TrainingSample.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "source": [
    "# Training and calculating sharpness\n",
    "\n",
    "In this notebook we provide two examples of training the model and calculating the sharpness of the minimum obtained. Both examples are on FashionMNIST dataset, with *SimpleBatch* architecture (neural network with 6 convolutional layers, 2 linear layers and batch normalization). The first example uses SGD optimizer, while the second example uses Sharpness Aware Minimization (SAM) with SGD optimizer as a background optimizer. Each of these can be easily changed in the notebook and this notebook is desinged to demonstrate the usage of functions. \n",
    "\n",
    "For systematical trainings and sharpness calculations, you can run either the bash files withinthe repository or refer to runnig main.py according to the README instructions."
   ],
   "metadata": {
    "id": "8t2By2_zONiN",
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "40zPtJsx_njC",
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "## Drive mouting and imports"
   ]
  },
  {
   "cell_type": "code",
   "source": [
    "# libraries import\n",
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.utils.data.dataloader as dataloader\n",
    "import torch.optim as optim\n",
    "from torch.utils.data import TensorDataset\n",
    "from torch.autograd import Variable\n",
    "from torchvision import transforms\n",
    "from torchvision.datasets import CIFAR10\n",
    "from torchvision.datasets import FashionMNIST\n",
    "\n",
    "import matplotlib.pyplot as plt\n",
    "import random\n",
    "from itertools import product\n",
    "import sys \n",
    "import os\n",
    "from datetime import datetime\n",
    "\n",
    "# setting the device\n",
    "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
    "print(device)"
   ],
   "metadata": {
    "id": "ftaYG9uWd8TZ",
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "outputId": "9cd6fa0c-a1af-452e-ea89-35a62317d15f",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": 1,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "cpu\n"
     ]
    }
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "id": "-IyZV96e_k_u",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [],
   "source": [
    "# import of models and other helpers\n",
    "from models import *\n",
    "from helpers import *\n",
    "\n",
    "# optimizer imports that can be used for additional runs\n",
    "from optimizers.adashift import AdaShift            # code taken from: https://github.com/MichaelKonobeev/adashift\n",
    "from optimizers.adabound import AdaBound            # code taken from: https://github.com/Luolc/AdaBound\n",
    "from optimizers.sam import SAM                      # code taken from: https://github.com/davda54/sam\n",
    "\n",
    "# import functions for calculating sharpness \n",
    "from sharpness.Minimum import effective as minimum_shaprness_eff        # code taken from: https://github.com/ibayashi-hikaru/minimum-sharpness\n",
    "\n",
    "# getting the path to checkpoint folder from helpers file\n",
    "checkpoints_dir = 'checkpoints_test'"
   ]
  },
  {
   "cell_type": "markdown",
   "source": [
    "Here, we specify which architecture, dataset and maximal epoch number we use. In order to do training for different architecture or dataset, only this cell should be changed."
   ],
   "metadata": {
    "id": "sw8wfrXgPRp8",
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "id": "TQeXXMEW3tOU",
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "outputId": "09b97175-b835-45bc-e7c2-7897d05ae465",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "checkpoints_test/FashionMNIST/SimpleBatch\n"
     ]
    }
   ],
   "source": [
    "ARCHITECTURE = 'SimpleBatch'                                    # Other possibilities: 'MiddleBatch', 'ComplexBatch'\n",
    "DATASET = 'FashionMNIST'                                        # Other possibilities: 'CIFAR10'\n",
    "MAX_EPOCH = 200                                                 # Our trainings are done for 200 epochs\n",
    "TRAIN_BATCH_SIZE = 2**7\n",
    "\n",
    "# location to store current configuration results\n",
    "store_dir = os.path.join(checkpoints_dir, f'{DATASET}/{ARCHITECTURE}')\n",
    "print(store_dir)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "vWpl_tWl_Qvn",
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "## Loading the dataset and creating dataloaders"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "id": "bS48TLHJ_Qvs",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz\n",
      "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-images-idx3-ubyte.gz to ./data/FashionMNIST/raw/train-images-idx3-ubyte.gz\n"
     ]
    },
    {
     "data": {
      "text/plain": "  0%|          | 0/26421880 [00:00<?, ?it/s]",
      "application/vnd.jupyter.widget-view+json": {
       "version_major": 2,
       "version_minor": 0,
       "model_id": "d3a8884694cc41bf93d30a9423f8b2cc"
      }
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting ./data/FashionMNIST/raw/train-images-idx3-ubyte.gz to ./data/FashionMNIST/raw\n",
      "\n",
      "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-labels-idx1-ubyte.gz\n",
      "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/train-labels-idx1-ubyte.gz to ./data/FashionMNIST/raw/train-labels-idx1-ubyte.gz\n"
     ]
    },
    {
     "data": {
      "text/plain": "  0%|          | 0/29515 [00:00<?, ?it/s]",
      "application/vnd.jupyter.widget-view+json": {
       "version_major": 2,
       "version_minor": 0,
       "model_id": "d72331e29f7d4af59d0129f165eff266"
      }
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting ./data/FashionMNIST/raw/train-labels-idx1-ubyte.gz to ./data/FashionMNIST/raw\n",
      "\n",
      "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-images-idx3-ubyte.gz\n",
      "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-images-idx3-ubyte.gz to ./data/FashionMNIST/raw/t10k-images-idx3-ubyte.gz\n"
     ]
    },
    {
     "data": {
      "text/plain": "  0%|          | 0/4422102 [00:00<?, ?it/s]",
      "application/vnd.jupyter.widget-view+json": {
       "version_major": 2,
       "version_minor": 0,
       "model_id": "cbc6beebe4a8453d8091b295ce9b8d5c"
      }
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting ./data/FashionMNIST/raw/t10k-images-idx3-ubyte.gz to ./data/FashionMNIST/raw\n",
      "\n",
      "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-labels-idx1-ubyte.gz\n",
      "Downloading http://fashion-mnist.s3-website.eu-central-1.amazonaws.com/t10k-labels-idx1-ubyte.gz to ./data/FashionMNIST/raw/t10k-labels-idx1-ubyte.gz\n"
     ]
    },
    {
     "data": {
      "text/plain": "  0%|          | 0/5148 [00:00<?, ?it/s]",
      "application/vnd.jupyter.widget-view+json": {
       "version_major": 2,
       "version_minor": 0,
       "model_id": "b95c462b70db48d18e4a66352ff4d052"
      }
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting ./data/FashionMNIST/raw/t10k-labels-idx1-ubyte.gz to ./data/FashionMNIST/raw\n",
      "\n"
     ]
    }
   ],
   "source": [
    "VAL_BATCH_SIZE = 1000\n",
    "\n",
    "if DATASET == 'CIFAR10':\n",
    "    #loading datasets\n",
    "    train_data =  CIFAR10('./data', train=True, download=True, transform=transforms.Compose([\n",
    "        transforms.ToTensor(), # ToTensor does min-max normalization. \n",
    "    ]), )\n",
    "\n",
    "    test_data = CIFAR10('./data', train=False, download=True, transform=transforms.Compose([\n",
    "        transforms.ToTensor(), # ToTensor does min-max normalization. \n",
    "    ]), )\n",
    "\n",
    "    #creating dataLoaders\n",
    "    train_loader = dataloader.DataLoader(train_data, shuffle=True, batch_size=TRAIN_BATCH_SIZE)\n",
    "    test_loader = dataloader.DataLoader(test_data, shuffle=False, batch_size=VAL_BATCH_SIZE)\n",
    "\n",
    "if DATASET == 'FashionMNIST':\n",
    "    #loading datasets\n",
    "    train_data =  FashionMNIST('./data', train=True, download=True, transform=transforms.Compose([\n",
    "        transforms.ToTensor(), # ToTensor does min-max normalization. \n",
    "    ]), )\n",
    "\n",
    "    test_data = FashionMNIST('./data', train=False, download=True, transform=transforms.Compose([\n",
    "        transforms.ToTensor(), # ToTensor does min-max normalization. \n",
    "    ]), )\n",
    "\n",
    "    #creating dataLoaders\n",
    "    train_loader = dataloader.DataLoader(train_data, shuffle=True, batch_size=TRAIN_BATCH_SIZE)\n",
    "    test_loader = dataloader.DataLoader(test_data, shuffle=False, batch_size=VAL_BATCH_SIZE)   "
   ]
  },
  {
   "cell_type": "code",
   "source": [
    "data = preprocess_data_for_sharpness(train_data, DATASET, device)"
   ],
   "metadata": {
    "id": "73ROsMFPUOw6",
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "outputId": "0c6c1994-85d3-4133-ab98-fb1305d6f0d4",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": 5,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Preporcessing dataset FashionMNIST in order to calculate sharpness...\n",
      "Time needed 0:00:08.131259\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/ina/Repos/OptML/helpers.py:214: UserWarning: To copy construct from a tensor, it is recommended to use sourceTensor.clone().detach() or sourceTensor.clone().detach().requires_grad_(True), rather than torch.tensor(sourceTensor).\n",
      "  y = torch.tensor(train_data.targets)\n"
     ]
    }
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "iV7B-lKD3P_x",
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "## Training model on SGD without SAM\n"
   ]
  },
  {
   "cell_type": "code",
   "source": [
    "# Getting the model based on the architecture and the dataset\n",
    "model = get_model(ARCHITECTURE, DATASET).to(device)\n",
    "\n",
    "# Specifying the optimizer. This can be changed to any optimizer that is supported by PyTorch.\n",
    "# In order to use AdaBound:   optimizer = AdaBound(model.parameters())\n",
    "# In order to use AdaShift:   optimizer = AdaShift(model.parameters())\n",
    "# In order to use PHB:        optimizer = torch.optim.SGD(model.parameters(), lr=0.1, momentum=0.8)\n",
    "# In order to use Adagrad:    optimizer = torch.optim.Adagrad(model.parameters())\n",
    "optimizer = torch.optim.SGD(model.parameters(), lr=0.1)\n",
    "optimizer_name = 'SGD'\n",
    "\n",
    "model = train(model, optimizer, train_loader=train_loader, device=device, epoch_num=MAX_EPOCH, max_nbr_epochs=MAX_EPOCH, path=optimizer_name, val_dataloader=test_loader, sam=False, dir_path=store_dir)"
   ],
   "metadata": {
    "id": "f52IUqRyQZDr",
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "outputId": "0d1e7ee4-1195-448c-9b0e-762e359ab9a0",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": null,
   "outputs": [
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "new version 2\n",
      " Train Epoch: 0 [60000/60000 (100%)]\tLoss: 0.379745\t accuracy: 86.46%\t Time: 0:00:06.316012\n",
      "\t Test loss: 0.324387663602829 \t Test accuracy: 88.04000091552734\n",
      "\n",
      " Train Epoch: 1 [60000/60000 (100%)]\tLoss: 0.258497\t accuracy: 90.68%\t Time: 0:00:06.137482\n",
      "\t Test loss: 0.2614754170179367 \t Test accuracy: 90.52999877929688\n",
      "\n",
      " Train Epoch: 2 [60000/60000 (100%)]\tLoss: 0.217269\t accuracy: 92.01%\t Time: 0:00:06.212931\n",
      "\t Test loss: 0.24827186167240142 \t Test accuracy: 91.0\n",
      "\n",
      " Train Epoch: 3 [60000/60000 (100%)]\tLoss: 0.187217\t accuracy: 93.18%\t Time: 0:00:06.872480\n",
      "\t Test loss: 0.2555343255400658 \t Test accuracy: 91.11000061035156\n",
      "\n",
      " Train Epoch: 4 [60000/60000 (100%)]\tLoss: 0.162010\t accuracy: 94.06%\t Time: 0:00:06.174041\n",
      "\t Test loss: 0.22924970090389252 \t Test accuracy: 91.86000061035156\n",
      "\n",
      " Train Epoch: 5 [60000/60000 (100%)]\tLoss: 0.139343\t accuracy: 94.91%\t Time: 0:00:06.108058\n",
      "\t Test loss: 0.23696358650922775 \t Test accuracy: 91.8699951171875\n",
      "\n",
      " Train Epoch: 6 [60000/60000 (100%)]\tLoss: 0.122351\t accuracy: 95.44%\t Time: 0:00:06.234192\n",
      "\t Test loss: 0.24584759324789046 \t Test accuracy: 91.97000122070312\n",
      "\n",
      " Train Epoch: 7 [60000/60000 (100%)]\tLoss: 0.105100\t accuracy: 96.09%\t Time: 0:00:06.172856\n",
      "\t Test loss: 0.2486617535352707 \t Test accuracy: 92.12999725341797\n",
      "\n",
