UL1680RSqXgJ6YJqJumvWPX4M6k9x9NKreIhNSJ7HJhDCgJ/JgfLiLiUFAiuJd3jvrLGuqcCF0XEfcVzh3Rf/QNVfujfAG4mtSf5M3BmRMyuMF8HEXEb6aJoDqmUND6/ruZUUnB8idRgqd5gAulzuRa4J+e12jlQ9DfSbY7rCmnX5rT2/YuI50htMj7CytLe10nndQf5s/g5qRHj31l58VDvZ/8V4H9zdfyhEXE96Z7t/5CqcP9OCsoREf8gfUc/Qfq+Hkz6PlXyrrxf/93pc7+IVEr91yrLXUA6dx8jXax9ts79qPuzjIiHSAHzi3k/5pAusCF9jyBVa/+F9F3qyd+Pf0W6f/4MqdQ8NefhKVK7jemk37KTgUMioq3a/IukIPw88AWq3E6s4kt5e4uAP7FqYeaDwOWdqvj7jFItg1nvk/QN4E0RcUIfbe/7wN8j4py+2F5huz8gteY9scuZBxBJ2wJ3R8Rg6/NgQFD62+ONwLoxyH5IJd1EahDYkxLzgCHpQtK5/PX+zkubXLMyl9SY7+/9kQd/sa0hlBrXfZjUerlPRMRn+mI7uVo+gHtJtzmOZ83rUGaNJOmDpJqvDYD/j9S50qAK7tY38nmxW5czNpCr6K3XSTqZVA39vxFxSxezD0YbkqpLXyZVC349Iv5cexEriU+RqmTvJ906cpewNmC5it7MzKyEXII3MzMrIQd4MzOzEnKAN+uCpLFKA4TU1RNhN9d9htJAIqWlNNhI517ZbDV053yRNEtSn/yDxQYmB3gbcCQtkPSPHFSfkvQLVe8xr6t1fVjSdV3PWV1ELMq9m1XrZKfXSdpCaRSrVXoBk3SppO/m15J0slKXscuVerCbpcJIW3m+f9HKkbaWSrpD0uclDe+rfapEadTDj1ZI/7SkOYX3XeY/d0BzodKIXy8qjUD2I3UxBnu+AFmRz7eXlEYr+2qneUJpZMYhhbSvSzqvxjpDq45Dv0tOn9XVsTHrKQd4G6jeHxHrk/5msjsVeg3Lwa3H53AjSuY9FRGPkXq/Oq6YrtQF50GkroAhdWJzCqmTkk1JnQN9iTR4UNsyR5E6Pfk1aZCcTUkdHo2he92tNsL5VO5O9bg8ra78S3obqROcx4EJuQvVfUj9oE9eZe2rejxfxK2f5/93FYaNzTanelfJlSwB9lbHscCnkjq0MWs4B3gb0HKgu4LU5WpbteMMSdeTutF9a+4m8udKY5o/lktWQ5XGnz8H2CuXzJ7P6zhP0n9LulzSy8B+kg6WdHsu+T0q6Yy2PEhqyqWutQp5+JrSeN/LJF2l1IVq2/x7SrpBqceyOyVNKUwbL+lvebmrqT5IB6QAd1yntKOBeyJinlJ3qp8gdaRxdUT8IyLeiIjrIuLDeXsi9Wx4ZkT8NA/CQUTcHxGfjIgHKm24zuMxVWkgomckTS9MXzcf4+ck3Uu6QKvmAmCyVo6bTv7cdgZmdiP/ZwDXR8RnImJxnufpiDg7IrrTM1lbN8w3sHL8+DbfJnVRXG//Ia+SusI9Ou/XUFKvch16a5O0t6RblcYqv1Uru53u8nypda51mu9teT0v5M+rOz0J2iDlAG8DmlI//weRRqdrcxxpkJORpG4uzyf1U/820pjm+wMnRMR80rCTN+bS2UaFdRwLzMjruI70n/bjSd36Hgx8vEIJruhYUvembyKNT/2fOb9bkLqs/Dpp7ID/BC5W6vgHUil0LumH+mvU7iP+UmCUpGIJ9DhWdof5buDRiKjVRew2pJLuxTXmqaSe4zE5r/89wFdyYAY4nTTAyFakbmCr7mMOxn+l44XM8azs3rPe/L+3jnnqImlrUun/pk6TLiF1//zhbqzul6ysoTiA1N1u+2A+uUbmT6SamE1JFzN/KpT6q54vdZxrRV8jjQa5Mel4/qgb+2CDlAO8DVS/yyXu60j9in+jMO28iLgn0lCvm5CGmD0lIl7Og0ucRddVqb+PiOsjYkVEvBIRsyJiXn5/F6mP7s6D1xT9IiL+nvsM/y1pcA9II+hdHhGX53VdTep/+yBJY0ml2S9HGilvNtX7Fyev+3/JASIHnnewckCeUXQcnhSlYXWfl/RKLhW3lfieLMxzYZ5nuaTONQRt267neHw11xrcSRpLYJec/q/AjIh4NtLIgD+sto9Ze01FvuXSzMpbEPXmf1SneU7O87wk6addbB9g8zz/i6Qq9Jvp2I89pN4Lv0y6mFmnjnUSETcAmygNYnQ8q/ZVfjDwQERcEBGv5zER7gPeX8f5UvVcq5CV10gjqm2ez/cetUuxwcEB3gaqwyMNnzkuIj6Rg12b4vjL40gDgzyhlWNq/w8rx1OvpsNY3kqjUP1VqYHWC6SSf63q82rjfo8DjlLHMb4nk0Z52xx4rm3gm6yrsc3PB/5VqTHZccCfY+XQu0vpNG55RIzJ+V6HNITn0jzpLYV5js61GbcBFdsf1Hk8emv89kuAtyiN5DWFNHTnnwr7WE/+l3aa58d5nrOpMHBMBY/n820DUq3FP1h5kdEuIi4nDSwyrY51trmANMDJfqRamaJaY5V3db7UOtc6+xzpfLhF0j2q0LDRyscB3gajYveLj5JG8xoVK8fU3iAidqgwb7V1QCoVXwZsGREbku7d1zteedGjwAXRcYzv9SLim6TRyjZWGkKyTc2xzSPiWlLwOoxUYiuWAP8CjJE0scYq7iONFvbBbu5HT45Ht8Zvj4jlpEZ0x5MuYi6MiFfz5Hrz/391zFOXSGOi/xp4f5VZvkQamWxElemdXUBqK3F53teiWmOVd3W+1DrXOu/TkxHxsYjYnDSK238pNUy0EnOAt0EtIp4g3Vv8nqQNJA2RtJWkturkp0hBcO3qawHSvfhnI+IVSZPo3hjvRb8iVa8eoNTQb7jSX6bGRMRCUhXqV5XGHZ9M9SBS9EvgW6SSZXsVbUTcT6qtuFDpb2Tr5oZcexfmCVIL+9MlfUxpnHvl6v7NamyzJ8fjt8AX8rbGAJ+sY5nzSS3jj6BQcu5G/s8A3inp+/neNEoNH7ejm5T+knk06X75KiJiFmko4lrtJ4rzP0K6vTG9wuTLgbdLOlZpPPJ/IzXu+2Md50vVc63CPh1VSH+OdIHbZ3/7tP7hAG9lcDypodu9pB+vi1hZTfkX0g/1k5Jqjcn8CeBMSctI43bXM+b4KvI957bxr5eQSln/j5XftWOBPYBnSY3ROt+TreSXpJLbbyKi89jj/0G6x/39vM7FpAZV/0aqSiYifkO6L/6hnJ9n8v61kO7xV9KT4/FVUlXyI6SLrwvqWGY28ALwWETcWpxQT/4jDce5J6kB2Z0539eTSshfrmP7m+f79S/lvG9CagtQzZfyPHXJ/2x4vEL6UuAQ0kXMUlJV+iGxcvzwqudLHeda0e7AzXn/LgM+nS88rMQ82IyZmVkJuQRvZmZWQg7wZlZ6kr6olV3RFh9X9HfezBrFVfRmZmYl5BK8mZlZCdXbp/KgMGrUqGhqaurvbJiZmfWJuXPnPhMRlbonLleAb2pqYs6cWt1ym5mZlYekqj1FuorezMyshBzgzczMSsgB3szMrIRKdQ/ezKxer732GosXL+aVV17p76yYdWn48OGMGTOGYcPqGRwxcYA3szXS4sWLGTlyJE1NTUirM3CgWd+ICJYuXcrixYsZP3583cu5it7M1kivvPIKm266qYO7DXiS2HTTTbtd29SwAC9pS0l/lTRf0j2SPl1hnimSXpB0R358pTDtQEn3S3pQ0mmNyqeZrbkc3G2wWJ1ztZEl+NeBz0bEdqRhHP9D0vYV5rs2InbNjzMB8pjWPwHeRxob+Zgqy5qZDUqnnnoqZ599dvv7Aw44gBNOOKH9/Wc/+1m+//3v11zH3nvv3eV2mpqaeOaZVUdKnjVrFjfccEM3clx7fQC33347krjyyis7pK+//vpdrreeeax7GhbgI+KJiLgtv14GzAe2qHPxScCDEfFwRLwKXEga99jMrBT23nvv9gC7YsUKnnnmGe6555726TfccAP77LNPzXWsToBus7oBvpaZM2cyefJkZs6c2avrtdXTJ/fgJTUBE4CbK0zeS9Kdkq6QtENO2wJ4tDDPYuq/ODAz632trdDUBEOGpOfW1h6tbp999mkPsPfccw877rgjI0eO5LnnnuOf//wn8+fPZ8KECQB85zvfYffdd2fnnXfm9NNPb19HW6l3xYoVfOITn2CHHXbgkEMO4aCDDuKiiy5qn+9HP/oRu+22GzvttBP33XcfCxYs4JxzzuGss85i11135dprr2XJkiUcccQR7L777uy+++5cf/31ACxdupT999+fCRMmcOKJJ1JtgLKI4KKLLuK8887jqquuqni/eNasWey777584AMfYPvtt+ekk05ixYoV7dOnT5/OLrvswp577slTTz0FwB/+8Af22GMPJkyYwHvf+972dOtawwO8pPWBi4FTIuLFTpNvA8ZFxC7Aj4DftS1WYVUVzypJ0yTNkTRnyZIlvZVtM7OVWlth2jRYuBAi0vO0aT0K8ptvvjlrrbUWixYt4oYbbmCvvfZijz324MYbb2TOnDnsvPPOrL322lx11VU88MAD3HLLLdxxxx3MnTuX2bNnd1jXJZdcwoIFC5g3bx4/+9nPuPHGGztMHzVqFLfddhsf//jH+e53v0tTUxMnnXQSp556KnfccQfvfOc7+fSnP82pp57KrbfeysUXX9x+u+CrX/0qkydP5vbbb+fQQw9l0aJFFffn+uuvZ/z48Wy11VZMmTKFyy+/vOJ8t9xyC9/73veYN28eDz30EJdccgkAL7/8MnvuuSd33nkn++67Lz/96U8BmDx5MjfddBO33347Rx99NN/+9rdX+5ivaRoa4CUNIwX31oi4pPP0iHgxIl7Kry8HhkkaRSqxb1mYdQzweKVtRERLREyMiImjR1fsb9/MrGemT4flyzumLV+e0nugrRTfFuD32muv9vdt99evuuoqrrrqKiZMmMBuu+3GfffdxwMPPNBhPddddx1HHXUUQ4YM4c1vfjP77bdfh+kf/OAHAXjHO97BggULKublmmuu4eSTT2bXXXfl0EMP5cUXX2TZsmXMnj2bD33oQwAcfPDBbLzxxhWXnzlzJkcffTQARx99dNVq+kmTJvHWt76VoUOHcswxx3DdddcBsPbaa3PIIYesks/FixdzwAEHsNNOO/Gd73ynw20Mq61h/4NXavL3c2B+RFRsKSLpzcBTERGSJpEuOJYCzwNbSxoPPAYcDRzbqLyamdVUpdRaNb1Obffh582bx4477siWW27J9773PTbYYAM++tGPAqnq+wtf+AInnnhi1fVUqzZvs8466wAwdOhQXn/99YrzrFixghtvvJF11113lWldteB+4403uPjii7nsssuYMWNG+/+2ly1bxsiRI2uuq+39sGHD2l8X8/nJT36Sz3zmMxx66KHMmjWLM844o2ZebKVGluD3AY4D3l34G9xBkk6SdFKe50jgbkl3Aj8Ejo7kdeBk4EpS47zfRoQv28ysf4wd2730Ou2zzz788Y9/ZJNNNmHo0KFssskmPP/889x4443stddeQGpdf+655/LSSy8B8Nhjj/H00093WM/kyZO5+OKLWbFiBU899RSzZs3qctsjR45k2bJl7e/3339/fvzjH7e/v+OOOwDYd999ac23Iq644gqee+65VdZ1zTXXsMsuu/Doo4+yYMECFi5cyBFHHMHvfve7Vea95ZZbeOSRR1ixYgW/+c1vmDx5cs18vvDCC2yxRWqCdf7553e5X7ZSI1vRXxcRioidC3+DuzwizomIc/I8P46IHSJil4jYMyJuKCx/eUS8PSK2iogZjcqnmVmXZsyAESM6po0YkdJ7YKedduKZZ55hzz337JC24YYbMmrUKCAF3mOPPZa99tqLnXbaiSOPPLJDYAY44ogjGDNmDDvuuCMnnngie+yxBxtuuGHNbb///e/n0ksvbW9k98Mf/rD93v/222/POeecA8Dpp5/O7Nmz2W233bjqqqsYW+GiZubMmXzgAx9YJU+//vWvV5l3r7324rTTTmPHHXdk/PjxqyzX2RlnnMFRRx3FO9/5zvZjYvVRV1U7g8nEiRPD48GbWT3mz5/PdtttV/8Cra3pnvuiRankPmMGNDc3LoPd9NJLL7H++uuzdOlSJk2axPXXX8+b3/zm/s5WB7NmzeK73/0uf/zjH/s7K4NSpXNW0tyImFhpfvdFb2ZWj+bmARXQOzvkkEN4/vnnefXVV/nyl7884IK79T0H+C4M8It2MzOAuu6797cpU6YwZcqU/s7GGsMBvoa2v762/Tum7a+v4CBvZmYDm0eTq6FBf301M9NA5JkAACAASURBVDNrOAf4Ghr011czM7OGc4CvoUF/fTUzM2s4B/gaGvTXVzMzDxfbSU+Hi73lllvYd9992Wabbdh222054YQTWN75HusgMGvWrPYue3vKAb6G5mZoaYFx40BKzy0tbmBnZj3n4WJ7z1NPPcVRRx3Ft771Le6//37mz5/PgQceuEqHQGsaB/guNDfDggWwYkV6dnA3WzP18mixHi62F4eL/clPfsLUqVPbu/eVxJFHHslmm23Gs88+y+GHH87OO+/MnnvuyV133QWkHvKmTp3K/vvvT1NTE5dccgmf+9zn2GmnnTjwwAN57bXXgFRj8fnPf55JkyYxadIkHnzwQQAWLlzIe97zHnbeeWfe8573tI+y9+EPf7jDsW/7jGbNmsWUKVM48sgj2XbbbWlubm4/ln/+85/ZdtttmTx5cvvoer3BAd7MrAsNGC3Ww8X24nCxd999N+94xzsqbu/0009nwoQJ3HXXXXzjG9/g+OOPb5/20EMP8ac//Ynf//73fOhDH2K//fZj3rx5rLvuuvzpT39qn2+DDTbglltu4eSTT+aUU04B4OSTT+b444/nrrvuorm5mU996lMVt190++23c/bZZ3Pvvffy8MMPc/311/PKK6/wsY99jD/84Q9ce+21PPnkk12up14O8GZmXWjUX2Y9XGzjh4u97rrrOO644wB497vfzdKlS3nhhRcAeN/73sewYcPYaaedeOONNzjwwAOBNB5A8Tgdc8wx7c9tF0833ngjxx6bBjk97rjj2vejlkmTJjFmzBiGDBnCrrvuyoIFC7jvvvsYP348W2+9NZLaj3VvcEc3ZmZdaNRfZj1cbO8MF7vDDjswd+5cDjvssFWmVTo2bdtpOy5DhgzpsP0hQ4Z0OE7FPFc7Fm3pa621Vvtth4jg1VdfbZ+nbXud97Gr47u6XII3M+tCo/4y6+Fie2e42JNPPpnzzz+fm2++uT3tV7/6FU8++WSH/M+aNYtRo0axwQYb1NxmZ7/5zW/an9s+l7333psLL7wQgNbW1vb9aGpqYu7cuQD8/ve/b7+XX822227LI488wkMPPQTQqw0UHeDNzLrQqL/MerjY3hkudrPNNuPCCy/kP//zP9lmm23YbrvtuPbaa9lggw0444wz2vfrtNNOW60x5f/5z3+yxx578IMf/ICzzjoLgB/+8If84he/YOedd+aCCy7gBz/4AQAf+9jH+Nvf/sakSZO4+eabWW+99Wque/jw4bS0tHDwwQczefJkxo0b1+38VePhYs1sjdTd4WIH+sBTHi62MZqampgzZ86AGIvew8WamTXAAB8t1sPF2ioc4M3MSsDDxTZGtX8dDAa+B29mZlZCDvBmtsYqUxskK7fVOVcd4M1sjTR8+HCWLl3qIG8DXlu/AsOHD+/Wcr4Hb2ZrpDFjxrB48WKWLFnS31kx69Lw4cMZM2ZMt5ZxgDezNdKwYcMYP358f2fDrGFcRW9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg0L8JK2lPRXSfMl3SPp0xXmaZZ0V37cIGmXwrQFkuZJukPSnEbl08zMrIzWauC6Xwc+GxG3SRoJzJV0dUTcW5jnEeBdEfGcpPcBLcAehen7RcQzDcyjmZlZKTUswEfEE8AT+fUySfOBLYB7C/PcUFjkJmBMo/JjZma2JumTe/CSmoAJwM01Zvt34IrC+wCukjRX0rTG5c7MzKx8GllFD4Ck9YGLgVMi4sUq8+xHCvCTC8n7RMTjkt4EXC3pvoiYXWHZacA0gLFjx/Z6/s3MzAajhpbgJQ0jBffWiLikyjw7Az8DDouIpW3pEfF4fn4auBSYVGn5iGiJiIkRMXH06NG9vQtmZmaDUiNb0Qv4OTA/Ir5fZZ6xwCXAcRHx90L6erlhHpLWA/YH7m5UXs3MzMqmkVX0+wDHAfMk3ZHTvgiMBYiIc4CvAJsC/5WuB3g9IiYCmwGX5rS1gF9HxJ8bmFczM7NSaWQr+usAdTHPCcAJFdIfBnZZdQkzMzOrh3uyMzMzKyEHeDMzsxJygDczMyshB3gzM7MScoA360JrKzQ1wZAh6bm1tb9zZGbWtYb3ZGc2mLW2wrRpsHx5er9wYXoP0Nzcf/kyM+uKS/BWkUutyfTpK4N7m+XLU7qZ2UDmErytwqXWlRYt6l66mdlA4RK8rcKl1pWqjV/kcY3MbKBzgLdVuNS60owZMGJEx7QRI1K6mdlA5gBvq3CpdaXmZmhpgXHjQErPLS1r3q0KMxt8HOBtFS61dtTcDAsWwIoV6dnB3cwGAwd4W4VLrWZmg59b0VtFzc0O6GZmg5lL8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXUsAAvaUtJf5U0X9I9kj5dYR5J+qGkByXdJWm3wrSpkh7Ij6mNyqeZmVkZrdXAdb8OfDYibpM0Epgr6eqIuLcwz/uArfNjD+C/gT0kbQKcDkwEIi97WUQ818D8mpmZlUbDSvAR8URE3JZfLwPmA1t0mu0w4JeR3ARsJOktwAHA1RHxbA7qVwMHNiqvZmZmZdMn9+AlNQETgJs7TdoCeLTwfnFOq5ZuZmZmdWh4gJe0PnAxcEpEvNh5coVFokZ6pfVPkzRH0pwlS5b0LLNmZmYl0dAAL2kYKbi3RsQlFWZZDGxZeD8GeLxG+ioioiUiJkbExNGjR/dOxs3MzAa5RraiF/BzYH5EfL/KbJcBx+fW9HsCL0TEE8CVwP6SNpa0MbB/TjMzM7M6NLIV/T7AccA8SXfktC8CYwEi4hzgcuAg4EFgOfCRPO1ZSV8Dbs3LnRkRzzYwr2ZmZqXSsAAfEddR+V56cZ4A/qPKtHOBcxuQNTMzs9JzT3ZmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZtZwra3Q1ARDhqTn1tb+zpFZ+TVsPHgzM0jBfNo0WL48vV+4ML0HaG7uv3yZlZ1L8GbWUNOnrwzubZYvT+lm1jgO8GbWUIsWdS/dzHqHA7yZNdTYsd1LN7Pe4QBvZg01YwaMGNExbcSIlG5mjeMAb2YN1dwMLS0wbhxI6bmlxQ3szBrNrejNrOGamx3QzfqaS/BmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ADfFY9zaWZmg5A7uqnF41yamdkg5RJ8LR7n0szMBikH+Fo8zqWZmQ1SDvC1eJxLMzMbpBzga/E4l2ZmNkg5wNficS7NzGyQciv6rnicSzMzG4RcgjczMyshB3gzM7MScoA3MzMrIQd4MzOzEnKAN7MueUgGs8HHrejNrCYPyWA2OCkias8gjQY+D2wPDG9Lj4h3NzZr3Tdx4sSYM2dOf2fDrFSamlJQ72zcOFiwoK9zY2ZFkuZGxMRK0+qpom8F5gPjga8CC4Bbey13ZjageUgGs8GpngC/aUT8HHgtIv4WER8F9mxwvsxsgPCQDGaDUz0B/rX8/ISkgyVNAMY0ME9mNoB4SAazwameRnZfl7Qh8FngR8AGwKkNzZWZDRhtDemmT0/V8mPHpuDuBnZmA1uXjewGEzeyMzOzNUmtRnZdluBzK/qPAU3F+fO9eDMzMxuA6qmi/z1wLXAN8EZjs2NmZma9oZ4APyIiPt/dFUs6FzgEeDoidqww/f8BbXfx1gK2A0ZHxLOSFgDLSBcUr1erfjAzM7PK6mlF/0dJB63Gus8DDqw2MSK+ExG7RsSuwBeAv0XEs4VZ9svTHdzNzMy6qWoJXtIyIAABX5T0Kiv/MhcRsUGtFUfEbElNdebjGGBmnfOamZlZF6oG+IgY2RcZkDSCVNI/ubh54CpJAfxPRLTUWH4aMA1grHveMDMzA+ocbEbSB4HJpMB7bUT8rhfz8H7g+k7V8/tExOOS3gRcLem+iJhdaeEc/Fsg/U2uF/NlZmY2aHV5D17SfwEnAfOAu4GTJP2kF/NwNJ2q5yPi8fz8NHApMKkXt2dmZlZ69ZTg3wXsGLlHHEnnk4J9j+Ue8t4FfKiQth4wJCKW5df7A2f2xvbMzMzWFPUE+PuBsUDbgJFbAnd1tZCkmcAUYJSkxcDpwDCAiDgnz/YB4KqIeLmw6GbApZLa8vfriPhzHfk0MzOzrJ4AvykwX9It+f3uwI2SLgOIiEMrLRQRx3S14og4j/R3umLaw8AudeTLzMzMqqgnwH+l4bkwMzOzXtVlgI+Iv/VFRszMzKz31NPRzSqTqKOjGzMzM+s//d7RjZmZmfW+evqibydpPUnNkv7UqAyZmZlZz9XT0c3akg6X9FvgCeC9wDldLGZmZmb9qNY9+H8hDQJzAPBX4AJgUkR8pI/yZmZmZqupViv6K4FrgckR8QiApB/0Sa7MzMysR2oF+HeQ+om/RtLDwIXA0D7JlZmZmfVI1XvwEXF7RHw+IrYCzgAmAGtLuiIP0WpmZmYDVF2t6CPi+og4GdgCOBvYq6G5MjMzsx6pazz4NhGxgnRv/srGZMfMzMx6Q7f+B29mZmaDgwO8mZlZCdX6H/wmtRaMiGd7PztmZmbWG2rdg59LGmxGwFjgufx6I2ARML7huTMzM7PVUutvcuMj4q2kBnXvj4hREbEpcAhwSV9l0PpJays0NcGQIem5tbW/c2RmZt1Qzz343SPi8rY3EXEF8K7GZcn6XWsrTJsGCxdCRHqeNs1B3sxsEKknwD8j6UuSmiSNkzQdWNrojFk/mj4dli/vmLZ8eUo3M7NBoZ4AfwwwGrg0P0bnNCurRYu6l25mZgNOlx3d5Nbyn5a0fkS81Ad5sv42dmyqlq+UbmZmg0I948HvLele4N78fhdJ/9XwnFn/mTEDRozomDZiREo3M7NBoZ4q+rNIY8IvBYiIO4F9G5kp62fNzdDSAuPGgZSeW1pSupmZDQp19UUfEY9KKia90Zjs2IDR3OyAbmY2iNUT4B+VtDcQktYGPgXMb2y2zMzMrCfqqaI/CfgP0lCxi4FdgU80MlNmZmbWM/WU4LeJiA51tZL2Aa5vTJbMzMysp+opwf+ozjQzMzMbIKoGeEl7SfosMFrSZwqPM4ChfZZDs/7mfvnNbBCqVUW/NrB+nmdkIf1F4MhGZspswGjrl7+t6962fvnB/zIwswFNEVF7BmlcRFTo1mzgmThxYsyZM6e/s2Fl0tRUuVe/ceNgwYK+zo2ZWQeS5kbExErT6mlkt46kFqCpOH9EvLt3smc2gLlffjMbpOoJ8P8LnAP8DHdwY2sa98tvZoNUPQH+9Yj474bnxGwgmjGj4z14cL/8ZjYo1GpFv4mkTYA/SPqEpLe0peV0s/Jzv/xmNkhVbWQn6REgAFWYHBHx1kZmbHW4kZ2Zma1JVquRXUSMb1yWzMzMrJG6vAcv6YMVkl8A5kXE072fJTMzM+upehrZ/TuwF/DX/H4KcBPwdklnRsQFDcqbmZmZraZ6+qJfAWwXEUdExBHA9sA/gT2Azzcyc2ZmVm7uCbpx6inBN0XEU4X3TwNvj4hnJb3WoHyZmVnJuSfoxqqnBH+tpD9KmippKvB7YLak9YDnG5s9MzMrq+nTO3YxAen99On9k5+yqacE/x/AEcA+pL/M/RK4ONL/6/ZrYN7MzKzE3BN0Y3UZ4HMgvyg/zMzMeoV7gm6sWj3ZXZefl0l6sfBYJunFvsuimZmV0YwZqefnIvcE3XtqdXQzOT+PrDaPmZnZ6mprSDd9eqqWHzs2BXc3sOsd9dyDR9JkYOuI+IWkUcDIiHiksVkzM7Oya252QG+ULlvRSzqd9H/3L+SktYFfNTJTZmZm1jP1/E3uA8ChwMsAEfE44Gp7MzOzAayeAP9qbkkfAPn/72ZmZjaA1RPgfyvpf4CNJH0MuAb4aVcLSTpX0tOS7q4yfYqkFyTdkR9fKUw7UNL9kh6UdFq9O2NmZmZJlwE+Ir5L+g/8xcA2wFci4kd1rPs84MAu5rk2InbNjzMBJA0FfgK8j9Tv/TGStq9je2Zmq8X9oVsZVW1FL+kU4Hrg9oi4Gri6OyuOiNmSmlYjT5OAByPi4ZyPC4HDgHtXY11mZjW5P3Qrq1ol+DHAD4CnJc2S9A1JB0vapBe3v5ekOyVdIWmHnLYF8GhhnsU5zcys17k/dCurWh3d/CeApLWBicDewEeBn0p6PiJ6Wm1+GzAuIl6SdBDwO2BrUn/3q2Sn2kokTQOmAYx1/4Zm1k3uD93Kqp5GdusCGwAb5sfjwM093XBEvBgRL+XXlwPDcic6i4EtC7OOydustp6WiJgYERNHjx7d02yZ2RqmWrnA5QUb7Gr1Rd8i6XrgN8BewA3AUTmYfqSnG5b0ZknKryflvCwFbgW2ljQ+1x4cDVzW0+2ZmVXi/tCtrGp1VTsWWAd4AHiMVLKue/x3STOBKcAoSYuB04FhABFxDnAk8HFJrwP/AI7O/7d/XdLJwJXAUODciLinm/tlZlYX94duZaUUU6tMTCXsHUj33/cGdgSeBW6MiNP7JIfdMHHixJgzZ05/Z8PMzKxPSJobERMrTas52EwuUd8t6Xnghfw4hPRXtgEX4M3MzCyp9T/4T5FK7fsAr5H+E38jcC4wr09yZ2ZmZqulVgm+idSD3akR8UTfZMfMzMx6Q63/wX+mLzNiZmZmvaee/8GbmZnZIOMAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBm1rXWVmhqgiFD0nNra3/nyMy6UHO4WDMzWlth2jRYvjy9X7gwvQdobu6/fJlZTS7Bm1lt06evDO5tli9P6WY2YDnAm1ltixZ1L93MBgQHeDOrbezY7qWb2YDgAG9mtc2YASNGdEwbMSKlm9mA5QBvZrU1N0NLC4wbB1J6bmlxAzuzAc6t6M2sa83NDuhmg4xL8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZtZ47sverM+5Fb2ZNZb7sjfrFy7Bm1ljuS97s37hAG9mjeW+7M36hQO8mTWW+7I36xcO8GbWWO7L3qxfOMCbWWO5L3uzfuFW9GbWeO7L3qzPuQRvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDQvwks6V9LSku6tMb5Z0V37cIGmXwrQFkuZJukPSnEbl0czMrKwaWYI/DziwxvRHgHdFxM7A14CWTtP3i4hdI2Jig/JnZmZWWms1asURMVtSU43pNxTe3gSMaVRezMzM1jQD5R78vwNXFN4HcJWkuZKm1VpQ0jRJcyTNWbJkSUMzaWZmNlg0rARfL0n7kQL85ELyPhHxuKQ3AVdLui8iZldaPiJayNX7EydOjIZn2MzMbBDo1xK8pJ2BnwGHRcTStvSIeDw/Pw1cCkzqnxyamZmtntZWaGqCIUPSc2tr326/3wK8pLHAJcBxEfH3Qvp6kka2vQb2Byq2xDczMxuIWlth2jRYuBAi0vO0aX0b5BXRmFptSTOBKcAo4CngdGAYQEScI+lnwBHAwrzI6xExUdJbSaV2SLcQfh0RM+rZ5sSJE2POHP+rzszM+ldTUwrqnY0bBwsW9N52JM2t9m+zhgX4/uAAb2ZmA8GQIank3pkEK1b03nZqBfiB0orezMysNMaO7V56IzjAm5mZ9bIZM2DEiI5pI0ak9L7iAG9mZtbLmpuhpSXdc5fSc0tLSu8r/f4/eDMzszJqbu7bgN6ZS/BmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2bW2prG9xwyJD335aDdZg3irmrNbM3W2grTpsHy5en9woXpPfRvP6NmPeQSvJmt2aZPXxnc2yxfntLNBjEHeDNbsy1a1L10s0HCAd7M1mxjx3Yv3WyQcIA3szXbjBkwYkTHtBEjUrrZIOYAb2ZrtuZmaGmBceNASs8tLW5gZ4OeW9GbmTU3O6Bb6bgEb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmb9p7UVmppgyJD03Nra3zkqDfdFb2Zm/aO1FaZNg+XL0/uFC9N78NgAvcAleDMz6x/Tp68M7m2WL0/p1mMO8GZm1j8WLepeunWLA7yZmfWPsWO7l27d4gBvZmb9Y8YMGDGiY9qIESndeswB3szM+kdzM7S0wLhxIKXnlhY3sOslbkVvZmb9p7nZAb1BXII3MzMrIQd4MzOzEnKANzMzKyEHeDMzsxJygDczMyshB3gzM7MScoA3MzMrIQd4MzOzEnKANzMzKyEHeDMzsxJygDczMyuhhgZ4SedKelrS3VWmS9IPJT0o6S5JuxWmTZX0QH5MbWQ+zczMyqbRJfjzgANrTH8fsHV+TAP+G0DSJsDpwB7AJOB0SRs3NKdmZmYl0tAAHxGzgWdrzHIY8MtIbgI2kvQW4ADg6oh4NiKeA66m9oWCmZmZFfT3PfgtgEcL7xfntGrpZmZmVof+DvCqkBY10lddgTRN0hxJc5YsWdKrmTMzMxus+jvALwa2LLwfAzxeI30VEdESERMjYuLo0aMbllEzM7PBpL8D/GXA8bk1/Z7ACxHxBHAlsL+kjXPjuv1zmpmZmdVhrUauXNJMYAowStJiUsv4YQARcQ5wOXAQ8CCwHPhInvaspK8Bt+ZVnRkRtRrrmZmZWUFDA3xEHNPF9AD+o8q0c4FzG5EvMzOzsuvvKnozMzNrAAd4MzOzEnKANzMzKyEHeDMzsxJygDczMyshB3gzM7MScoA3MzMrIQd4MzOzEnKANzMzKyEHeDMzsxJygDczMyshB3gzM7MScoA3MzMrIQd4MzOzEnKANzMzKyEHeDMzsxJygDczMyshB3gzM7MScoA3MzMrIQd4MzOzEnKANzMzKyEHeDMzsxJygDczMyshB3gzM7MScoA3MzMrIQd4MzOzEnKANzMzKyEHeDMzsxJygDczMyshB3gzM7MScoA3MzMrIQd4MzOzEnKANzMzKyEHeDMzsxJygDczMyshB3gzM7MScoA3MzNrhNZWaGqCIUPSc2trn25+rT7dmpmZ2ZqgtRWmTYPly9P7hQvTe4Dm5j7JgkvwZmZmvW369JXBvc3y5Sm9jzjAm5mZ9bZFi7qX3gAO8GZmZr1t7NjupTeAA7yZmVlvmzEDRozomDZiRErvIw7wZmZmva25GVpaYNw4kNJzS0ufNbADt6I3MzNrjObmPg3onbkEb2ZmVkIO8GZmZiXkAG9mZlZCDvBmZmYl5ABvZmZWQg7wZmZmJeQAb2ZmVkINDfCSDpR0v6QHJZ1WYfpZku7Ij79Ler4w7Y3CtMsamU8zM7OyaVhHN5KGAj8B/gVYDNwq6bKIuLdtnog4tTD/J4EJhVX8IyJ2bVT+zMzMyqyRJfhJwIMR8XBEvApcCBxWY/5jgJkNzI+Zmdkao5EBfgvg0cL7xTltFZLGAeOBvxSSh0uaI+kmSYdX24ikaXm+OUuWLOmNfJuZmQ16jQzwqpAWVeY9GrgoIt4opI2NiInAscDZkraqtGBEtETExIiYOHr06J7l2MzMrCQaGeAXA1sW3o8BHq8y79F0qp6PiMfz88PALDrenzczM7MaGhngbwW2ljRe0tqkIL5Ka3hJ2wAbAzcW0jaWtE5+PQrYB7i387JmZmZWWcNa0UfE65JOBq4EhgLnRsQ9ks4E5kREW7A/BrgwIorV99sB/yNpBeki5JvF1vdmZmZWmzrG1cFt4sSJMWfOnP7OhpmZWZ+QNDe3V1uFe7IzMzMroVKV4CUtARb2dz562Sjgmf7ORB9b0/bZ+1tua9r+wpq3z/25v+MiouJfyEoV4MtI0pxq1S9ltabts/e33Na0/YU1b58H6v66it7MzKyEHODNzMxKyAF+4Gvp7wz0gzVtn72/5bam7S+sefs8IPfX9+DNzMxKyCV4MzOzEnKA70eSTpV0j6S7Jc2UNFzSuyXdltPOl1Sxt0FJUyU9kB9T+zrvq6OH+/uGpDvyY5UujwciSZ/O+3WPpFNy2iaSrs6f29WSNq6y7KD7fKHH+1yWz/io/H6FpKotqyUdKOl+SQ9KOq3vcr36eri/CyTNy5/voOiRrMr+fkfSfZLuknSppI2qLNv/n29E+NEPD9LQuY8A6+b3vwU+Shpi9+057Uzg3yssuwnwcH7eOL/euL/3qVH7m6e91N/70M393RG4GxhB6hL6GmBr4NvAaXme04BvleHz7ek+l+wz3g7YhjRI1sQqyw4FHgLeCqwN3Als39/7fC9qSQAABsJJREFU1Kj9zcsvAEb19370wv7uD6yV5/lWle/wgPh8XYLvX2sB6+ZS6wjgZeCfEfH3PP1q4IgKyx0AXB0Rz0bEc3m+A/siwz20uvs7GG0H3BQRyyPideBvwAeAw4Dz8zznA4dXWHawfr492efBqOL+RsT8iLi/i2UnAQ9GxMMR8SpwIek4DWQ92d/BqNr+XpXfA9xEGim1swHx+TrA95OIeAz4LrAIeAJ4gVSqHVao5jqSjkPuttmCVPJtszinDVg93F+A4ZLmSLpJ0mAIEHcD+0raVNII4CDSvm0WEU8A5Oc3VVh20H2+WU/2GcrzGddjMH7GPdlfgACukjRX0rSG5LB31bO/HwWuqLDsgPh8///27i3EqiqO4/j3Z5N2oSwrsxAvSVfsouUFJYM0u1BGDRVSEFYvSUkRVCSRUj1EkQRhWloPIRFCZUXk9UELu1GOM6VdLBEzNRjsqqX172GtYY7TmXGaM5czu98HDrPPPnvts/6zhlln7b3O+ndZNjlrW74PeS0wHNgDLAVuJqXVnaeULncFcKBc8TL7qvrrEBXGCzAkInZIOg1YI6k+IrZ0Q9U7JCI2SXqCNPr+lXSJrrXYWup17QsVxwxu46pu405o34m5fQcCKyVtjoi1XVHXznCoeCXNzs+XlCleFe3rEXzPmQJ8FxE/RsR+4DVgQkSsj4iLI2IssBb4ukzZ7Rz8SXIwsKPLa1yZSuIlInbkn9+S7vWN6p5qd1xELI6I0RExCWgkxbZL0ikA+efuMkV7Y/sCFcVcpDZuj17ZxhXEW9q+u4HXSZexq1pr8eaJr1cDN0e+6d5CVbSvO/iesw0YL+koSQImA5vyp1vyiPYBYEGZssuBqZKOzyPjqXlfNetwvDnOfnn7RGAi8EW31byDSmIbAlwPvAK8CTTNir8VWFamaG9sX6DjMResjdvjY+B0ScMl9SVdyar6bw50NF5JR0s6pmmb9Dfd0FX17Czl4pV0Bel/1bSI+L2VotXRvt09q8+Pg2ZazgU2k/7QXwb6AU8Cm4AvgXtKjr0IWFTy/Dbgm/yY0dOxdGW8wASgnnSJrJ5WZtpX2wNYR+qk6oDJed8JwGrSSGA1MKAo7VtJzAVr4+tII7g/gF3A8rz/VOCdkrJXAV+RZlvP7ulYujJe0mzyuvz4vJfH+w3p/vqG/FhQre3rlezMzMwKyJfozczMCsgdvJmZWQG5gzczMysgd/BmZmYF5A7ezMysgNzBm/VSeQnNpuxrOyV9X/K87384z22SBrXxel9JjZIe7Zyam1l38NfkzApA0hxSNranOlD2PeCuiNjQyuvTgPuBgRFxRkUVbbseNdGcxMPMKuQRvFkBKeWT/yiP5udL6iOpRtLLOSd3g6RZkm4CLgBebWPkPx14mrTk7JiS9xgnab2kOkkf5lUKayTNy+ffKGlmPna7ct5sSeMlrcrbj0laKGkl8JKkEZLWSfosJyUZV/J+D+W610l6XNKZkj4qef3s0udm/3dONmNWMJJGklYXmxARByQ9T1oqcwspH/e5+bjjImKPpLtpZQSflxW9BJgBDCJ19h9LOoKUArM2Ij6V1J+0ktlM0ope50fEX5IGtKPKo4BJEbFPKWvXZXn7LFJ62XGSrgGuBMZGxF5JAyKiUdI+SSMjoiHX8aWO/t7MisYjeLPimQKMAT6RtIHUQY8gLbF5pqRnJF1OStl7KNNIuen3kTIA1krqQ8qVvS0iPgWIiJ8i4q/83gvyNhHR2I73WJbPD2n54sWSGkgfIM4pienFiNjb4ryLgRmSaoAbaP9a8GaF5xG8WfGI1Bk+/K8XpPNII+FZQC1wqLzc00kj6K35+UBgEvAz5dNfqpX9B2geUBzR4rXfSrbvI63zfQtwOClNZ1vnXQo8BLwPrI+IPW3EYva/4hG8WfGsAm7MWdmaZtsPkXQSaWLtUuARYHQ+/hfgmJYnyZnsxgGDI2JYRAwjfTCYTkoYMlTS6HzssZIOA1YAd+ZtSi7RbwUuzNu1bdS9P/BDpNm/t9KcV3sFcLukI0vPGymb1xrgWXx53uwg7uDNCiYi6kmZ+1ZJ2kjqHE8m5ademy/bv0Aa+ULqGBeVmWRXS7o8v79k3xuk+/t/kzr65yTV5ffoBywEdgIb8/4bc7k5wHxJ64A/26j+s8Adkj4AhpLu6xMRbwPv0nzb4d6SMkuA/aRMdWaW+WtyZtarSXoQ6BcRc3u6LmbVxPfgzazXkvQW6crEpT1dF7Nq4xG8mZlZAfkevJmZWQG5gzczMysgd/BmZmYF5A7ezMysgNzBm5mZFZA7eDMzswL6B7SP2vEa9HAUAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 576x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "plt.rcParams['figure.figsize'] = [8,8]\n",
    "\n",
    "x = []\n",
    "y1, y2 = [], []\n",
    "for modelname, accuracy in accuracies5.items():\n",
    "    x.append(accuracy)\n",
    "    summary = [d[\"summary\"] for d in data if d[\"name\"] == modelname]\n",
    "    y1.append(summary[0][\"alpha_weighted\"])\n",
    "    y2.append(summary[0][\"alpha_weighted_compound\"])\n",
    "plt.scatter(x,y1,label=\"Weighted Alpha\", color='r')\n",
    "plt.scatter(x,y2,label=\"Weighted Alpha Compound\", color='b')\n",
    "\n",
    "plt.legend()\n",
    "plt.title(r\"Test Accuracy vs (Weighted Alpha and Weighted Alpha compound)\"+\"\\nPretrained VGG and VGG_BN Models\")\n",
    "plt.xlabel(r\"Test Accuracy\")\n",
    "plt.ylabel(r\"Weighted Alpha\");"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'0.2.1'"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": []
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 4. Detailed Analysis of VGG\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "ww.__version__\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 117,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "import logging\n",
    "logging.basicConfig(level=logging.INFO)\n",
    "logger = logging.getLogger(__name__)\n",
    "logger.setLevel(logging.DEBUG)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 125,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:__main__:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:__main__:Analyzing model\n",
      "DEBUG:__main__:Layer 1: VGG(\n",
      "  (features): Sequential(\n",
      "    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (1): ReLU(inplace)\n",
      "    (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "    (3): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (4): ReLU(inplace)\n",
      "    (5): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "    (6): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (7): ReLU(inplace)\n",
      "    (8): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (9): ReLU(inplace)\n",
      "    (10): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "    (11): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (12): ReLU(inplace)\n",
      "    (13): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (14): ReLU(inplace)\n",
      "    (15): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "    (16): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (17): ReLU(inplace)\n",
      "    (18): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (19): ReLU(inplace)\n",
      "    (20): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  )\n",
      "  (classifier): Sequential(\n",
      "    (0): Linear(in_features=25088, out_features=4096, bias=True)\n",
      "    (1): ReLU(inplace)\n",
      "    (2): Dropout(p=0.5)\n",
      "    (3): Linear(in_features=4096, out_features=4096, bias=True)\n",
      "    (4): ReLU(inplace)\n",
      "    (5): Dropout(p=0.5)\n",
      "    (6): Linear(in_features=4096, out_features=1000, bias=True)\n",
      "  )\n",
      ")\n",
      "DEBUG:__main__:Layer 1: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 2: Sequential(\n",
      "  (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (1): ReLU(inplace)\n",
      "  (2): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (3): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (4): ReLU(inplace)\n",
      "  (5): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (6): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (7): ReLU(inplace)\n",
      "  (8): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (9): ReLU(inplace)\n",
      "  (10): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (11): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (12): ReLU(inplace)\n",
      "  (13): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (14): ReLU(inplace)\n",
      "  (15): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (16): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (17): ReLU(inplace)\n",
      "  (18): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (19): ReLU(inplace)\n",
      "  (20): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      ")\n",
      "DEBUG:__main__:Layer 2: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 3: Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 64x3 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 3: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 2/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 3/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 4/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 5/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 6/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 7/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 8/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 9/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:Layer 4: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 4: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 5: MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "DEBUG:__main__:Layer 5: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 6: Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 128x64 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 6: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,128): Alpha: 1.6501384162432091, Alpha Weighted: 1.07664760445317, D: 0.1694854835629266\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,128): Lognorm: 0.7435316443443298\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,128): Alpha: 1.572974868785374, Alpha Weighted: 1.166384992154907, D: 0.17973868667331738\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,128): Lognorm: 0.7791202068328857\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,128): Alpha: 1.4121040631114676, Alpha Weighted: 0.8954582243928534, D: 0.19428105159218895\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,128): Lognorm: 0.7536869049072266\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,128): Alpha: 1.434591216620384, Alpha Weighted: 0.9448701338353103, D: 0.1813393068641459\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,128): Lognorm: 0.7681989669799805\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,128): Alpha: 1.6184991505482107, Alpha Weighted: 1.5610023153628745, D: 0.1623258197962626\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,128): Lognorm: 0.8472697138786316\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,128): Alpha: 1.3854856636971848, Alpha Weighted: 1.0132790756883576, D: 0.1890513411313855\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,128): Lognorm: 0.7852299809455872\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,128): Alpha: 1.5692041854056904, Alpha Weighted: 0.9040592047004868, D: 0.1819836653967607\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,128): Lognorm: 0.7200042009353638\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,128): Alpha: 1.446611999098628, Alpha Weighted: 1.0071459510259362, D: 0.18158004443579462\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,128): Lognorm: 0.7804905772209167\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,128): Alpha: 1.6136188154479758, Alpha Weighted: 1.0025022923615543, D: 0.17297153107921293\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,128): Lognorm: 0.7450646758079529\n",
      "DEBUG:__main__:Layer 7: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 7: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 8: MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "DEBUG:__main__:Layer 8: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 9: Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 256x128 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 9: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,256): Alpha: 2.682211513725723, Alpha Weighted: 1.1110503676258303, D: 0.15839147318472652\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,256): Lognorm: 0.7695422768592834\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,256): Alpha: 1.6245754806972994, Alpha Weighted: 0.9849721678153283, D: 0.15038340682532747\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,256): Lognorm: 0.8349188566207886\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,256): Alpha: 1.7481680268299262, Alpha Weighted: 0.7086751866266146, D: 0.15976205513386332\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,256): Lognorm: 0.7710186839103699\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,256): Analyzing ...\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "DEBUG:__main__:    Weight matrix 4/9 (128,256): Alpha: 2.576077238558769, Alpha Weighted: 1.6520572774911981, D: 0.133132943439495\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,256): Lognorm: 0.8396487832069397\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,256): Alpha: 2.5337249999767604, Alpha Weighted: 2.2415200586217683, D: 0.12604047603275326\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,256): Lognorm: 0.9509152770042419\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,256): Alpha: 1.615083583562476, Alpha Weighted: 1.0727975403406367, D: 0.12912543499597612\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,256): Lognorm: 0.8459222912788391\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,256): Alpha: 1.664928709624092, Alpha Weighted: 0.6363944562361387, D: 0.1549378605089503\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,256): Lognorm: 0.7692914009094238\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,256): Alpha: 1.6911656531221961, Alpha Weighted: 1.0206931799586925, D: 0.13113812678865122\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,256): Lognorm: 0.8332030773162842\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,256): Alpha: 2.0503673036718295, Alpha Weighted: 0.8569681941704108, D: 0.14425798012094093\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,256): Lognorm: 0.7771602272987366\n",
      "DEBUG:__main__:Layer 10: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 10: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 11: Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 256x256 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 11: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,256): Alpha: 2.210368775449256, Alpha Weighted: 0.8721198307613643, D: 0.12885903121779885\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,256): Lognorm: 0.8579453229904175\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,256): Alpha: 2.025263486777846, Alpha Weighted: 1.025637347986162, D: 0.12252831337042036\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,256): Lognorm: 0.8855762481689453\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,256): Alpha: 2.048448118154022, Alpha Weighted: 0.8159200676805229, D: 0.1220413714161323\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,256): Lognorm: 0.8663251996040344\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,256): Alpha: 2.658269741880663, Alpha Weighted: 1.3211440029743788, D: 0.12169937788427815\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,256): Lognorm: 0.8791447281837463\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,256): Alpha: 1.8004535746567867, Alpha Weighted: 1.3681391905858296, D: 0.11590087071900723\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,256): Lognorm: 0.9453065395355225\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,256): Alpha: 2.042517105751384, Alpha Weighted: 1.1107899450173098, D: 0.12374522941566979\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,256): Lognorm: 0.886420488357544\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,256): Alpha: 2.6040952619496194, Alpha Weighted: 1.038414125171048, D: 0.12428432704444348\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,256): Lognorm: 0.8589674830436707\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,256): Alpha: 1.9744271620481515, Alpha Weighted: 0.9784463993718218, D: 0.12141239669808901\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,256): Lognorm: 0.8895105123519897\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,256): Alpha: 2.0731955814562264, Alpha Weighted: 0.8033697506343951, D: 0.13215320344674464\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,256): Lognorm: 0.8611611127853394\n",
