testing_fastai.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Testing out fastai"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "from fastai.vision.all import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "path = '../cv_data/cifar10/train'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "50000"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "files = get_image_files(path)\n",
    "len(files)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "fastcore.foundation.L"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "type(files)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Dataloading take 2\n",
    "cifar10 = '../cv_data/cifar10'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "dls = ImageDataLoaders.from_folder(cifar10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAgAAAAIHCAYAAADpfeRCAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/Il7ecAAAACXBIWXMAAAsTAAALEwEAmpwYAACmD0lEQVR4nO39d7Rl2XXeh8619z75nHvPjVV1763cXZ3RDTS6EYhEACZIkZSYRIqiqEeloTFMWk/yG5asQFmWFZ7ek61nPsqk5WdLFm0GkSIlMQcEErHRAZ1D5XRzOjnuvdf7owpWzfnNul2AOhxgz98YPYB1ap4d1l5rnVV7fvVN570nwzAMwzCyRfB2X4BhGIZhGG89tgEwDMMwjAxiGwDDMAzDyCC2ATAMwzCMDGIbAMMwDMPIILYBMAzDMIwMYhsAgXPuHufcV5xzbefcX3m7r8cwvh6cc5edcx9XPv+gc+61N+JYhvFmYmvxm0/0dl/ABPLXiegz3vt3vt0XYhhvNN77zxLRPW/3dRjGHWBr8ZuMvQFAjhPRS9ofOOfCt/haDOMtwzlnfyEwJglbi99kbANwC865TxHRtxLRTzvnOs65n3fO/Yxz7recc10i+lbn3H3Ouc845xrOuZecc3/8lu/POed+3TnXcs496Zz7B865z71tN2Rknceccy875/adc//SOVd0zn3EOXf9qwE3X+//Defc80TUdc5Fzrkfdc5dcc7tOuf+9tt4/UZGsbX4rcE2ALfgvf8oEX2WiH7Ce18lohER/Wki+odEVCOiJ4jo14no94hokYj+CyL6P51zX32l+s+JqEtEh4no/3bzP8N4u/gRIvoEEZ0mojNE9HduE/fDRPSdRFS/GfczRPSjRLRERHNEtPJmX6hh3IqtxW8NtgF4ff699/7z3vuUiB4hoioR/T+99yPv/aeI6DeI6IdvvpL6fiL6b7z3Pe/9y0T0v79tV20YRD/tvb/mvd+jGwvnD98m7qduxvWJ6AeI6De893/kvR8S0U8SUfoWXa9hHIStxW8wtgF4fa7d8v+XiOjazQH4Va4Q0TIRLdANUeW123zXMN5qbh1/V+jG+H29uKVb2977LhHtvvGXZhhfM7YWv8HYBuD1ubVc4hoRHXXO3dpvx4holYi2iSgm/rr06Jt/eYZxW24df8foxvjVuHWMr9/6PedcmW6kAQzj7cbW4jcY2wB8bTxBN/JKf905l3POfYSIvpuIftF7nxDRrxLR33POlZ1z9xLRn33brtQwiH7cObfinJslor9FRL90B9/5FSL6LufcB5xzeSL6+2TrhDF52Fr8BmAT+2vAez8ioj9ORN9BRDtE9D8R0Z/13r96M+QniGiaiDaI6OeI6BeIaPg2XKphEBH9PN0QSV28+d8/eL0veO9fIqIfv/nddSLaJ6LrB37JMN5ibC1+Y3De+9ePMr4unHP/hIgOe+9NgWoYhvE2YWuxjr0BeANxzt3rnHuHu8HjRPQXiOjX3u7rMgzDyBK2Ft8Z5vz1xlKjG6+alohoi4j+eyL692/rFRmGYWQPW4vvAEsBGIZhGEYGsRSAYRiGYWQQ2wAYhmEYRgY5UAMwPT3N8gM5JV1QLvI9RCVXgpj2aMDapQrGJMJstL4wCzEuGbP2uIf/qqPb5eciByFE+RxrBkphqWK1zNqFWgVipurTrN1oNyAmSfgFbJ9X/kVVOuLNAPdl+aDA2v1+D2K8uI1jdx2HmLHokO1r63g5SZe163X0gem3eN/Hilvs+tqG1vtvKT/3s7/EBu2oh/c7bl9m7aiTQEzBH+IxVIYYebMRHoam5viHQSkHMcNoirXbEfa/z/Ex0mqch5he42V+feM9iJk58mHWDisn8FxhkbfTBsQkw1dYOwgOQwxFi6w5inH5aexww7bW/irElMp11q7UFINDz5/P7g7Ou/GIj+EgwHXgH/7UX3vbxzAR0d/9q3+VjeNmbwwxacTHUr5UgJhSno+bQauNxxnw9WhuFtfiUczPH7sRxBw9fYKfuzYNMVHEfwsuvHYZYr7wh59h7YX5OsTkc/zeS8Uixoh1f+HIIsR0xG9VrYoxd9/9EGtfvnIFYl584RnWLoQxxKxdF2N9dxNivu0D72btqQo+05eucaPOMMKYn/6X/1odx/YGwDAMwzAyiG0ADMMwDCODHJgCyDm+PwgifIsQiVeYoyG+9yyV+Glq0/j6tNPn3xt6fGVSLeRZO+liCiBJ+HGcwz1OKeCvq4MAX6cFI57u2LmOr2f6/Tprl+cW8HraHdauFPC+cgnvn0YPX6cNxdvJqICvecpzNdYulDHVMm7z1/s5JbXR3ezzczcx3VARr5l8OBFvSoFiwPuk2OlDTHyOd2771ZcgptXkr/Ryyj+eqYhXjKVyDWLGZfEgS0o6rMifSb5UhRhf5WmCHOH4LOT4sy7O46vyYvUIa6c5fHWapHzM8vorN3DBEdHGuZAm4l7HOH+LET9/Ye6Uci4x9hKc46MRf7VdKuCaUxDpyjSd3KKHpVqdtaMK3vP69jZrt3axhtPSEk/NHL37DMS89jJP5wxD/JkYiZxtuT4PMVF1hrUTZS2Oe3yMtnY2IKZA/LV8ORhAzHjAn3dviOfaG/LxtreH5ypP83RHMMbjjHv7rH3qJKZaGw2ebrt68VWIKVbE2pvOQEyzw++1WlRSjwO+PocFJfd4G+wNgGEYhmFkENsAGIZhGEYGsQ2AYRiGYWQQ2wAYhmEYRgY5UARYKXHBko9RJFMocmHeSBH2wD/yH6HIbXqaC6b2lH+fmoRcZFXWziUEfvkQ/521G3JhnitiN0yJf/u+cGoFYrY7TdYedjsQM1vhoo2klIeYYsL7Oappoi/eP9PTdYhxQiTZ3N2CmET4B2j/XjYR/564O0aRZN/zz/Ij/DfUk4Bvc8FOroHCKL/XYO20jP9eedDmYy0KUPRYP8KFb8sP3gsxQZmPB0VjRP0hF/6MxPUREfUvvcbaaQNFgMUifybhiSmIcW6HXx9aDpAX/8Y6TXBOpTEXgY0JVZLxmAu+kjGKlYJAiJwUcelo2BPtFsRInWK+iGLLXJ7PM0XbODG09rjAL1dE8a6LucA16TUgphBxseax48cg5tXXuGCt0+9CTK3Mz3/0yGm8npSvq629NYjZ3RD+FQP0a7h3iU+SIjUhJlcVz1eZn3v7fA5v7e1DTLvHjx200TfksuPj7cjJ+yHmxBIX9LU3cX0shEJ8nODvUKfHx3q3h8+9luPzKBk3IOZ22BsAwzAMw8ggtgEwDMMwjAxiGwDDMAzDyCAHagBSkccLlSRZIjyhXYS5v3TMP+spBj6FAs/h5CLMmXiR3+/HeJxI5FVCJRfkBvyzkZL7a+5yDUKk5MH3hElDkCqmEQV+jcU5TLLmQp7X2b62AzEk8pyDBuavRqKfkxGa3szUuZagVENjiUEo+nmoGL8Iw5juCPtnEiiX+TOauvsIxFRPiWdSwBz36ss8V5nsYv9XRf2Iyj1oYhMJrUvP4XxxMc/pTacY4y9yH/H2V16AmPblq6w9fBoNjkYXuXYgPXUUYugY178Ec+iPTo6b88gaGERETphHFcuodQnF/NXspXyLGxP1eqh1GYj8dy2Hz70Q8DEcKHVMJoVozPVF3RbWdRgLTZIbYe7+2ML7WXu2ijnl40e5YVRNeU6B0INVAtR1bW9xj/xzrzwLMT1R66GOEimarfL5mHO4HuWFlKmn6MOqJb4+O8KTBcJ0aBG9vGi08yJrX+1tQ8ww5ReUG+Hzeud9d7H22QvYh2vX+Pycnca6DMUav8gdpabA7bA3AIZhGIaRQWwDYBiGYRgZxDYAhmEYhpFBbANgGIZhGBnkYBGgEPLklapQUSQ+y6N4r7/HxShhEUUlY+LfCyLFwEcIT4YeRTvFAr9mn2L1vUgI2FyI+yA35MKrYg8rUJVGXKzV6mPVvGTEhU9BDqv4NbpctNEfoLFJRPxcUQEFLHGP32s6xnsPhYYkX8Zn6gP+vZoizMk5HjMqTKYRUG2KC2Tmp9AMR1bxyyl9W5zhxh57ayjUHLW4UCt2yhgWFecKORx7hRx/Jm6IzzGe5uKt8QyqldItPs6LTTTXGmxz45VhD+8raXORnbv/PogpHTvJ2ueuoDAqEULB+oIiyJwSVeW8UnYx4H3WH6IB1/XrXPB4NMRnWijWWTtXUBRfE8LJE1yYd/HSZYjpCrOqQ9N4PxsXzrJ2opiXRdI8K8W1Lxnwz14+9yWI6Y/4OpYboTDPC5OhJuoWqZDydb6sVLvrirXOK79VsRhLy0ewgmGeuBBvPMI5UyjxeZXkFalqnx/n6DJWx1xaEKLhChoKbe9+hbV/93Ofg5higd9XVTF3ux32BsAwDMMwMohtAAzDMAwjg9gGwDAMwzAyyIEaAOl9k89huMzVV+cwv18ThSt6I8zh5EVRoQGm0ykSOcTSPOZVxgNe7GU0wMIR73v34/w7CZrYvHbuAmtXFG2Di/h9FMqoSchXec45GuKN1fN8H9YsYQ5nZ8CTY1NKP0d5HpMvYP/4Kn+GAyUvNzXDrznpYWIuLyxa8gHmuyeBsMRNfgpV7FshAVDK2BB58Uz2FNOnzX3+bAcec4OLAS80VFU0GBEJo5NYMVlKeM4z7qPp01hoUpwyznPyXEoRrtFrl3h7F2PogQZr7m+jOU9D6Hy67ZMQc/eD38LazuG8y+d5brtQxFzuaMz7dX31IsQUCvxZFEtYBGpSmD52grWLHVxHGle5nmN+/jDEJH3+7IY71yDm9DyfM+M+apKurV9m7f2NqxBTrPA1vaiYsi3M8nWsO8bZV5iqs/ZMSdHW5Pj3mu0GxMi6cF7RSI2FERUVZiCGCtI8Cc2UpsRvQT6P60W7JYrJxajZOH3fo6x9RRFJXLsoDL6+BlM2ewNgGIZhGBnENgCGYRiGkUFsA2AYhmEYGcQ2AIZhGIaRQQ4UAcYJFwjFRQx3Qug07KCorCaMO/JVFHGMHY+JlUpNlSp3sSkUUTQxHnFRSb+DlZFqFSH2GaPRxdI8P9dmswEx2/tCDFVBk5lcxO+r10FR4rTjfVjOo1nQkbll1i4EKJapLvHz5+ex8uB2R1zzGJ9FlON9mCpmSpUa78OtBgqFJoF4yAUxaaKIFfN8XI8T7NvLVzdY+yvPvQwxVy6ts/ax5SWIqa9xkepdJw9BzIkV/tzKNRwPaZeLgQaKoKnbEoIhRTgm6+0phTEp9dzUxDVQcDjc4eYx01UUfHUqvJ8bkXbNZ1i7XMLqZ7mQ90e1iiLA+QVeiXFn+zWIaQuDo+kZfBaTQjfl93z01L0Y0+DPu6qItnOOr88z04opm6ikd/nKJYjZa/JxPDeLa199igtn01Qx8Bnw+an495AXQtBhgKK7ktDuTddQPFoXJmD9AYrlWkJMmyQ4I4YdPh9CaYZHRHEiK1biWC/1xXzoo8CvMeJmQadP3QMxUcSvcXPtOsTcDnsDYBiGYRgZxDYAhmEYhpFBbANgGIZhGBnkQA1ApcLzTpUShs8tLLJ2f6QYkrR53jkZYY613ed5+GJ5EWLmD/M8+OLiMsTs7fBcbWMLcy+vXDrP2lrRjMUFbqLjSpiHjYVpxPU9zO+vXeF5HafkajuO53ByJcyfThfqrF1Q8tSjDu/DbuM8xMSisEwhh7myXIcXCCkqxhtRjucJa7k7N594K+k2+b20c5iHdDHXmzTaI4j5yjPcbOPlVy5ATKPBz7W13YCYRJj6XLvvBMS87z0PsfZD96NhjpvlufG0hCZUA1E8K+mi1oWE9sY5/DtBKDQ8Lsaxl9vaY+35Bh4nmeH93KmUIabfXmPtQh5jIsc/Kxbw3g8dWmHtRlOZCwnXRIxjTSMxGSQ9vq7KojVERItV3r91xbysJHRKsWIO9cKLL7L2/h7qqCpT/BnMzmBefmGaxxTyqOtq9/k6koS4zkYVrokZxZiXjztck1AkpUBPwudnqYzXE1X5b8HW9i7ETE9z/dPsAhqudbr8vra3GhAz6IvfAkWAs7vBi2r1E7zm2RJ/ppWl43ig22BvAAzDMAwjg9gGwDAMwzAyiG0ADMMwDCOD2AbAMAzDMDLIgSLAcokLDnwPRTKtDhe5leewopb3XN0wVCqXBcSFZlGI4p9Dy0dZ+6Mf+hDEPPvZz7L29cE+xJwWJhr3n7kLYpzQOe02UAzy0osvsHbl6hWI2WpwcVS/r4j3hPhjqAhzGjtcDBIUUJg36nORS6hU3yrGXAQ3HmJMSXyvP0Rzp74oFVmrKKKbCaDZ5GO2IB8sEaXCLGh9vwMx60LQ1+4qYtcRF2a1FNOn/oAL8dIYxVyyatp8AYU/bneHtRur2xDTl9eY4r07oTwKtP5JZYyiVhLGVDllXFVHQvTkcG4mi3wOpVWsaJfmuMFMmMP+qU1xc6BqtQ4xkTDK8R4FupNCfiQEvk183jTiY31hEY2oqMwFY888/xyErG5wQ6vFWRRZFvPCJG6EJjZJzPs3KODPzfxcnR93Cs3LckJ0t7ODosQ9IVzOORT7es/HaKFYh5hKmY+tII/mUImofJqvo5lSPuT9Ux9hFdJcyD9r9xoQU0j5HNlVRO3LwnAsLd/5WmxvAAzDMAwjg9gGwDAMwzAyiG0ADMMwDCODHKgB6ImCI5FSGMFv8xz3YIT5aydya6MB5pQLIucfhGhikY94ziRSTCNGIqd6bB5NGj7+/g+y9vQi5p1eeoEbv9RnsCjJe9/9OI+poW7h7IWzrL21h1qCjQbPFWsFa4r+9fdqZWEGkyhFfHLCHGS/jc+iKXJlshgHEVEw5nkmX8Cc2yQQFvm9xIR552ab51f3mziuTt17P2tXp2cgZvsqf9Z7a3sQ097n/VQaoN6g8yo/zkur1yAm3OLjKNzHcRUIkxUlvQ8agMi/fv4wVTQAY/FRWzlOv8cvYLSKBaQ6n+c56WIBc7CVUzznHyhFuPJlnpetz6KWIBfx4zjCcT4pLCzw6y/kcemuiPmfm0KDs4bQYaztoE6FhPHTrFLox3k+biuKqU5OaFe8YjKViLVmOMbfj0is6Wm/DTGzwlTqkLJe12b4Oj9QzIKubvH1oVRGTdvTL/H5ufr0qxBTFnn4d584AjH1Kh+jJcUoabbC+2fbo5lXr98S7Ts3tLI3AIZhGIaRQWwDYBiGYRgZxDYAhmEYhpFBbANgGIZhGBnkQBGgFPsMlO1CSZhppH0Ug8VjLuIYdtA0orjABRpBDgUaa5evs/Yf9D8HMa2dVda+dwaFMNPTXNQyNYUil0OLXHzU2NuCmCkRMxjifR0RIsT1zQ2I+fVP/yFrJ6QYmxT4NXqHYpmgyB9Q0kKxTCCEgZUiCp/2ezwmcHg9U47HuAn1UPF5XqUsCfAZ+RGvQDfqoDgtcrzC1t33YoW+Rx/m4yHeRNVd54UGa7fOoQhreP0qa/dHKPwRPiPKiKE7krTlhRBKEwqSiBl7DGoQ/wyllkRjsZ64IR4nWOXzY++ZZ/BAeS4GnrrnXggJA373tQqKgeW9BorYdVLY2eL9cnQJxZE7QmydFtHAZ+EQX2fzJaziNxJGVEmMQuFcwNf5IMRn6UN+nEIF1+JEmFMNezj3QifNvNA8qzzNjZ+K1XmImZrm9761iYZCQ8cFwLMFFAQfFqLIfaUC7rEVLto8uoyi4YKYtd0+ir87Qz4m84q4sSDEjQ2l4uztsDcAhmEYhpFBbANgGIZhGBnENgCGYRiGkUEO1ADkcjwnEUSYD6kd5bmo5jbmnafy/HvtgWL2IIx/nFLIZjzi5hPjFuZwlkXea34Rcy+HHjjDP1CuZ6ZeZ+39jTWICUVudH4W8zPbovhLtYB5ufe/6x2s/cql6xAz6PI8WC/BvJwX+aKyYsaSjIURTYS5u27M94VVUbCDiKha58Yb7bXLEDMJjDzv7zYdg5hCnuf0VpZfgZhel4+1NI9757C8zI+7/FGIKa1+hbXzOTQRaaVcF+Ac5gZD+VGK5jzDlD9rpQ4VlYQ5S8EpY0YU8+p51Pl0pXmUYrKCiw3mKp04zuD8eYjpiHmWm8M5XjrMC6RUqovKucT15bBgy6Rw5RIfJ+EYDaRqU3xObm9chZjhGjeVirtoIJUL+Hooi1wREcXCQGoqwflQEGudU4yApmo8Jsop61Gbz4e5KXyWxQI37AlzONr22rzIUX+4AzGVkOfl61JsQ0QffhfXA72v8hDEDEb8Ppqb6xCzs8+VMi+9ioZfFzf4c07LONY7W9xwbH4eTYduh70BMAzDMIwMYhsAwzAMw8ggtgEwDMMwjAxiGwDDMAzDyCAHGwENuUin10JDkvIcN5K4654TELN9nQuoggGKbQrCcCFyeGmFEhdoHF2sQ8yxCv9s/jBW+qvOclOQIEXhSbvJxYz1w2gkkgpx1FCpvtcWJi7zC3gcn3JBXzBGkdVWg5tN9BUBVZjj/bOlVJEbJvx7hTyKtWYrokpaisLOqMCFWFMLWG1tEkhF/6cOx94w4MKaJPcuiCkUuRjNjVE8lexxUU88xHONO1zMifJToqDAv1dK8RnlIv5ZJ0ahVk+It7pKpbW+GLOaFU4qTH5ixS1ICmIDxSwIP1MEh2JYRz2sbDY4e461W4qwNnovV0kWimh4E4S8n10wuUZA1Qq/1lYbrZba7W3W7nQxptfjgrqVeVxnW+JnQQoziYiiiAsOh/jTQCHxZ5BTKrxGwkBIW49GEf/edA2N27z4u2wa4PpYrfJrzjVxPsyX+e9Q2eO8Go14HzYU47DnX77C2uvbGFOb4uvOMMJxPL/CBYdXNvCZLkzzqoKjIQrEb4e9ATAMwzCMDGIbAMMwDMPIILYBMAzDMIwMcqAGoCwK8uSUXJAXufJIyes1BrwAS0J4nFye553DCC9tRhRCOH58CWIWQ3GuRCnIss/NFRaPrUDM4bvvYu0YvVioscvzwNtKXq4rDHwqitmIj6WhEJo9DERK65H774OYRx5+N2v/q//5f4aY186/yNpVpdiFE4U+Rn3FCGSdmxXlpqYhZhJIUp7nixOlbE7K98HDFGPaY65xKIxwnJcavE+isy9AjL/ATTtKTSwkEshiLErOvSlMfnaVcd4XRlChYsQiZ5kyzCkU5kA5QnMUEuZA2DsKijFRIHQ1TtHnjDZ4YZxRDxPQ/XUeM/PQOyCmfOo0a0d1nHeTQlTk8/TJp56GmLtP8vWwXsM1dDzgc3l+EfPpK0LX1djDglWJ58fORThnooCPt8ChtkkWbRr0cK2JAv5Z6tHAZyzmda6orEee58o9YUxnyK9xcw/X9Ocv8/z+Ey9dhpitTb7uf/zjH4OYRx59nLVfvXAFYoI2X/jLU2g2F4mJ1FCu+XbYGwDDMAzDyCC2ATAMwzCMDGIbAMMwDMPIILYBMAzDMIwMcqAIMAi4IOfEITTViXPcvODSuUsQM9jnIpJkjCKieMzFDlphLi9EiHGCRg61OW72cPUyGrZcPs8NcqYOo4nNSF6jRxOLVIisioqRQ7XChScDxTFj5dgJ1l7bQgOfgPh9xD0Uj508cZK1/9yP/wTE/OzP/DRrr1/AanRlYQQUtxUzjCE3thi27tx84q0kEULWfF4RPTpRiVIRK+0K05BeE8Vp6QY/Tu4VFOO4Nn/+mumTT/jYayUYsyeEgn1FdZeEwhxFMddKxbm8oswriM8Kyt8bvFgrlOKEIPpzilRQFgENlOuJhOlRvI+isGG7wdo7e7gOzPS4GLj2wIMQMym8coGvq7udLsScEYLeRDGxCYRZWFRAg6SpIh8nlTLGbO1wMWvg8IEnCR/r0jyIiKhY4LLTOMQxmoz4b0NXEVsPYmFe1Mffhpyo+lquogAykqJExeDIF7nwfXUfK1ZWyvOsfebkKYg5cZRXD43yKKS8fI2b6FX20JRtJNyzAsLndTvsDYBhGIZhZBDbABiGYRhGBrENgGEYhmFkkAM1ALFIPTalGw0RjbzIq8ygUcG4y/MWe0r+eiQKl1RSzHuORQ5nt4X5kEQU2wkitDa5dokbttz18P0QE0Z8b5SPMF857PE8+MI83nslxw1INtfXIOb61cusPVaKCi3Ua6w92NuCmE/91q+x9tGHH4OYex54J2tfVTQAsShOVKmjICMMRM4tVlxdJoBkzMdIkK9CjNQAeCU3XQj5OPK7mIeMz19g7bxSPIvEuJa6FiKiVGpdlH26zLhGEcYkCb+PZoJ52rw4vcz3ExFJ/6CcYijkSfYhnkt6+mj9nAoRQKSICUJxjZpOIBQna23ivBu/8Cxrx1KAMEGs7vJx7Is4jqMS/ywKUCcQC0+f/hD7d7rKtQRT8zWI8Tk+/7fXVyFmNOLrfBhibjoShX7KFaVoUyzuNcHfj95I/BYoGrLZJW701MbuIT/kY3s8xLG1t8PN1Iohar8euJf/prTauBZ84fNfYu3jp45DzCFR8C5UDJeurK6ztiysdxD2BsAwDMMwMohtAAzDMAwjg9gGwDAMwzAyiG0ADMMwDCODHFwNsMKNG9ojNHspznHhW1pQKo7VhUiih8fpjrkiY0YxlkiFCLCjqDhaHS70KJbQ7CGUQiNlGxSGXJySJCjMO3b8BL+eDTQk2b7OBYfOodCoXhYimwBFN+MKr1zVamKFrqvXzrL2l194EWL2mlxsmShVF0dC8BiEKMgMhHgnJDQdmQRKxMdaMdmEGEd8jAyV/p8u8phAEf44YR4VFrBvvRDmpYrJTyQq+yn6U4qFUHSgGO90hH5pX4mRcqGyMj69qEYo20REXsyptIAqrJGTVRfxekhWMIzR0CUQ9x4p1yOGJ8Upjs+BEK4lJXymk8JdD/BKn2dfwUqTm8Kc5/CpeYhJ+0IsF+Na3BvwZ1eexqp5lVlhmBPj+jgQAvFmF8V71Srv80DREsslKh8qvw1Dvh72RBVWIqJRh/9WDQcoOOz2+G/K2WvrELO9w+/j0Qew0uTx07zS5GYHRYCdPf570RthzNzyEda+pPzGDIRwsVqsQ8ztsDcAhmEYhpFBbANgGIZhGBnENgCGYRiGkUEO1ADIoiiaTUYYiLxeH/PyU3M8pzpqK4VjxvwzzZMjVxRFagLcv1zb5zmuQ7OHIObwbJ1/oOTBUrE3KlfQeCMs81yZV0wjqqLgRLeL5kUrK9wA4rJiqrHb4cYz3isJVJG/rSnJ4+ttngMvVjG/Fwc8n9cbYv502OcxhfJk7iVzIR8zBYcGPsWQF4pp0l0QEziez0xz2LdxhWfUhx3Mg+cHvC+1gjhhyMdRXkmMJsTn5l4fc7ANYcbTy+MzkocuKcY7MptaDpVrzvEDDUI81zjH+8cFuPyUhmJcjfG+glSeC0KoK65RCaEk4fO+t7uhRE0Gd4uc8v4O5qZ3pTnYaVz7jizzwmfrylozHPInvre9DTG1Gl/TazMLEJMvcvOsXhPNy3oD/gxmqmjcFgb8esYj1BIMe2LN6uI4Lhd5zGiMx2mJ6j9tRVxz7DQ3+bn/gYcgZneP95nrY37/4YcfYe3ZWZwPfWEcNh7gWpwXRZbc1/D3+slctQ3DMAzDeFOxDYBhGIZhZBDbABiGYRhGBrENgGEYhmFkkANFgCQqDx2aQWOJoaiSlptCgdCUENC5Y4pBSo/LdEZ9FGiMR1wMEvdRMNIQwsVivgwxx4R48A8//QcQc987uLlDRZYyI6J4wK8xV8IqTLNzc6y9sXkdYlJxG+NYEXoEvH9chOcaCSOJ2jRWjnro3vtYuzfCc33lVWEgFCgSqoSL4vqJJrN6+/EhN/8IQhxX5ZA/E+/QCCgVArpWGUU93SneB1GC1SF761yEGQ9R5JaK6nbFHO7Tp8RYW1Mq/XU9P3bq8BkNhc/OYIzjvCPaJWXVqBX5vE8GaOAj/VtyStEyX+bHKSvixkiYJ/UU85h9Ybg0lM5ARDQW/eyGDTzQhFCvcMOchx9C4dlLT3+OtROP97x8fJm1t5WqonHMH1ROWWtCzwdBOcTnNCMqmJ5vNCDmtz/P15rH3vUIxJDn6/6+MDwiItpf4+LzlSNzEFM5JOcIjtGhEDxv7bUgZvHYA6ydhPg71Onxa+618TiNBv9sahqFlFtCgDnq4Xo9P8+/1x/gGnc77A2AYRiGYWQQ2wAYhmEYRgaxDYBhGIZhZJCDiwGVebGEWDEAceIItQrm3KM8N0RZPD0DMa1tnmncubILMaUGN+ooO8yrVMJF1u52Mcf1/Iv82J986imI+X6RG3/41EmIKc/w+yjXsPBQo8z7Y+HwEsT4Ec9pzs7gcZqeX0+nhTnoKWE61BuiwdHR5aP8+lpojHPpyjnW7nax8FCxzPNesZ/MvWTsuA5i7LBvfcCf0VzpKsQ0PZ8L4Szm3KNj3FQpncFCP0mfj9lY0WCkI55PDVKcd0GXf68mk/lE5EUuVzPXuuL5NQ6UxyhH2kDRwwxH/It5pchRLeTXHEXYh36arxXdEHUseTGs47HSzyIl3VWMmxJZFEwpMjMpfOp3fou1pX6BiCgQl3/lOpoFnTzB88UnTt8NMc889Sxr54toFjYzz9e+Q9OoNxgITdKzr1yDmC+/zNfidAEH4PYWz+9fO4/3dXyRr6v1E6hXEzIzihyOmymxXgcB5vdjUbBqdh71BlIOduEc9s/6doO113bQcMmJ31znUI9xaJEXDNrdR43E7ZjMVdswDMMwjDcV2wAYhmEYRgaxDYBhGIZhZBDbABiGYRhGBjlQBJiL+B/niiiIGE9xUcKghdXuhmMu4pg5eRRiZue50CTpSvsRor0mFwGOeyja2d7iAq6pOlbxy0dcMLV/dRNiXnjiS6x9YhbNYboxF+LMr2BM9RCvyHVEMS15+vPcwCOKcF/mhBlPpBibzAhR4s7eDsR0m1x0E3g08Fg5wq/55XMotozHvO/jdFL3klxEE3sU0Yw9F5pNuysQ0xuusHY/h2LX8AgXFVUWsN9GI/69ToRCzdE2F0/FYxSwpW0+hqdG+BzzomLkwOF8CUQVv9ThcxTDnIaxUjFQ6KkipVJnWwzZssfJUBgKQyHFhMqV+fNKYhRzNZu8f0Z9jIkK/BorBVzfJoW9LS5mnjl8GGLKZb6Gvnb+PMQkIy7ofeyxRyDmzAMPs/YLL74GMedX+dpy5uQxiNkWJjo7Q1yLP/IdP8zap06chpjlY/yap2bwvtyAx3hF9Bk4/nwLylgv1rnYd2UJ+9kLoWBeWa+9GEuHl7F/Fpe5udNeE4XvZWEAdeUSCiD7QgBcLGAV0tsxqau2YRiGYRhvIrYBMAzDMIwMYhsAwzAMw8ggBxcDqvJ8SHu/ASFBXRQi2EZjGd/kOc3hAE1sjt3DjXamSphnzPV5vrRSQpOQ8ZjnnbxT3E+GPB94pIj7oLzj+cnEY1edf+kF1r7we1hU6P3vf5y1D88tQkx9mRfoaO2hJuEdj/PjDJqotQhF3rW4gbkgP+a54rXVSxBTDIXegDA32hWuI+MRPtNJwCUix55i36bxRdYeFzAvv9ngJjbbe5