      " Train Epoch: 8 [60000/60000 (100%)]\tLoss: 0.090020\t accuracy: 96.77%\t Time: 0:00:06.155411\n",
      "\t Test loss: 0.27641472965478897 \t Test accuracy: 91.72999572753906\n",
      "\n",
      " Train Epoch: 9 [60000/60000 (100%)]\tLoss: 0.074638\t accuracy: 97.27%\t Time: 0:00:06.067835\n",
      "\t Test loss: 0.26530072242021563 \t Test accuracy: 92.22000122070312\n",
      "\n",
      " Train Epoch: 10 [60000/60000 (100%)]\tLoss: 0.064212\t accuracy: 97.64%\t Time: 0:00:06.165391\n",
      "\t Test loss: 0.28607737123966215 \t Test accuracy: 91.8499984741211\n",
      "\n",
      " Train Epoch: 11 [60000/60000 (100%)]\tLoss: 0.052745\t accuracy: 98.09%\t Time: 0:00:06.099154\n",
      "\t Test loss: 0.2927433207631111 \t Test accuracy: 91.94999694824219\n",
      "\n",
      " Train Epoch: 12 [60000/60000 (100%)]\tLoss: 0.042727\t accuracy: 98.41%\t Time: 0:00:06.142994\n",
      "\t Test loss: 0.33663240969181063 \t Test accuracy: 92.25\n",
      "\n",
      " Train Epoch: 13 [60000/60000 (100%)]\tLoss: 0.036211\t accuracy: 98.68%\t Time: 0:00:06.148901\n",
      "\t Test loss: 0.33118622899055483 \t Test accuracy: 92.15999603271484\n",
      "\n",
      " Train Epoch: 14 [60000/60000 (100%)]\tLoss: 0.030318\t accuracy: 98.92%\t Time: 0:00:06.096683\n",
      "\t Test loss: 0.34282347857952117 \t Test accuracy: 91.75999450683594\n",
      "\n",
      " Train Epoch: 15 [60000/60000 (100%)]\tLoss: 0.025354\t accuracy: 99.12%\t Time: 0:00:06.203493\n",
      "\t Test loss: 0.38046678602695466 \t Test accuracy: 91.93000030517578\n",
      "\n",
      " Train Epoch: 16 [60000/60000 (100%)]\tLoss: 0.026110\t accuracy: 99.06%\t Time: 0:00:06.384160\n",
      "\t Test loss: 0.413285493850708 \t Test accuracy: 91.57999420166016\n",
      "\n",
      " Train Epoch: 17 [60000/60000 (100%)]\tLoss: 0.021081\t accuracy: 99.26%\t Time: 0:00:06.517120\n",
      "\t Test loss: 0.3901343882083893 \t Test accuracy: 91.97999572753906\n",
      "\n",
      " Train Epoch: 18 [60000/60000 (100%)]\tLoss: 0.018013\t accuracy: 99.37%\t Time: 0:00:06.064202\n",
      "\t Test loss: 0.39350261688232424 \t Test accuracy: 92.29000091552734\n",
      "\n",
      " Train Epoch: 19 [60000/60000 (100%)]\tLoss: 0.009204\t accuracy: 99.73%\t Time: 0:00:06.207876\n",
      "\t Test loss: 0.4119370460510254 \t Test accuracy: 92.08999633789062\n",
      "\n",
      " Train Epoch: 20 [60000/60000 (100%)]\tLoss: 0.004157\t accuracy: 99.90%\t Time: 0:00:06.151054\n",
      "\t Test loss: 0.41050626039505006 \t Test accuracy: 92.30999755859375\n",
      "\n",
      " Train Epoch: 21 [60000/60000 (100%)]\tLoss: 0.003569\t accuracy: 99.92%\t Time: 0:00:06.113181\n",
      "\t Test loss: 0.42936062514781953 \t Test accuracy: 92.47000122070312\n",
      "\n",
      " Train Epoch: 22 [60000/60000 (100%)]\tLoss: 0.001613\t accuracy: 99.98%\t Time: 0:00:06.112760\n",
      "\t Test loss: 0.4451223939657211 \t Test accuracy: 92.54999542236328\n",
      "\n",
      " Train Epoch: 23 [60000/60000 (100%)]\tLoss: 0.001306\t accuracy: 99.99%\t Time: 0:00:06.427041\n",
      "\t Test loss: 0.46118242740631105 \t Test accuracy: 92.58999633789062\n",
      "\n",
      " Train Epoch: 24 [60000/60000 (100%)]\tLoss: 0.000582\t accuracy: 100.00%\t Time: 0:00:06.234942\n",
      "\t Test loss: 0.46854046285152434 \t Test accuracy: 92.70999908447266\n",
      "\n",
      " Train Epoch: 25 [60000/60000 (100%)]\tLoss: 0.000454\t accuracy: 100.00%\t Time: 0:00:06.135436\n",
      "\t Test loss: 0.4643065005540848 \t Test accuracy: 92.47999572753906\n",
      "\n",
      " Train Epoch: 26 [60000/60000 (100%)]\tLoss: 0.000309\t accuracy: 100.00%\t Time: 0:00:06.147495\n",
      "\t Test loss: 0.4731369405984879 \t Test accuracy: 92.57999420166016\n",
      "\n",
      " Train Epoch: 27 [60000/60000 (100%)]\tLoss: 0.000282\t accuracy: 100.00%\t Time: 0:00:06.129451\n",
      "\t Test loss: 0.4774056255817413 \t Test accuracy: 92.55999755859375\n",
      "\n",
      " Train Epoch: 28 [60000/60000 (100%)]\tLoss: 0.000260\t accuracy: 100.00%\t Time: 0:00:06.190133\n",
      "\t Test loss: 0.48504171669483187 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 29 [60000/60000 (100%)]\tLoss: 0.000225\t accuracy: 100.00%\t Time: 0:00:06.096598\n",
      "\t Test loss: 0.48450961112976076 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 30 [60000/60000 (100%)]\tLoss: 0.000241\t accuracy: 100.00%\t Time: 0:00:06.118444\n",
      "\t Test loss: 0.4883306622505188 \t Test accuracy: 92.58999633789062\n",
      "\n",
      " Train Epoch: 31 [60000/60000 (100%)]\tLoss: 0.000203\t accuracy: 100.00%\t Time: 0:00:06.067722\n",
      "\t Test loss: 0.49005786776542665 \t Test accuracy: 92.5199966430664\n",
      "\n",
      " Train Epoch: 32 [60000/60000 (100%)]\tLoss: 0.000175\t accuracy: 100.00%\t Time: 0:00:06.127925\n",
      "\t Test loss: 0.4913718640804291 \t Test accuracy: 92.58999633789062\n",
      "\n",
      " Train Epoch: 33 [60000/60000 (100%)]\tLoss: 0.000148\t accuracy: 100.00%\t Time: 0:00:06.211589\n",
      "\t Test loss: 0.49753284752368926 \t Test accuracy: 92.52999877929688\n",
      "\n",
      " Train Epoch: 34 [60000/60000 (100%)]\tLoss: 0.000152\t accuracy: 100.00%\t Time: 0:00:06.179880\n",
      "\t Test loss: 0.4993539959192276 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 35 [60000/60000 (100%)]\tLoss: 0.000127\t accuracy: 100.00%\t Time: 0:00:06.105337\n",
      "\t Test loss: 0.4998589545488358 \t Test accuracy: 92.69999694824219\n",
      "\n",
      " Train Epoch: 36 [60000/60000 (100%)]\tLoss: 0.000186\t accuracy: 100.00%\t Time: 0:00:06.163509\n",
      "\t Test loss: 0.5088812977075576 \t Test accuracy: 92.61000061035156\n",
      "\n",
      " Train Epoch: 37 [60000/60000 (100%)]\tLoss: 0.000151\t accuracy: 100.00%\t Time: 0:00:06.246681\n",
      "\t Test loss: 0.5068963944911957 \t Test accuracy: 92.48999786376953\n",
      "\n",
      " Train Epoch: 38 [60000/60000 (100%)]\tLoss: 0.000137\t accuracy: 100.00%\t Time: 0:00:06.216365\n",
      "\t Test loss: 0.5104755640029908 \t Test accuracy: 92.57999420166016\n",
      "\n",
      " Train Epoch: 39 [60000/60000 (100%)]\tLoss: 0.000112\t accuracy: 100.00%\t Time: 0:00:06.265903\n",
      "\t Test loss: 0.5104041934013367 \t Test accuracy: 92.54999542236328\n",
      "\n",
      " Train Epoch: 40 [60000/60000 (100%)]\tLoss: 0.000134\t accuracy: 100.00%\t Time: 0:00:06.078392\n",
      "\t Test loss: 0.5087833136320115 \t Test accuracy: 92.61000061035156\n",
      "\n",
      " Train Epoch: 41 [60000/60000 (100%)]\tLoss: 0.000117\t accuracy: 100.00%\t Time: 0:00:06.134207\n",
      "\t Test loss: 0.5113711148500443 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 42 [60000/60000 (100%)]\tLoss: 0.000098\t accuracy: 100.00%\t Time: 0:00:06.204309\n",
      "\t Test loss: 0.5184780031442642 \t Test accuracy: 92.56999969482422\n",
      "\n",
      " Train Epoch: 43 [60000/60000 (100%)]\tLoss: 0.000115\t accuracy: 100.00%\t Time: 0:00:06.108733\n",
      "\t Test loss: 0.5163725972175598 \t Test accuracy: 92.54999542236328\n",
      "\n",
      " Train Epoch: 44 [60000/60000 (100%)]\tLoss: 0.000115\t accuracy: 100.00%\t Time: 0:00:06.088242\n",
      "\t Test loss: 0.5177141040563583 \t Test accuracy: 92.56999969482422\n",
      "\n",
      " Train Epoch: 45 [60000/60000 (100%)]\tLoss: 0.000093\t accuracy: 100.00%\t Time: 0:00:06.128325\n",
      "\t Test loss: 0.5213184803724289 \t Test accuracy: 92.56999969482422\n",
      "\n",
      " Train Epoch: 46 [60000/60000 (100%)]\tLoss: 0.000080\t accuracy: 100.00%\t Time: 0:00:06.191910\n",
      "\t Test loss: 0.5221878468990326 \t Test accuracy: 92.6199951171875\n",
      "\n",
      " Train Epoch: 47 [60000/60000 (100%)]\tLoss: 0.000075\t accuracy: 100.00%\t Time: 0:00:06.142348\n",
      "\t Test loss: 0.5265421003103257 \t Test accuracy: 92.61000061035156\n",
      "\n",
      " Train Epoch: 48 [60000/60000 (100%)]\tLoss: 0.000071\t accuracy: 100.00%\t Time: 0:00:06.084149\n",
      "\t Test loss: 0.5248075872659683 \t Test accuracy: 92.57999420166016\n",
      "\n",
      " Train Epoch: 49 [60000/60000 (100%)]\tLoss: 0.000079\t accuracy: 100.00%\t Time: 0:00:06.116992\n",
      "\t Test loss: 0.5278496354818344 \t Test accuracy: 92.58999633789062\n",
      "\n",
      " Train Epoch: 50 [60000/60000 (100%)]\tLoss: 0.000072\t accuracy: 100.00%\t Time: 0:00:06.203565\n",
      "\t Test loss: 0.5286787658929825 \t Test accuracy: 92.55999755859375\n",
      "\n",
      " Train Epoch: 51 [60000/60000 (100%)]\tLoss: 0.000074\t accuracy: 100.00%\t Time: 0:00:06.173974\n",
      "\t Test loss: 0.5308913826942444 \t Test accuracy: 92.5199966430664\n",
      "\n",
      " Train Epoch: 52 [60000/60000 (100%)]\tLoss: 0.000070\t accuracy: 100.00%\t Time: 0:00:06.114141\n",
      "\t Test loss: 0.5332953453063964 \t Test accuracy: 92.58999633789062\n",
      "\n",
      " Train Epoch: 53 [60000/60000 (100%)]\tLoss: 0.000067\t accuracy: 100.00%\t Time: 0:00:06.206770\n",
      "\t Test loss: 0.5314970076084137 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 54 [60000/60000 (100%)]\tLoss: 0.000099\t accuracy: 100.00%\t Time: 0:00:06.213529\n",
      "\t Test loss: 0.5345382928848267 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 55 [60000/60000 (100%)]\tLoss: 0.000087\t accuracy: 100.00%\t Time: 0:00:06.241142\n",
      "\t Test loss: 0.533430603146553 \t Test accuracy: 92.54999542236328\n",
      "\n",
      " Train Epoch: 56 [60000/60000 (100%)]\tLoss: 0.000072\t accuracy: 100.00%\t Time: 0:00:07.293219\n",
      "\t Test loss: 0.5366463512182236 \t Test accuracy: 92.61000061035156\n",
      "\n",
      " Train Epoch: 57 [60000/60000 (100%)]\tLoss: 0.000062\t accuracy: 100.00%\t Time: 0:00:06.239215\n",
      "\t Test loss: 0.5369058459997177 \t Test accuracy: 92.58999633789062\n",
      "\n",
      " Train Epoch: 58 [60000/60000 (100%)]\tLoss: 0.000060\t accuracy: 100.00%\t Time: 0:00:06.261461\n",
      "\t Test loss: 0.5392028987407684 \t Test accuracy: 92.58999633789062\n",
      "\n",
      " Train Epoch: 59 [60000/60000 (100%)]\tLoss: 0.000059\t accuracy: 100.00%\t Time: 0:00:06.363649\n",
      "\t Test loss: 0.5400536477565765 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 60 [60000/60000 (100%)]\tLoss: 0.000066\t accuracy: 100.00%\t Time: 0:00:06.224904\n",
      "\t Test loss: 0.5399689555168152 \t Test accuracy: 92.61000061035156\n",
      "\n",
      " Train Epoch: 61 [60000/60000 (100%)]\tLoss: 0.000055\t accuracy: 100.00%\t Time: 0:00:06.148478\n",
      "\t Test loss: 0.5416390389204025 \t Test accuracy: 92.57999420166016\n",
      "\n",
      " Train Epoch: 62 [60000/60000 (100%)]\tLoss: 0.000054\t accuracy: 100.00%\t Time: 0:00:06.115291\n",
      "\t Test loss: 0.5434554845094681 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 63 [60000/60000 (100%)]\tLoss: 0.000052\t accuracy: 100.00%\t Time: 0:00:06.208395\n",
      "\t Test loss: 0.5427346378564835 \t Test accuracy: 92.50999450683594\n",
      "\n",
      " Train Epoch: 64 [60000/60000 (100%)]\tLoss: 0.000050\t accuracy: 100.00%\t Time: 0:00:06.181799\n",
      "\t Test loss: 0.5437881737947464 \t Test accuracy: 92.5199966430664\n",
      "\n",
      " Train Epoch: 65 [60000/60000 (100%)]\tLoss: 0.000053\t accuracy: 100.00%\t Time: 0:00:06.106355\n",
      "\t Test loss: 0.5460168123245239 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 66 [60000/60000 (100%)]\tLoss: 0.000053\t accuracy: 100.00%\t Time: 0:00:06.277644\n",
      "\t Test loss: 0.5454131126403808 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 67 [60000/60000 (100%)]\tLoss: 0.000051\t accuracy: 100.00%\t Time: 0:00:06.148318\n",
      "\t Test loss: 0.5463641613721848 \t Test accuracy: 92.6199951171875\n",
      "\n",
      " Train Epoch: 68 [60000/60000 (100%)]\tLoss: 0.000041\t accuracy: 100.00%\t Time: 0:00:06.145802\n",
      "\t Test loss: 0.5453376978635788 \t Test accuracy: 92.6199951171875\n",
      "\n",
      " Train Epoch: 69 [60000/60000 (100%)]\tLoss: 0.000056\t accuracy: 100.00%\t Time: 0:00:06.044102\n",
      "\t Test loss: 0.5492742478847503 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 70 [60000/60000 (100%)]\tLoss: 0.000051\t accuracy: 100.00%\t Time: 0:00:06.085050\n",
      "\t Test loss: 0.5484433889389038 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 71 [60000/60000 (100%)]\tLoss: 0.000043\t accuracy: 100.00%\t Time: 0:00:06.030451\n",
      "\t Test loss: 0.5468575119972229 \t Test accuracy: 92.66999816894531\n",
      "\n",