      "DEBUG:__main__:Layer 12: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 12: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 13: MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "DEBUG:__main__:Layer 13: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 14: Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 512x256 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 14: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,512): Alpha: 2.6517226518393358, Alpha Weighted: 1.2570193700750756, D: 0.10811389451521652\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,512): Lognorm: 0.9264131784439087\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,512): Alpha: 2.1681148371722547, Alpha Weighted: 1.216737308211145, D: 0.1237200121600533\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,512): Lognorm: 0.9532338976860046\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,512): Alpha: 2.689081100241649, Alpha Weighted: 1.2802702235446664, D: 0.09975582428211116\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,512): Lognorm: 0.9262917041778564\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,512): Alpha: 2.055719150635971, Alpha Weighted: 1.261199135786308, D: 0.11025163408581629\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,512): Lognorm: 0.9493745565414429\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,512): Alpha: 3.8241649495814745, Alpha Weighted: 3.263764804819505, D: 0.09906517298550521\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,512): Lognorm: 1.0106232166290283\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,512): Alpha: 2.8416595280214443, Alpha Weighted: 1.7285444834356298, D: 0.10146035935224873\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,512): Lognorm: 0.9507067203521729\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,512): Alpha: 2.5426597762813685, Alpha Weighted: 1.1927620002824646, D: 0.11033070120208638\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,512): Lognorm: 0.9216506481170654\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,512): Alpha: 2.89949125033221, Alpha Weighted: 1.6334759855622025, D: 0.11620853244221552\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,512): Lognorm: 0.9460031390190125\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,512): Alpha: 2.267163007122237, Alpha Weighted: 1.1108198962364197, D: 0.11398771955144887\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,512): Lognorm: 0.922294020652771\n",
      "DEBUG:__main__:Layer 15: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 15: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 16: Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 512x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 16: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Alpha: 3.1496461609658, Alpha Weighted: 2.028053336585761, D: 0.05121125656456915\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Lognorm: 0.9826978445053101\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Alpha: 3.5179699537499443, Alpha Weighted: 2.7261393586125564, D: 0.06635147012063258\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Lognorm: 0.9950791001319885\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Alpha: 3.147034393982754, Alpha Weighted: 1.9897800581420975, D: 0.053711552648875394\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Lognorm: 0.9823008179664612\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Alpha: 2.7469945928755637, Alpha Weighted: 2.139472885682934, D: 0.05741479262466076\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Lognorm: 0.9806458353996277\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Alpha: 2.40742308849171, Alpha Weighted: 2.3368157721467253, D: 0.062303956634809976\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Lognorm: 1.024043083190918\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Alpha: 3.302657328288202, Alpha Weighted: 2.5299423152717266, D: 0.0500698108501777\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Lognorm: 0.9784256815910339\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Alpha: 3.195929353845168, Alpha Weighted: 1.9390562610333255, D: 0.05217670162546484\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Lognorm: 0.9802641868591309\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Alpha: 3.35401990153947, Alpha Weighted: 2.6749296775487785, D: 0.04650916556371215\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Lognorm: 0.9925371408462524\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Alpha: 2.967192312297823, Alpha Weighted: 1.8252350268175732, D: 0.060972319563294686\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Lognorm: 0.9784339666366577\n",
      "DEBUG:__main__:Layer 17: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 17: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 18: MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "DEBUG:__main__:Layer 18: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 19: Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 512x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 19: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Alpha: 2.7750601689566214, Alpha Weighted: 1.4194197913482802, D: 0.08158088074983577\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Lognorm: 0.9942262768745422\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Alpha: 3.650765336346804, Alpha Weighted: 2.226924039689674, D: 0.07941522981723148\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Lognorm: 1.0120388269424438\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Alpha: 3.1718399966146964, Alpha Weighted: 1.712985180361673, D: 0.07556401606411633\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Lognorm: 0.9953451752662659\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Alpha: 3.875876983334032, Alpha Weighted: 2.3812682394650935, D: 0.07539057571708585\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Lognorm: 0.9894785284996033\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Alpha: 2.234561722169161, Alpha Weighted: 1.878436945236891, D: 0.07883971568998421\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Lognorm: 1.0499871969223022\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Alpha: 3.2023158968607195, Alpha Weighted: 2.0257131075044965, D: 0.0716379196572251\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Lognorm: 0.9896095991134644\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Alpha: 2.886601508718468, Alpha Weighted: 1.5256245302635651, D: 0.07946961191450125\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Lognorm: 0.9956439733505249\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Alpha: 2.993324436814876, Alpha Weighted: 1.8340192950958008, D: 0.07260432852083687\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Lognorm: 1.0118615627288818\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Alpha: 2.9316370008659227, Alpha Weighted: 1.5953353569053834, D: 0.07331758439484826\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Lognorm: 0.9950817823410034\n",
      "DEBUG:__main__:Layer 20: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 20: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 21: Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 512x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 21: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Alpha: 3.7174833225794837, Alpha Weighted: 3.1915906215451173, D: 0.03975874804808616\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Lognorm: 0.9724318981170654\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Alpha: 3.6541367272376895, Alpha Weighted: 3.643474373641885, D: 0.04704344460392029\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Lognorm: 0.9977350234985352\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Alpha: 3.7384614958464444, Alpha Weighted: 3.356180931310959, D: 0.03906594730890983\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Lognorm: 0.976440966129303\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Alpha: 3.515367343187722, Alpha Weighted: 3.2881311154300485, D: 0.029411764705883137\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Lognorm: 0.9728533625602722\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Alpha: 2.775349822902953, Alpha Weighted: 2.983763554011817, D: 0.055357745535253855\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Lognorm: 1.0238560438156128\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Alpha: 3.4376664436846047, Alpha Weighted: 3.236642486139822, D: 0.03492502465135561\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Lognorm: 0.9753591418266296\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Alpha: 3.7856467710955197, Alpha Weighted: 3.202841126253127, D: 0.031724841231911494\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Lognorm: 0.9722591638565063\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Alpha: 3.409718294273313, Alpha Weighted: 3.3248414445314327, D: 0.044012552185762\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Lognorm: 0.9990035891532898\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Alpha: 3.6552995501344165, Alpha Weighted: 3.110290662804049, D: 0.03151384888698627\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Lognorm: 0.9731558561325073\n",
      "DEBUG:__main__:Layer 22: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 22: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 23: MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "DEBUG:__main__:Layer 23: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 24: Sequential(\n",
      "  (0): Linear(in_features=25088, out_features=4096, bias=True)\n",
      "  (1): ReLU(inplace)\n",
      "  (2): Dropout(p=0.5)\n",
      "  (3): Linear(in_features=4096, out_features=4096, bias=True)\n",
      "  (4): ReLU(inplace)\n",
      "  (5): Dropout(p=0.5)\n",
      "  (6): Linear(in_features=4096, out_features=1000, bias=True)\n",
      ")\n",
      "DEBUG:__main__:Layer 24: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 25: Linear(in_features=25088, out_features=4096, bias=True)\n",
      "DEBUG:__main__:Layer 25: Analyzing 1 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,25088): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,25088): Alpha: 2.3250657702590294, Alpha Weighted: 3.5838094337568434, D: 0.03256343090025382\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "DEBUG:__main__:    Weight matrix 1/1 (4096,25088): Lognorm: 1.758880853652954\n",
      "DEBUG:__main__:Layer 26: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 26: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 27: Dropout(p=0.5)\n",
      "DEBUG:__main__:Layer 27: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 28: Linear(in_features=4096, out_features=4096, bias=True)\n",
      "DEBUG:__main__:Layer 28: Analyzing 1 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,4096): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,4096): Alpha: 2.167513738036675, Alpha Weighted: 3.8585291581042744, D: 0.030891252864063246\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,4096): Lognorm: 1.6416341066360474\n",
      "DEBUG:__main__:Layer 29: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 29: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 30: Dropout(p=0.5)\n",
      "DEBUG:__main__:Layer 30: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 31: Linear(in_features=4096, out_features=1000, bias=True)\n",
      "DEBUG:__main__:Layer 31: Analyzing 1 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (1000,4096): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (1000,4096): Alpha: 2.8256544917686948, Alpha Weighted: 4.999376215087901, D: 0.03977309921022354\n",
      "DEBUG:__main__:    Weight matrix 1/1 (1000,4096): Lognorm: 1.5924100875854492\n",
      "INFO:__main__:### Printing results ###\n",
      "DEBUG:__main__:Layer 5: Norm compound: 5.895910263061523\n",
      "DEBUG:__main__:Layer 5: LogNorm compound: 0.7691774368286133\n",
      "DEBUG:__main__:Layer 5: Alpha compound: 1.522580930995347\n",
      "DEBUG:__main__:Layer 5: Alpha Weighted compound: 1.0634833104417167\n",
      "DEBUG:__main__:Layer 8: Norm compound: 6.683967113494873\n",
      "DEBUG:__main__:Layer 8: LogNorm compound: 0.8212912082672119\n",
      "DEBUG:__main__:Layer 8: Alpha compound: 2.0207002788632304\n",
      "DEBUG:__main__:Layer 8: Alpha Weighted compound: 1.1427920476540687\n",
      "DEBUG:__main__:Layer 10: Norm compound: 7.619507789611816\n",
      "DEBUG:__main__:Layer 10: LogNorm compound: 0.8811509013175964\n",
      "DEBUG:__main__:Layer 10: Alpha compound: 2.159670978680439\n",
      "DEBUG:__main__:Layer 10: Alpha Weighted compound: 1.0371089622425371\n",
      "DEBUG:__main__:Layer 13: Norm compound: 8.830634117126465\n",
      "DEBUG:__main__:Layer 13: LogNorm compound: 0.9451767802238464\n",
      "DEBUG:__main__:Layer 13: Alpha compound: 2.6599751390253275\n",
      "DEBUG:__main__:Layer 13: Alpha Weighted compound: 1.5493992453281573\n",
      "DEBUG:__main__:Layer 15: Norm compound: 9.73856258392334\n",
      "DEBUG:__main__:Layer 15: LogNorm compound: 0.9882696866989136\n",
      "DEBUG:__main__:Layer 15: Alpha compound: 3.0876518984484926\n",
      "DEBUG:__main__:Layer 15: Alpha Weighted compound: 2.243269410204609\n",
      "DEBUG:__main__:Layer 18: Norm compound: 10.09455394744873\n",
      "DEBUG:__main__:Layer 18: LogNorm compound: 1.0036969184875488\n",
      "DEBUG:__main__:Layer 18: Alpha compound: 3.0802203389645886\n",
      "DEBUG:__main__:Layer 18: Alpha Weighted compound: 1.8444140539856508\n",
      "DEBUG:__main__:Layer 20: Norm compound: 9.663440704345703\n",
      "DEBUG:__main__:Layer 20: LogNorm compound: 0.9847883582115173\n",
      "DEBUG:__main__:Layer 20: Alpha compound: 3.5210144189935715\n",
      "DEBUG:__main__:Layer 20: Alpha Weighted compound: 3.2597507017409177\n",
      "DEBUG:__main__:Layer 24: Norm: 57.39590072631836\n",
      "DEBUG:__main__:Layer 24: LogNorm: 1.758880853652954\n",
      "DEBUG:__main__:Layer 24: Alpha: 2.3250657702590294\n",
      "DEBUG:__main__:Layer 24: Alpha Weighted: 3.5838094337568434\n",
      "DEBUG:__main__:Layer 27: Norm: 43.81613540649414\n",
      "DEBUG:__main__:Layer 27: LogNorm: 1.6416341066360474\n",
      "DEBUG:__main__:Layer 27: Alpha: 2.167513738036675\n",
      "DEBUG:__main__:Layer 27: Alpha Weighted: 3.8585291581042744\n",
      "DEBUG:__main__:Layer 30: Norm: 39.12101745605469\n",