hTDmJehKMS9iAmrPPvBXejJqF4hI/rQQtz5XuX+fXsN/F62qLQUCPB4/RFHtRrXjjye4oRkMvxL6YhzqnGiMfsSOEAEUVN3mczczjHSRx7THicVOgEXKyYmIlD+9Lk/n3oPlH858jSMsTk8nyN+sw2GstsN3i+eDDGez56/ARrzy6dgpjf+/QXWPvCDhpazc7zvPe7l9EA7uSpe1l7qoamQ1ev8vlZyOM1txr8dycfHYGYnBy2Y7zmSo1f43QRx9+5a3zNfOyR+yBmVODjbTBCM7VOlz+LhXksHnZkiRscTdfR4KjR4WtvOX/nWpbJHfGGYRiGYbxp2AbAMAzDMDKIbQAMwzAMI4PYBsAwDMMwMsjBRkA5LgIcxmgsM5XjgoOogsIzKb85NI9ikEGzwdp7eygmXBTVCWXFJSKiyhFu3ODbfYipigpPnSIe6Mzdd/HvlNDkZ29njbU/+6nfg5gXnvsya9enUORy8m5ufjFTQ5OZ81tcvPbgXfdATDXHBSu7jQbENNv8Gdam8XqG13jVrrzodyKi3SYXtVSjg/WkbxfjPjfOGFEDYvpFLmSlIVb32u9yAdv1HcXkp89FdzMVrNBXFiq7XE6pZFfjMZ0IRXeXd3nMdh9jpFFVjJ465ISgKlLMeWTRPueVAwlSmPVEacI/G4+VqoJCuJgr4biKSuIDpWKgF0KoIIdGSfLG4kgRDE8IBTEHd7ZxLa4LI7CFw2g+M1XifT4zh2K5aq3O23lcj77jO7+XtVfXNiBmOBTi7xEqTEdDvj67aRTCjYWAs9lAce3iIS6WK1freD2xmI8e5+f+PjfjqZbxmvMBN7bb20Jh8fwcX0MqDgWZl67zqqu9RDHP6vL+OXrXGYgpii7Lpbim3A57A2AYhmEYGcQ2AIZhGIaRQWwDYBiGYRgZ5MDE7bwopDNbQPOJXshzSkeOYX6/MpZVQLAoxI4wGZpfwfxV0OP5kN19zL3EAc9/DLcaEBPM8PzM4eUliJF1UwadLsTsba2ydimPecZ4wA17VptbEHPu4ldYu1bBnHuQ49qKc6ewaMbjD/MiHvPzmN8PiZtfbGyhgU93xPvZp2iME3RFPyvFXyaBpMuvfRShlqMZ8Dzo/irmBuMOf24yd0lEtNPl08k7fI4kcrCBUpyoL3Kn+x3UG/TFlBoXMJ8uTX1CpfiJE4ZXoVLYJITcOJ4rFUM/HSpVr0TBqEgp2CK1A50B5jPDVBhV5ZXCL6J/tOEpvyb7YpLI52RhNtRaLSxwk5hQMcyJh3w90gzOSuU6aw/GmAdvt7kGaDRCU518jo9tT3iccpkLOgpKQaYTJ7kR0ZVrWKyruXuZtdd3cb12NT6W5sqoHQmEqVROMQWbnub38eUnPgcxp04+wNoVxWxOGuLtNrAA3jjlg3R1HbUWp+46wdpTudfX6HyVyVy1DcMwDMN4U7ENgGEYhmFkENsAGIZhGEYGsQ2AYRiGYWSQA0WArR4341H8Nig/zQV9xxex2t3FL3KRW6QYMLiAix2mF7DqUffCWX6cAAVClUqdtR98DA1z0pCf68zKCYiZq3JxyvNPfR5itje4yUwhj92ZCGOVahU7sTjioo3RACtHJV3+2YvPoADSjRus/W0f/iDE1Ba5MPCzTz4DMf0WN8OYkk4wRNQT9zpUKmtNAqnjotRRoDyjmCvYSq4BMYfre6zdUkxNdofcMGV7IB1riDoxP39EKCb0QjTbT1GoGef5sX1REwHyZyKFrUREaSyq+ClCQRLVIdNQERnl+PdSwjlO4rMgxPnrxHwZxXjR6ZB/5j1ecyTElmmE884JYZ0UDk4SpTIX/U3XpiCmXuefTc2g4PX55/lavL2LwrN6nQvohmMUNxfy/HoS5TnVqnw+lKso8PNCqdps7kFMqcrv4+F3Pgwxzz3NzYEuXD4HMaMZPk7ieezDKWHCVq6gSPfeM8dZ+9JVnJ9yeQhzdYhZOsFNffaffRpiCgPeH7LiKxFRl/u2qcZtt8PeABiGYRhGBrENgGEYhmFkENsAGIZhGEYGOVADEI95fjJfR5Ofqek6aw8GmIuMRzzXN9hBM5zSLC/iM+hjXqUS8nzM7BE08OkNeL7q2GksntBOeG40VXImjUGDtZ97+csQ027z/FmlgOYcfVHMIR1izi0vtAypUpNkuijysIoZy72nuWGGVhPiqRd4buylV16DmNmiyB2HeF+eeM5tqjqZe8mRyP26Ij7rCAxTMFeZG/PPpoeo05ju8X67sod90hE6gZzHh5QTfRs6TE6HeT4XQqXYTRDKvCzmaaUPjCIToCTlQUmAfeg97+eEsGCLD4WpiqLhkYM2UIoKuZD3qyvi86Iy7580QD3GKBA6gUQpGDQheBgn+KR2dnnBmcoUGq6VqzzvvbmLJlPt9iusHSqaj/o8r0AzHqHmI0l4/zZbDYjp9Lmxjfw9ISIqC93UwiHUh7338cdY+9WvQAjtXuX3RWMsOHc04mOppKwFtRk+/k+fWoCY9pB/r9vH/lmY5X34HZ/4NojZuHaJtXtdNDjqCY1GSxTWO4jJXLUNwzAMw3hTsQ2AYRiGYWQQ2wAYhmEYRgaxDYBhGIZhZJADRYCLC1yYF5RREFEWFZ/Wr2Clpr6oiBcoApaoyI+jGTA0hdBkoBiADLpcZPeFp56EGF/j4piZAgqEin0utkg8CoSKoj/GfaVqnhBSpjGK90YJFzwWlH7OiSptVaWiYme7wdq/99J5iNn1QsxWwyEQpLzvm218XsUc/6yoVJGbBFq73KypVK1DzLjMjUb6illQlPIxUgxRBDhX5sK3y3solhsII6BBguO86LjoshKhCLNY4sKswRDH5zgW4lJFqBUkXGQX5PDe05CPGZ9qQl9+fk2ASgEXQLpQGXsJn0NBpBhMiTUnLaOhixcV7ShBMWEY8DnuCQW6k8LyMq9YubGGVeGe/go39brvwQcg5tgxbmJz9mUUAa9d5VVODx85AjH5kjCikqUnieiqqNo3UqoKzgnDN60i4/oad7rZ2roGMTnic2RxDg3pykMueFy7dhZi1oXg8cTxExATD3lMdR5/PyJRGTboNiCmFHBTtiCHxk2Fad73QVUR8na4WdD1q5ch5nZM5qptGIZhGMabim0ADMMwDCOD2AbAMAzDMDLIgRqAnjD16TewAM1oxPPXuT4WN6GxNABBY4muKAKRy2NMkvBjD1Pcv0xVeV6l3W1DzKjFPyssLUPMWBgudBtoGpEKc45Q2U95YS7S66ORQ+R5njNXwDxPnBeGNkplpu11nru7eukSxOw5ns8tjPB6pAlTEJUhZmGB9/NgiMZNk0C5zK89UcxeOk1e/Mgl2CflEs8Plwo4zg/X+XOrr+KYudLlc2qs5KYHor/7is9NJGbuKId5yHGB5zxTj/dFjvdHmEe9QRqInLuSlnee90+YoB7GB/zYXjOYEoW6AqcUDBKFh0gpkCJlCs4r10NijSnMQcyksDjHTdiuKHM7EAXV1lbXIGZ3nWtiWvs4RqvCLOjy5QsQMz3D53+lis9SLDU0VUfDnJkZboaTV0yHFufqrP3k089CzPoqv8ZTh/BZVoUmplbEnPvaVW6m1O/h+DtxDzdcO76IxkSHD/HnVfA4z7eEZmtzdRViKBTGYcr8rAqNTlWJuR32BsAwDMMwMohtAAzDMAwjg9gGwDAMwzAyiG0ADMMwDCODHCgCbApxVDLCCl/9NhfUzQ2wutRCiYuI4iKan8QxNykpt9CUIxAGPv0cih1GQ65QmqmiOEpotWjcQ3FUIoQUiVIVatTm9xrkFcGS2GJNVTCmluP31RwqQkpBo7GLn+1xkSbeOVGlw+91mKI5jBRg5qdQcNjr8n4exJNZSa0yza89TnHsjRMhPBqjGYnL8WcSKGOvJqrm3T2Nop6NNd5vrQ6Kg/o5PtaiMj7JUAh/EkLToXGBC7Vij/NXVhFMIqUUpVDUpV4ZnyJGCluJiHzCx1rqcOy5gN9XUMCx5yL+DJ1mLDbmomKnXLMXzytXRPOYSaEm1r5mYw9iLp7lxjaPvutRiCkI4WM6UJ6lGG/5AgrzykL0l1Oe08KRFdZeOXEKYoZCPLx65SLE7Gzx6rEVqN5JtDTPhYtpiDEbYy76y9dxzsyUuHjv4mW8nnNrXCj4qCIUPHXqKGvP1XG9WMzzsb6wjNVtpVBwfw8NyHo9Pq87XfwNvh32BsAwDMMwMohtAAzDMAwjg9gGwDAMwzAyyIEagJ1VXnQhVAq+zM7wnIlW6KdY4qfZG2N+MJAFeVLMKefLPKcZKgVH1jYus7aL0aThxDw3pEgHeD3tkOcnozwW33Flfo2D8Q7EBKJwSZBgH7ZEznm/gzfWaPG8U1nJg42Jf2+xhEVSzhzmz6ud4Ll2hSai0VLyTh3eP0Xp+jEhhCE3gIlymKuMRP48UcZwkHBNileKQ0XCxObUch1iXlnnmplGE/N1XWHEJIvxEBEFwtgmVcy1ElEgJVXMo7zj9+oUA58g5tcYpaglCIR2wHm8Hi+MsyjQHIV4M03w3p0Ya4FTdALC4MhFeJxIGOeUC5P796Gm0Bv5FMfofWfuZe3FGVz7khF/vrUarhF7+1xfcPreeyBmboGvI2sbuPZFBb5G5HJoKObEuN3ZxeNcX+eFj6aqmLu/+xhf0/OKBmC3y/tsrBjJUcx/Y44p6/54zO/r7PktiNnY4vP85PFDEHPy5Al+fduo60iFedFIKei1O+CfdZS5dzsmd8QbhmEYhvGmYRsAwzAMw8ggtgEwDMMwjAxiGwDDMAzDyCAHigCnhCGJCxRhnjAXGaI2hfZH/Dh7bTT5ObLEq0IpWjnqNfj3DuUUsYMwzOi00eSndogbLnjUEFFziwtPui2sdhcIAdWwr1R2a3OjjRA9ZigRTyFW7r0iHIUKRRQ1Vcvc6KIW4I0FEb+A4QBFXwMhiiwqIqtKRVSIi5UbmwC83OPekUAGY6KCMI1RSuI5YX4zKGD1s0qZ93c+wn7rdHlM0mpAjC8JcZIi0BWaRMp7nJxeCOEU7S05IUQKFPOoSHRZqgrz+PlDh/Ml9byfRz2tqqAQLhYUoWCR90e5gsKxUKwfuaKyEEwIq2tcaNbp4TMoVbiA7do6itNCMSjyJeyXWFRCPbSEBjU9UfW118fn5EJ+nGvX1yGmWuPCwECpELl8jBsIHV3G6q1Xn/8Sa5cUoeqxOhc8tsf489cp8DW0OlWHmDDi42RtYwNiArFen7+8DTGjMX8+86LCIhHR1UuvsfbqOpqLRXPcwOrw0RMQczvsDYBhGIZhZBDbABiGYRhGBrENgGEYhmFkkAM1AMmY59oCrxTvEEfIKzmckThOKY/FTQ5Ncw3AoK8U6En2WXsYYc4uEZ85pdBPq8tNGmiMuch8gd/H2ONxQlEApz6N+bT+kOfGRor3CeX4Psw7TMROl3ieMx1jjksWW2kq+dOmyIH3NUMbx/PSY8VAJifysL0YNRKTwFgILIIQiwEFTuSCA6WMkhCK9MZoajLs8WMPGniuvNCETAeKbiTPv9fpYMGWQBio5PP4jMKQ37tmeJWKMeNTvB4X8v7wAZqjeGGURCHqKAJhVBUoY8+J84dOcfsSeoegiGtOWOHzpaCYx+SESVdRiZkUQjEm63OHIebIMi++k6SoL5E55IpSLG3hxEnWrk7PQcyFC7xITk5Z0/NCpxTmUKuRz/N5dPquByGm2+IGOcuHUZPQbdzP2rurlyGmF/OxpZnW9Vv8N+bI0aMQEwrTurUdnDPDMR/b/RjH8dkLF1g7OX4EYvIF/gxXDtchZkz8XLO4NN0WewNgGIZhGBnENgCGYRiGkUFsA2AYhmEYGcQ2AIZhGIaRQZz3k2ngYhiGYRjGm4e9ATAMwzCMDGIbAMMwDMPIILYBMAzDMIwMYhsAwzAMw8ggtgEwDMMwjAxiGwDDMAzDyCC2ATAMwzCMDGIbAMMwDMPIILYBMAzDMIwMYhsAwzAMw8ggtgEwDMMwjAxiGwDDMAzDyCC2ATAMwzCMDJLpDYBz7rJz7uNv93UYxjcSzrnPOOf+4m3+7JhzruOcC18v1jAO4nbrs3Pug865196IY2WdTG8ADOMblUn9YfXeX/XeV733ydt9LcY3J977z3rv73m7r+ObAdsA/CfinIve7mswDMMwbD3+WrENANEjzrnnnXNN59wvOeeKRETOub/knDvvnNtzzv0H59zSV7/gnPPOuR93zp0jonPuBv/MObd18zjPO+cevBlbcM79U+fcVefcpnPuZ51zpbfpXo0Jwzn3XzvnLjjn2s65l51z33vz87/nnPs/bok7cXPcRc65f0hEHySin775uv2nb8a83zn35M0x+KRz7v23fP8zzrl/4Jz7ws3v/Lpzbs45938651o340/cEn/bY93ktHPuyzf//N8752bldd7mfv+8c+4V59y+c+53nXPH36i+NL4peezmvNh3zv1L51zROfcR59z1rwbcfL3/N5xzzxNR9+Yc+VHn3BXn3K5z7m+/jdc/0dgGgOgHiejbiegkEb2DiH7MOfdRIvrHN//sCBFdIaJfFN/7HiJ6DxHdT0TfRkQfIqIzRFQnoh8iot2bcf/k5uePENFdRLRMRH/3TboX4xuPC3Tjx3yaiP5bIvo/nHNHDvqC9/5vE9Fniegnbr5u/4mbP8C/SUQ/RURzRPQ/ENFvOufmbvnqnyKiH6UbY/A0EX2RiP4lEc0S0StE9N8QEd3hsf4sEf15Iloiovhm7IE4576HiP4WEX0fES3cvIdfeL3vGZnmR4joE3RjvJ4hor9zm7gfJqLvpBvr7xki+hm6MdaX6MYYXnmzL/QbEdsAEP2U937Ne79HRL9ON36of4SI/jfv/TPe+yER/U0iet+tf0Mion/svd/z3veJaExENSK6l4ic9/4V7/26c84R0V8ior92M7ZNRP+IbizEhkHe+1++Of5S7/0vEdE5Inr86zjUdxLROe/9z3nvY+/9LxDRq0T03bfE/Evv/QXvfZOIfpuILnjv/8B7HxPRLxPRO7+GY/2c9/5F732XiH6SiH7wq8K/A/jLdGPevHLznP+IbryBs7cAxu34ae/9tZvr8z+kGz/0Gj91M65PRD9ARL/hvf+jm+v3TxJR+hZd7zcUtgEg2rjl//eIqEo3do1Xvvqh975DN/5Gv3xL7LVb/vxTRPTTRPTPiWjTOfcvnHNTdONvOWUieto513DONYjod25+bhjknPuzzrlnbxkfDxLR/NdxKDZmb3KF+JjdvOX/95V29Ws41jXxZzl6/es+TkT/4y33ukdEThzXMG5FjrOlO4hbIr4+d+k/vpE1bsE2ADprdGOxIiIi51yFbrxGWr0lxt/6Be/9T3nvHyWiB+jGK6j/ioh26MbC+oD3vn7zv2nvfZWMzHPzb77/CxH9BBHNee/rRPQi3fhR7NKNzeNXOSy+7kWbjdmbHCM+Zu+UOznWUfFnY7ox3g/iGhH95VvmQt17X/Lef+HruEYjG8hxtnabuFvnw/qt33POlenG+m0IbAOg8/NE9Oecc4845wp041XlE977y1qwc+4x59x7nHM5urFwD4go8d6ndGOB/2fOucWbscvOuU+8JXdhTDoVurFwbRMROef+HN14A0BE9CwRfcjd+Hf103QjDXUrm0R06pb2bxHRGefcn74pgvohuqFP+Y2v47ru5Fh/xjl3/83F9e8T0a/cwT/9+1ki+pvOuQeIiJxz0865P/l1XJ+RHX7cObdyU5fyt4jol+7gO79CRN/lnPuAcy5PN8an/dYpWKcoeO8/STfyRv+WbuwmT9PBefspuvFDv083XlPtEtE/vflnf4OIzhPRl5xzLSL6AyKyf8NqkPf+ZSL67+mGGG+TiB4ios/f/LPfpxuL3fNE9DThD/n/SEQ/cFMd/VPe+10i+i4i+n/QjfH314nou7z3r/e3cu267uRYP0dE/4pupNCKRPRX7uC4v0Y3RLG/eHMuvEhE3/G1Xp+RKX6eiH6PiC7e/O8fvN4XvPcvEdGP3/zuOt1Yl68f+KWM4ryXbxINwzAMw/hmx94AGIZhGEYGsQ2AYRiGYWQQ2wAYhmEYRgaxDYBhGIZhZJADCyf8/P/3TzOFoIN/ekxUyOVYOwgcHsjxfUYQ4r4jF/FLSWI0bkpT/lm304GYcqXM2rk83mIkz++Vc8lbVbSScRyzdpLgcQbjwYHfISIaDcesPRyMIUaKNQulAsR8/o/OsfbGs1ch5rGP38faR+5CT6Iw5H2mPS9y/Hp2tvYg5K//vz+rDIa3lq+8dJFdaBDlIOaGYeN/RJXFOtnUot6g2/06hLnO4Xfk1ehX9/WcC8dDt9tj7e2NTYiR5wqUYSWHWi5QzAWdnGfKvbvXfxZ3IoD+yEc+/LaPYSKiz/32L3DfkWQEMeloyNoR9BNRKPpqLL5DRDQW689IWbPGjv+LTx/hOpsrVli7kK9BTBrzB56M+hCTiPU5VZ5ICoMJB1ck1rVciGtBKsaEV34b5FzTfvOCkH8WKmtoGPKxnSMc64Ew1/TKuHYBv69UiXnvx75bHcf2BsAwDMMwMohtAAzDMAwjg9gGwDAMwzAyyIEagGtbPB8TRpijKEQ8HxLlcE8h86WhkqMoFXg+Jh5hHrzX5/n0isj3ExE1NhusPT2NeadA3IeWC/SpzAtjLkjm7qVGgQjz+WNF29Dr83xeEqOjapLwz6I8Hqfb5efKKUnW7pAfZ7eF+b0A8sl4nFSIJFav9yBmEsgViqyt5n1lTk85DuaUlVzc13htt0Vco5bOlh8FSl5efs+pV+hfN0bmQXs9zNP22y3WDlUpkNACKTGh+FDP5b8xf2+5A5nAxJCmIuee4BoRJ0KTlGJMIJ53quiWegO+znZ7OLfbQ/7ZSFn78gW+Pk9PYa2oqWqdtbXfmFDkwXHFIhqL82t6rNjzb/pE047wsaX9nuWEjkiZeoTzSokQ1xxrWgLxTe+VdUc4cKdKzO2wNwCGYRiGkUFsA2AYhmEYGcQ2AIZhGIaRQWwDYBiGYRgZ5EARYFv47DhCUUkoDQ8iRXgixD+awC8fcUOK4RCNLqJcnrV32ihOCcWeZm1jG2IKlRJrO8WkQYplNMuQsbwPj8eJx1x4MlYEfu22EFsqCiopAkwJDTxaTf5ZWVGnNNr8OMNtlNTkhLBTU7AkQkDTi9FUYxJwKX9GmrEHmvwoIXIA3IEw705Ugeq5Xu/chAI21X9LzAX1OHdwQR1h8rO7jRWGx2M+rkJFgArmScr1pGJc3Yme6esvaCrFlpOrCkzFuqGJ90ZirUkVAx8pGOt3uxDT3N9nbbn2EBElYo0KQvwp8UIovLeLa3G/w88/P49CwXzx9YW88ho1sbV0d3PK4JICv8DlIcYH/Nj6+OMx6tASn8nnR6RMEeVAIRga3fnf6+0NgGEYhmFkENsAGIZhGEYGsQ2AYRiGYWSQgzUAbZGHV5IduRzPsepFUjjjGGPSMc8zVmpViBnFwjBHyU1JI