      " Train Epoch: 72 [60000/60000 (100%)]\tLoss: 0.000050\t accuracy: 100.00%\t Time: 0:00:06.166624\n",
      "\t Test loss: 0.5483483850955964 \t Test accuracy: 92.68999481201172\n",
      "\n",
      " Train Epoch: 73 [60000/60000 (100%)]\tLoss: 0.000059\t accuracy: 100.00%\t Time: 0:00:06.114518\n",
      "\t Test loss: 0.5752911984920501 \t Test accuracy: 92.48999786376953\n",
      "\n",
      " Train Epoch: 74 [60000/60000 (100%)]\tLoss: 0.000075\t accuracy: 100.00%\t Time: 0:00:06.009990\n",
      "\t Test loss: 0.548921400308609 \t Test accuracy: 92.5199966430664\n",
      "\n",
      " Train Epoch: 75 [60000/60000 (100%)]\tLoss: 0.000051\t accuracy: 100.00%\t Time: 0:00:06.051732\n",
      "\t Test loss: 0.5513844758272171 \t Test accuracy: 92.54999542236328\n",
      "\n",
      " Train Epoch: 76 [60000/60000 (100%)]\tLoss: 0.000050\t accuracy: 100.00%\t Time: 0:00:06.084102\n",
      "\t Test loss: 0.5516646951436996 \t Test accuracy: 92.56999969482422\n",
      "\n",
      " Train Epoch: 77 [60000/60000 (100%)]\tLoss: 0.000047\t accuracy: 100.00%\t Time: 0:00:06.128250\n",
      "\t Test loss: 0.5531105726957322 \t Test accuracy: 92.58999633789062\n",
      "\n",
      " Train Epoch: 78 [60000/60000 (100%)]\tLoss: 0.000098\t accuracy: 100.00%\t Time: 0:00:06.065869\n",
      "\t Test loss: 0.5591896176338196 \t Test accuracy: 92.65999603271484\n",
      "\n",
      " Train Epoch: 79 [60000/60000 (100%)]\tLoss: 0.000226\t accuracy: 100.00%\t Time: 0:00:06.048606\n",
      "\t Test loss: 0.5432758182287216 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 80 [60000/60000 (100%)]\tLoss: 0.000075\t accuracy: 100.00%\t Time: 0:00:06.020986\n",
      "\t Test loss: 0.5432236909866333 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 81 [60000/60000 (100%)]\tLoss: 0.000067\t accuracy: 100.00%\t Time: 0:00:06.127109\n",
      "\t Test loss: 0.5394491702318192 \t Test accuracy: 92.69999694824219\n",
      "\n",
      " Train Epoch: 82 [60000/60000 (100%)]\tLoss: 0.000099\t accuracy: 100.00%\t Time: 0:00:06.153926\n",
      "\t Test loss: 0.5509782284498215 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 83 [60000/60000 (100%)]\tLoss: 0.000067\t accuracy: 100.00%\t Time: 0:00:06.097894\n",
      "\t Test loss: 0.5564783424139023 \t Test accuracy: 92.72999572753906\n",
      "\n",
      " Train Epoch: 84 [60000/60000 (100%)]\tLoss: 0.000051\t accuracy: 100.00%\t Time: 0:00:06.160342\n",
      "\t Test loss: 0.5537560254335403 \t Test accuracy: 92.68000030517578\n",
      "\n",
      " Train Epoch: 85 [60000/60000 (100%)]\tLoss: 0.000049\t accuracy: 100.00%\t Time: 0:00:06.127173\n",
      "\t Test loss: 0.5543896168470382 \t Test accuracy: 92.69999694824219\n",
      "\n",
      " Train Epoch: 86 [60000/60000 (100%)]\tLoss: 0.000049\t accuracy: 100.00%\t Time: 0:00:06.068498\n",
      "\t Test loss: 0.554761016368866 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 87 [60000/60000 (100%)]\tLoss: 0.000041\t accuracy: 100.00%\t Time: 0:00:06.106853\n",
      "\t Test loss: 0.5589666932821273 \t Test accuracy: 92.66999816894531\n",
      "\n",
      " Train Epoch: 88 [60000/60000 (100%)]\tLoss: 0.000049\t accuracy: 100.00%\t Time: 0:00:06.056234\n",
      "\t Test loss: 0.557737848162651 \t Test accuracy: 92.72999572753906\n",
      "\n",
      " Train Epoch: 89 [60000/60000 (100%)]\tLoss: 0.000055\t accuracy: 100.00%\t Time: 0:00:05.979125\n",
      "\t Test loss: 0.5573846936225891 \t Test accuracy: 92.72999572753906\n",
      "\n",
      " Train Epoch: 90 [60000/60000 (100%)]\tLoss: 0.000048\t accuracy: 100.00%\t Time: 0:00:06.150204\n",
      "\t Test loss: 0.5597957760095597 \t Test accuracy: 92.68999481201172\n",
      "\n",
      " Train Epoch: 91 [60000/60000 (100%)]\tLoss: 0.000039\t accuracy: 100.00%\t Time: 0:00:06.023012\n",
      "\t Test loss: 0.559625443816185 \t Test accuracy: 92.68000030517578\n",
      "\n",
      " Train Epoch: 92 [60000/60000 (100%)]\tLoss: 0.000035\t accuracy: 100.00%\t Time: 0:00:06.004882\n",
      "\t Test loss: 0.5620987623929977 \t Test accuracy: 92.7699966430664\n",
      "\n",
      " Train Epoch: 93 [60000/60000 (100%)]\tLoss: 0.000033\t accuracy: 100.00%\t Time: 0:00:06.052245\n",
      "\t Test loss: 0.5636122524738312 \t Test accuracy: 92.66999816894531\n",
      "\n",
      " Train Epoch: 94 [60000/60000 (100%)]\tLoss: 0.000034\t accuracy: 100.00%\t Time: 0:00:06.044023\n",
      "\t Test loss: 0.5655857592821121 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 95 [60000/60000 (100%)]\tLoss: 0.000031\t accuracy: 100.00%\t Time: 0:00:06.093906\n",
      "\t Test loss: 0.5642242223024369 \t Test accuracy: 92.68999481201172\n",
      "\n",
      " Train Epoch: 96 [60000/60000 (100%)]\tLoss: 0.000033\t accuracy: 100.00%\t Time: 0:00:06.828881\n",
      "\t Test loss: 0.5644757658243179 \t Test accuracy: 92.69999694824219\n",
      "\n",
      " Train Epoch: 97 [60000/60000 (100%)]\tLoss: 0.000041\t accuracy: 100.00%\t Time: 0:00:06.018357\n",
      "\t Test loss: 0.5628064543008804 \t Test accuracy: 92.65999603271484\n",
      "\n",
      " Train Epoch: 98 [60000/60000 (100%)]\tLoss: 0.000033\t accuracy: 100.00%\t Time: 0:00:06.053611\n",
      "\t Test loss: 0.5636581182479858 \t Test accuracy: 92.70999908447266\n",
      "\n",
      " Train Epoch: 99 [60000/60000 (100%)]\tLoss: 0.000035\t accuracy: 100.00%\t Time: 0:00:06.143169\n",
      "\t Test loss: 0.566396564245224 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 100 [60000/60000 (100%)]\tLoss: 0.000032\t accuracy: 100.00%\t Time: 0:00:06.032763\n",
      "\t Test loss: 0.564967480301857 \t Test accuracy: 92.68999481201172\n",
      "\n",
      " Train Epoch: 101 [60000/60000 (100%)]\tLoss: 0.000029\t accuracy: 100.00%\t Time: 0:00:06.058733\n",
      "\t Test loss: 0.568510240316391 \t Test accuracy: 92.68999481201172\n",
      "\n",
      " Train Epoch: 102 [60000/60000 (100%)]\tLoss: 0.000065\t accuracy: 100.00%\t Time: 0:00:06.114428\n",
      "\t Test loss: 0.5695555061101913 \t Test accuracy: 92.44999694824219\n",
      "\n",
      " Train Epoch: 103 [60000/60000 (100%)]\tLoss: 0.000036\t accuracy: 100.00%\t Time: 0:00:06.207174\n",
      "\t Test loss: 0.5669927716255188 \t Test accuracy: 92.54000091552734\n",
      "\n",
      " Train Epoch: 104 [60000/60000 (100%)]\tLoss: 0.000035\t accuracy: 100.00%\t Time: 0:00:06.148720\n",
      "\t Test loss: 0.5671198785305023 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 105 [60000/60000 (100%)]\tLoss: 0.000031\t accuracy: 100.00%\t Time: 0:00:06.200745\n",
      "\t Test loss: 0.5690486758947373 \t Test accuracy: 92.54000091552734\n",
      "\n",
      " Train Epoch: 106 [60000/60000 (100%)]\tLoss: 0.000032\t accuracy: 100.00%\t Time: 0:00:06.177121\n",
      "\t Test loss: 0.5706786960363388 \t Test accuracy: 92.57999420166016\n",
      "\n",
      " Train Epoch: 107 [60000/60000 (100%)]\tLoss: 0.000028\t accuracy: 100.00%\t Time: 0:00:06.153004\n",
      "\t Test loss: 0.5720565319061279 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 108 [60000/60000 (100%)]\tLoss: 0.000027\t accuracy: 100.00%\t Time: 0:00:06.249454\n",
      "\t Test loss: 0.5706721335649491 \t Test accuracy: 92.57999420166016\n",
      "\n",
      " Train Epoch: 109 [60000/60000 (100%)]\tLoss: 0.000031\t accuracy: 100.00%\t Time: 0:00:06.354120\n",
      "\t Test loss: 0.5729069292545319 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 110 [60000/60000 (100%)]\tLoss: 0.000027\t accuracy: 100.00%\t Time: 0:00:06.196537\n",
      "\t Test loss: 0.5746269792318344 \t Test accuracy: 92.57999420166016\n",
      "\n",
      " Train Epoch: 111 [60000/60000 (100%)]\tLoss: 0.000029\t accuracy: 100.00%\t Time: 0:00:06.104224\n",
      "\t Test loss: 0.574618124961853 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 112 [60000/60000 (100%)]\tLoss: 0.000024\t accuracy: 100.00%\t Time: 0:00:06.084865\n",
      "\t Test loss: 0.5724717766046524 \t Test accuracy: 92.69999694824219\n",
      "\n",
      " Train Epoch: 113 [60000/60000 (100%)]\tLoss: 0.000026\t accuracy: 100.00%\t Time: 0:00:06.050375\n",
      "\t Test loss: 0.571234792470932 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 114 [60000/60000 (100%)]\tLoss: 0.000026\t accuracy: 100.00%\t Time: 0:00:06.118758\n",
      "\t Test loss: 0.5742225497961044 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 115 [60000/60000 (100%)]\tLoss: 0.000025\t accuracy: 100.00%\t Time: 0:00:06.071429\n",
      "\t Test loss: 0.5747802257537842 \t Test accuracy: 92.70999908447266\n",
      "\n",
      " Train Epoch: 116 [60000/60000 (100%)]\tLoss: 0.000023\t accuracy: 100.00%\t Time: 0:00:06.050746\n",
      "\t Test loss: 0.5762748301029206 \t Test accuracy: 92.70999908447266\n",
      "\n",
      " Train Epoch: 117 [60000/60000 (100%)]\tLoss: 0.000025\t accuracy: 100.00%\t Time: 0:00:06.211764\n",
      "\t Test loss: 0.5760311841964721 \t Test accuracy: 92.61000061035156\n",
      "\n",
      " Train Epoch: 118 [60000/60000 (100%)]\tLoss: 0.000022\t accuracy: 100.00%\t Time: 0:00:06.085294\n",
      "\t Test loss: 0.5801340788602829 \t Test accuracy: 92.6199951171875\n",
      "\n",
      " Train Epoch: 119 [60000/60000 (100%)]\tLoss: 0.000025\t accuracy: 100.00%\t Time: 0:00:06.117867\n",
      "\t Test loss: 0.5772035151720047 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 120 [60000/60000 (100%)]\tLoss: 0.000037\t accuracy: 100.00%\t Time: 0:00:06.075978\n",
      "\t Test loss: 0.5783930838108062 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 121 [60000/60000 (100%)]\tLoss: 0.000029\t accuracy: 100.00%\t Time: 0:00:06.104758\n",
      "\t Test loss: 0.579199469089508 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 122 [60000/60000 (100%)]\tLoss: 0.000027\t accuracy: 100.00%\t Time: 0:00:06.045889\n",
      "\t Test loss: 0.5827281802892685 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 123 [60000/60000 (100%)]\tLoss: 0.000030\t accuracy: 100.00%\t Time: 0:00:06.059830\n",
      "\t Test loss: 0.5779795825481415 \t Test accuracy: 92.68999481201172\n",
      "\n",
      " Train Epoch: 124 [60000/60000 (100%)]\tLoss: 0.000028\t accuracy: 100.00%\t Time: 0:00:06.028840\n",
      "\t Test loss: 0.5781209081411361 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 125 [60000/60000 (100%)]\tLoss: 0.000025\t accuracy: 100.00%\t Time: 0:00:06.043876\n",
      "\t Test loss: 0.5788948446512222 \t Test accuracy: 92.70999908447266\n",
      "\n",
      " Train Epoch: 126 [60000/60000 (100%)]\tLoss: 0.000021\t accuracy: 100.00%\t Time: 0:00:06.103750\n",
      "\t Test loss: 0.5820209711790085 \t Test accuracy: 92.6199951171875\n",
      "\n",
      " Train Epoch: 127 [60000/60000 (100%)]\tLoss: 0.000028\t accuracy: 100.00%\t Time: 0:00:06.098668\n",
      "\t Test loss: 0.5828913688659668 \t Test accuracy: 92.55999755859375\n",
      "\n",
      " Train Epoch: 128 [60000/60000 (100%)]\tLoss: 0.000024\t accuracy: 100.00%\t Time: 0:00:06.157566\n",
      "\t Test loss: 0.5800959527492523 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 129 [60000/60000 (100%)]\tLoss: 0.000036\t accuracy: 100.00%\t Time: 0:00:06.068615\n",
      "\t Test loss: 0.5853619903326035 \t Test accuracy: 92.54000091552734\n",
      "\n",
      " Train Epoch: 130 [60000/60000 (100%)]\tLoss: 0.000022\t accuracy: 100.00%\t Time: 0:00:06.096277\n",
      "\t Test loss: 0.5837894171476364 \t Test accuracy: 92.58999633789062\n",
      "\n",
      " Train Epoch: 131 [60000/60000 (100%)]\tLoss: 0.000026\t accuracy: 100.00%\t Time: 0:00:06.067250\n",
      "\t Test loss: 0.5841598600149155 \t Test accuracy: 92.5\n",
      "\n",
      " Train Epoch: 132 [60000/60000 (100%)]\tLoss: 0.000029\t accuracy: 100.00%\t Time: 0:00:06.038329\n",
      "\t Test loss: 0.588387343287468 \t Test accuracy: 92.65999603271484\n",
      "\n",
      " Train Epoch: 133 [60000/60000 (100%)]\tLoss: 0.000024\t accuracy: 100.00%\t Time: 0:00:06.037472\n",
      "\t Test loss: 0.5827661782503128 \t Test accuracy: 92.55999755859375\n",
      "\n",
      " Train Epoch: 134 [60000/60000 (100%)]\tLoss: 0.000024\t accuracy: 100.00%\t Time: 0:00:06.043930\n",
      "\t Test loss: 0.5863658517599106 \t Test accuracy: 92.55999755859375\n",
      "\n",
      " Train Epoch: 135 [60000/60000 (100%)]\tLoss: 0.000034\t accuracy: 100.00%\t Time: 0:00:06.074038\n",
      "\t Test loss: 0.5854810684919357 \t Test accuracy: 92.54000091552734\n",
      "\n",
      " Train Epoch: 136 [60000/60000 (100%)]\tLoss: 0.000022\t accuracy: 100.00%\t Time: 0:00:06.090437\n",
      "\t Test loss: 0.5871375769376754 \t Test accuracy: 92.5199966430664\n",