      "DEBUG:__main__:Layer 30: LogNorm: 1.5924100875854492\n",
      "DEBUG:__main__:Layer 30: Alpha: 2.8256544917686948\n",
      "DEBUG:__main__:Layer 30: Alpha Weighted: 4.999376215087901\n",
      "INFO:__main__:Norm: min: 5.2481255531311035, max: 57.39590072631836, avg: 10.107156753540039\n",
      "INFO:__main__:Norm compound: min: 5.895910263061523, max: 57.39590072631836, avg: 19.885963439941406\n",
      "INFO:__main__:LogNorm: min: 0.7200042009353638, max: 1.758880853652954, avg: 0.9474983215332031\n",
      "INFO:__main__:LogNorm compound: min: 0.7691774368286133, max: 1.758880853652954, avg: 1.1386475563049316\n",
      "INFO:__main__:Alpha: min: 1.3854856636971848, max: 3.875876983334032, avg: 2.5724933311485363\n",
      "INFO:__main__:Alpha compound: min: 1.522580930995347, max: 3.5210144189935715, avg: 2.5370047984035393\n",
      "INFO:__main__:Alpha Weighted: min: 0.6363944562361387, max: 4.999376215087901, avg: 1.8439950665352716\n",
      "INFO:__main__:Alpha Weighted compound: min: 1.0371089622425371, max: 4.999376215087901, avg: 2.4581932538546676\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>norm</th>\n",
       "      <th>norm_compound</th>\n",
       "      <th>lognorm</th>\n",
       "      <th>lognorm_compound</th>\n",
       "      <th>alpha</th>\n",
       "      <th>alpha_compound</th>\n",
       "      <th>alpha_weighted</th>\n",
       "      <th>alpha_weighted_compound</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>10.107157</td>\n",
       "      <td>19.885963</td>\n",
       "      <td>0.947498</td>\n",
       "      <td>1.138648</td>\n",
       "      <td>2.572493</td>\n",
       "      <td>2.537005</td>\n",
       "      <td>1.843995</td>\n",
       "      <td>2.458193</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "        norm  norm_compound   lognorm  lognorm_compound     alpha  \\\n",
       "0  10.107157      19.885963  0.947498          1.138648  2.572493   \n",
       "\n",
       "   alpha_compound  alpha_weighted  alpha_weighted_compound  \n",
       "0        2.537005        1.843995                 2.458193  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:__main__:### Printing results ###\n",
      "DEBUG:__main__:Layer 5: Norm compound: 5.895910263061523\n",
      "DEBUG:__main__:Layer 5: LogNorm compound: 0.7691774368286133\n",
      "DEBUG:__main__:Layer 5: Alpha compound: 1.522580930995347\n",
      "DEBUG:__main__:Layer 5: Alpha Weighted compound: 1.0634833104417167\n",
      "DEBUG:__main__:Layer 8: Norm compound: 6.683967113494873\n",
      "DEBUG:__main__:Layer 8: LogNorm compound: 0.8212912082672119\n",
      "DEBUG:__main__:Layer 8: Alpha compound: 2.0207002788632304\n",
      "DEBUG:__main__:Layer 8: Alpha Weighted compound: 1.1427920476540687\n",
      "DEBUG:__main__:Layer 10: Norm compound: 7.619507789611816\n",
      "DEBUG:__main__:Layer 10: LogNorm compound: 0.8811509013175964\n",
      "DEBUG:__main__:Layer 10: Alpha compound: 2.159670978680439\n",
      "DEBUG:__main__:Layer 10: Alpha Weighted compound: 1.0371089622425371\n",
      "DEBUG:__main__:Layer 13: Norm compound: 8.830634117126465\n",
      "DEBUG:__main__:Layer 13: LogNorm compound: 0.9451767802238464\n",
      "DEBUG:__main__:Layer 13: Alpha compound: 2.6599751390253275\n",
      "DEBUG:__main__:Layer 13: Alpha Weighted compound: 1.5493992453281573\n",
      "DEBUG:__main__:Layer 15: Norm compound: 9.73856258392334\n",
      "DEBUG:__main__:Layer 15: LogNorm compound: 0.9882696866989136\n",
      "DEBUG:__main__:Layer 15: Alpha compound: 3.0876518984484926\n",
      "DEBUG:__main__:Layer 15: Alpha Weighted compound: 2.243269410204609\n",
      "DEBUG:__main__:Layer 18: Norm compound: 10.09455394744873\n",
      "DEBUG:__main__:Layer 18: LogNorm compound: 1.0036969184875488\n",
      "DEBUG:__main__:Layer 18: Alpha compound: 3.0802203389645886\n",
      "DEBUG:__main__:Layer 18: Alpha Weighted compound: 1.8444140539856508\n",
      "DEBUG:__main__:Layer 20: Norm compound: 9.663440704345703\n",
      "DEBUG:__main__:Layer 20: LogNorm compound: 0.9847883582115173\n",
      "DEBUG:__main__:Layer 20: Alpha compound: 3.5210144189935715\n",
      "DEBUG:__main__:Layer 20: Alpha Weighted compound: 3.2597507017409177\n",
      "DEBUG:__main__:Layer 24: Norm: 57.39590072631836\n",
      "DEBUG:__main__:Layer 24: LogNorm: 1.758880853652954\n",
      "DEBUG:__main__:Layer 24: Alpha: 2.3250657702590294\n",
      "DEBUG:__main__:Layer 24: Alpha Weighted: 3.5838094337568434\n",
      "DEBUG:__main__:Layer 27: Norm: 43.81613540649414\n",
      "DEBUG:__main__:Layer 27: LogNorm: 1.6416341066360474\n",
      "DEBUG:__main__:Layer 27: Alpha: 2.167513738036675\n",
      "DEBUG:__main__:Layer 27: Alpha Weighted: 3.8585291581042744\n",
      "DEBUG:__main__:Layer 30: Norm: 39.12101745605469\n",
      "DEBUG:__main__:Layer 30: LogNorm: 1.5924100875854492\n",
      "DEBUG:__main__:Layer 30: Alpha: 2.8256544917686948\n",
      "DEBUG:__main__:Layer 30: Alpha Weighted: 4.999376215087901\n",
      "INFO:__main__:Norm: min: 5.2481255531311035, max: 57.39590072631836, avg: 10.107156753540039\n",
      "INFO:__main__:Norm compound: min: 5.895910263061523, max: 57.39590072631836, avg: 19.885963439941406\n",
      "INFO:__main__:LogNorm: min: 0.7200042009353638, max: 1.758880853652954, avg: 0.9474983215332031\n",
      "INFO:__main__:LogNorm compound: min: 0.7691774368286133, max: 1.758880853652954, avg: 1.1386475563049316\n",
      "INFO:__main__:Alpha: min: 1.3854856636971848, max: 3.875876983334032, avg: 2.5724933311485363\n",
      "INFO:__main__:Alpha compound: min: 1.522580930995347, max: 3.5210144189935715, avg: 2.5370047984035393\n",
      "INFO:__main__:Alpha Weighted: min: 0.6363944562361387, max: 4.999376215087901, avg: 1.8439950665352716\n",
      "INFO:__main__:Alpha Weighted compound: min: 1.0371089622425371, max: 4.999376215087901, avg: 2.4581932538546676\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>layer_type</th>\n",
       "      <th>N</th>\n",
       "      <th>M</th>\n",
       "      <th>level</th>\n",
       "      <th>comment</th>\n",
       "      <th>norm</th>\n",
       "      <th>lognorm</th>\n",
       "      <th>alpha</th>\n",
       "      <th>alpha_weighted</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>layer_id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>128</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>5.54028</td>\n",
       "      <td>0.743532</td>\n",
       "      <td>1.65014</td>\n",
       "      <td>1.07665</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>128</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>6.0134</td>\n",
       "      <td>0.77912</td>\n",
       "      <td>1.57297</td>\n",
       "      <td>1.16638</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>128</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>5.67136</td>\n",
       "      <td>0.753687</td>\n",
       "      <td>1.4121</td>\n",
       "      <td>0.895458</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>128</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>5.86407</td>\n",
       "      <td>0.768199</td>\n",
       "      <td>1.43459</td>\n",
       "      <td>0.94487</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>128</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>7.03509</td>\n",
       "      <td>0.84727</td>\n",
       "      <td>1.6185</td>\n",
       "      <td>1.561</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         layer_type    N   M        level      comment     norm   lognorm  \\\n",
       "layer_id                                                                    \n",
       "5            CONV2D  128  64  LEVEL.SLICE  Slice level  5.54028  0.743532   \n",
       "5            CONV2D  128  64  LEVEL.SLICE  Slice level   6.0134   0.77912   \n",
       "5            CONV2D  128  64  LEVEL.SLICE  Slice level  5.67136  0.753687   \n",
       "5            CONV2D  128  64  LEVEL.SLICE  Slice level  5.86407  0.768199   \n",
       "5            CONV2D  128  64  LEVEL.SLICE  Slice level  7.03509   0.84727   \n",
       "\n",
       "            alpha alpha_weighted  \n",
       "layer_id                          \n",
       "5         1.65014        1.07665  \n",
       "5         1.57297        1.16638  \n",
       "5          1.4121       0.895458  \n",
       "5         1.43459        0.94487  \n",
       "5          1.6185          1.561  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model_11 = models.vgg11(pretrained=True)\n",
    "watcher = ww.WeightWatcher(model=model_11, logger=logger)\n",
    "results_11 = watcher.analyze(alphas=True)\n",
    "summary_11 = watcher.get_summary(pandas=True)\n",
    "display(summary_11)\n",
    "details_11 = watcher.get_details(results=results_11)\n",
    "details_11.drop(columns=['slice', 'slice_count'], inplace=True)\n",
    "details_11.dropna(inplace=True)\n",
    "display(details_11.head())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 139,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>layer_type</th>\n",
       "      <th>N</th>\n",
       "      <th>M</th>\n",
       "      <th>level</th>\n",
       "      <th>comment</th>\n",
       "      <th>norm</th>\n",
       "      <th>lognorm</th>\n",
       "      <th>alpha</th>\n",
       "      <th>alpha_weighted</th>\n",
       "      <th>NxM</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>layer_id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>128</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>5.54028</td>\n",
       "      <td>0.743532</td>\n",
       "      <td>1.65014</td>\n",
       "      <td>1.07665</td>\n",
       "      <td>8192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>128</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>6.0134</td>\n",
       "      <td>0.77912</td>\n",
       "      <td>1.57297</td>\n",
       "      <td>1.16638</td>\n",
       "      <td>8192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>128</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>5.67136</td>\n",
       "      <td>0.753687</td>\n",
       "      <td>1.4121</td>\n",
       "      <td>0.895458</td>\n",
       "      <td>8192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>128</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>5.86407</td>\n",
       "      <td>0.768199</td>\n",
       "      <td>1.43459</td>\n",
       "      <td>0.94487</td>\n",
       "      <td>8192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>128</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>7.03509</td>\n",
       "      <td>0.84727</td>\n",
       "      <td>1.6185</td>\n",
       "      <td>1.561</td>\n",
       "      <td>8192</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         layer_type    N   M        level      comment     norm   lognorm  \\\n",
       "layer_id                                                                    \n",
       "5            CONV2D  128  64  LEVEL.SLICE  Slice level  5.54028  0.743532   \n",
       "5            CONV2D  128  64  LEVEL.SLICE  Slice level   6.0134   0.77912   \n",
       "5            CONV2D  128  64  LEVEL.SLICE  Slice level  5.67136  0.753687   \n",
       "5            CONV2D  128  64  LEVEL.SLICE  Slice level  5.86407  0.768199   \n",
       "5            CONV2D  128  64  LEVEL.SLICE  Slice level  7.03509   0.84727   \n",
       "\n",
       "            alpha alpha_weighted   NxM  \n",
       "layer_id                                \n",
       "5         1.65014        1.07665  8192  \n",
       "5         1.57297        1.16638  8192  \n",
       "5          1.4121       0.895458  8192  \n",
       "5         1.43459        0.94487  8192  \n",
       "5          1.6185          1.561  8192  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "details_11['NxM'] = pd.to_numeric(details_11.N * details_11.M)\n",
    "display(details_11.head())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 140,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0, 0.5, 'alpha')"
      ]
     },
     "execution_count": 140,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x = np.log10(details_11.NxM.to_numpy())\n",
    "y = details_11.alpha.to_numpy()\n",
    "plt.scatter(x,y)\n",
    "plt.title(r\"VGG 11  layer size vs alpha ($\\alpha$)\")\n",
    "plt.xlabel('NxM')\n",
    "plt.ylabel('alpha')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 141,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0, 0.5, 'alpha')"
      ]
     },
     "execution_count": 141,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x = details_11.index.to_numpy()\n",
    "y = details_11.alpha.to_numpy()\n",
    "plt.scatter(x,y)\n",
    "plt.title(r\"VGG 11  layer size vs depth\")\n",
    "plt.xlabel('depth')\n",
    "plt.ylabel('alpha')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 126,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:__main__:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:__main__:Analyzing model\n",
      "DEBUG:__main__:Layer 1: VGG(\n",
      "  (features): Sequential(\n",
      "    (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (1): ReLU(inplace)\n",
      "    (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (3): ReLU(inplace)\n",
      "    (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "    (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (6): ReLU(inplace)\n",
      "    (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (8): ReLU(inplace)\n",
      "    (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "    (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (11): ReLU(inplace)\n",
      "    (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (13): ReLU(inplace)\n",
      "    (14): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "    (15): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (16): ReLU(inplace)\n",
      "    (17): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (18): ReLU(inplace)\n",
      "    (19): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "    (20): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (21): ReLU(inplace)\n",
      "    (22): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "    (23): ReLU(inplace)\n",
      "    (24): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  )\n",
      "  (classifier): Sequential(\n",
      "    (0): Linear(in_features=25088, out_features=4096, bias=True)\n",
      "    (1): ReLU(inplace)\n",
      "    (2): Dropout(p=0.5)\n",
      "    (3): Linear(in_features=4096, out_features=4096, bias=True)\n",
      "    (4): ReLU(inplace)\n",
      "    (5): Dropout(p=0.5)\n",
      "    (6): Linear(in_features=4096, out_features=1000, bias=True)\n",
      "  )\n",
      ")\n",
      "DEBUG:__main__:Layer 1: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 2: Sequential(\n",
      "  (0): Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (1): ReLU(inplace)\n",
      "  (2): Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (3): ReLU(inplace)\n",
      "  (4): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (5): Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (6): ReLU(inplace)\n",
      "  (7): Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (8): ReLU(inplace)\n",