      "\n",
      " Train Epoch: 137 [60000/60000 (100%)]\tLoss: 0.000021\t accuracy: 100.00%\t Time: 0:00:06.966737\n",
      "\t Test loss: 0.5877095341682435 \t Test accuracy: 92.54999542236328\n",
      "\n",
      " Train Epoch: 138 [60000/60000 (100%)]\tLoss: 0.000024\t accuracy: 100.00%\t Time: 0:00:06.009725\n",
      "\t Test loss: 0.5894282460212708 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 139 [60000/60000 (100%)]\tLoss: 0.000026\t accuracy: 100.00%\t Time: 0:00:06.107984\n",
      "\t Test loss: 0.5868011116981506 \t Test accuracy: 92.6199951171875\n",
      "\n",
      " Train Epoch: 140 [60000/60000 (100%)]\tLoss: 0.000021\t accuracy: 100.00%\t Time: 0:00:06.087088\n",
      "\t Test loss: 0.5895364552736282 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 141 [60000/60000 (100%)]\tLoss: 0.000024\t accuracy: 100.00%\t Time: 0:00:06.056780\n",
      "\t Test loss: 0.5876668214797973 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 142 [60000/60000 (100%)]\tLoss: 0.000019\t accuracy: 100.00%\t Time: 0:00:06.094286\n",
      "\t Test loss: 0.5895857512950897 \t Test accuracy: 92.61000061035156\n",
      "\n",
      " Train Epoch: 143 [60000/60000 (100%)]\tLoss: 0.000020\t accuracy: 100.00%\t Time: 0:00:06.125026\n",
      "\t Test loss: 0.5887399971485138 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 144 [60000/60000 (100%)]\tLoss: 0.000020\t accuracy: 100.00%\t Time: 0:00:06.264682\n",
      "\t Test loss: 0.5871270805597305 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 145 [60000/60000 (100%)]\tLoss: 0.000020\t accuracy: 100.00%\t Time: 0:00:06.180625\n",
      "\t Test loss: 0.5898298382759094 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 146 [60000/60000 (100%)]\tLoss: 0.000019\t accuracy: 100.00%\t Time: 0:00:06.074820\n",
      "\t Test loss: 0.5901375949382782 \t Test accuracy: 92.57999420166016\n",
      "\n",
      " Train Epoch: 147 [60000/60000 (100%)]\tLoss: 0.000019\t accuracy: 100.00%\t Time: 0:00:06.076002\n",
      "\t Test loss: 0.5874303698539733 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 148 [60000/60000 (100%)]\tLoss: 0.000017\t accuracy: 100.00%\t Time: 0:00:06.080093\n",
      "\t Test loss: 0.5905827432870865 \t Test accuracy: 92.66999816894531\n",
      "\n",
      " Train Epoch: 149 [60000/60000 (100%)]\tLoss: 0.000016\t accuracy: 100.00%\t Time: 0:00:06.003488\n",
      "\t Test loss: 0.5900115817785263 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 150 [60000/60000 (100%)]\tLoss: 0.000019\t accuracy: 100.00%\t Time: 0:00:06.068486\n",
      "\t Test loss: 0.5932327002286911 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 151 [60000/60000 (100%)]\tLoss: 0.000054\t accuracy: 100.00%\t Time: 0:00:06.066453\n",
      "\t Test loss: 0.5952422857284546 \t Test accuracy: 92.56999969482422\n",
      "\n",
      " Train Epoch: 152 [60000/60000 (100%)]\tLoss: 0.000032\t accuracy: 100.00%\t Time: 0:00:06.184554\n",
      "\t Test loss: 0.5939153373241425 \t Test accuracy: 92.5199966430664\n",
      "\n",
      " Train Epoch: 153 [60000/60000 (100%)]\tLoss: 0.000022\t accuracy: 100.00%\t Time: 0:00:06.076413\n",
      "\t Test loss: 0.5925319582223892 \t Test accuracy: 92.5199966430664\n",
      "\n",
      " Train Epoch: 154 [60000/60000 (100%)]\tLoss: 0.000019\t accuracy: 100.00%\t Time: 0:00:06.116914\n",
      "\t Test loss: 0.5942920118570327 \t Test accuracy: 92.5199966430664\n",
      "\n",
      " Train Epoch: 155 [60000/60000 (100%)]\tLoss: 0.000023\t accuracy: 100.00%\t Time: 0:00:06.029599\n",
      "\t Test loss: 0.5947288542985916 \t Test accuracy: 92.57999420166016\n",
      "\n",
      " Train Epoch: 156 [60000/60000 (100%)]\tLoss: 0.000020\t accuracy: 100.00%\t Time: 0:00:06.037923\n",
      "\t Test loss: 0.5972042083740234 \t Test accuracy: 92.54000091552734\n",
      "\n",
      " Train Epoch: 157 [60000/60000 (100%)]\tLoss: 0.000019\t accuracy: 100.00%\t Time: 0:00:06.088074\n",
      "\t Test loss: 0.5948121458292007 \t Test accuracy: 92.6199951171875\n",
      "\n",
      " Train Epoch: 158 [60000/60000 (100%)]\tLoss: 0.000025\t accuracy: 100.00%\t Time: 0:00:06.061945\n",
      "\t Test loss: 0.595933935046196 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 159 [60000/60000 (100%)]\tLoss: 0.000022\t accuracy: 100.00%\t Time: 0:00:06.087756\n",
      "\t Test loss: 0.5966085702180862 \t Test accuracy: 92.61000061035156\n",
      "\n",
      " Train Epoch: 160 [60000/60000 (100%)]\tLoss: 0.000019\t accuracy: 100.00%\t Time: 0:00:05.999960\n",
      "\t Test loss: 0.596185314655304 \t Test accuracy: 92.58999633789062\n",
      "\n",
      " Train Epoch: 161 [60000/60000 (100%)]\tLoss: 0.000018\t accuracy: 100.00%\t Time: 0:00:06.084498\n",
      "\t Test loss: 0.5967938750982285 \t Test accuracy: 92.55999755859375\n",
      "\n",
      " Train Epoch: 162 [60000/60000 (100%)]\tLoss: 0.000021\t accuracy: 100.00%\t Time: 0:00:06.035734\n",
      "\t Test loss: 0.5980557382106781 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 163 [60000/60000 (100%)]\tLoss: 0.000016\t accuracy: 100.00%\t Time: 0:00:06.114438\n",
      "\t Test loss: 0.5986964166164398 \t Test accuracy: 92.6199951171875\n",
      "\n",
      " Train Epoch: 164 [60000/60000 (100%)]\tLoss: 0.000016\t accuracy: 100.00%\t Time: 0:00:06.076113\n",
      "\t Test loss: 0.5988890647888183 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 165 [60000/60000 (100%)]\tLoss: 0.000018\t accuracy: 100.00%\t Time: 0:00:06.081259\n",
      "\t Test loss: 0.6011576890945435 \t Test accuracy: 92.58999633789062\n",
      "\n",
      " Train Epoch: 166 [60000/60000 (100%)]\tLoss: 0.000016\t accuracy: 100.00%\t Time: 0:00:06.153581\n",
      "\t Test loss: 0.6010808736085892 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 167 [60000/60000 (100%)]\tLoss: 0.000148\t accuracy: 100.00%\t Time: 0:00:06.043501\n",
      "\t Test loss: 0.5954079985618591 \t Test accuracy: 92.6199951171875\n",
      "\n",
      " Train Epoch: 168 [60000/60000 (100%)]\tLoss: 0.000080\t accuracy: 100.00%\t Time: 0:00:06.177115\n",
      "\t Test loss: 0.5936545699834823 \t Test accuracy: 92.57999420166016\n",
      "\n",
      " Train Epoch: 169 [60000/60000 (100%)]\tLoss: 0.000036\t accuracy: 100.00%\t Time: 0:00:06.139269\n",
      "\t Test loss: 0.5917010813951492 \t Test accuracy: 92.68000030517578\n",
      "\n",
      " Train Epoch: 170 [60000/60000 (100%)]\tLoss: 0.000061\t accuracy: 100.00%\t Time: 0:00:06.155119\n",
      "\t Test loss: 0.5930016845464706 \t Test accuracy: 92.66999816894531\n",
      "\n",
      " Train Epoch: 171 [60000/60000 (100%)]\tLoss: 0.000047\t accuracy: 100.00%\t Time: 0:00:06.070617\n",
      "\t Test loss: 0.5926021993160248 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 172 [60000/60000 (100%)]\tLoss: 0.000024\t accuracy: 100.00%\t Time: 0:00:06.143276\n",
      "\t Test loss: 0.592509788274765 \t Test accuracy: 92.70999908447266\n",
      "\n",
      " Train Epoch: 173 [60000/60000 (100%)]\tLoss: 0.000025\t accuracy: 100.00%\t Time: 0:00:06.090342\n",
      "\t Test loss: 0.5950492680072784 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 174 [60000/60000 (100%)]\tLoss: 0.000035\t accuracy: 100.00%\t Time: 0:00:06.102653\n",
      "\t Test loss: 0.5966584056615829 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 175 [60000/60000 (100%)]\tLoss: 0.000031\t accuracy: 100.00%\t Time: 0:00:06.143478\n",
      "\t Test loss: 0.5930318892002105 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 176 [60000/60000 (100%)]\tLoss: 0.000025\t accuracy: 100.00%\t Time: 0:00:06.028171\n",
      "\t Test loss: 0.5934894472360611 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 177 [60000/60000 (100%)]\tLoss: 0.000022\t accuracy: 100.00%\t Time: 0:00:06.075282\n",
      "\t Test loss: 0.598437386751175 \t Test accuracy: 92.69999694824219\n",
      "\n",
      " Train Epoch: 178 [60000/60000 (100%)]\tLoss: 0.000017\t accuracy: 100.00%\t Time: 0:00:06.955617\n",
      "\t Test loss: 0.5977778375148773 \t Test accuracy: 92.65999603271484\n",
      "\n",
      " Train Epoch: 179 [60000/60000 (100%)]\tLoss: 0.000018\t accuracy: 100.00%\t Time: 0:00:06.128344\n",
      "\t Test loss: 0.5968911081552506 \t Test accuracy: 92.61000061035156\n",
      "\n",
      " Train Epoch: 180 [60000/60000 (100%)]\tLoss: 0.000018\t accuracy: 100.00%\t Time: 0:00:06.076857\n",
      "\t Test loss: 0.6003297001123429 \t Test accuracy: 92.5999984741211\n",
      "\n",
      " Train Epoch: 181 [60000/60000 (100%)]\tLoss: 0.000024\t accuracy: 100.00%\t Time: 0:00:06.036090\n",
      "\t Test loss: 0.5975831329822541 \t Test accuracy: 92.61000061035156\n",
      "\n",
      " Train Epoch: 182 [60000/60000 (100%)]\tLoss: 0.000018\t accuracy: 100.00%\t Time: 0:00:06.015973\n",
      "\t Test loss: 0.5973942548036575 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 183 [60000/60000 (100%)]\tLoss: 0.000019\t accuracy: 100.00%\t Time: 0:00:06.030413\n",
      "\t Test loss: 0.6005059659481049 \t Test accuracy: 92.6199951171875\n",
      "\n",
      " Train Epoch: 184 [60000/60000 (100%)]\tLoss: 0.000017\t accuracy: 100.00%\t Time: 0:00:06.187943\n",
      "\t Test loss: 0.5991847008466721 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 185 [60000/60000 (100%)]\tLoss: 0.000025\t accuracy: 100.00%\t Time: 0:00:06.091530\n",
      "\t Test loss: 0.5997264981269836 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 186 [60000/60000 (100%)]\tLoss: 0.000019\t accuracy: 100.00%\t Time: 0:00:06.108565\n",
      "\t Test loss: 0.6016536176204681 \t Test accuracy: 92.64999389648438\n",
      "\n",
      " Train Epoch: 187 [60000/60000 (100%)]\tLoss: 0.000019\t accuracy: 100.00%\t Time: 0:00:06.205554\n",
      "\t Test loss: 0.6033084869384766 \t Test accuracy: 92.66999816894531\n",
      "\n",
      " Train Epoch: 188 [60000/60000 (100%)]\tLoss: 0.000021\t accuracy: 100.00%\t Time: 0:00:06.108147\n",
      "\t Test loss: 0.5977592468261719 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 189 [60000/60000 (100%)]\tLoss: 0.000016\t accuracy: 100.00%\t Time: 0:00:06.086971\n",
      "\t Test loss: 0.5977938324213028 \t Test accuracy: 92.66999816894531\n",
      "\n",
      " Train Epoch: 190 [60000/60000 (100%)]\tLoss: 0.000018\t accuracy: 100.00%\t Time: 0:00:06.076880\n",
      "\t Test loss: 0.600655061006546 \t Test accuracy: 92.63999938964844\n",
      "\n",
      " Train Epoch: 191 [60000/60000 (100%)]\tLoss: 0.000015\t accuracy: 100.00%\t Time: 0:00:06.037900\n",
      "\t Test loss: 0.6028425902128219 \t Test accuracy: 92.6199951171875\n",
      "\n",
      " Train Epoch: 192 [60000/60000 (100%)]\tLoss: 0.000019\t accuracy: 100.00%\t Time: 0:00:06.045791\n",
      "\t Test loss: 0.6010464280843735 \t Test accuracy: 92.68000030517578\n",
      "\n",
      " Train Epoch: 193 [60000/60000 (100%)]\tLoss: 0.000014\t accuracy: 100.00%\t Time: 0:00:06.120608\n",
      "\t Test loss: 0.6032459676265717 \t Test accuracy: 92.66999816894531\n",
      "\n",
      " Train Epoch: 194 [60000/60000 (100%)]\tLoss: 0.000016\t accuracy: 100.00%\t Time: 0:00:06.016612\n",
      "\t Test loss: 0.6053977280855178 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 195 [60000/60000 (100%)]\tLoss: 0.000013\t accuracy: 100.00%\t Time: 0:00:06.008593\n",
      "\t Test loss: 0.604275107383728 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 196 [60000/60000 (100%)]\tLoss: 0.000014\t accuracy: 100.00%\t Time: 0:00:06.168497\n",
      "\t Test loss: 0.6042879730463028 \t Test accuracy: 92.65999603271484\n",
      "\n",
      " Train Epoch: 197 [60000/60000 (100%)]\tLoss: 0.000016\t accuracy: 100.00%\t Time: 0:00:06.003102\n",
      "\t Test loss: 0.6016179233789444 \t Test accuracy: 92.62999725341797\n",
      "\n",
      " Train Epoch: 198 [60000/60000 (100%)]\tLoss: 0.000013\t accuracy: 100.00%\t Time: 0:00:06.144099\n",
      "\t Test loss: 0.6043727189302445 \t Test accuracy: 92.65999603271484\n",
      "\n",
      " Train Epoch: 199 [60000/60000 (100%)]\tLoss: 0.000014\t accuracy: 100.00%\t Time: 0:00:06.024010\n",
      "\t Test loss: 0.6056303650140762 \t Test accuracy: 92.61000061035156\n",
      "\n"
     ]
    }
   ]
  },
  {
   "cell_type": "code",
   "source": [
    "lr = 0.1 if DATASET == 'FashionMNIST' else 1\n",
    "num_epochs = 100000\n",
    "batch_size = 128\n",
    "\n",
    "computed = False\n",
    "\n",
    "path = os.path.join(store_dir, 'epoch200', optimizer_name + '.pt')\n",
    "checkpoint = torch.load(path, map_location=torch.device('cpu'))\n",
    "model = checkpoint['state_dict']\n",
    "model = get_model(ARCHITECTURE, DATASET).to(device)\n",
    "model.load_state_dict(checkpoint['state_dict'])\n",
    "\n",
    "\n",
    "while not computed:\n",
    "    try:\n",
    "        # Calculating the sharpness. Returns an error if the learning rate is too big\n",
    "        sharpnesses, losses = minimum_shaprness_eff(data, model, batch_size, lr, num_epochs=num_epochs, optimizer_file=path)\n",
    "\n",
    "        # storing the sharpness\n",
    "        sharpness_path = os.path.join(store_dir, 'epoch200', optimizer_name + '_sharpness.pt')\n",
    "        checkpoint = {'sharpnesses':sharpnesses, 'sharpness':sharpnesses[-1], 'losses': losses}\n",
    "        torch.save(checkpoint, sharpness_path)\n",
    "\n",
    "\n",
    "        computed = True\n",
    "        print(f'Sharpness: {sharpnesses[-1]}')\n",
    "    except:\n",
    "        # Error is returned if the learning rate is too big, so in that case learning rate is set to be twice smaller and number of epochs are set to be twice as bigger\n",
    "        computed = False\n",
    "        lr /= 2.0\n",
    "        num_epochs *= 2\n",
    "        print(f'Use smaller stepsize than {lr}')"
   ],
   "metadata": {
    "id": "daRK_9SnQYiq",
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "outputId": "001beaa9-4d6e-43d7-a498-657b86e89391",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": null,
   "outputs": [
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "\t Calculating Hessian\n"
     ]
    },
    {
     "output_type": "stream",
     "name": "stderr",
     "text": [
      "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py:1033: UserWarning: Using a non-full backward hook when the forward contains multiple autograd Nodes is deprecated and will be removed in future versions. This hook will be missing some grad_input. Please use register_full_backward_hook to get the documented behavior.\n",
      "  warnings.warn(\"Using a non-full backward hook when the forward contains multiple autograd Nodes \"\n"
     ]
    },
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "\tFinished the diag calculation. Time needed: 0:00:59.205600. Computing sharpness...\n",
      " \t\t epoch:100000\t processed 100.0%\t loss:0.011053657309945586 \t minimum sharpness: 108900.66487971791 \t Time needed 0:01:45.861625\n",
      "--------------------------------------------------\n",
      "Sharpness: 108900.66487971791\n"
     ]
    }
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "id": "1bARzz1n3amq",
    "pycharm": {
     "name": "#%% md\n"
    }
   },
   "source": [
    "## Training with SAM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "id": "L-36bmvfB4wO",
    "colab": {
     "base_uri": "https://localhost:8080/"
    },
    "outputId": "5928d34d-5dbf-48e5-e73a-5fa33b395a05",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "outputs": [
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "new version 2\n",
      " Train Epoch: 0 [60000/60000 (100%)]\tLoss: 0.461048\t accuracy: 83.57%\t Time: 0:00:09.832774\n",
      "\t Test loss: 0.3120079696178436 \t Test accuracy: 88.6199951171875\n",
      "\n",
      " Train Epoch: 1 [60000/60000 (100%)]\tLoss: 0.311164\t accuracy: 88.77%\t Time: 0:00:10.131190\n",
      "\t Test loss: 0.26722029894590377 \t Test accuracy: 90.3499984741211\n",
      "\n",
      " Train Epoch: 2 [60000/60000 (100%)]\tLoss: 0.273540\t accuracy: 90.00%\t Time: 0:00:09.600028\n",
      "\t Test loss: 0.2555205300450325 \t Test accuracy: 90.5\n",
      "\n",
      " Train Epoch: 3 [60000/60000 (100%)]\tLoss: 0.245635\t accuracy: 91.10%\t Time: 0:00:09.443238\n",
      "\t Test loss: 0.2255650833249092 \t Test accuracy: 91.83999633789062\n",
      "\n",
      " Train Epoch: 4 [60000/60000 (100%)]\tLoss: 0.224806\t accuracy: 91.83%\t Time: 0:00:09.344006\n",
      "\t Test loss: 0.21657753735780716 \t Test accuracy: 92.25\n",
      "\n",
      " Train Epoch: 5 [60000/60000 (100%)]\tLoss: 0.203977\t accuracy: 92.59%\t Time: 0:00:09.324913\n",
      "\t Test loss: 0.21508579403162004 \t Test accuracy: 92.06999969482422\n",
      "\n",
      " Train Epoch: 6 [60000/60000 (100%)]\tLoss: 0.188953\t accuracy: 93.06%\t Time: 0:00:09.337152\n",
      "\t Test loss: 0.20805060267448425 \t Test accuracy: 92.48999786376953\n",
      "\n",
      " Train Epoch: 7 [60000/60000 (100%)]\tLoss: 0.171920\t accuracy: 93.72%\t Time: 0:00:09.274472\n",
      "\t Test loss: 0.19853134155273439 \t Test accuracy: 93.04000091552734\n",
      "\n",
      " Train Epoch: 8 [60000/60000 (100%)]\tLoss: 0.157221\t accuracy: 94.31%\t Time: 0:00:09.376995\n",
      "\t Test loss: 0.20485417991876603 \t Test accuracy: 92.65999603271484\n",
      "\n",
      " Train Epoch: 9 [60000/60000 (100%)]\tLoss: 0.143114\t accuracy: 94.76%\t Time: 0:00:09.411449\n",
      "\t Test loss: 0.2101579874753952 \t Test accuracy: 92.43000030517578\n",
      "\n",
      " Train Epoch: 10 [60000/60000 (100%)]\tLoss: 0.132252\t accuracy: 95.19%\t Time: 0:00:09.349373\n",
      "\t Test loss: 0.2144463464617729 \t Test accuracy: 92.33999633789062\n",
      "\n",
      " Train Epoch: 11 [60000/60000 (100%)]\tLoss: 0.119548\t accuracy: 95.68%\t Time: 0:00:09.382098\n",
      "\t Test loss: 0.21231337487697602 \t Test accuracy: 92.79999542236328\n",
      "\n",
      " Train Epoch: 12 [60000/60000 (100%)]\tLoss: 0.107594\t accuracy: 96.15%\t Time: 0:00:09.367931\n",
      "\t Test loss: 0.2064189851284027 \t Test accuracy: 93.18000030517578\n",
      "\n",
      " Train Epoch: 13 [60000/60000 (100%)]\tLoss: 0.098619\t accuracy: 96.42%\t Time: 0:00:09.335694\n",
      "\t Test loss: 0.2078656867146492 \t Test accuracy: 93.18999481201172\n",
      "\n",
      " Train Epoch: 14 [60000/60000 (100%)]\tLoss: 0.089453\t accuracy: 96.78%\t Time: 0:00:09.321829\n",
      "\t Test loss: 0.21648130267858506 \t Test accuracy: 92.97999572753906\n",
      "\n",
      " Train Epoch: 15 [60000/60000 (100%)]\tLoss: 0.081988\t accuracy: 97.05%\t Time: 0:00:09.361415\n",
      "\t Test loss: 0.2228183701634407 \t Test accuracy: 92.86000061035156\n",
      "\n",
      " Train Epoch: 16 [60000/60000 (100%)]\tLoss: 0.072515\t accuracy: 97.40%\t Time: 0:00:09.350140\n",
      "\t Test loss: 0.22507776618003844 \t Test accuracy: 92.97999572753906\n",
      "\n",
      " Train Epoch: 17 [60000/60000 (100%)]\tLoss: 0.065936\t accuracy: 97.64%\t Time: 0:00:09.315508\n",
      "\t Test loss: 0.23177945017814636 \t Test accuracy: 92.82999420166016\n",
      "\n",
      " Train Epoch: 18 [60000/60000 (100%)]\tLoss: 0.059667\t accuracy: 97.85%\t Time: 0:00:09.449717\n",
      "\t Test loss: 0.23996212631464003 \t Test accuracy: 93.02999877929688\n",
      "\n",
      " Train Epoch: 19 [60000/60000 (100%)]\tLoss: 0.056506\t accuracy: 97.95%\t Time: 0:00:09.409806\n",
      "\t Test loss: 0.23465668261051179 \t Test accuracy: 93.18999481201172\n",
      "\n",
      " Train Epoch: 20 [60000/60000 (100%)]\tLoss: 0.050390\t accuracy: 98.18%\t Time: 0:00:09.308188\n",
      "\t Test loss: 0.24818700850009917 \t Test accuracy: 93.15999603271484\n",
      "\n",
      " Train Epoch: 21 [60000/60000 (100%)]\tLoss: 0.047718\t accuracy: 98.37%\t Time: 0:00:09.311120\n",
      "\t Test loss: 0.25298127830028533 \t Test accuracy: 92.98999786376953\n",
      "\n",
      " Train Epoch: 22 [60000/60000 (100%)]\tLoss: 0.042268\t accuracy: 98.45%\t Time: 0:00:09.273689\n",
      "\t Test loss: 0.258545458316803 \t Test accuracy: 93.15999603271484\n",
      "\n",
      " Train Epoch: 23 [60000/60000 (100%)]\tLoss: 0.039920\t accuracy: 98.55%\t Time: 0:00:09.331988\n",
      "\t Test loss: 0.2619346663355827 \t Test accuracy: 92.86000061035156\n",
      "\n",
      " Train Epoch: 24 [60000/60000 (100%)]\tLoss: 0.039446\t accuracy: 98.60%\t Time: 0:00:09.315591\n",
      "\t Test loss: 0.2656273618340492 \t Test accuracy: 92.94999694824219\n",
      "\n",
      " Train Epoch: 25 [60000/60000 (100%)]\tLoss: 0.035839\t accuracy: 98.73%\t Time: 0:00:09.391559\n",
      "\t Test loss: 0.26606987714767455 \t Test accuracy: 93.22999572753906\n",
      "\n",
      " Train Epoch: 26 [60000/60000 (100%)]\tLoss: 0.036163\t accuracy: 98.76%\t Time: 0:00:09.333709\n",
      "\t Test loss: 0.2683510735630989 \t Test accuracy: 92.7699966430664\n",
      "\n",
      " Train Epoch: 27 [60000/60000 (100%)]\tLoss: 0.033953\t accuracy: 98.81%\t Time: 0:00:09.316212\n",
      "\t Test loss: 0.28209550827741625 \t Test accuracy: 92.72999572753906\n",
      "\n",
      " Train Epoch: 28 [60000/60000 (100%)]\tLoss: 0.030333\t accuracy: 98.91%\t Time: 0:00:09.303017\n",
      "\t Test loss: 0.27880283892154695 \t Test accuracy: 93.16999816894531\n",
      "\n",
      " Train Epoch: 29 [60000/60000 (100%)]\tLoss: 0.029653\t accuracy: 98.93%\t Time: 0:00:09.308316\n",
      "\t Test loss: 0.2830014884471893 \t Test accuracy: 92.73999786376953\n",
      "\n",
      " Train Epoch: 30 [60000/60000 (100%)]\tLoss: 0.026990\t accuracy: 99.04%\t Time: 0:00:09.268159\n",
      "\t Test loss: 0.27988036870956423 \t Test accuracy: 92.93999481201172\n",
      "\n",
      " Train Epoch: 31 [60000/60000 (100%)]\tLoss: 0.026938\t accuracy: 99.00%\t Time: 0:00:09.933610\n",
      "\t Test loss: 0.28468252420425416 \t Test accuracy: 93.25\n",
      "\n",
      " Train Epoch: 32 [60000/60000 (100%)]\tLoss: 0.025780\t accuracy: 99.07%\t Time: 0:00:09.271695\n",
      "\t Test loss: 0.2974221333861351 \t Test accuracy: 92.97999572753906\n",
      "\n",
      " Train Epoch: 33 [60000/60000 (100%)]\tLoss: 0.023818\t accuracy: 99.13%\t Time: 0:00:09.356552\n",
      "\t Test loss: 0.29851331263780595 \t Test accuracy: 92.90999603271484\n",
      "\n",
      " Train Epoch: 34 [60000/60000 (100%)]\tLoss: 0.023589\t accuracy: 99.16%\t Time: 0:00:09.376362\n",
      "\t Test loss: 0.30359736680984495 \t Test accuracy: 92.68000030517578\n",
      "\n",
      " Train Epoch: 35 [60000/60000 (100%)]\tLoss: 0.022188\t accuracy: 99.17%\t Time: 0:00:09.370207\n",
      "\t Test loss: 0.295995269715786 \t Test accuracy: 93.13999938964844\n",
      "\n",
      " Train Epoch: 36 [60000/60000 (100%)]\tLoss: 0.020867\t accuracy: 99.26%\t Time: 0:00:09.338497\n",
      "\t Test loss: 0.3031784549355507 \t Test accuracy: 93.05999755859375\n",
      "\n",
      " Train Epoch: 37 [60000/60000 (100%)]\tLoss: 0.019709\t accuracy: 99.29%\t Time: 0:00:09.326902\n",
      "\t Test loss: 0.31597876995801927 \t Test accuracy: 93.18999481201172\n",
      "\n",
      " Train Epoch: 38 [60000/60000 (100%)]\tLoss: 0.021536\t accuracy: 99.23%\t Time: 0:00:09.332577\n",
      "\t Test loss: 0.3041060879826546 \t Test accuracy: 92.87999725341797\n",
      "\n",
      " Train Epoch: 39 [60000/60000 (100%)]\tLoss: 0.019421\t accuracy: 99.31%\t Time: 0:00:09.358529\n",
      "\t Test loss: 0.30559062361717226 \t Test accuracy: 93.06999969482422\n",
      "\n",
      " Train Epoch: 40 [60000/60000 (100%)]\tLoss: 0.017381\t accuracy: 99.36%\t Time: 0:00:09.300420\n",
      "\t Test loss: 0.3121866136789322 \t Test accuracy: 93.22000122070312\n",
      "\n",
      " Train Epoch: 41 [60000/60000 (100%)]\tLoss: 0.018301\t accuracy: 99.33%\t Time: 0:00:09.353812\n",
      "\t Test loss: 0.31289210319519045 \t Test accuracy: 93.14999389648438\n",
      "\n",
      " Train Epoch: 42 [60000/60000 (100%)]\tLoss: 0.017144\t accuracy: 99.37%\t Time: 0:00:09.360380\n",
      "\t Test loss: 0.3192660450935364 \t Test accuracy: 93.14999389648438\n",
      "\n",
      " Train Epoch: 43 [60000/60000 (100%)]\tLoss: 0.017454\t accuracy: 99.36%\t Time: 0:00:09.339063\n",
      "\t Test loss: 0.3097944900393486 \t Test accuracy: 93.15999603271484\n",
      "\n",
      " Train Epoch: 44 [60000/60000 (100%)]\tLoss: 0.016593\t accuracy: 99.39%\t Time: 0:00:09.241771\n",
      "\t Test loss: 0.3190863445401192 \t Test accuracy: 93.15999603271484\n",
      "\n",
      " Train Epoch: 45 [60000/60000 (100%)]\tLoss: 0.015049\t accuracy: 99.46%\t Time: 0:00:09.266894\n",
      "\t Test loss: 0.3195186257362366 \t Test accuracy: 93.25999450683594\n",
      "\n",
      " Train Epoch: 46 [60000/60000 (100%)]\tLoss: 0.014184\t accuracy: 99.47%\t Time: 0:00:09.294775\n",
      "\t Test loss: 0.3198965162038803 \t Test accuracy: 93.04000091552734\n",
      "\n",
      " Train Epoch: 47 [60000/60000 (100%)]\tLoss: 0.014259\t accuracy: 99.45%\t Time: 0:00:09.369349\n",
      "\t Test loss: 0.33562563806772233 \t Test accuracy: 93.06999969482422\n",
      "\n",