      "  (9): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (10): Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (11): ReLU(inplace)\n",
      "  (12): Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (13): ReLU(inplace)\n",
      "  (14): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (15): Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (16): ReLU(inplace)\n",
      "  (17): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (18): ReLU(inplace)\n",
      "  (19): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (20): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (21): ReLU(inplace)\n",
      "  (22): Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (23): ReLU(inplace)\n",
      "  (24): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      ")\n",
      "DEBUG:__main__:Layer 2: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 3: Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 64x3 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 3: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 2/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 3/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 4/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 5/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 6/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 7/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 8/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:    Weight matrix 9/9 (3,64): Skipping: too small (<50)\n",
      "DEBUG:__main__:Layer 4: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 4: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 5: Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 64x64 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 5: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,64): Alpha: 1.9985715582411618, Alpha Weighted: 0.5308489491249853, D: 0.18598435864486917\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,64): Lognorm: 0.555422306060791\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,64): Alpha: 1.5227655741250756, Alpha Weighted: 0.7031175814665804, D: 0.15918353760026382\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,64): Lognorm: 0.5787814259529114\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,64): Alpha: 2.1369659865342907, Alpha Weighted: 0.6382023369779065, D: 0.16491783981447594\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,64): Lognorm: 0.53458172082901\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,64): Alpha: 1.4775758478961944, Alpha Weighted: 0.6623067662111037, D: 0.16073579713978559\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,64): Lognorm: 0.5797208547592163\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,64): Alpha: 1.5488983071019056, Alpha Weighted: 0.8708144624228942, D: 0.1493814620980075\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,64): Lognorm: 0.626145601272583\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,64): Alpha: 1.55225151192464, Alpha Weighted: 0.621203297781999, D: 0.16765236741203737\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,64): Lognorm: 0.5915738940238953\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,64): Alpha: 1.4229916456933371, Alpha Weighted: 0.4182560668624704, D: 0.19553251096327984\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,64): Lognorm: 0.5425663590431213\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,64): Alpha: 1.4870416517218874, Alpha Weighted: 0.6945968813193277, D: 0.16496928322995708\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,64): Lognorm: 0.5852140784263611\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,64): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,64): Alpha: 1.4267629740682501, Alpha Weighted: 0.37454269723312644, D: 0.20325861360671726\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,64): Lognorm: 0.5504339933395386\n",
      "DEBUG:__main__:Layer 6: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 6: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 7: MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "DEBUG:__main__:Layer 7: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 8: Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 128x64 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 8: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,128): Alpha: 1.8107393371856193, Alpha Weighted: 0.7929709580783333, D: 0.145003078083912\n",
      "DEBUG:__main__:    Weight matrix 1/9 (64,128): Lognorm: 0.6680006384849548\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,128): Alpha: 1.772993868234496, Alpha Weighted: 1.044640941862756, D: 0.1453439114249766\n",
      "DEBUG:__main__:    Weight matrix 2/9 (64,128): Lognorm: 0.7057045102119446\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (64,128): Alpha: 1.8114713567892062, Alpha Weighted: 0.8059383018120009, D: 0.15762453151811828\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "DEBUG:__main__:    Weight matrix 3/9 (64,128): Lognorm: 0.6612639427185059\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,128): Alpha: 1.7804244107953249, Alpha Weighted: 0.9158587100252071, D: 0.1557509944471459\n",
      "DEBUG:__main__:    Weight matrix 4/9 (64,128): Lognorm: 0.6933040618896484\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,128): Alpha: 1.7223668128083331, Alpha Weighted: 1.2801291360661946, D: 0.14315025688733707\n",
      "DEBUG:__main__:    Weight matrix 5/9 (64,128): Lognorm: 0.7414924502372742\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,128): Alpha: 1.7330248755447897, Alpha Weighted: 0.9110641458044347, D: 0.14202299805469754\n",
      "DEBUG:__main__:    Weight matrix 6/9 (64,128): Lognorm: 0.6887338161468506\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,128): Alpha: 1.8129096745427962, Alpha Weighted: 1.036491422395262, D: 0.1380976024427068\n",
      "DEBUG:__main__:    Weight matrix 7/9 (64,128): Lognorm: 0.6727420091629028\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,128): Alpha: 1.7591482598610289, Alpha Weighted: 0.9669423925995486, D: 0.1393481636301292\n",
      "DEBUG:__main__:    Weight matrix 8/9 (64,128): Lognorm: 0.6912813782691956\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,128): Alpha: 1.8891461770425502, Alpha Weighted: 0.822944189948913, D: 0.16650105038326868\n",
      "DEBUG:__main__:    Weight matrix 9/9 (64,128): Lognorm: 0.6467894315719604\n",
      "DEBUG:__main__:Layer 9: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 9: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 10: Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 128x128 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 10: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,128): Alpha: 1.8570219612853736, Alpha Weighted: 0.4250346664819842, D: 0.15544349756901876\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,128): Lognorm: 0.7174193859100342\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,128): Alpha: 1.9010923121147156, Alpha Weighted: 0.6002314329093935, D: 0.15090251087721718\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,128): Lognorm: 0.7336912751197815\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,128): Alpha: 3.6536047227195163, Alpha Weighted: 1.0583658869687238, D: 0.15351074278406585\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,128): Lognorm: 0.7086058855056763\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,128): Alpha: 1.7262487614890296, Alpha Weighted: 0.4218873424575699, D: 0.17243243233990202\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,128): Lognorm: 0.7316402196884155\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,128): Alpha: 4.82206051976946, Alpha Weighted: 2.2655205577904214, D: 0.14172919039989396\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,128): Lognorm: 0.7713156938552856\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,128): Alpha: 2.004595277053026, Alpha Weighted: 0.5872489432244166, D: 0.16553362840440905\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,128): Lognorm: 0.7348038554191589\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,128): Alpha: 1.9531349404262532, Alpha Weighted: 0.509380743516003, D: 0.1566485003494653\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,128): Lognorm: 0.7069422006607056\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,128): Alpha: 1.959105001452026, Alpha Weighted: 0.43972431347878904, D: 0.16397838208896398\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,128): Lognorm: 0.7316706776618958\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,128): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,128): Alpha: 1.8342964803521693, Alpha Weighted: 0.46273930910770666, D: 0.16355890803466455\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,128): Lognorm: 0.7158021926879883\n",
      "DEBUG:__main__:Layer 11: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 11: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 12: MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "DEBUG:__main__:Layer 12: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 13: Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 256x128 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 13: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,256): Alpha: 2.5156263315427014, Alpha Weighted: 1.0518319493150474, D: 0.10143020362332034\n",
      "DEBUG:__main__:    Weight matrix 1/9 (128,256): Lognorm: 0.767629086971283\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,256): Alpha: 2.451452327104103, Alpha Weighted: 1.3212482861828114, D: 0.11681686474542585\n",
      "DEBUG:__main__:    Weight matrix 2/9 (128,256): Lognorm: 0.8146148920059204\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,256): Alpha: 3.0291383971766956, Alpha Weighted: 1.1853419152641411, D: 0.1295866906581885\n",
      "DEBUG:__main__:    Weight matrix 3/9 (128,256): Lognorm: 0.7666121125221252\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,256): Alpha: 2.8392635931972205, Alpha Weighted: 1.5765672492033074, D: 0.1086031022794709\n",
      "DEBUG:__main__:    Weight matrix 4/9 (128,256): Lognorm: 0.814735472202301\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,256): Alpha: 4.882618427826814, Alpha Weighted: 3.6244402885396063, D: 0.14285714285714257\n",
      "DEBUG:__main__:    Weight matrix 5/9 (128,256): Lognorm: 0.898858368396759\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,256): Alpha: 2.6335172889477008, Alpha Weighted: 1.3063947877860587, D: 0.10870686136859381\n",
      "DEBUG:__main__:    Weight matrix 6/9 (128,256): Lognorm: 0.8094183802604675\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,256): Alpha: 2.6176186602074143, Alpha Weighted: 0.9261289249369303, D: 0.11393706295861206\n",
      "DEBUG:__main__:    Weight matrix 7/9 (128,256): Lognorm: 0.7631824612617493\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,256): Alpha: 2.586947761932789, Alpha Weighted: 1.4324399280097222, D: 0.13113972609376623\n",
      "DEBUG:__main__:    Weight matrix 8/9 (128,256): Lognorm: 0.8044846057891846\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,256): Alpha: 3.14734192478364, Alpha Weighted: 1.1551267747085086, D: 0.11083817736730506\n",
      "DEBUG:__main__:    Weight matrix 9/9 (128,256): Lognorm: 0.7636407017707825\n",
      "DEBUG:__main__:Layer 14: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 14: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 15: Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 256x256 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 15: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,256): Alpha: 2.4906044072096485, Alpha Weighted: 0.8338422690802988, D: 0.1192682466854233\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,256): Lognorm: 0.8347761034965515\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,256): Alpha: 2.74644122687605, Alpha Weighted: 1.319816794724483, D: 0.09733492380390774\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,256): Lognorm: 0.8510901927947998\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "DEBUG:__main__:    Weight matrix 3/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,256): Alpha: 3.072085288559462, Alpha Weighted: 1.0791189177059968, D: 0.09978831257031451\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,256): Lognorm: 0.8321912884712219\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,256): Alpha: 3.04434009703647, Alpha Weighted: 1.5885952171494162, D: 0.09122357270973214\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,256): Lognorm: 0.8481708765029907\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,256): Alpha: 3.423834630957338, Alpha Weighted: 2.4744256947845185, D: 0.06957180457999956\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,256): Lognorm: 0.9025205969810486\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,256): Alpha: 3.1933722378218037, Alpha Weighted: 1.546795863739017, D: 0.08670652456400962\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,256): Lognorm: 0.8415446281433105\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,256): Alpha: 2.5797551305867876, Alpha Weighted: 0.8874103468881296, D: 0.10902770718007143\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,256): Lognorm: 0.8381807804107666\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,256): Alpha: 3.0865068573284464, Alpha Weighted: 1.5002476477585363, D: 0.10738600672168619\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,256): Lognorm: 0.8552940487861633\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,256): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,256): Alpha: 2.7554707777336134, Alpha Weighted: 1.0248411875342611, D: 0.09562037875936025\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,256): Lognorm: 0.835618257522583\n",
      "DEBUG:__main__:Layer 16: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 16: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 17: MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "DEBUG:__main__:Layer 17: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 18: Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 512x256 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 18: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,512): Alpha: 2.339918418834043, Alpha Weighted: 1.0091318822760382, D: 0.11919282218587512\n",
      "DEBUG:__main__:    Weight matrix 1/9 (256,512): Lognorm: 0.9165790677070618\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,512): Alpha: 2.0322164304588775, Alpha Weighted: 1.0873146989220235, D: 0.11662828334888398\n",
      "DEBUG:__main__:    Weight matrix 2/9 (256,512): Lognorm: 0.9380613565444946\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,512): Alpha: 2.7086147148873967, Alpha Weighted: 1.260077451735826, D: 0.10192457018398393\n",
      "DEBUG:__main__:    Weight matrix 3/9 (256,512): Lognorm: 0.9178951382637024\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,512): Alpha: 2.174964107939772, Alpha Weighted: 1.1613203374998577, D: 0.11920908133771457\n",
      "DEBUG:__main__:    Weight matrix 4/9 (256,512): Lognorm: 0.9282858967781067\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,512): Alpha: 3.944479224615485, Alpha Weighted: 3.036396421458769, D: 0.07692307692307654\n",