      " Train Epoch: 48 [60000/60000 (100%)]\tLoss: 0.014705\t accuracy: 99.42%\t Time: 0:00:09.301212\n",
      "\t Test loss: 0.3297470986843109 \t Test accuracy: 92.97999572753906\n",
      "\n",
      " Train Epoch: 49 [60000/60000 (100%)]\tLoss: 0.014312\t accuracy: 99.46%\t Time: 0:00:09.360780\n",
      "\t Test loss: 0.32556616216897966 \t Test accuracy: 93.30999755859375\n",
      "\n",
      " Train Epoch: 50 [60000/60000 (100%)]\tLoss: 0.013575\t accuracy: 99.47%\t Time: 0:00:09.316601\n",
      "\t Test loss: 0.33325162529945374 \t Test accuracy: 93.12999725341797\n",
      "\n",
      " Train Epoch: 51 [60000/60000 (100%)]\tLoss: 0.012578\t accuracy: 99.53%\t Time: 0:00:09.358912\n",
      "\t Test loss: 0.3309946686029434 \t Test accuracy: 93.14999389648438\n",
      "\n",
      " Train Epoch: 52 [60000/60000 (100%)]\tLoss: 0.012334\t accuracy: 99.60%\t Time: 0:00:09.348685\n",
      "\t Test loss: 0.3356934651732445 \t Test accuracy: 93.22000122070312\n",
      "\n",
      " Train Epoch: 53 [60000/60000 (100%)]\tLoss: 0.012240\t accuracy: 99.53%\t Time: 0:00:09.248661\n",
      "\t Test loss: 0.3412322700023651 \t Test accuracy: 93.14999389648438\n",
      "\n",
      " Train Epoch: 54 [60000/60000 (100%)]\tLoss: 0.012364\t accuracy: 99.54%\t Time: 0:00:09.335524\n",
      "\t Test loss: 0.3414233192801476 \t Test accuracy: 93.0199966430664\n",
      "\n",
      " Train Epoch: 55 [60000/60000 (100%)]\tLoss: 0.010118\t accuracy: 99.59%\t Time: 0:00:09.296231\n",
      "\t Test loss: 0.33687679618597033 \t Test accuracy: 92.93999481201172\n",
      "\n",
      " Train Epoch: 56 [60000/60000 (100%)]\tLoss: 0.010143\t accuracy: 99.62%\t Time: 0:00:09.342644\n",
      "\t Test loss: 0.3473061889410019 \t Test accuracy: 93.22000122070312\n",
      "\n",
      " Train Epoch: 57 [60000/60000 (100%)]\tLoss: 0.009803\t accuracy: 99.62%\t Time: 0:00:09.306327\n",
      "\t Test loss: 0.3463709056377411 \t Test accuracy: 93.12999725341797\n",
      "\n",
      " Train Epoch: 58 [60000/60000 (100%)]\tLoss: 0.010499\t accuracy: 99.58%\t Time: 0:00:09.434058\n",
      "\t Test loss: 0.34769490361213684 \t Test accuracy: 93.22999572753906\n",
      "\n",
      " Train Epoch: 59 [60000/60000 (100%)]\tLoss: 0.010854\t accuracy: 99.55%\t Time: 0:00:09.325937\n",
      "\t Test loss: 0.3480085551738739 \t Test accuracy: 93.38999938964844\n",
      "\n",
      " Train Epoch: 60 [60000/60000 (100%)]\tLoss: 0.009134\t accuracy: 99.65%\t Time: 0:00:10.048541\n",
      "\t Test loss: 0.3598631590604782 \t Test accuracy: 93.27999877929688\n",
      "\n",
      " Train Epoch: 61 [60000/60000 (100%)]\tLoss: 0.009537\t accuracy: 99.63%\t Time: 0:00:09.281219\n",
      "\t Test loss: 0.3599990904331207 \t Test accuracy: 93.19999694824219\n",
      "\n",
      " Train Epoch: 62 [60000/60000 (100%)]\tLoss: 0.009669\t accuracy: 99.64%\t Time: 0:00:09.260246\n",
      "\t Test loss: 0.35194430947303773 \t Test accuracy: 93.27999877929688\n",
      "\n",
      " Train Epoch: 63 [60000/60000 (100%)]\tLoss: 0.009455\t accuracy: 99.62%\t Time: 0:00:09.253679\n",
      "\t Test loss: 0.36437949538230896 \t Test accuracy: 93.00999450683594\n",
      "\n",
      " Train Epoch: 64 [60000/60000 (100%)]\tLoss: 0.007765\t accuracy: 99.68%\t Time: 0:00:09.315269\n",
      "\t Test loss: 0.3657155454158783 \t Test accuracy: 93.29000091552734\n",
      "\n",
      " Train Epoch: 65 [60000/60000 (100%)]\tLoss: 0.008964\t accuracy: 99.67%\t Time: 0:00:09.337128\n",
      "\t Test loss: 0.3581859201192856 \t Test accuracy: 93.22000122070312\n",
      "\n",
      " Train Epoch: 66 [60000/60000 (100%)]\tLoss: 0.009737\t accuracy: 99.63%\t Time: 0:00:09.405087\n",
      "\t Test loss: 0.35865474343299864 \t Test accuracy: 93.25\n",
      "\n",
      " Train Epoch: 67 [60000/60000 (100%)]\tLoss: 0.007547\t accuracy: 99.72%\t Time: 0:00:09.547166\n",
      "\t Test loss: 0.3570860832929611 \t Test accuracy: 93.33999633789062\n",
      "\n",
      " Train Epoch: 68 [60000/60000 (100%)]\tLoss: 0.007145\t accuracy: 99.68%\t Time: 0:00:09.330461\n",
      "\t Test loss: 0.3655595451593399 \t Test accuracy: 93.18000030517578\n",
      "\n",
      " Train Epoch: 69 [60000/60000 (100%)]\tLoss: 0.008940\t accuracy: 99.67%\t Time: 0:00:09.432739\n",
      "\t Test loss: 0.3614060044288635 \t Test accuracy: 93.31999969482422\n",
      "\n",
      " Train Epoch: 70 [60000/60000 (100%)]\tLoss: 0.008845\t accuracy: 99.65%\t Time: 0:00:09.390816\n",
      "\t Test loss: 0.3672251045703888 \t Test accuracy: 93.16999816894531\n",
      "\n",
      " Train Epoch: 71 [60000/60000 (100%)]\tLoss: 0.007774\t accuracy: 99.70%\t Time: 0:00:09.287584\n",
      "\t Test loss: 0.36474879682064054 \t Test accuracy: 93.3499984741211\n",
      "\n",
      " Train Epoch: 72 [60000/60000 (100%)]\tLoss: 0.007639\t accuracy: 99.67%\t Time: 0:00:09.321475\n",
      "\t Test loss: 0.35896356105804444 \t Test accuracy: 93.29999542236328\n",
      "\n",
      " Train Epoch: 73 [60000/60000 (100%)]\tLoss: 0.006698\t accuracy: 99.74%\t Time: 0:00:09.371419\n",
      "\t Test loss: 0.35025014281272887 \t Test accuracy: 93.23999786376953\n",
      "\n",
      " Train Epoch: 74 [60000/60000 (100%)]\tLoss: 0.006346\t accuracy: 99.72%\t Time: 0:00:09.357987\n",
      "\t Test loss: 0.37449328005313876 \t Test accuracy: 93.38999938964844\n",
      "\n",
      " Train Epoch: 75 [60000/60000 (100%)]\tLoss: 0.006878\t accuracy: 99.72%\t Time: 0:00:09.312526\n",
      "\t Test loss: 0.3716102600097656 \t Test accuracy: 93.18999481201172\n",
      "\n",
      " Train Epoch: 76 [60000/60000 (100%)]\tLoss: 0.007186\t accuracy: 99.70%\t Time: 0:00:09.319189\n",
      "\t Test loss: 0.3708547502756119 \t Test accuracy: 93.27999877929688\n",
      "\n",
      " Train Epoch: 77 [60000/60000 (100%)]\tLoss: 0.007063\t accuracy: 99.72%\t Time: 0:00:09.357602\n",
      "\t Test loss: 0.3772864371538162 \t Test accuracy: 93.14999389648438\n",
      "\n",
      " Train Epoch: 78 [60000/60000 (100%)]\tLoss: 0.006182\t accuracy: 99.72%\t Time: 0:00:09.341837\n",
      "\t Test loss: 0.37167791426181795 \t Test accuracy: 93.37999725341797\n",
      "\n",
      " Train Epoch: 79 [60000/60000 (100%)]\tLoss: 0.005693\t accuracy: 99.77%\t Time: 0:00:09.291335\n",
      "\t Test loss: 0.37761319875717164 \t Test accuracy: 93.25\n",
      "\n",
      " Train Epoch: 80 [60000/60000 (100%)]\tLoss: 0.007354\t accuracy: 99.71%\t Time: 0:00:09.270314\n",
      "\t Test loss: 0.37298767268657684 \t Test accuracy: 93.19999694824219\n",
      "\n",
      " Train Epoch: 81 [60000/60000 (100%)]\tLoss: 0.007477\t accuracy: 99.72%\t Time: 0:00:09.283907\n",
      "\t Test loss: 0.37060480415821073 \t Test accuracy: 93.22999572753906\n",
      "\n",
      " Train Epoch: 82 [60000/60000 (100%)]\tLoss: 0.006524\t accuracy: 99.74%\t Time: 0:00:09.334848\n",
      "\t Test loss: 0.37784287333488464 \t Test accuracy: 93.36000061035156\n",
      "\n",
      " Train Epoch: 83 [60000/60000 (100%)]\tLoss: 0.005871\t accuracy: 99.76%\t Time: 0:00:09.295226\n",
      "\t Test loss: 0.3944758266210556 \t Test accuracy: 93.30999755859375\n",
      "\n",
      " Train Epoch: 84 [60000/60000 (100%)]\tLoss: 0.005332\t accuracy: 99.81%\t Time: 0:00:09.285315\n",
      "\t Test loss: 0.38701200783252715 \t Test accuracy: 93.29999542236328\n",
      "\n",
      " Train Epoch: 85 [60000/60000 (100%)]\tLoss: 0.005328\t accuracy: 99.78%\t Time: 0:00:09.252280\n",
      "\t Test loss: 0.3849873661994934 \t Test accuracy: 93.29000091552734\n",
      "\n",
      " Train Epoch: 86 [60000/60000 (100%)]\tLoss: 0.004738\t accuracy: 99.82%\t Time: 0:00:09.271400\n",
      "\t Test loss: 0.37511946856975553 \t Test accuracy: 93.37999725341797\n",
      "\n",
      " Train Epoch: 87 [60000/60000 (100%)]\tLoss: 0.005275\t accuracy: 99.78%\t Time: 0:00:09.269518\n",
      "\t Test loss: 0.38675864934921267 \t Test accuracy: 93.29000091552734\n",
      "\n",
      " Train Epoch: 88 [60000/60000 (100%)]\tLoss: 0.005081\t accuracy: 99.79%\t Time: 0:00:09.314209\n",
      "\t Test loss: 0.38761823177337645 \t Test accuracy: 93.29000091552734\n",
      "\n",
      " Train Epoch: 89 [60000/60000 (100%)]\tLoss: 0.005658\t accuracy: 99.76%\t Time: 0:00:09.332965\n",
      "\t Test loss: 0.39428270161151885 \t Test accuracy: 93.29999542236328\n",
      "\n",
      " Train Epoch: 90 [60000/60000 (100%)]\tLoss: 0.005445\t accuracy: 99.77%\t Time: 0:00:10.057441\n",
      "\t Test loss: 0.39164571464061737 \t Test accuracy: 93.33999633789062\n",
      "\n",
      " Train Epoch: 91 [60000/60000 (100%)]\tLoss: 0.004932\t accuracy: 99.83%\t Time: 0:00:09.300570\n",
      "\t Test loss: 0.3919877946376801 \t Test accuracy: 93.41999816894531\n",
      "\n",
      " Train Epoch: 92 [60000/60000 (100%)]\tLoss: 0.004805\t accuracy: 99.81%\t Time: 0:00:09.293940\n",
      "\t Test loss: 0.39280240833759306 \t Test accuracy: 93.36000061035156\n",
      "\n",
      " Train Epoch: 93 [60000/60000 (100%)]\tLoss: 0.005189\t accuracy: 99.79%\t Time: 0:00:09.282728\n",
      "\t Test loss: 0.39918738305568696 \t Test accuracy: 93.23999786376953\n",
      "\n",
      " Train Epoch: 94 [60000/60000 (100%)]\tLoss: 0.004558\t accuracy: 99.81%\t Time: 0:00:09.232582\n",
      "\t Test loss: 0.3951236218214035 \t Test accuracy: 93.32999420166016\n",
      "\n",
      " Train Epoch: 95 [60000/60000 (100%)]\tLoss: 0.003701\t accuracy: 99.85%\t Time: 0:00:09.322664\n",
      "\t Test loss: 0.408807584643364 \t Test accuracy: 93.23999786376953\n",
      "\n",
      " Train Epoch: 96 [60000/60000 (100%)]\tLoss: 0.004463\t accuracy: 99.81%\t Time: 0:00:09.304266\n",
      "\t Test loss: 0.40878275632858274 \t Test accuracy: 93.27999877929688\n",
      "\n",
      " Train Epoch: 97 [60000/60000 (100%)]\tLoss: 0.004084\t accuracy: 99.83%\t Time: 0:00:09.263469\n",
      "\t Test loss: 0.4117244064807892 \t Test accuracy: 93.22999572753906\n",
      "\n",
      " Train Epoch: 98 [60000/60000 (100%)]\tLoss: 0.004955\t accuracy: 99.81%\t Time: 0:00:09.390013\n",
      "\t Test loss: 0.4092125505208969 \t Test accuracy: 93.25999450683594\n",
      "\n",
      " Train Epoch: 99 [60000/60000 (100%)]\tLoss: 0.004994\t accuracy: 99.79%\t Time: 0:00:09.358087\n",
      "\t Test loss: 0.4076308935880661 \t Test accuracy: 93.33999633789062\n",
      "\n",
      " Train Epoch: 100 [60000/60000 (100%)]\tLoss: 0.005438\t accuracy: 99.77%\t Time: 0:00:09.425125\n",
      "\t Test loss: 0.40811256468296053 \t Test accuracy: 93.25\n",
      "\n",
      " Train Epoch: 101 [60000/60000 (100%)]\tLoss: 0.003754\t accuracy: 99.86%\t Time: 0:00:09.416537\n",
      "\t Test loss: 0.41559899151325225 \t Test accuracy: 93.18999481201172\n",
      "\n",
      " Train Epoch: 102 [60000/60000 (100%)]\tLoss: 0.003794\t accuracy: 99.86%\t Time: 0:00:09.402298\n",
      "\t Test loss: 0.4127135783433914 \t Test accuracy: 93.13999938964844\n",
      "\n",
      " Train Epoch: 103 [60000/60000 (100%)]\tLoss: 0.003486\t accuracy: 99.88%\t Time: 0:00:09.375060\n",
      "\t Test loss: 0.41267190873622894 \t Test accuracy: 93.30999755859375\n",
      "\n",
      " Train Epoch: 104 [60000/60000 (100%)]\tLoss: 0.004111\t accuracy: 99.83%\t Time: 0:00:09.410248\n",
      "\t Test loss: 0.42278688549995425 \t Test accuracy: 93.27999877929688\n",
      "\n",
      " Train Epoch: 105 [60000/60000 (100%)]\tLoss: 0.003740\t accuracy: 99.86%\t Time: 0:00:09.481978\n",
      "\t Test loss: 0.41974216401576997 \t Test accuracy: 93.33999633789062\n",
      "\n",
      " Train Epoch: 106 [60000/60000 (100%)]\tLoss: 0.004068\t accuracy: 99.84%\t Time: 0:00:09.610759\n",
      "\t Test loss: 0.4199068546295166 \t Test accuracy: 93.48999786376953\n",
      "\n",
      " Train Epoch: 107 [60000/60000 (100%)]\tLoss: 0.003016\t accuracy: 99.86%\t Time: 0:00:09.547814\n",
      "\t Test loss: 0.43008445799350736 \t Test accuracy: 93.43000030517578\n",
      "\n",
      " Train Epoch: 108 [60000/60000 (100%)]\tLoss: 0.002864\t accuracy: 99.89%\t Time: 0:00:09.513945\n",
      "\t Test loss: 0.42405453622341155 \t Test accuracy: 93.22999572753906\n",
      "\n",
      " Train Epoch: 109 [60000/60000 (100%)]\tLoss: 0.002547\t accuracy: 99.90%\t Time: 0:00:09.454783\n",
      "\t Test loss: 0.42555989921092985 \t Test accuracy: 93.25999450683594\n",
      "\n",
      " Train Epoch: 110 [60000/60000 (100%)]\tLoss: 0.003202\t accuracy: 99.88%\t Time: 0:00:09.516128\n",
      "\t Test loss: 0.4292365849018097 \t Test accuracy: 93.29999542236328\n",
      "\n",
      " Train Epoch: 111 [60000/60000 (100%)]\tLoss: 0.004130\t accuracy: 99.84%\t Time: 0:00:09.478109\n",