      "DEBUG:__main__:    Weight matrix 5/9 (256,512): Lognorm: 0.9806009531021118\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,512): Alpha: 3.025388001310995, Alpha Weighted: 1.6204499479644785, D: 0.10638330254655604\n",
      "DEBUG:__main__:    Weight matrix 6/9 (256,512): Lognorm: 0.9300090670585632\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,512): Alpha: 2.3007343415207795, Alpha Weighted: 0.9941728113716969, D: 0.11341638954201105\n",
      "DEBUG:__main__:    Weight matrix 7/9 (256,512): Lognorm: 0.912216305732727\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,512): Alpha: 1.9946395540113593, Alpha Weighted: 1.0130027659383336, D: 0.11187633149873716\n",
      "DEBUG:__main__:    Weight matrix 8/9 (256,512): Lognorm: 0.9304941296577454\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,512): Alpha: 2.9786896198905066, Alpha Weighted: 1.218162287228671, D: 0.11800518041856178\n",
      "DEBUG:__main__:    Weight matrix 9/9 (256,512): Lognorm: 0.9159024953842163\n",
      "DEBUG:__main__:Layer 19: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 19: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 20: Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 512x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 20: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Alpha: 3.1574304512457396, Alpha Weighted: 1.959673491800216, D: 0.05004403502379673\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Lognorm: 0.9726711511611938\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Alpha: 3.2358627573946226, Alpha Weighted: 2.5448653806116566, D: 0.0540045531688389\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Lognorm: 0.9830425977706909\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Alpha: 3.1313316250195684, Alpha Weighted: 2.0400345156060835, D: 0.05733112693393483\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Lognorm: 0.9723473787307739\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Alpha: 3.2813711964193457, Alpha Weighted: 2.531529735972014, D: 0.04870584911629261\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Lognorm: 0.9659870266914368\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Alpha: 2.480002026824974, Alpha Weighted: 2.3742308660866014, D: 0.0528351150810798\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Lognorm: 1.0074034929275513\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Alpha: 3.4296669398802138, Alpha Weighted: 2.6592346950418024, D: 0.05915618705536088\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Lognorm: 0.9684156179428101\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Alpha: 3.1932014166852474, Alpha Weighted: 1.883564697669941, D: 0.056019200873784536\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Lognorm: 0.9694008231163025\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Alpha: 3.4969867946215785, Alpha Weighted: 2.705355840373922, D: 0.05143822160779635\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Lognorm: 0.9819566607475281\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Alpha: 3.19090675071667, Alpha Weighted: 2.010219124891505, D: 0.061848384665032896\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Lognorm: 0.9707570672035217\n",
      "DEBUG:__main__:Layer 21: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 21: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 22: MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "DEBUG:__main__:Layer 22: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 23: Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 512x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 23: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Alpha: 3.196730945929455, Alpha Weighted: 1.7552654556969438, D: 0.05404486438920553\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Lognorm: 0.9932413101196289\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Alpha: 3.586872770204918, Alpha Weighted: 2.1468361875739195, D: 0.08012886952137144\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Lognorm: 1.0082613229751587\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Alpha: 3.0779288592378506, Alpha Weighted: 1.5116604314948319, D: 0.0627208911312393\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Lognorm: 0.9908903241157532\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Alpha: 3.787068655365929, Alpha Weighted: 2.410067408902185, D: 0.0587585215825307\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Lognorm: 0.9824333190917969\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Alpha: 2.319694534673407, Alpha Weighted: 1.867081459274312, D: 0.08181707853284692\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Lognorm: 1.0340932607650757\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Alpha: 3.1356638417110205, Alpha Weighted: 1.976421956686364, D: 0.06750535259362916\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Lognorm: 0.978428065776825\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Alpha: 2.878874846977443, Alpha Weighted: 1.4185104675004347, D: 0.07264423335227188\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Lognorm: 0.9884415864944458\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Alpha: 3.0839524846717596, Alpha Weighted: 1.7792875309962597, D: 0.06842032282217742\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Lognorm: 0.9988676309585571\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Alpha: 3.1636296409755826, Alpha Weighted: 1.4732428325666085, D: 0.07122925214535514\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Lognorm: 0.9857199192047119\n",
      "DEBUG:__main__:Layer 24: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 24: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 25: Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "DEBUG:__main__:Pytorch tensor shape detected: 512x512 (NxM), 3x3 (i,j)\n",
      "DEBUG:__main__:Layer 25: Analyzing 9 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Alpha: 3.7468951617220485, Alpha Weighted: 3.211596539693416, D: 0.033182616374172524\n",
      "DEBUG:__main__:    Weight matrix 1/9 (512,512): Lognorm: 0.9672555923461914\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Alpha: 3.7153278422776888, Alpha Weighted: 3.5843350458213554, D: 0.03767336108718222\n",
      "DEBUG:__main__:    Weight matrix 2/9 (512,512): Lognorm: 0.9875938892364502\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Alpha: 3.7927679841651027, Alpha Weighted: 3.1733240555913227, D: 0.054952447500068835\n",
      "DEBUG:__main__:    Weight matrix 3/9 (512,512): Lognorm: 0.9652756452560425\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Alpha: 3.5677494262480494, Alpha Weighted: 3.3017193632752635, D: 0.03672940846227407\n",
      "DEBUG:__main__:    Weight matrix 4/9 (512,512): Lognorm: 0.9651980996131897\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Alpha: 3.356123388625047, Alpha Weighted: 3.485673644695026, D: 0.05544787773259918\n",
      "DEBUG:__main__:    Weight matrix 5/9 (512,512): Lognorm: 1.0066924095153809\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Alpha: 3.629195789248568, Alpha Weighted: 3.2481857556059928, D: 0.03350392075153297\n",
      "DEBUG:__main__:    Weight matrix 6/9 (512,512): Lognorm: 0.961090624332428\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Alpha: 3.755606692335275, Alpha Weighted: 3.1718733447125915, D: 0.03545458814282754\n",
      "DEBUG:__main__:    Weight matrix 7/9 (512,512): Lognorm: 0.9657158255577087\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Alpha: 3.560065781142996, Alpha Weighted: 3.4027957409950074, D: 0.02702702702702775\n",
      "DEBUG:__main__:    Weight matrix 8/9 (512,512): Lognorm: 0.98719322681427\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Alpha: 3.7644313977699615, Alpha Weighted: 3.1703163208379115, D: 0.04301552503776196\n",
      "DEBUG:__main__:    Weight matrix 9/9 (512,512): Lognorm: 0.9637725353240967\n",
      "DEBUG:__main__:Layer 26: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 26: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 27: MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "DEBUG:__main__:Layer 27: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 28: Sequential(\n",
      "  (0): Linear(in_features=25088, out_features=4096, bias=True)\n",
      "  (1): ReLU(inplace)\n",
      "  (2): Dropout(p=0.5)\n",
      "  (3): Linear(in_features=4096, out_features=4096, bias=True)\n",
      "  (4): ReLU(inplace)\n",
      "  (5): Dropout(p=0.5)\n",
      "  (6): Linear(in_features=4096, out_features=1000, bias=True)\n",
      ")\n",
      "DEBUG:__main__:Layer 28: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 29: Linear(in_features=25088, out_features=4096, bias=True)\n",
      "DEBUG:__main__:Layer 29: Analyzing 1 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,25088): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,25088): Alpha: 2.3381178188805567, Alpha Weighted: 3.636844506019609, D: 0.03273875341892374\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,25088): Lognorm: 1.7561644315719604\n",
      "DEBUG:__main__:Layer 30: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 30: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 31: Dropout(p=0.5)\n",
      "DEBUG:__main__:Layer 31: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 32: Linear(in_features=4096, out_features=4096, bias=True)\n",
      "DEBUG:__main__:Layer 32: Analyzing 1 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,4096): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,4096): Alpha: 2.186124848459917, Alpha Weighted: 3.8836067216212076, D: 0.03052670033349769\n",
      "DEBUG:__main__:    Weight matrix 1/1 (4096,4096): Lognorm: 1.6374969482421875\n",
      "DEBUG:__main__:Layer 33: ReLU(inplace)\n",
      "DEBUG:__main__:Layer 33: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 34: Dropout(p=0.5)\n",
      "DEBUG:__main__:Layer 34: Skipping (Layer not supported)\n",
      "DEBUG:__main__:Layer 35: Linear(in_features=4096, out_features=1000, bias=True)\n",
      "DEBUG:__main__:Layer 35: Analyzing 1 weight matrices...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (1000,4096): Analyzing ...\n",
      "DEBUG:__main__:    Weight matrix 1/1 (1000,4096): Alpha: 2.7221347634801365, Alpha Weighted: 4.799173592745528, D: 0.03692774936449139\n",
      "DEBUG:__main__:    Weight matrix 1/1 (1000,4096): Lognorm: 1.5858654975891113\n",
      "INFO:__main__:### Printing results ###\n",
      "DEBUG:__main__:Layer 4: Norm compound: 3.7364306449890137\n",
      "DEBUG:__main__:Layer 4: LogNorm compound: 0.5716044902801514\n",
      "DEBUG:__main__:Layer 4: Alpha compound: 1.6193138952563046\n",
      "DEBUG:__main__:Layer 4: Alpha Weighted compound: 0.6126543377111549\n",
      "DEBUG:__main__:Layer 7: Norm compound: 4.856048107147217\n",
      "DEBUG:__main__:Layer 7: LogNorm compound: 0.6854791641235352\n",
      "DEBUG:__main__:Layer 7: Alpha compound: 1.788024974756016\n",
      "DEBUG:__main__:Layer 7: Alpha Weighted compound: 0.9529977998436279\n",
      "DEBUG:__main__:Layer 9: Norm compound: 5.350394248962402\n",
      "DEBUG:__main__:Layer 9: LogNorm compound: 0.7279879450798035\n",
      "DEBUG:__main__:Layer 9: Alpha compound: 2.412351108517952\n",
      "DEBUG:__main__:Layer 9: Alpha Weighted compound: 0.7522370217705565\n",
      "DEBUG:__main__:Layer 12: Norm compound: 6.344095706939697\n",
      "DEBUG:__main__:Layer 12: LogNorm compound: 0.8003528714179993\n",
      "DEBUG:__main__:Layer 12: Alpha compound: 2.9670583014132306\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "DEBUG:__main__:Layer 12: Alpha Weighted compound: 1.508835567105126\n",
      "DEBUG:__main__:Layer 14: Norm compound: 7.068286895751953\n",
      "DEBUG:__main__:Layer 14: LogNorm compound: 0.8488208055496216\n",
      "DEBUG:__main__:Layer 14: Alpha compound: 2.9324900726788465\n",
      "DEBUG:__main__:Layer 14: Alpha Weighted compound: 1.361677104373851\n",
      "DEBUG:__main__:Layer 17: Norm compound: 8.520411491394043\n",
      "DEBUG:__main__:Layer 17: LogNorm compound: 0.9300048351287842\n",
      "DEBUG:__main__:Layer 17: Alpha compound: 2.611071601496579\n",
      "DEBUG:__main__:Layer 17: Alpha Weighted compound: 1.3777809560439662\n",
      "DEBUG:__main__:Layer 19: Norm compound: 9.485450744628906\n",
      "DEBUG:__main__:Layer 19: LogNorm compound: 0.9768868684768677\n",
      "DEBUG:__main__:Layer 19: Alpha compound: 3.1774177732008844\n",
      "DEBUG:__main__:Layer 19: Alpha Weighted compound: 2.3009675942281937\n",
      "DEBUG:__main__:Layer 22: Norm compound: 9.905954360961914\n",
      "DEBUG:__main__:Layer 22: LogNorm compound: 0.9955974221229553\n",
      "DEBUG:__main__:Layer 22: Alpha compound: 3.136712953305263\n",
      "DEBUG:__main__:Layer 22: Alpha Weighted compound: 1.8153748589657621\n",
      "DEBUG:__main__:Layer 24: Norm compound: 9.433554649353027\n",
      "DEBUG:__main__:Layer 24: LogNorm compound: 0.9744208455085754\n",
      "DEBUG:__main__:Layer 24: Alpha compound: 3.654240384837193\n",
      "DEBUG:__main__:Layer 24: Alpha Weighted compound: 3.3055355345808763\n",
      "DEBUG:__main__:Layer 28: Norm: 57.03801345825195\n",
      "DEBUG:__main__:Layer 28: LogNorm: 1.7561644315719604\n",
      "DEBUG:__main__:Layer 28: Alpha: 2.3381178188805567\n",
      "DEBUG:__main__:Layer 28: Alpha Weighted: 3.636844506019609\n",
      "DEBUG:__main__:Layer 31: Norm: 43.40072250366211\n",
      "DEBUG:__main__:Layer 31: LogNorm: 1.6374969482421875\n",
      "DEBUG:__main__:Layer 31: Alpha: 2.186124848459917\n",
      "DEBUG:__main__:Layer 31: Alpha Weighted: 3.8836067216212076\n",
      "DEBUG:__main__:Layer 34: Norm: 38.5359001159668\n",
      "DEBUG:__main__:Layer 34: LogNorm: 1.5858654975891113\n",
      "DEBUG:__main__:Layer 34: Alpha: 2.7221347634801365\n",
      "DEBUG:__main__:Layer 34: Alpha Weighted: 4.799173592745528\n",
      "INFO:__main__:Norm: min: 3.4243783950805664, max: 57.03801345825195, avg: 8.586669921875\n",
      "INFO:__main__:Norm compound: min: 3.7364306449890137, max: 57.03801345825195, avg: 16.972938537597656\n",
      "INFO:__main__:LogNorm: min: 0.53458172082901, max: 1.7561644315719604, avg: 0.8640468120574951\n",
      "INFO:__main__:LogNorm compound: min: 0.5716044902801514, max: 1.7561644315719604, avg: 1.0408902168273926\n",
      "INFO:__main__:Alpha: min: 1.4229916456933371, max: 4.882618427826814, avg: 2.6896965121426315\n",
      "INFO:__main__:Alpha compound: min: 1.6193138952563046, max: 3.654240384837193, avg: 2.6287548746902396\n",
      "INFO:__main__:Alpha Weighted: min: 0.37454269723312644, max: 4.799173592745528, avg: 1.6453829975237424\n",
      "INFO:__main__:Alpha Weighted compound: min: 0.6126543377111549, max: 4.799173592745528, avg: 2.1923071329174553\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>norm</th>\n",
       "      <th>norm_compound</th>\n",
       "      <th>lognorm</th>\n",
       "      <th>lognorm_compound</th>\n",
       "      <th>alpha</th>\n",
       "      <th>alpha_compound</th>\n",
       "      <th>alpha_weighted</th>\n",
       "      <th>alpha_weighted_compound</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>8.58667</td>\n",
       "      <td>16.972939</td>\n",
       "      <td>0.864047</td>\n",
       "      <td>1.04089</td>\n",
       "      <td>2.689697</td>\n",
       "      <td>2.628755</td>\n",
       "      <td>1.645383</td>\n",
       "      <td>2.192307</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "      norm  norm_compound   lognorm  lognorm_compound     alpha  \\\n",
       "0  8.58667      16.972939  0.864047           1.04089  2.689697   \n",
       "\n",
       "   alpha_compound  alpha_weighted  alpha_weighted_compound  \n",
       "0        2.628755        1.645383                 2.192307  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:__main__:### Printing results ###\n",
      "DEBUG:__main__:Layer 4: Norm compound: 3.7364306449890137\n",
      "DEBUG:__main__:Layer 4: LogNorm compound: 0.5716044902801514\n",
      "DEBUG:__main__:Layer 4: Alpha compound: 1.6193138952563046\n",
      "DEBUG:__main__:Layer 4: Alpha Weighted compound: 0.6126543377111549\n",
      "DEBUG:__main__:Layer 7: Norm compound: 4.856048107147217\n",
      "DEBUG:__main__:Layer 7: LogNorm compound: 0.6854791641235352\n",
      "DEBUG:__main__:Layer 7: Alpha compound: 1.788024974756016\n",
      "DEBUG:__main__:Layer 7: Alpha Weighted compound: 0.9529977998436279\n",
      "DEBUG:__main__:Layer 9: Norm compound: 5.350394248962402\n",
      "DEBUG:__main__:Layer 9: LogNorm compound: 0.7279879450798035\n",
      "DEBUG:__main__:Layer 9: Alpha compound: 2.412351108517952\n",
      "DEBUG:__main__:Layer 9: Alpha Weighted compound: 0.7522370217705565\n",
      "DEBUG:__main__:Layer 12: Norm compound: 6.344095706939697\n",
      "DEBUG:__main__:Layer 12: LogNorm compound: 0.8003528714179993\n",
      "DEBUG:__main__:Layer 12: Alpha compound: 2.9670583014132306\n",
      "DEBUG:__main__:Layer 12: Alpha Weighted compound: 1.508835567105126\n",
      "DEBUG:__main__:Layer 14: Norm compound: 7.068286895751953\n",
      "DEBUG:__main__:Layer 14: LogNorm compound: 0.8488208055496216\n",
      "DEBUG:__main__:Layer 14: Alpha compound: 2.9324900726788465\n",
      "DEBUG:__main__:Layer 14: Alpha Weighted compound: 1.361677104373851\n",
      "DEBUG:__main__:Layer 17: Norm compound: 8.520411491394043\n",
      "DEBUG:__main__:Layer 17: LogNorm compound: 0.9300048351287842\n",
      "DEBUG:__main__:Layer 17: Alpha compound: 2.611071601496579\n",
      "DEBUG:__main__:Layer 17: Alpha Weighted compound: 1.3777809560439662\n",
      "DEBUG:__main__:Layer 19: Norm compound: 9.485450744628906\n",
      "DEBUG:__main__:Layer 19: LogNorm compound: 0.9768868684768677\n",
      "DEBUG:__main__:Layer 19: Alpha compound: 3.1774177732008844\n",
      "DEBUG:__main__:Layer 19: Alpha Weighted compound: 2.3009675942281937\n",
      "DEBUG:__main__:Layer 22: Norm compound: 9.905954360961914\n",
      "DEBUG:__main__:Layer 22: LogNorm compound: 0.9955974221229553\n",
      "DEBUG:__main__:Layer 22: Alpha compound: 3.136712953305263\n",
      "DEBUG:__main__:Layer 22: Alpha Weighted compound: 1.8153748589657621\n",
      "DEBUG:__main__:Layer 24: Norm compound: 9.433554649353027\n",
      "DEBUG:__main__:Layer 24: LogNorm compound: 0.9744208455085754\n",
      "DEBUG:__main__:Layer 24: Alpha compound: 3.654240384837193\n",
      "DEBUG:__main__:Layer 24: Alpha Weighted compound: 3.3055355345808763\n",
      "DEBUG:__main__:Layer 28: Norm: 57.03801345825195\n",
      "DEBUG:__main__:Layer 28: LogNorm: 1.7561644315719604\n",
      "DEBUG:__main__:Layer 28: Alpha: 2.3381178188805567\n",
      "DEBUG:__main__:Layer 28: Alpha Weighted: 3.636844506019609\n",
      "DEBUG:__main__:Layer 31: Norm: 43.40072250366211\n",
      "DEBUG:__main__:Layer 31: LogNorm: 1.6374969482421875\n",
      "DEBUG:__main__:Layer 31: Alpha: 2.186124848459917\n",
      "DEBUG:__main__:Layer 31: Alpha Weighted: 3.8836067216212076\n",
      "DEBUG:__main__:Layer 34: Norm: 38.5359001159668\n",
      "DEBUG:__main__:Layer 34: LogNorm: 1.5858654975891113\n",
      "DEBUG:__main__:Layer 34: Alpha: 2.7221347634801365\n",
      "DEBUG:__main__:Layer 34: Alpha Weighted: 4.799173592745528\n",
      "INFO:__main__:Norm: min: 3.4243783950805664, max: 57.03801345825195, avg: 8.586669921875\n",
      "INFO:__main__:Norm compound: min: 3.7364306449890137, max: 57.03801345825195, avg: 16.972938537597656\n",
      "INFO:__main__:LogNorm: min: 0.53458172082901, max: 1.7561644315719604, avg: 0.8640468120574951\n",
      "INFO:__main__:LogNorm compound: min: 0.5716044902801514, max: 1.7561644315719604, avg: 1.0408902168273926\n",
      "INFO:__main__:Alpha: min: 1.4229916456933371, max: 4.882618427826814, avg: 2.6896965121426315\n",
      "INFO:__main__:Alpha compound: min: 1.6193138952563046, max: 3.654240384837193, avg: 2.6287548746902396\n",
      "INFO:__main__:Alpha Weighted: min: 0.37454269723312644, max: 4.799173592745528, avg: 1.6453829975237424\n",
      "INFO:__main__:Alpha Weighted compound: min: 0.6126543377111549, max: 4.799173592745528, avg: 2.1923071329174553\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>layer_type</th>\n",
       "      <th>N</th>\n",
       "      <th>M</th>\n",
       "      <th>level</th>\n",
       "      <th>comment</th>\n",
       "      <th>norm</th>\n",
       "      <th>lognorm</th>\n",
       "      <th>alpha</th>\n",
       "      <th>alpha_weighted</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>layer_id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.59271</td>\n",
       "      <td>0.555422</td>\n",
       "      <td>1.99857</td>\n",
       "      <td>0.530849</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.79124</td>\n",
       "      <td>0.578781</td>\n",
       "      <td>1.52277</td>\n",
       "      <td>0.703118</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.42438</td>\n",
       "      <td>0.534582</td>\n",
       "      <td>2.13697</td>\n",
       "      <td>0.638202</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.79945</td>\n",
       "      <td>0.579721</td>\n",
       "      <td>1.47758</td>\n",
       "      <td>0.662307</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>4.2281</td>\n",
       "      <td>0.626146</td>\n",
       "      <td>1.5489</td>\n",
       "      <td>0.870814</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         layer_type   N   M        level      comment     norm   lognorm  \\\n",
       "layer_id                                                                   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  3.59271  0.555422   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  3.79124  0.578781   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  3.42438  0.534582   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  3.79945  0.579721   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level   4.2281  0.626146   \n",
       "\n",
       "            alpha alpha_weighted  \n",
       "layer_id                          \n",
       "4         1.99857       0.530849  \n",
       "4         1.52277       0.703118  \n",
       "4         2.13697       0.638202  \n",
       "4         1.47758       0.662307  \n",
       "4          1.5489       0.870814  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model_13 = models.vgg13(pretrained=True)\n",
    "watcher = ww.WeightWatcher(model=model_13, logger=logger)\n",
    "results_13 = watcher.analyze(alphas=True)\n",
    "summary_13 = watcher.get_summary(pandas=True)\n",
    "display(summary_13)\n",
    "details_13 = watcher.get_details(results=results_13)\n",
    "details_13.drop(columns=['slice', 'slice_count'], inplace=True)\n",
    "details_13.dropna(inplace=True)\n",
    "display(details_13.head())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 127,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 19:29:36,744 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:weightwatcher.weightwatcher:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 19:29:36,745 INFO Analyzing model\n",
      "INFO:weightwatcher.weightwatcher:Analyzing model\n",
      "2019-12-24 19:34:24,506 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 19:34:26,095 INFO Norm: min: 3.203004837036133, max: 51.2126579284668, avg: 7.801441669464111\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 3.203004837036133, max: 51.2126579284668, avg: 7.801441669464111\n",
      "2019-12-24 19:34:26,096 INFO Norm compound: min: 3.56429123878479, max: 51.2126579284668, avg: 14.09792709350586\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 3.56429123878479, max: 51.2126579284668, avg: 14.09792709350586\n",
      "2019-12-24 19:34:26,097 INFO LogNorm: min: 0.5055575966835022, max: 1.7093772888183594, avg: 0.8381870985031128\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.5055575966835022, max: 1.7093772888183594, avg: 0.8381870985031128\n",
      "2019-12-24 19:34:26,098 INFO LogNorm compound: min: 0.5506355166435242, max: 1.7093772888183594, avg: 0.9792784452438354\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.5506355166435242, max: 1.7093772888183594, avg: 0.9792784452438354\n",
      "2019-12-24 19:34:26,100 INFO Alpha: min: 1.525053187957775, max: 4.343339437091599, avg: 2.89233801062374\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.525053187957775, max: 4.343339437091599, avg: 2.89233801062374\n",
      "2019-12-24 19:34:26,101 INFO Alpha compound: min: 1.7535254182659035, max: 3.8419123407107967, avg: 2.767717801865902\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.7535254182659035, max: 3.8419123407107967, avg: 2.767717801865902\n",
      "2019-12-24 19:34:26,102 INFO Alpha Weighted: min: 0.35369982858727583, max: 3.7137959804872724, avg: 1.405072984118814\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: 0.35369982858727583, max: 3.7137959804872724, avg: 1.405072984118814\n",
      "2019-12-24 19:34:26,102 INFO Alpha Weighted compound: min: 0.4626615096713753, max: 3.7137959804872724, avg: 1.7930749068678844\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: 0.4626615096713753, max: 3.7137959804872724, avg: 1.7930749068678844\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>norm</th>\n",
       "      <th>norm_compound</th>\n",
       "      <th>lognorm</th>\n",
       "      <th>lognorm_compound</th>\n",
       "      <th>alpha</th>\n",
       "      <th>alpha_compound</th>\n",
       "      <th>alpha_weighted</th>\n",
       "      <th>alpha_weighted_compound</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>7.801442</td>\n",
       "      <td>14.097927</td>\n",
       "      <td>0.838187</td>\n",
       "      <td>0.979278</td>\n",
       "      <td>2.892338</td>\n",
       "      <td>2.767718</td>\n",
       "      <td>1.405073</td>\n",
       "      <td>1.793075</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       norm  norm_compound   lognorm  lognorm_compound     alpha  \\\n",
       "0  7.801442      14.097927  0.838187          0.979278  2.892338   \n",
       "\n",
       "   alpha_compound  alpha_weighted  alpha_weighted_compound  \n",
       "0        2.767718        1.405073                 1.793075  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 19:34:26,136 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 19:34:27,653 INFO Norm: min: 3.203004837036133, max: 51.2126579284668, avg: 7.801441669464111\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 3.203004837036133, max: 51.2126579284668, avg: 7.801441669464111\n",
      "2019-12-24 19:34:27,654 INFO Norm compound: min: 3.56429123878479, max: 51.2126579284668, avg: 14.09792709350586\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 3.56429123878479, max: 51.2126579284668, avg: 14.09792709350586\n",
      "2019-12-24 19:34:27,655 INFO LogNorm: min: 0.5055575966835022, max: 1.7093772888183594, avg: 0.8381870985031128\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.5055575966835022, max: 1.7093772888183594, avg: 0.8381870985031128\n",
      "2019-12-24 19:34:27,656 INFO LogNorm compound: min: 0.5506355166435242, max: 1.7093772888183594, avg: 0.9792784452438354\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.5506355166435242, max: 1.7093772888183594, avg: 0.9792784452438354\n",
      "2019-12-24 19:34:27,657 INFO Alpha: min: 1.525053187957775, max: 4.343339437091599, avg: 2.89233801062374\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.525053187957775, max: 4.343339437091599, avg: 2.89233801062374\n",
      "2019-12-24 19:34:27,658 INFO Alpha compound: min: 1.7535254182659035, max: 3.8419123407107967, avg: 2.767717801865902\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.7535254182659035, max: 3.8419123407107967, avg: 2.767717801865902\n",
      "2019-12-24 19:34:27,659 INFO Alpha Weighted: min: 0.35369982858727583, max: 3.7137959804872724, avg: 1.405072984118814\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: 0.35369982858727583, max: 3.7137959804872724, avg: 1.405072984118814\n",
      "2019-12-24 19:34:27,660 INFO Alpha Weighted compound: min: 0.4626615096713753, max: 3.7137959804872724, avg: 1.7930749068678844\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: 0.4626615096713753, max: 3.7137959804872724, avg: 1.7930749068678844\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>layer_type</th>\n",
       "      <th>N</th>\n",
       "      <th>M</th>\n",
       "      <th>level</th>\n",
       "      <th>comment</th>\n",
       "      <th>norm</th>\n",
       "      <th>lognorm</th>\n",
       "      <th>alpha</th>\n",
       "      <th>alpha_weighted</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>layer_id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.28156</td>\n",
       "      <td>0.516081</td>\n",
       "      <td>3.14242</td>\n",
       "      <td>1.03563</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.78437</td>\n",
       "      <td>0.577994</td>\n",
       "      <td>2.69269</td>\n",
       "      <td>1.43223</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.5147</td>\n",
       "      <td>0.545889</td>\n",
       "      <td>1.79157</td>\n",
       "      <td>0.60329</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.59088</td>\n",
       "      <td>0.555201</td>\n",
       "      <td>2.64471</td>\n",
       "      <td>1.39963</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>4.12886</td>\n",
       "      <td>0.61583</td>\n",
       "      <td>2.85539</td>\n",
       "      <td>1.8627</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         layer_type   N   M        level      comment     norm   lognorm  \\\n",
       "layer_id                                                                   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  3.28156  0.516081   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  3.78437  0.577994   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level   3.5147  0.545889   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  3.59088  0.555201   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  4.12886   0.61583   \n",
       "\n",
       "            alpha alpha_weighted  \n",
       "layer_id                          \n",
       "4         3.14242        1.03563  \n",
       "4         2.69269        1.43223  \n",
       "4         1.79157        0.60329  \n",
       "4         2.64471        1.39963  \n",