      "\t Test loss: 0.4228162169456482 \t Test accuracy: 93.06999969482422\n",
      "\n",
      " Train Epoch: 112 [60000/60000 (100%)]\tLoss: 0.003928\t accuracy: 99.84%\t Time: 0:00:09.493449\n",
      "\t Test loss: 0.43294128179550173 \t Test accuracy: 93.25999450683594\n",
      "\n",
      " Train Epoch: 113 [60000/60000 (100%)]\tLoss: 0.003097\t accuracy: 99.87%\t Time: 0:00:09.533542\n",
      "\t Test loss: 0.43482327163219453 \t Test accuracy: 93.14999389648438\n",
      "\n",
      " Train Epoch: 114 [60000/60000 (100%)]\tLoss: 0.003333\t accuracy: 99.87%\t Time: 0:00:09.473288\n",
      "\t Test loss: 0.43082004487514497 \t Test accuracy: 93.23999786376953\n",
      "\n",
      " Train Epoch: 115 [60000/60000 (100%)]\tLoss: 0.003839\t accuracy: 99.84%\t Time: 0:00:09.465365\n",
      "\t Test loss: 0.43327498733997344 \t Test accuracy: 93.22999572753906\n",
      "\n",
      " Train Epoch: 116 [60000/60000 (100%)]\tLoss: 0.003570\t accuracy: 99.86%\t Time: 0:00:09.538697\n",
      "\t Test loss: 0.43573940694332125 \t Test accuracy: 93.33999633789062\n",
      "\n",
      " Train Epoch: 117 [60000/60000 (100%)]\tLoss: 0.003220\t accuracy: 99.86%\t Time: 0:00:09.418317\n",
      "\t Test loss: 0.443584731221199 \t Test accuracy: 93.3499984741211\n",
      "\n",
      " Train Epoch: 118 [60000/60000 (100%)]\tLoss: 0.003118\t accuracy: 99.88%\t Time: 0:00:09.429484\n",
      "\t Test loss: 0.4393545240163803 \t Test accuracy: 93.39999389648438\n",
      "\n",
      " Train Epoch: 119 [60000/60000 (100%)]\tLoss: 0.003104\t accuracy: 99.87%\t Time: 0:00:10.213200\n",
      "\t Test loss: 0.43042248785495757 \t Test accuracy: 93.2699966430664\n",
      "\n",
      " Train Epoch: 120 [60000/60000 (100%)]\tLoss: 0.003324\t accuracy: 99.88%\t Time: 0:00:09.400090\n",
      "\t Test loss: 0.44196840822696687 \t Test accuracy: 93.40999603271484\n",
      "\n",
      " Train Epoch: 121 [60000/60000 (100%)]\tLoss: 0.003846\t accuracy: 99.85%\t Time: 0:00:09.357984\n",
      "\t Test loss: 0.437187647819519 \t Test accuracy: 93.22999572753906\n",
      "\n",
      " Train Epoch: 122 [60000/60000 (100%)]\tLoss: 0.003168\t accuracy: 99.87%\t Time: 0:00:09.588392\n",
      "\t Test loss: 0.43979112803936005 \t Test accuracy: 93.32999420166016\n",
      "\n",
      " Train Epoch: 123 [60000/60000 (100%)]\tLoss: 0.003395\t accuracy: 99.87%\t Time: 0:00:09.512102\n",
      "\t Test loss: 0.44317795932292936 \t Test accuracy: 93.38999938964844\n",
      "\n",
      " Train Epoch: 124 [60000/60000 (100%)]\tLoss: 0.003390\t accuracy: 99.85%\t Time: 0:00:09.499492\n",
      "\t Test loss: 0.44284993708133696 \t Test accuracy: 93.5199966430664\n",
      "\n",
      " Train Epoch: 125 [60000/60000 (100%)]\tLoss: 0.003621\t accuracy: 99.87%\t Time: 0:00:09.562116\n",
      "\t Test loss: 0.43577209413051604 \t Test accuracy: 93.5\n",
      "\n",
      " Train Epoch: 126 [60000/60000 (100%)]\tLoss: 0.002896\t accuracy: 99.88%\t Time: 0:00:09.532083\n",
      "\t Test loss: 0.4430164784193039 \t Test accuracy: 93.43000030517578\n",
      "\n",
      " Train Epoch: 127 [60000/60000 (100%)]\tLoss: 0.002633\t accuracy: 99.90%\t Time: 0:00:09.509023\n",
      "\t Test loss: 0.4488104432821274 \t Test accuracy: 93.29999542236328\n",
      "\n",
      " Train Epoch: 128 [60000/60000 (100%)]\tLoss: 0.001955\t accuracy: 99.93%\t Time: 0:00:09.776599\n",
      "\t Test loss: 0.4528578549623489 \t Test accuracy: 93.27999877929688\n",
      "\n",
      " Train Epoch: 129 [60000/60000 (100%)]\tLoss: 0.001747\t accuracy: 99.93%\t Time: 0:00:09.598965\n",
      "\t Test loss: 0.4538241297006607 \t Test accuracy: 93.30999755859375\n",
      "\n",
      " Train Epoch: 130 [60000/60000 (100%)]\tLoss: 0.002478\t accuracy: 99.90%\t Time: 0:00:09.756776\n",
      "\t Test loss: 0.44560582637786866 \t Test accuracy: 93.5199966430664\n",
      "\n",
      " Train Epoch: 131 [60000/60000 (100%)]\tLoss: 0.002639\t accuracy: 99.90%\t Time: 0:00:09.626397\n",
      "\t Test loss: 0.451632359623909 \t Test accuracy: 93.47000122070312\n",
      "\n",
      " Train Epoch: 132 [60000/60000 (100%)]\tLoss: 0.002092\t accuracy: 99.92%\t Time: 0:00:09.515152\n",
      "\t Test loss: 0.4488516688346863 \t Test accuracy: 93.29000091552734\n",
      "\n",
      " Train Epoch: 133 [60000/60000 (100%)]\tLoss: 0.001388\t accuracy: 99.96%\t Time: 0:00:09.484489\n",
      "\t Test loss: 0.4570650726556778 \t Test accuracy: 93.38999938964844\n",
      "\n",
      " Train Epoch: 134 [60000/60000 (100%)]\tLoss: 0.001589\t accuracy: 99.94%\t Time: 0:00:09.560250\n",
      "\t Test loss: 0.4589389204978943 \t Test accuracy: 93.30999755859375\n",
      "\n",
      " Train Epoch: 135 [60000/60000 (100%)]\tLoss: 0.001780\t accuracy: 99.94%\t Time: 0:00:09.611339\n",
      "\t Test loss: 0.4525360196828842 \t Test accuracy: 93.48999786376953\n",
      "\n",
      " Train Epoch: 136 [60000/60000 (100%)]\tLoss: 0.002332\t accuracy: 99.92%\t Time: 0:00:09.537915\n",
      "\t Test loss: 0.46410867273807527 \t Test accuracy: 93.40999603271484\n",
      "\n",
      " Train Epoch: 137 [60000/60000 (100%)]\tLoss: 0.001503\t accuracy: 99.95%\t Time: 0:00:09.601598\n",
      "\t Test loss: 0.4590067744255066 \t Test accuracy: 93.3699951171875\n",
      "\n",
      " Train Epoch: 138 [60000/60000 (100%)]\tLoss: 0.002142\t accuracy: 99.91%\t Time: 0:00:09.528700\n",
      "\t Test loss: 0.4646662950515747 \t Test accuracy: 93.32999420166016\n",
      "\n",
      " Train Epoch: 139 [60000/60000 (100%)]\tLoss: 0.001681\t accuracy: 99.93%\t Time: 0:00:09.493583\n",
      "\t Test loss: 0.4564815104007721 \t Test accuracy: 93.44999694824219\n",
      "\n",
      " Train Epoch: 140 [60000/60000 (100%)]\tLoss: 0.002054\t accuracy: 99.92%\t Time: 0:00:09.492182\n",
      "\t Test loss: 0.46638724505901336 \t Test accuracy: 93.3499984741211\n",
      "\n",
      " Train Epoch: 141 [60000/60000 (100%)]\tLoss: 0.002446\t accuracy: 99.90%\t Time: 0:00:09.449618\n",
      "\t Test loss: 0.46491379141807554 \t Test accuracy: 93.22999572753906\n",
      "\n",
      " Train Epoch: 142 [60000/60000 (100%)]\tLoss: 0.001882\t accuracy: 99.93%\t Time: 0:00:09.519320\n",
      "\t Test loss: 0.46468465924263 \t Test accuracy: 93.30999755859375\n",
      "\n",
      " Train Epoch: 143 [60000/60000 (100%)]\tLoss: 0.001290\t accuracy: 99.96%\t Time: 0:00:09.539614\n",
      "\t Test loss: 0.4620191156864166 \t Test accuracy: 93.43000030517578\n",
      "\n",
      " Train Epoch: 144 [60000/60000 (100%)]\tLoss: 0.001715\t accuracy: 99.93%\t Time: 0:00:09.476552\n",
      "\t Test loss: 0.47348441183567047 \t Test accuracy: 93.32999420166016\n",
      "\n",
      " Train Epoch: 145 [60000/60000 (100%)]\tLoss: 0.001296\t accuracy: 99.95%\t Time: 0:00:09.518245\n",
      "\t Test loss: 0.4692289590835571 \t Test accuracy: 93.3499984741211\n",
      "\n",
      " Train Epoch: 146 [60000/60000 (100%)]\tLoss: 0.001609\t accuracy: 99.94%\t Time: 0:00:09.491765\n",
      "\t Test loss: 0.46615682542324066 \t Test accuracy: 93.29999542236328\n",
      "\n",
      " Train Epoch: 147 [60000/60000 (100%)]\tLoss: 0.001759\t accuracy: 99.93%\t Time: 0:00:09.478107\n",
      "\t Test loss: 0.46311271786689756 \t Test accuracy: 93.37999725341797\n",
      "\n",
      " Train Epoch: 148 [60000/60000 (100%)]\tLoss: 0.001421\t accuracy: 99.95%\t Time: 0:00:09.590809\n",
      "\t Test loss: 0.4695258170366287 \t Test accuracy: 93.38999938964844\n",
      "\n",
      " Train Epoch: 149 [60000/60000 (100%)]\tLoss: 0.001732\t accuracy: 99.94%\t Time: 0:00:09.962467\n",
      "\t Test loss: 0.46802498400211334 \t Test accuracy: 93.36000061035156\n",
      "\n",
      " Train Epoch: 150 [60000/60000 (100%)]\tLoss: 0.002093\t accuracy: 99.92%\t Time: 0:00:09.514167\n",
      "\t Test loss: 0.4736798942089081 \t Test accuracy: 93.25999450683594\n",
      "\n",
      " Train Epoch: 151 [60000/60000 (100%)]\tLoss: 0.001781\t accuracy: 99.93%\t Time: 0:00:09.505056\n",
      "\t Test loss: 0.4760215163230896 \t Test accuracy: 93.25\n",
      "\n",
      " Train Epoch: 152 [60000/60000 (100%)]\tLoss: 0.001210\t accuracy: 99.97%\t Time: 0:00:09.518970\n",
      "\t Test loss: 0.47661066949367525 \t Test accuracy: 93.43000030517578\n",
      "\n",
      " Train Epoch: 153 [60000/60000 (100%)]\tLoss: 0.001370\t accuracy: 99.95%\t Time: 0:00:09.542597\n",
      "\t Test loss: 0.48387548327445984 \t Test accuracy: 93.2699966430664\n",
      "\n",
      " Train Epoch: 154 [60000/60000 (100%)]\tLoss: 0.001759\t accuracy: 99.94%\t Time: 0:00:09.504894\n",
      "\t Test loss: 0.4755702495574951 \t Test accuracy: 93.25999450683594\n",
      "\n",
      " Train Epoch: 155 [60000/60000 (100%)]\tLoss: 0.001299\t accuracy: 99.95%\t Time: 0:00:09.476918\n",
      "\t Test loss: 0.4823435515165329 \t Test accuracy: 93.3699951171875\n",
      "\n",
      " Train Epoch: 156 [60000/60000 (100%)]\tLoss: 0.001845\t accuracy: 99.93%\t Time: 0:00:09.491638\n",
      "\t Test loss: 0.4797886490821838 \t Test accuracy: 93.32999420166016\n",
      "\n",
      " Train Epoch: 157 [60000/60000 (100%)]\tLoss: 0.001658\t accuracy: 99.94%\t Time: 0:00:09.493441\n",
      "\t Test loss: 0.47898799180984497 \t Test accuracy: 93.23999786376953\n",
      "\n",
      " Train Epoch: 158 [60000/60000 (100%)]\tLoss: 0.001860\t accuracy: 99.94%\t Time: 0:00:09.532343\n",
      "\t Test loss: 0.4795629233121872 \t Test accuracy: 93.3699951171875\n",
      "\n",
      " Train Epoch: 159 [60000/60000 (100%)]\tLoss: 0.002063\t accuracy: 99.93%\t Time: 0:00:09.535227\n",
      "\t Test loss: 0.48379203379154206 \t Test accuracy: 93.3699951171875\n",
      "\n",
      " Train Epoch: 160 [60000/60000 (100%)]\tLoss: 0.001464\t accuracy: 99.95%\t Time: 0:00:09.545256\n",
      "\t Test loss: 0.48315146565437317 \t Test accuracy: 93.36000061035156\n",
      "\n",
      " Train Epoch: 161 [60000/60000 (100%)]\tLoss: 0.001288\t accuracy: 99.94%\t Time: 0:00:09.515149\n",
      "\t Test loss: 0.48493083715438845 \t Test accuracy: 93.41999816894531\n",
      "\n",
      " Train Epoch: 162 [60000/60000 (100%)]\tLoss: 0.002047\t accuracy: 99.90%\t Time: 0:00:09.523379\n",
      "\t Test loss: 0.4789374828338623 \t Test accuracy: 93.2699966430664\n",
      "\n",
      " Train Epoch: 163 [60000/60000 (100%)]\tLoss: 0.000955\t accuracy: 99.97%\t Time: 0:00:09.482120\n",
      "\t Test loss: 0.48651803135871885 \t Test accuracy: 93.45999908447266\n",
      "\n",
      " Train Epoch: 164 [60000/60000 (100%)]\tLoss: 0.001085\t accuracy: 99.96%\t Time: 0:00:09.678957\n",
      "\t Test loss: 0.4825561702251434 \t Test accuracy: 93.25\n",
      "\n",
      " Train Epoch: 165 [60000/60000 (100%)]\tLoss: 0.001108\t accuracy: 99.96%\t Time: 0:00:09.518950\n",
      "\t Test loss: 0.5003556042909623 \t Test accuracy: 93.32999420166016\n",
      "\n",
      " Train Epoch: 166 [60000/60000 (100%)]\tLoss: 0.001013\t accuracy: 99.96%\t Time: 0:00:09.499386\n",
      "\t Test loss: 0.4991769134998322 \t Test accuracy: 93.33999633789062\n",
      "\n",
      " Train Epoch: 167 [60000/60000 (100%)]\tLoss: 0.001288\t accuracy: 99.94%\t Time: 0:00:09.599390\n",
      "\t Test loss: 0.5020662009716034 \t Test accuracy: 93.30999755859375\n",
      "\n",
      " Train Epoch: 168 [60000/60000 (100%)]\tLoss: 0.001129\t accuracy: 99.97%\t Time: 0:00:09.541531\n",
      "\t Test loss: 0.5101614683866501 \t Test accuracy: 93.20999908447266\n",
      "\n",
      " Train Epoch: 169 [60000/60000 (100%)]\tLoss: 0.001071\t accuracy: 99.96%\t Time: 0:00:09.526811\n",
      "\t Test loss: 0.5105735778808593 \t Test accuracy: 93.31999969482422\n",
      "\n",
      " Train Epoch: 170 [60000/60000 (100%)]\tLoss: 0.000837\t accuracy: 99.97%\t Time: 0:00:09.545031\n",
      "\t Test loss: 0.5088773131370544 \t Test accuracy: 93.32999420166016\n",
      "\n",
      " Train Epoch: 171 [60000/60000 (100%)]\tLoss: 0.000973\t accuracy: 99.97%\t Time: 0:00:09.562944\n",
      "\t Test loss: 0.49551807045936586 \t Test accuracy: 93.40999603271484\n",
      "\n",
      " Train Epoch: 172 [60000/60000 (100%)]\tLoss: 0.000867\t accuracy: 99.98%\t Time: 0:00:09.517565\n",
      "\t Test loss: 0.505307388305664 \t Test accuracy: 93.45999908447266\n",
      "\n",
      " Train Epoch: 173 [60000/60000 (100%)]\tLoss: 0.000874\t accuracy: 99.97%\t Time: 0:00:09.528023\n",
      "\t Test loss: 0.49806272983551025 \t Test accuracy: 93.5\n",
      "\n",
      " Train Epoch: 174 [60000/60000 (100%)]\tLoss: 0.000659\t accuracy: 99.98%\t Time: 0:00:09.562428\n",
      "\t Test loss: 0.5043317556381226 \t Test accuracy: 93.3699951171875\n",
      "\n",
      " Train Epoch: 175 [60000/60000 (100%)]\tLoss: 0.000641\t accuracy: 99.99%\t Time: 0:00:09.523303\n",
      "\t Test loss: 0.5092911750078202 \t Test accuracy: 93.40999603271484\n",
      "\n",
      " Train Epoch: 176 [60000/60000 (100%)]\tLoss: 0.000395\t accuracy: 99.99%\t Time: 0:00:09.595521\n",
      "\t Test loss: 0.5109501361846924 \t Test accuracy: 93.47999572753906\n",
      "\n",
      " Train Epoch: 177 [60000/60000 (100%)]\tLoss: 0.000445\t accuracy: 100.00%\t Time: 0:00:09.514351\n",
      "\t Test loss: 0.5134999096393585 \t Test accuracy: 93.43999481201172\n",
      "\n",
      " Train Epoch: 178 [60000/60000 (100%)]\tLoss: 0.000935\t accuracy: 99.97%\t Time: 0:00:10.267449\n",
      "\t Test loss: 0.511391070485115 \t Test accuracy: 93.47000122070312\n",
      "\n",
      " Train Epoch: 179 [60000/60000 (100%)]\tLoss: 0.000639\t accuracy: 99.99%\t Time: 0:00:09.491346\n",
      "\t Test loss: 0.5124162316322327 \t Test accuracy: 93.41999816894531\n",
      "\n",
      " Train Epoch: 180 [60000/60000 (100%)]\tLoss: 0.000477\t accuracy: 99.99%\t Time: 0:00:09.538351\n",
      "\t Test loss: 0.5157376080751419 \t Test accuracy: 93.31999969482422\n",
      "\n",
      " Train Epoch: 181 [60000/60000 (100%)]\tLoss: 0.000719\t accuracy: 99.98%\t Time: 0:00:09.559162\n",
      "\t Test loss: 0.5194885849952697 \t Test accuracy: 93.45999908447266\n",
      "\n",
      " Train Epoch: 182 [60000/60000 (100%)]\tLoss: 0.000535\t accuracy: 99.99%\t Time: 0:00:09.534085\n",
      "\t Test loss: 0.5128354221582413 \t Test accuracy: 93.40999603271484\n",
      "\n",
      " Train Epoch: 183 [60000/60000 (100%)]\tLoss: 0.000407\t accuracy: 99.99%\t Time: 0:00:09.583695\n",
      "\t Test loss: 0.5177047729492188 \t Test accuracy: 93.52999877929688\n",
      "\n",
      " Train Epoch: 184 [60000/60000 (100%)]\tLoss: 0.000226\t accuracy: 100.00%\t Time: 0:00:09.537331\n",
      "\t Test loss: 0.5204257160425186 \t Test accuracy: 93.48999786376953\n",
      "\n",
      " Train Epoch: 185 [60000/60000 (100%)]\tLoss: 0.000246\t accuracy: 100.00%\t Time: 0:00:09.477034\n",
      "\t Test loss: 0.5193239152431488 \t Test accuracy: 93.43000030517578\n",
      "\n",
      " Train Epoch: 186 [60000/60000 (100%)]\tLoss: 0.000540\t accuracy: 99.98%\t Time: 0:00:09.488280\n",
      "\t Test loss: 0.5213761746883392 \t Test accuracy: 93.38999938964844\n",
      "\n",
      " Train Epoch: 187 [60000/60000 (100%)]\tLoss: 0.001122\t accuracy: 99.96%\t Time: 0:00:09.501423\n",
      "\t Test loss: 0.5209697097539902 \t Test accuracy: 93.43000030517578\n",
      "\n",
      " Train Epoch: 188 [60000/60000 (100%)]\tLoss: 0.000388\t accuracy: 99.99%\t Time: 0:00:09.533397\n",
      "\t Test loss: 0.5173685163259506 \t Test accuracy: 93.43999481201172\n",
      "\n",
      " Train Epoch: 189 [60000/60000 (100%)]\tLoss: 0.000515\t accuracy: 99.98%\t Time: 0:00:09.523869\n",
      "\t Test loss: 0.5196193248033524 \t Test accuracy: 93.41999816894531\n",
      "\n",
      " Train Epoch: 190 [60000/60000 (100%)]\tLoss: 0.001001\t accuracy: 99.97%\t Time: 0:00:09.622504\n",
      "\t Test loss: 0.5216112703084945 \t Test accuracy: 93.41999816894531\n",
      "\n",
      " Train Epoch: 191 [60000/60000 (100%)]\tLoss: 0.000571\t accuracy: 99.98%\t Time: 0:00:09.575853\n",
      "\t Test loss: 0.523773518204689 \t Test accuracy: 93.29999542236328\n",
      "\n",
      " Train Epoch: 192 [60000/60000 (100%)]\tLoss: 0.001225\t accuracy: 99.95%\t Time: 0:00:09.511558\n",
      "\t Test loss: 0.5169226408004761 \t Test accuracy: 93.39999389648438\n",
      "\n",
      " Train Epoch: 193 [60000/60000 (100%)]\tLoss: 0.000758\t accuracy: 99.97%\t Time: 0:00:09.551434\n",
      "\t Test loss: 0.5187291175127029 \t Test accuracy: 93.29999542236328\n",
      "\n",
      " Train Epoch: 194 [60000/60000 (100%)]\tLoss: 0.000708\t accuracy: 99.98%\t Time: 0:00:09.555560\n",
      "\t Test loss: 0.5206519484519958 \t Test accuracy: 93.31999969482422\n",
      "\n",
      " Train Epoch: 195 [60000/60000 (100%)]\tLoss: 0.000777\t accuracy: 99.98%\t Time: 0:00:09.525462\n",
      "\t Test loss: 0.5217676252126694 \t Test accuracy: 93.5\n",
      "\n",
      " Train Epoch: 196 [60000/60000 (100%)]\tLoss: 0.000957\t accuracy: 99.97%\t Time: 0:00:09.481147\n",
      "\t Test loss: 0.5247333139181137 \t Test accuracy: 93.47000122070312\n",
      "\n",
      " Train Epoch: 197 [60000/60000 (100%)]\tLoss: 0.000969\t accuracy: 99.97%\t Time: 0:00:09.518456\n",
      "\t Test loss: 0.5157062143087388 \t Test accuracy: 93.41999816894531\n",
      "\n",
      " Train Epoch: 198 [60000/60000 (100%)]\tLoss: 0.000911\t accuracy: 99.97%\t Time: 0:00:09.478582\n",
      "\t Test loss: 0.5178808122873306 \t Test accuracy: 93.45999908447266\n",
      "\n",
      " Train Epoch: 199 [60000/60000 (100%)]\tLoss: 0.000769\t accuracy: 99.97%\t Time: 0:00:09.511959\n",
      "\t Test loss: 0.5241554617881775 \t Test accuracy: 93.45999908447266\n",
      "\n"
     ]
    }
   ],
   "source": [
    "model = get_model(ARCHITECTURE, DATASET).to(device)\n",
    "\n",
    "# Specifying the background-optimizer of SAM. This can be changed to any optimizer that is supported by PyTorch. Examples:\n",
    "# In order to use PHB:        optimizer = SAM(model.parameters(), torch.optim.SGD, lr=0.1, momentum=0.8)\n",
    "# In order to use Adam:       optimizer = SAM(model.parameters(), torch.optim.Adam)\n",
    "# In order to use AdaBound:   optimizer = SAM(model.parameters(), AdaBound)\n",
    "# In order to use AdaShift:   optimizer = SAM(model.parameters(), AdaShift)\n",
    "optimizer = SAM(model.parameters(), torch.optim.SGD, lr=0.1)\n",
    "\n",
    "sam_optimizer_name = 'SAM_' + optimizer_name\n",
    "model = train(model, optimizer, train_loader=train_loader, device=device, epoch_num=MAX_EPOCH, max_nbr_epochs=MAX_EPOCH, path=sam_optimizer_name, val_dataloader=test_loader, sam=True, dir_path=store_dir)"
   ]
  },
  {
   "cell_type": "code",
   "source": [
    "lr = 0.1 if DATASET == 'FashionMNIST' else 1\n",
    "num_epochs = 100000\n",
    "computed = False\n",
    "batch_size = 128\n",
    "\n",
    "path = os.path.join(store_dir, 'epoch200', sam_optimizer_name + '.pt')\n",
    "checkpoint = torch.load(path, map_location=torch.device('cpu'))\n",
    "model = checkpoint['state_dict']\n",
    "model = get_model(ARCHITECTURE, DATASET).to(device)\n",
    "model.load_state_dict(checkpoint['state_dict'])\n",
    "\n",
    "\n",
    "while not computed:\n",
    "    try:\n",
    "        # Calculating the sharpness. Returns an error if the learning rate is too big\n",
    "        sharpnesses, losses = minimum_shaprness_eff(data, model, batch_size, lr, num_epochs=num_epochs, optimizer_file=path)\n",
    "\n",
    "        # storing the sharpness\n",
    "        sharpness_path = os.path.join(store_dir, 'epoch200', sam_optimizer_name + '_sharpness.pt')\n",
    "        checkpoint = {'sharpnesses':sharpnesses, 'sharpness':sharpnesses[-1], 'losses': losses}\n",
    "        torch.save(checkpoint, sharpness_path)\n",
    "\n",
    "        # Plotting sharpness as an objective function being minimized during time\n",
    "        plt.ylabel('Sharpness')\n",
    "        plt.xlabel('Epoch [x100]')\n",
    "        plt.plot(sharpnesses)\n",
    "        plt.show()\n",
    "\n",
    "\n",
    "        computed = True\n",
    "        print(f'Sharpness: {sharpnesses[-1]}')\n",
    "    except:\n",
    "        # Error is returned if the learning rate is too big, so in that case learning rate is set to be twice smaller and number of epochs are set to be twice as bigger\n",
    "        computed = False\n",
    "        lr /= 2.0\n",
    "        num_epochs *= 2\n",
    "        print(f'Use smaller stepsize than {lr}')"
   ],
   "metadata": {
    "id": "iPLBIguqajZW",
    "colab": {
     "base_uri": "https://localhost:8080/",
     "height": 424
    },
    "outputId": "f072fc44-6f2f-463f-b3ea-d96d29934bf6",
    "pycharm": {
     "name": "#%%\n"
    }
   },
   "execution_count": null,
   "outputs": [
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "\t Calculating Hessian\n"
     ]
    },
    {
     "output_type": "stream",
     "name": "stderr",
     "text": [
      "/usr/local/lib/python3.7/dist-packages/torch/nn/modules/module.py:1033: UserWarning: Using a non-full backward hook when the forward contains multiple autograd Nodes is deprecated and will be removed in future versions. This hook will be missing some grad_input. Please use register_full_backward_hook to get the documented behavior.\n",
      "  warnings.warn(\"Using a non-full backward hook when the forward contains multiple autograd Nodes \"\n"
     ]
    },
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "\tFinished the diag calculation. Time needed: 0:01:01.459540. Computing sharpness...\n",
      " \t\t epoch:100000\t processed 100.0%\t loss:0.016041555268829076 \t minimum sharpness: 28238.661864613856 \t Time needed 0:01:46.752965\n",
      "--------------------------------------------------\n"
     ]
    },
    {
     "output_type": "display_data",
     "data": {
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ],
      "image/png": "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\n"
     },
     "metadata": {
      "needs_background": "light"
     }
    },
    {
     "output_type": "stream",
     "name": "stdout",
     "text": [
      "Sharpness: 28238.661864613856\n"
     ]
    }
   ]
  },
  {
   "cell_type": "markdown",
   "source": [
    "## Loading the data"
   ],
   "metadata": {
    "id": "Th3AcINbffKZ",
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "markdown",
   "source": [
    "Here we demonstrate how to load the data. We compare here SGD and SAM with SGD. Plots used in the report are shown in the *Data analysis* notebook."
   ],
   "metadata": {
    "id": "A-5nn6CHfic_",
    "pycharm": {
     "name": "#%% md\n"
    }
   }
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Minimum sharpness of SGD: 108900.66487971791\n",
      "Minimum sharpness of SAM: 28238.661864613856\n"
     ]
    },
    {
     "data": {
      "text/plain": "<Figure size 1332x756 with 2 Axes>",
      "image/png": "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\n"
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Loading information about SGD training\n",
    "checkpoint_sgd = torch.load(os.path.join(store_dir, 'epoch200', optimizer_name + '.pt'), map_location=torch.device('cpu'))\n",
    "losses_sgd = checkpoint_sgd['training_loss']\n",
    "acc_sgd = checkpoint_sgd['validation_accuracy']\n",
    "sharpness_sgd = torch.load(os.path.join(store_dir, 'epoch200', optimizer_name + '_sharpness.pt'))['sharpness']\n",
    "\n",
    "# Loading information about SAM SGD training\n",
    "#checkpoint_sam = torch.load(os.path.join(store_dir, 'epoch200', 'SAM_SGD.pt'), map_location=torch.device('cpu'))\n",
    "checkpoint_sam = torch.load(os.path.join(store_dir, 'epoch200', sam_optimizer_name + '.pt'), map_location=torch.device('cpu'))\n",
    "losses_sam = checkpoint_sam['training_loss']\n",
    "acc_sam = checkpoint_sam['validation_accuracy']\n",
    "sharpness_sam = torch.load(os.path.join(store_dir, 'epoch200', sam_optimizer_name + '_sharpness.pt'))['sharpness']\n",
    "\n",
    "# Plotting both\n",
    "fig, ax = plt.subplots(2,1)\n",
    "fig.set_size_inches(18.5, 10.5, forward=True)\n",
    "\n",
    "ax[0].set_title(\"Training loss comparison\")\n",
    "ax[0].loglog(losses_sgd, label='SGD')\n",
    "ax[0].loglog(losses_sam, label='SAM')\n",
    "ax[0].set_xlabel('Epoch')\n",
    "ax[0].set_ylabel('Training loss')\n",
    "ax[0].legend()\n",
    "\n",
    "ax[1].set_title(\"Test accuracy comparison\")\n",
    "ax[1].loglog(acc_sgd, label='SGD')\n",
    "ax[1].loglog(acc_sam, label='SAM')\n",
    "ax[1].set_xlabel('Epoch')\n",
    "ax[1].set_ylabel('Test accuracy')\n",
    "ax[1].legend()\n",
    "\n",
    "# Writing sharpness\n",
    "print(f'Minimum sharpness of SGD: {sharpness_sgd}')\n",
    "print(f'Minimum sharpness of SAM: {sharpness_sam}')"
   ],
   "metadata": {
    "collapsed": false,
    "pycharm": {
     "name": "#%%\n"
    }
   }
  }
 ],
 "metadata": {
  "colab": {
   "collapsed_sections": [],
   "name": "Training.ipynb",
   "provenance": []
  },
  "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": "3.7.6"
  },
  "gpuClass": "standard",
  "accelerator": "GPU"
 },
 "nbformat": 4,
 "nbformat_minor": 0
}