       "4         2.85539         1.8627  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model_16 = models.vgg16(pretrained=True)\n",
    "watcher = ww.WeightWatcher(model=model_16)\n",
    "results_16 = watcher.analyze(alphas=True)\n",
    "summary_16 = watcher.get_summary(pandas=True)\n",
    "display(summary_16)\n",
    "details_16 = watcher.get_details(results=results_16)\n",
    "details_16.drop(columns=['slice', 'slice_count'], inplace=True)\n",
    "details_16.dropna(inplace=True)\n",
    "display(details_16.head())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 128,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 19:34:29,861 INFO \n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "INFO:weightwatcher.weightwatcher:\n",
      "\n",
      "python      version 3.7.3 (default, Mar 27 2019, 16:54:48) \n",
      "[Clang 4.0.1 (tags/RELEASE_401/final)]\n",
      "numpy       version 1.16.4\n",
      "tensforflow version 1.13.1\n",
      "keras       version 2.2.4\n",
      "2019-12-24 19:34:29,862 INFO Analyzing model\n",
      "INFO:weightwatcher.weightwatcher:Analyzing model\n",
      "2019-12-24 19:40:14,064 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 19:40:16,103 INFO Norm: min: 3.2178001403808594, max: 47.974002838134766, avg: 7.2684478759765625\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 3.2178001403808594, max: 47.974002838134766, avg: 7.2684478759765625\n",
      "2019-12-24 19:40:16,104 INFO Norm compound: min: 3.559237241744995, max: 47.974002838134766, avg: 12.337833404541016\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 3.559237241744995, max: 47.974002838134766, avg: 12.337833404541016\n",
      "2019-12-24 19:40:16,106 INFO LogNorm: min: 0.507559061050415, max: 1.6810059547424316, avg: 0.8185057640075684\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.507559061050415, max: 1.6810059547424316, avg: 0.8185057640075684\n",
      "2019-12-24 19:40:16,107 INFO LogNorm compound: min: 0.5500112771987915, max: 1.6810059547424316, avg: 0.9365273714065552\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.5500112771987915, max: 1.6810059547424316, avg: 0.9365273714065552\n",
      "2019-12-24 19:40:16,109 INFO Alpha: min: 1.4433522167552777, max: 5.027317242480103, avg: 2.964672695949022\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.4433522167552777, max: 5.027317242480103, avg: 2.964672695949022\n",
      "2019-12-24 19:40:16,109 INFO Alpha compound: min: 1.5696912807219614, max: 4.10398537172971, avg: 2.8479756155799745\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.5696912807219614, max: 4.10398537172971, avg: 2.8479756155799745\n",
      "2019-12-24 19:40:16,111 INFO Alpha Weighted: min: 0.2145263550533989, max: 3.591568654707832, avg: 1.158888231628616\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: 0.2145263550533989, max: 3.591568654707832, avg: 1.158888231628616\n",
      "2019-12-24 19:40:16,112 INFO Alpha Weighted compound: min: 0.3858703649968457, max: 3.591568654707832, avg: 1.5002341935342836\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: 0.3858703649968457, max: 3.591568654707832, avg: 1.5002341935342836\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>norm</th>\n",
       "      <th>norm_compound</th>\n",
       "      <th>lognorm</th>\n",
       "      <th>lognorm_compound</th>\n",
       "      <th>alpha</th>\n",
       "      <th>alpha_compound</th>\n",
       "      <th>alpha_weighted</th>\n",
       "      <th>alpha_weighted_compound</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>7.268448</td>\n",
       "      <td>12.337833</td>\n",
       "      <td>0.818506</td>\n",
       "      <td>0.936527</td>\n",
       "      <td>2.964673</td>\n",
       "      <td>2.847976</td>\n",
       "      <td>1.158888</td>\n",
       "      <td>1.500234</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       norm  norm_compound   lognorm  lognorm_compound     alpha  \\\n",
       "0  7.268448      12.337833  0.818506          0.936527  2.964673   \n",
       "\n",
       "   alpha_compound  alpha_weighted  alpha_weighted_compound  \n",
       "0        2.847976        1.158888                 1.500234  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "2019-12-24 19:40:16,138 INFO ### Printing results ###\n",
      "INFO:weightwatcher.weightwatcher:### Printing results ###\n",
      "2019-12-24 19:40:18,147 INFO Norm: min: 3.2178001403808594, max: 47.974002838134766, avg: 7.2684478759765625\n",
      "INFO:weightwatcher.weightwatcher:Norm: min: 3.2178001403808594, max: 47.974002838134766, avg: 7.2684478759765625\n",
      "2019-12-24 19:40:18,148 INFO Norm compound: min: 3.559237241744995, max: 47.974002838134766, avg: 12.337833404541016\n",
      "INFO:weightwatcher.weightwatcher:Norm compound: min: 3.559237241744995, max: 47.974002838134766, avg: 12.337833404541016\n",
      "2019-12-24 19:40:18,150 INFO LogNorm: min: 0.507559061050415, max: 1.6810059547424316, avg: 0.8185057640075684\n",
      "INFO:weightwatcher.weightwatcher:LogNorm: min: 0.507559061050415, max: 1.6810059547424316, avg: 0.8185057640075684\n",
      "2019-12-24 19:40:18,150 INFO LogNorm compound: min: 0.5500112771987915, max: 1.6810059547424316, avg: 0.9365273714065552\n",
      "INFO:weightwatcher.weightwatcher:LogNorm compound: min: 0.5500112771987915, max: 1.6810059547424316, avg: 0.9365273714065552\n",
      "2019-12-24 19:40:18,152 INFO Alpha: min: 1.4433522167552777, max: 5.027317242480103, avg: 2.964672695949022\n",
      "INFO:weightwatcher.weightwatcher:Alpha: min: 1.4433522167552777, max: 5.027317242480103, avg: 2.964672695949022\n",
      "2019-12-24 19:40:18,153 INFO Alpha compound: min: 1.5696912807219614, max: 4.10398537172971, avg: 2.8479756155799745\n",
      "INFO:weightwatcher.weightwatcher:Alpha compound: min: 1.5696912807219614, max: 4.10398537172971, avg: 2.8479756155799745\n",
      "2019-12-24 19:40:18,154 INFO Alpha Weighted: min: 0.2145263550533989, max: 3.591568654707832, avg: 1.158888231628616\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted: min: 0.2145263550533989, max: 3.591568654707832, avg: 1.158888231628616\n",
      "2019-12-24 19:40:18,155 INFO Alpha Weighted compound: min: 0.3858703649968457, max: 3.591568654707832, avg: 1.5002341935342836\n",
      "INFO:weightwatcher.weightwatcher:Alpha Weighted compound: min: 0.3858703649968457, max: 3.591568654707832, avg: 1.5002341935342836\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>layer_type</th>\n",
       "      <th>N</th>\n",
       "      <th>M</th>\n",
       "      <th>level</th>\n",
       "      <th>comment</th>\n",
       "      <th>norm</th>\n",
       "      <th>lognorm</th>\n",
       "      <th>alpha</th>\n",
       "      <th>alpha_weighted</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>layer_id</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.34719</td>\n",
       "      <td>0.524681</td>\n",
       "      <td>1.44335</td>\n",
       "      <td>0.374001</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.65029</td>\n",
       "      <td>0.562327</td>\n",
       "      <td>1.52803</td>\n",
       "      <td>0.60588</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.29998</td>\n",
       "      <td>0.518511</td>\n",
       "      <td>2.13373</td>\n",
       "      <td>0.535882</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>3.55341</td>\n",
       "      <td>0.550646</td>\n",
       "      <td>1.51759</td>\n",
       "      <td>0.565788</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>CONV2D</td>\n",
       "      <td>64</td>\n",
       "      <td>64</td>\n",
       "      <td>LEVEL.SLICE</td>\n",
       "      <td>Slice level</td>\n",
       "      <td>4.22471</td>\n",
       "      <td>0.625797</td>\n",
       "      <td>1.55643</td>\n",
       "      <td>0.895931</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         layer_type   N   M        level      comment     norm   lognorm  \\\n",
       "layer_id                                                                   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  3.34719  0.524681   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  3.65029  0.562327   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  3.29998  0.518511   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  3.55341  0.550646   \n",
       "4            CONV2D  64  64  LEVEL.SLICE  Slice level  4.22471  0.625797   \n",
       "\n",
       "            alpha alpha_weighted  \n",
       "layer_id                          \n",
       "4         1.44335       0.374001  \n",
       "4         1.52803        0.60588  \n",
       "4         2.13373       0.535882  \n",
       "4         1.51759       0.565788  \n",
       "4         1.55643       0.895931  "
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "model_19 = models.vgg19(pretrained=True)\n",
    "watcher = ww.WeightWatcher(model=model_19)\n",
    "results_19 = watcher.analyze(alphas=True)\n",
    "summary_19 = watcher.get_summary(pandas=True)\n",
    "display(summary_19)\n",
    "details_19 = watcher.get_details(results=results_19)\n",
    "details_19.drop(columns=['slice', 'slice_count'], inplace=True)\n",
    "details_19.dropna(inplace=True)\n",
    "display(details_19.head())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 131,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Linear(in_features=25088, out_features=4096, bias=True)\n",
      "Linear(in_features=4096, out_features=4096, bias=True)\n",
      "Linear(in_features=4096, out_features=1000, bias=True)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "11"
      ]
     },
     "execution_count": 131,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "import torch\n",
    "valid_types = [torch.nn.modules.conv.Conv2d, torch.nn.modules.linear.Linear]\n",
    "\n",
    "num_layers = 0\n",
    "for m in model_11.modules():\n",
    "    if(type(m) in valid_types):\n",
    "        num_layers +=1\n",
    "        print(m)\n",
    "        \n",
    "num_layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 132,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Linear(in_features=25088, out_features=4096, bias=True)\n",
      "Linear(in_features=4096, out_features=4096, bias=True)\n",
      "Linear(in_features=4096, out_features=1000, bias=True)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "13"
      ]
     },
     "execution_count": 132,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "num_layers = 0\n",
    "for m in model_13.modules():\n",
    "    if(type(m) in valid_types):\n",
    "        num_layers +=1\n",
    "        print(m)\n",
    "        \n",
    "num_layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 133,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Linear(in_features=25088, out_features=4096, bias=True)\n",
      "Linear(in_features=4096, out_features=4096, bias=True)\n",
      "Linear(in_features=4096, out_features=1000, bias=True)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "16"
      ]
     },
     "execution_count": 133,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "num_layers = 0\n",
    "for m in model_16.modules():\n",
    "    if(type(m) in valid_types):\n",
    "        num_layers +=1\n",
    "        print(m)\n",
    "        \n",
    "num_layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 134,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Conv2d(3, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(64, 64, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(64, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(128, 128, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(128, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(256, 256, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(256, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Conv2d(512, 512, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "Linear(in_features=25088, out_features=4096, bias=True)\n",
      "Linear(in_features=4096, out_features=4096, bias=True)\n",
      "Linear(in_features=4096, out_features=1000, bias=True)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "19"
      ]
     },
     "execution_count": 134,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "num_layers = 0\n",
    "for m in model_19.modules():\n",
    "    if(type(m) in valid_types):\n",
    "        num_layers +=1\n",
    "        print(m)\n",
    "        \n",
    "num_layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 142,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0, 0.5, 'alpha')"
      ]
     },
     "execution_count": 142,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "x = details_11.index.to_numpy()\n",
    "y = details_11.alpha.to_numpy()\n",
    "plt.scatter(x,y)\n",
    "\n",
    "x = details_19.index.to_numpy()\n",
    "y = details_19.alpha.to_numpy()\n",
    "plt.scatter(x,y)\n",
    "\n",
    "plt.title(r\"VGG 11 vs VGG 19  layer size vs depth\")\n",
    "plt.xlabel('depth')\n",
    "plt.ylabel('alpha')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 160,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'alpha')"
      ]
     },
     "execution_count": 160,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y = details_11.alpha.to_numpy()\n",
    "plt.hist(y, bins=20, alpha=0.5, color='blue', label='vgg11', density=True);\n",
    "y = details_19.alpha.to_numpy()\n",
    "plt.hist(y, bins=20, alpha=0.5, color='green', label='vgg19', density=True);\n",
    "plt.legend()\n",
    "plt.xlabel(\"alpha\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 159,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'alpha')"
      ]
     },
     "execution_count": 159,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y = details_11.alpha.to_numpy()\n",
    "plt.hist(y, bins=20, alpha=0.5, color='blue', label='vgg11', density=True);\n",
    "y = details_13.alpha.to_numpy()\n",
    "plt.hist(y, bins=20, alpha=0.5, color='purple', label='vgg13', density=True);\n",
    "y = details_16.alpha.to_numpy()\n",
    "plt.hist(y, bins=20, alpha=0.5, color='teal', label='vgg16', density=True);\n",
    "y = details_19.alpha.to_numpy()\n",
    "plt.hist(y, bins=20, alpha=0.5, color='green', label='vgg19', density=True);\n",
    "plt.legend()\n",
    "plt.xlabel(\"alpha\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 161,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Text(0.5, 0, 'lognorm')"
      ]
     },
     "execution_count": 161,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y = details_11.lognorm.to_numpy()\n",
    "plt.hist(y, bins=20, alpha=0.5, color='blue', label='vgg11', density=True);\n",
    "y = details_19.lognorm.to_numpy()\n",
    "plt.hist(y, bins=20, alpha=0.5, color='green', label='vgg19', density=True);\n",
    "plt.legend()\n",
    "plt.xlabel(\"lognorm\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### TODO:  redo using neeew design, easier design"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 5. Conclusion\n",
    "\n",
    "WeightWatcher helps you choose the best pretrained model for your needs.\n",
    "\n",
    "You can use WeightWatcher to compare several pretrained models and choose the one with the lowest Log Norm."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}