5HREPMqjSbPYRYrWFiHRK7M5TDGifNruSBpyhDHqH/oD3m+KAxfP8elFRUainstK+ZFLWE6lHZRS5DP3UExINHPw5GSc5sAApmLU/pEmt+oRkDiQ6kRuXkg0byD5L12LnEg1QjoDs6lFQh6PbRiIz1hBBMrOpZIMXCBY99Bsl7G3Fl+/+sWARx47klCmoxJIy4izHuPhji3h8Lkp99DDcBY5KILBcyDB6L4T0cxhyqWeEy5VISYVoPrDSKlqNDMHNcFxMrzxv5RdAtCp6IsBaCt0ApWyc+0Qj+yGJBacE5c80j5bUjS19cbyOJtmhbtdtgbAMMwDMPIILYBMAzDMIwMYhsAwzAMw8ggtgEwDMMwjAxyoAiwIQxqolCp1DQQlf4U1cRoyIUnQynGIKKKEOIlbRSwoAJCqeInP1MEVGPpMaQY5nzw2z7B2k9+/nPKcbiAxnu8LynWGcd4QSMhPIkUgwpptBJr5hxKRS5JnPJ7HY6V6lLiGUaKMExqWlL/+iKwt4NQCOFgfBBWP9RMdaToT9fyyQ+/PmHe12NIo37jTkR34lxSAEaEplza9d2JgE6Knt5M0Z089gR7/NwR8vJB+EVYya6vmKn1B3xdbXfQTG3c4w5wM9EMxOzt77L21evXIaZSrrD2kUMLEFPMcROb/T00mRoJwXOphhVeC2VeeTBSzNRk9T9NRC6rLCbKfEjF76AmwJXVObXhJ489GuFvnhzHmvgz9vx7iv7xttgbAMMwDMPIILYBMAzDMIwMYhsAwzAMw8ggB2oAEpFv0AwPUpH3TsdoZiCLHBRLZYiRiddUy5pAXk+5HuHukCp5Rln85/jRZYj5xEe/g7X7jSbE/P6nPs3avQHmcOZmeB4sGWGCZiSMVTQzlpEw2tE0AD6RBSiUokLSmEjRDcg6I14zbpKJpuDAofS2EQWvbywjc/6qEdDXkUPWznVnh7mDykPis0Q5mTSLknlJIizU1e0OIGakFO+SgDpHGVd3ks98O019JrkYkFx7NcOcfp9/5pWRnC/ytXc2j+Y8fsRN2LR+KQlN0qFDSxBTzPPrWZibhZhIHHq/hYZC8knm82hMlBNagiiH1xwJ3cQ4wHEdinkVBqhtSmHtxbEeBvx6nCIsknNEK94kdTNgZkRKoSitENJtsDcAhmEYhpFBbANgGIZhGBnENgCGYRiGkUFsA2AYhmEYGeRA5VYxL4QVihFQOubCN5+ikKEqKvsFimAsdPx7XhE+eSFiCtQyafwaY6UKUyyEi60Gmk985V/9DGtfv34FYq6ut1j77KV1iHns4eOsffwImmrE4poD5ZqlGGUw0gRdsoqcIoQRqpt8hMeRhd1yiugmFiLA2ONznwSkL5UqzBP9pMrH/IHNrx7pwKb+vTsRq+GBYiH0aTRaENMTFdruRAQ4Vir9SbGeNu1A0HQHIkBdqPf1VANUjiK/eAfP4usonviWkcvxSan13VAIsHOKwC8U4sExuKIRJcTncqPRgJhYmOrMzM5BzEy9ztq1Cl7PzvYG/878IYhZPnqCn1sxXBsl/Jo1gXgk1tBAGxRirGtVLpVfJvzkDqo3ennNWgXDEX8+4wE+r3jEjxMr5kW3w94AGIZhGEYGsQ2AYRiGYWQQ2wAYhmEYRgY52AhIOMKMh2jS4FKedypVShAjs4GKhwVJDwRtZyKzKNL058b1iAyNkiurV7ghxUr9EYh59ZXnWHu0fDfEVA8fYe3Ws/8aYrZ2edGMEyuYKwtE4YpEycPKvJe8TSLURMiiPkREqciDuQjz+15oCTQdhdRjhIo+ZBKQuUCvjBlNbwIxMr2v5Yu/Dh8ZpxiNyCHb2G9DTKvFC1GpZj2QB7+D+3zdCL3oVJp87YV+tMJMd1TA6HU/UI6txnAm1waIKAj4/O/1MBcsizZFuQLERFLPo4y/cpHHVKdQtzQUa1QQ4bnywvAtFGY9RET9fV6M6OrlTYiJoynWnpufghj4lUkVkx9xr1IPRYS5e6V7UEujjVmpARgrhX4G/N59iiZciSgQlI7wuacDUVRIKQJ1O+wNgGEYhmFkENsAGIZhGEYGsQ2AYRiGYWQQ2wAYhmEYRgY5WAQoRH+pUomoVuPV7kJFaOSECUI6RKOCSKiqNLEWiNwU2U6aCPGHIliaLx/j5x6h+cSRR76PteeqWO3qP3z6H7F2sVSDmKXFu1h73OtAjO+Ka1Qq/Y2F+APuk4iGfS4YGSmVB32XC0+GmjAsJw2FcJgkQoUYTqiCSgr8vFTz0Z1WgXtjTH6kXigZ4vjsdvm821dEgNIISBN8Ojmn7qT4nmacJeevZmpyB+rBOxIGfh3HuRNDoTuT+E2uE9BYiMh6vR7EREJkV6pUIcYJsW6uoJh8DbkYzSn9Wy4IMXGIAr8gkhX6MCYnhIoXL12CmN1dbtT20ANnIGZhYZqfW/n9yIlr1Cr9SXGvtjZ4Oa+U+ZAIg6W434WYeMifYZyieM8LdbxTKgYGwnzPxyYCNAzDMAzjAGwDYBiGYRgZxDYAhmEYhpFBDtQAfOv7HmdtrZiITJFEgZIzEXlFLdMm0zH5AhaOGA5kHkUxupGFIpScb5XuZe3NKw2IWbn7UdZ+5flXIaYU8PzVt37bD0HMPcdWWPvu+asQMxjwfL6mbSDRP6HyLDb2eJ6wpuW7qyIHGOIQKOb5Z9IYiAjNnUgpZDERyAGq5fRkfys5T5ln1jLK0iBnOEDzj77QaXQ6aK41knk/xXUICmEpeUj5UaIVFRKzMVXGTCCLASkxd5I9l32oFgz6Oux4VA2HuEZp8HLjXPx6lGE+MXTaXAfS7aJpTLnKDc4KSn5fmvEkSmGdoVhXNZMp2eea9isnxm3O4TOYK/NrvGtlEWL29/dY+8Irz0NM0j/Kjztbh5hYrJmRolsIvSgYpKyPXvSPNo5TkYf3I3xe6Zh/phXSC6UxUaLMNDGvvhZLNnsDYBiGYRgZxDYAhmEYhpFBbANgGIZhGBnENgCGYRiGkUEOFAF+6QufZe18TpEXCJFCqChppJHESDltfX6etY8cOw4x1y6+zI8bKiJAIewJHQo93nXXu1h7fW8DYvY2t1l7v7sHMUTcBGkx3MUQd5g1z77wAoSMPBfChIowp1Lk/ZoqYqmowEVAz7+KphqlOn+GhTxW8SIhfLnvXe+EkCvXrrF2EKCAZRKQAst+H0V3UsGmiXoSacChiJ6kWKrXxXMlQsSjiufEHNLEtwkIA/FIbiw+66OYqyTMooIAjzMQQq00pwijRJepBdK8qMyoiS3FfWjHkd8bjFBsGYjjVMsoKgYh26S6WRFRp8NFgGP5bAnN07T7iXJiLCmiskisCYFSMTQQD1xbsygW5kVtXB+bm1wUXXT4LGfKfM1qt1sQs3GFr3UFwt+PcplXJ9RWLBfz/nGKpC6VFUYVAW4o5qc06yEicsLMLYhfv6JnrM2HRIiG78RN66vnvONIwzAMwzC+abANgGEYhmFkENsAGIZhGEYGOVADUJ+f4cER5tOdtIRR0g+yUEQ3xn1HucLzM7UKFtapFvlnXtm+xDKzo+RedndXWbuQK0HM9vp18Z11iHnsoQdYOxh9EWKWQ/69ayHmlPp93oflIsZc2eCahH4fi4HM13hurBhhUYh8rc7a3mFuNAKth5LfG/CcpHOTaQS0s821GwNNAyDzzoq+Qua4NZ1ALI1utDy4yLkOEzzOaMxzgcUY+3Z+wMd1qYXFRsYtXniqp5jHNEVhrpGiq0lOcB1L6cQRiInAl0gpkCL6LFaKXsl+dUpxokjMoSRR8quioFWUxzkVCq1LXtE2TAojUQzIe7zWwYCvCTnFTM1Fr6/DiEQ/RIHSLyK/P+7h+Gvvb7H23hauofvbm/w4Qxyj4wFfx9xYKYQmjMi2N1DXtbCwwNrlEq77Xhw6Hil6IGEyFWjrhTTI80oRH6kBUMYxGplhiPwsUubw7bA3AIZhGIaRQWwDYBiGYRgZxDYAhmEYhpFBbANgGIZhGBnkQNVLAOIPFNIkqawKhaKJQAjEpIjnBtLMAAVCkahk5RRjoqI4fz7A61mYrbL2Ey8+BTFn155m7dosdtUjh06x9qCICo0rL3yFtUtL0xBT9Fz4srqLQph+kx+76KsQc2WPC/NWjmNlrRDMWCCEcuK5B4pIsigEmeORJq57++l0uDhJNfkRar1R/Poxmtg1FgK/viJWCoVYabqHQs3ZNS74DFso+Oz2+Ri5NEQDlXVhTLQ5xjm1Jw2FFLHcUWEes1JUhKMV/r1IES6Wi8LwKsK/f0gjsUCp2CYrjuaLUxBTLHAzG03gF0VCBJifXBFgLExiYqV/U8fHiWrgo5m5CUS3UIC1P2k04gLTfm8fYpr7XODX2EMRYCR+G3IOhXD9mI/1vDImAiFQb+w2IGY05HPt+HE0C6qItS4e4FqcjPg1JoqQNxUCV6/9nsFneBz5G+cjfH5OjO0wMBGgYRiGYRgHYBsAwzAMw8ggtgEwDMMwjAxyYNJLZmN8jPlKF/CcRLGIeSeZD4nHmC9OSeS0lUIW44TnOYMRJmJrIlc7jZdDo5QXsnnfYcxplnMN1v588wTEnH3591j76IySP108xNq/89wqxLwicr7gOkNE71zkOfdSER/dsMDvvTfG53VikZu4pIrJx0jkuNT8aYHfa65YgZhJ4NXL3BBEmtEQYeZNy+mBH4d0+lCCFps7EPJwg4/9qSEe58tdnl99rofzZU2IN9wMakLiKW6udXUDC1oVRB4+n8d7n064tqR77TLE0OE6a9ZyeJzZ+gpruxDHlRP5y5yiF5KfSeMaIqK8yJVq5ijyXE51WZlMxspaHAuNVlRQiiQJw55AyRfnQ/69YYpaln67ydstHFvdDtcFaOt+JIvmKBoy+Sz7A8yn94QJ0mCI19xs83kljaCIiE4dO8HauQLqDeIxXx/jEeoWEmHU5hO8Hlk3T5o0ERE5J37AtBj5kRkBGYZhGIZxELYBMAzDMIwMYhsAwzAMw8ggtgEwDMMwjAxyoAhwcZqb1uQCFN3JalKxYjaSCrOgcgWVeS7hApESKBuIlqbrrN1fRUFd3G2w9kZhHmKOnuEil9puE2LmK/w4p/w1iIn7XPh2leoQ8xt/+DJrr+60IKYg1CAr02WIiQZcULOfoqFQucJNLEoxilN8j997kqLRRTrmQpx0gAIWL4Q4qSIUmgQ29rjwR/HvAYFfoIxzKTxzDsVppR4fR+6zfwAxrX0hVrr/MYj54iFu4DRYWYCY5SlufhMpd3bl/Eus3bv8ZYjpi+prlTqOq/34BGs/LoRSRESl6WXWDupozpOP+LjOF3D5kf2cU4xPQhETyAdIuC5pVe+8EAw7dXRMBtI0Sau2GDkp6CxATCrueTxWqu+JynWDHq5Z/Tb/bNxvQ0wi1oRAlowkovGIixm9VuJVCBU7/Q6E7Ozyz6Q4nYioWOT9cfHSZYiR4+TEsZMQkyuI3y/lN2/U5WvoaISCTClM9ZrBkRjbXhH4eSmYf32vp/94/DsPNQzDMAzjmwXbABiGYRhGBrENgGEYhmFkkAM1AL1IGO+ANRDRVJXnVYqKsUxnwD9rttHEYr7GzXjqy5hDfOGFNdY+0sHiEt09nmOdf8f9EDMM+DWf28P81SMRzwW9ewFNLJ7v8eMk08cgJoxfYe3ZMubllg/NsXZ3Yw1iLvZ5Tqk+V4OYOOS5qHaCubKHT93L2sUyHqfX4/deO4w6iuqY57tTxZjkGwbIpyp5NhGSyDwgEfmzV1j7177yRYjZEgm6/2J5GWJOPHCatV0ec4P7MZ+b3R4WDFp/9Xkec/YFiDkzVWftD80fgpi7S3x8ljFtTJWrfA7Fc4chJprhYy1QDK9k7tYrpkxemDnhqkRE0tRHSe/LnL+fYA1AVBBmZV3U3MhiaYUCrjWx6LvBWNEJCcO14QjPJTUIgWKqky+KwjqK3kCaLzmPCey++N5IFuYioqG4r1Qp6CUdc3yK9/7SuddYO1QKXx07dJS1tcJXSYmvD7GitYqlJkIrRiWeaaoYAUkNQBre+Ti2NwCGYRiGkUFsA2AYhmEYGcQ2AIZhGIaRQWwDYBiGYRgZ5EAR4AcfeTdra9WywGBD0R/khDhNkyi0B1x44oYoKkuaXEgxW0UB2+4uj0lKGDO/zEVW6csvQsyoucnapSkUpyTiEmXVKiKi+45zYWBjhGKQ5u4mfCb50R/7GGt/5guvQMzIc6OV8Rh7+vgMN5k5cXQOYlIhzsorYqJHlu7m50YdzEQgRU9OMVCRZhupw5gEvDYwptlssPZ0BSskJvc/ws/tUUy49fQXWPuFGEWq9R4ffCuytBgRvbfBzVp+9K5HIebhMw+y9vEZFCUWO1xg+IUxGmc98cSTrL0QfAhi7vlj387ao74iHJXmPMp9xcI8ximiMCdNVpTnLgmUc00KYY4L6oII+07eolMMkqRnjOLNQ7LIahgqxm15EaRUFUzHstwdHkea6mgeNvkhF+sdOYJC1fLULGvvKuZuezu8YmG/rwjzHF/IiuWrELM4t8TamtgyKPHPQodiQhKVD31B+Y0Rh04jHMdpwI+TfA1iVnsDYBiGYRgZxDYAhmEYhpFBbANgGIZhGBnkQA1Aa+Mca4dKnocSYaag5J0iUeRA8YygQshzXLurWIDi6nlukNPPYYGFZo/ndb747/89xHzLx7ixxdzScYjZfuVZ1j4+xPuqlPk1N3Jo2FKr8Txwb6cLMcMONxn6zm97P8Q8dIbrFqoi50VE9NSzl1n7g++7B2LmZrhOYKQk750wc5K5bSKiUonHJH5C86fSbEa7TJH7lRoIIiKKeV4Ns5lEg4h/+uDKCYi5erjO2nvn0czqQTEejoaY732/mFNTBdS6lA+fYe3Y4X3FrQZrn1UMnc6VeP9cq6K24er9XDvQT/E4p7ZEDnaE5kUtoZFZUkyHwik+73ytCjGBvFdl7ZL5cE3jNCmMUz7fcnl8BmORK08Uk59cjt9jXjGWSWJRbEnoD4hwvo8VMxwvcv75Io5Rl/LvBcq4IVGcqChNkYioOs21TPki9s9QaKLWdy5DTE8Ux6pO4e/QTpNrcg7PzkCME0ZAYaRoAISFVZpXfl9zIr+vhEjN0tcix7I3AIZhGIaRQWwDYBiGYRgZxDYAhmEYhpFBbANgGIZhGBnkQBFgOuQiHa8JEIQYRDOfGIvKZZE0iCCixPOYjd1diLm6tcPapVmUYg3lnqbfgJjnfv9XWDuXL0OMLLb3wAArYh2r86B/+7u/DTFTNS4QGQ2wQp8T1ZvqBVQ+FanB2oslFHR94HEu+nr0tCaE4SJEP0QhlhdVswLFRCVucAGLDzQLj7efVIhUPWG/RREX1OUjnBalPB9rU9MoaFpa4UK48LknIabxuS+xdsGjicjDJT5m/ABlPa0RFyedG21AzLk+f9YXPAqsHl/gxlD7j2D1zEt1brxS8jgX8kIQK8VURETrLz/H2t0EK2zmV3gVwfK/+zLETNV4pdCNb38PxNAyv6/RAAXDYzEW0mRC3ayIqN/n1+YCFBzHCX8uI6WKX1lU6BtrFRllhT5NtS0MaUYB9i8FPKZYqUNIOubfSwdo4BPl+PlbXRTmuSHvn0IB176ZeS4UrO2jwdZcnov1Fg4tQsx2o8HPVcS1ryo+y5cUoWrArzkJUbSZiOejeF5BlUNlibst9gbAMAzDMDKIbQAMwzAMI4PYBsAwDMMwMsiBGgBZSIUSpZCKOESSYh4jEoU5UqW4xGjEz9VqYX6wKfI8A1m1gogKOX6uK03M6x2d4p8dymNutNHjOaS9IeaUTk7ze83HmN+/vMrzTNUSHidOhRGNYphxdY+fq9tGQ6GwwvN7F65iPm0sTG8CxQhE+qE4pURHKAvopJgDfASP/JZTLPIce7WEOfdpYSRTK+MzmhLmN4WpOsQMGtzUJ1Jyp0dSfv6mks98SeTPezHOqd/sNFj7iRHmM9My17Z84MMfgZia57nkVlcpwiUMZVKlaE5OaED6HZy/6xGfH9vbaxCTF/dF69cg5hNn+Zjd9qg32P2eD7N2rBRI8Ul6YHuS6LS5VqdcQt2SF0tdt4vPYHqKj+MwUgr0iAPFSrc4UfgsX0CzIDmSnPIMpLlct70HMSRMr8pl1D+0e1zvEMeobXLiXAuHFiBmZZkXbysr60Wvu8/aF65cgJhAFBU6cmgeYg4d5p/llHUnkcZcY/w9C4Vxm+L3dVvsDYBhGIZhZBDbABiGYRhGBrENgGEYhmFkENsAGIZhGEYGObga4C4Xmo3HKPSSVeE04iEXMoSyDBcReSEwbDdQ5LYvfC12+ij+SceiStp0HWIqNSEU3EchVkEYm5wboTAv1+XXKMWORES7wjQilcJKIlqpcyFO1LgOMY3RJmtPTaPoxnl+Pc097EMSQpxySalSJcQy3qEIMFXudRK599QJ1q5VUTxVKvG+DKSxBhGlQhgVhtgnsmBku4TjvDbkxx70UTiaS7kocUGpxnZGVBd7MsJznTh5krU/8V1/DGIKbT72c7/3GYghISaLyzhmnKhsNlbESlcurPKYV1+CmGf6fJx/wM1BzEeL3JwlfepZiGmc5uZFdOIwxCTCBCsea/NlMghkUctUqVwoTNk6ihCz0+NreD6v1LUUgl6nOMukQhkYKOZZLuafDYd4PfKnIA7wOKlYfwJlrEcRH28ba2iM1RaGWvnSFMTs7W6xdlN2PBGVyvwawxBj1jf4OL5w+QrE1OvTrL10YgliFhe5UHC6goZCstpuItWgB/CNsYobhmEYhvGGYhsAwzAMw8ggtgEwDMMwjAxyYAK/EvPiO1odgrkCz6MM+oqRiCgcoaSUyeX5h3Ef83H5Es9X5SLMvVza4Xm97/ogmj08ejf/7DNfQUOSL77G7/0zz+LdPx3ya/zSZbxmWURmQylA8S33nmDthRrm5USah2bRM4KiAs/9BBU0scjneT45TtBkRt6p03LionBKEEzmXnJujueQlVpV5MWHit8VeeGOFMZYaGV2yPtyS9FJpCNRGGuoGGcFfA4tLx6DmD9z/F7WrnU2IeaqmHf5ZgNi9nJ8DjV6qEmIEj5fxkr/OGEANlb+btFevcTa37KOedpgmms07q1MQ4ysiBI3UMOz9RVedCmcPQoxThQgS1J8ppOCNN4ipUCXLArTV3LuG1u8yNrsHPavJ94PqWLuJpcEr80rETNOMTediJioiBod6UzWaaHBWVtoWbRrnqnXWfvUXfdCjNS5JQnq3ipifS4pxkQrR7kGZWNjB2KuXefGYU898wLEFAtcb3NyeQVi7j7B14epKdSr3Y7JXLUNwzAMw3hTsQ2AYRiGYWQQ2wAYhmEYRgaxDYBhGIZhZJADRYBzM7Os3djfxyDHxRZTMyg8Czw/jVeEZyMhKisoVZhiUa1rrAhhFma4KPGD7z0FMVN5vu/5/nkUwoyi86z9jlOzEHN9kwv6XtxEA5+8MJJIUlRAPnz3DGsvzqEQJvFcmDMeoQBSenrklYphBWFg0+8pwkVhgjQcojjKCVHSsI9imUlAivc0UxPcBSsCKyHoK732MoSc+tzTrP1CC8f5lqj05xQDlRnx3FZOn4SYqSIXGf2Zi3hfr6y+xtoXfv+zEPNFoV8aXl6FmDML/FybFTQmGguzL00EOCPmtHvoAYj5jjY/zok+3ldL1Jnr5HFOtYSYseKxylwhJwzKlKqXk0JrnxvUFKszEOOEORSFuIZu7XKxXFupGFiv8DlTyCnGbaLS5VARxQ4GfE1IlWkVCKVgEOA1p6K83cijSHqnyedVPMZx88Dp06x9990nIKbf52v6KMb10QnjHxeguHF2ll/j0jL+xpw8zcW1V6+jKPbieV4N8/IlrDzoEj6O7zp1GmJuh70BMAzDMIwMYhsAwzAMw8ggtgEwDMMwjAxyoAbAi6IQ5ZJSOCIWOZxIKRxDyevGFEQuNJUuEkTU6vE802UlD/6RB3iuZaGqFC4ROaUohzmcH/go1w7M1hRtw0PclOELr6FGIhTnGvTxXMfmuatPlMN7j0T/FHJoPhEEPFc3UrQW8uwz05ibGop8bjGP50q97PvXLwr1dhA5fsfpGAtISROTNMFxlYiqJekF1ACEa1wDshDg2LsmClhFijPRdIHPs8rOLsREJ5dZe252HmJOrHEdy5UXX4OYRBT2OazUEXloi+eNR8q5Vsd8bnrFGCoVIpXN62heFLV5/xQd9uG5af7Z7scehZjRNB+PuRGax8h5H8eTqWMhIsqFolhaC9eagtAFzMwtQkxOrCNXlZzybsr1EysrhyAmFcfpDLDv+n2uLwgVDUBFGN1IbRERUavNj9Puot4gyPM19NAiFvo5dJjn3F2KpnWOhEaH8L58IicJrheSfAHva3GB33u1imZVS4tce9Zt4zUXIq4ZyheUAm+3wd4AGIZhGEYGsQ2AYRiGYWQQ2wAYhmEYRgaxDYBhGIZhZJADlVtpysUNhQKKAJ0QdshqSkREOSH+SWMUTYRSNKSUZJPX01ZiqkUhhFPEKaUKF00UAryv+QoXenS7eJwo4mKU7hhFdw8e4WKUVguNN6TxDFT+IqJatcramqGNNL3Rtnc5IeiLlBh5dtnvRESpuGbvFfXYBBALY4/BCPs/EKXNoB+JKI34GLmgCAWLbV6JclGZCz1RES0EMSXRa8LYY3oLxXKn7hb9rwh/xgU+zu8a4TM6VONCse375iAmqPCxd3IfKwaeC/k155SBtTvmff+utXWIGT/6Dtb+xQV8Fpsdbo4SDF+EmDnPx3k1UspnSgKt3ulkIKuKbu/gmOj3eP8enqtDzNwhLhTcX8Xn9MLLV1i71cM5c2jpMGvD+k1EIzEfuvsNiNmRVUULKLZ2OX7vWuXREyePs/axlSMQU6nw4wyHWJnV+4Fo4xxOPF/nA2W9TlN+jUmsxIi5Hyjrjlz3ywW891AIZcMARdu3w94AGIZhGEYGsQ2AYRiGYWQQ2wAYhmEYRgY5UANQEUVJvFJ8J5/jh0hSxaRB5GycU/YdolJEIcJ8eiTMb1ZqeJz5KZ7rq1YwNzoY8txUEGA3yAI4yRjvXX4SRJh7ObHA+3Avh8cpFfn3cjnUJMg8NbjXEFES8z7TTDXGA56rHSjPKxL57nYHC2KQ42ZO6jOdABqiSNFI0Wl4MfakERIRgVhiS8lDni3yY093MHfaFqKZRCl69Yeej89hHmMWXn2JtdMcGjqleZ4/bKWoAbhnyD+b3sb8/isbO6x9pIaFsVyZj6NeiGNPmlmt/uBHIaZV5mPt7NNPQszZV9ZYu91Ec6fjZ7i2ofPOwxDzrndwI69CfjLNrIiIQmHyVVTqFrWb/Dm1leJk00d4PxxfxKJCPvcu1j53+QrEdM5d5sdZwv4tpPyag1AxFBMxxRqO44VlbnpVnapBTLHM50hBSYOPB9zQKklxfrpQ5PdDxRRM6H9UzZC4L0peP0bxbaOx+N3RfoecGLZReOdr8WSu2oZhGIZhvKnYBsAwDMMwMohtAAzDMAwjg9gGwDAMwzAyyIGqlyjkSpNOBwVC8ZjHRDlUp4xFpbBU8dvICQHd4nQJYh44wgV+S7N1iDl5lItRpJiPiGg0liY2KPqKhaBOih2JiEaiGmFOuiIR0eIsv+YiodKjWODHjmMUa7WaXMBSraCxSSieV14x1RiP+L2OY7zmvhAK1mp4rmaTC9VyBUWVNAHsNLnZh1OeUS4QgkZFRBOJmOkFFD1tz/Ixu7GKZi1t0U3dKUXUU+HmUfvvexBirj3BzXDqSvXMvnj8V6GKGVFugZ9r9tAyxKwU+BxancG5cCKUwlq8nmKVi7de3boMMZ/7rS+xdr+N/TMzx4Vi73r8DMTc+wC/j/l5FNaK6UJjpX8mBS/MuGansdpdRaxR404DYnZWuSg2yqNI+sw997F2QRF97q9dZe16Edfr6XqdtcM5NJnyQsxcP4oV8SpynVfmp6zkGI9RuCy/lVOEql4uY4GyrglxdaqI472cj3BgNMSLFYO8WPxWpZpBnvTQU2Juh70BMAzDMIwMYhsAwzAMw8ggtgEwDMMwjAxyoAZAGsBoOYpE5KZixc1gJMxGlJQJxTE/Vy7B3P1jJ3jubz+HuanTD/J86ZVzT0NMInJ9faXQj8yfF4tKsYsRv5FIMboY9Pl9STMUIqJOlxuZxCPNTIl/r+0xx5WI4hK5HF5Ppcxzfh7sjIgS8QyTFK9Z9k9O0UhMAnmR6A0VDUAg+mAkjKKIiHqikE3g1yBmvMRznBd3sNjIVk8UCVGKOuWEsUjzyBLEPFm+zNr37axCTOGjH2LtklIPZ0doZi4EOPZeu/g8a+9t47kc8TmUL6P+pHmF9+vq6j7EHD22wNr33nccYpZX+Lyfm1fGXsB1LJrJilyIlPpOE4PU5eQV462imJNe0TQMxdrSG+FNV2Z4Z506hs9gUOaaCtdrQUxVFB6T6y4RUSiKWJVKqEmATL3yA+JklGKU5sTfd2V+nQhN2MJIKYAX8PuQxkBERLH4XUw1IyCRq49x6lEsx600GCLUAMTKmnI77A2AYRiGYWQQ2wAYhmEYRgaxDYBhGIZhZBDbABiGYRhGBjlQudXrc+GJT1FJU65w4YlmBDQaS/GHogIUog3FV4LmhNHKs6+iEOtLz/BbKiomP70eF9CVlIpsjUaTteuKqGl1T5gFRYrZgxD9LU7hnqvV48cpKkYXXSHIrNawIpYX5iu9FgoFu10uxAoVo4tAfNbrYLU1WWmwp1UMnABiIU7qDVGsFDo+RmLfgJhRssHa1SkUCu6scNHTl7ewstko5mP4uEPBTru1zdrOlSHmhWNHWPu5vfMQ8977eczF3WsQc+HJT7H29etY+W1jg19PWat6eZqL9xbKeM1HjvCYM+L6iIgOL/H+mZ+ehxgvhFDDMY7PvjC88h4FaHkh8MrJsmoTxHDMx0m+oInThCg5wjUrV+D9u72vCFUvXmLtlcNoerU8zdcfh6ei1v4ea/cUcXN9ihsaBVolOyH6c4RrVhTwMTkco1huc32Xn8vj79nhCh9voSLaJs/vwznFQEoIHseKwk8K5mPFlE2KBwPl7+ypFDwqYv3bYW8ADMMwDCOD2AbAMAzDMDKIbQAMwzAMI4McmPSSRhLzC1jMIUkwxy6pT/F8UX/Qh5hYGLTkq7g36QgzjPesYJ6xs8l1AfkpzFdOFXnCSsu9VHI8V5YLsauW6zwP94EHZiCmLrQMh+Ywd9cfCCOiMsbkhWFGohQMShNZwAjvHVJsmCqjMBA5VqWgkjS/0Aw8JoFG4wXWDvNNiCkW+L2Uyzj2pvKiQFIFNRgjoTconVK0LiORd1YqY8WXhR5GKSSyv8jNgX61+SWI+fIv/Dxrb26gBiAV5lErx9Bc673fcjdrLy2htmF2gT//ShlFPBUxRlIlBwtmLVrxE2ESJgs1ERHVinXWHseo2RjG/JmG4Z0XUXmrkdN9pOR5SwW+HkaKViMqcjeoqVApKtbnmopecxdiup4/37zHNWJvj3+vMcTfisLCImtXFUOxUK5jyvNut3mhuuvX1yEmSPixV45h4aFRyu+9uYcF8IZDYdymuExVqvxZyKJHRERi6qkxUi6XKgu2F2vI1+JnZW8ADMMwDCOD2AbAMAzDMDKIbQAMwzAMI4PYBsAwDMMwMsiBIsCyqBw36KF4zwnBmFaBbjTgApGCYq5QiLiQIQzRWaJa5se+/3gVYlJZKSrBPY4UuQWK+GcsVDf5EgoOIyFiOqZ056AnRCQOxTLVIr8eF+A114SoagxloojGQ/5ZTjEmGgqDFM24KRTn90r1rXjEzxUoleYmgekFbuAT5vBeSjk+zgOn9EnI+98ppjEnT3ATkZPH0UCl0+VzKB2imLN+b52fO78FMYEwqjp2QnkAETdnuuserCp4731chHXmXjTnmRFKVs1UZwRCKBSyjkdc7OoUYW1BGNwkSlVQ+JpUUxFRKkzLtKqX0vBKE8ROCkPRvzmPfVfO8zUzLOAaGor+na7icWoJnw/dRgNitoXoLlKqE5YOH2Ntp1Sy22xxQd30El5PqcjX/SCH54rE2luYxvFXDPnvRar8Vg3FGtrq4Fi/cOEqa09PoyD4WFH8NimGX7KCYaTMB/mT4mXpP0JTNvc1jGN7A2AYhmEYGcQ2AIZhGIaRQWwDYBiGYRgZ5EANQCHPcyRxouTaxGeDMeoEZL66XEXTmLzIRQ/6aBrR6vJ8UUUp0FMs8GsO85jP9am8D9wHFYo8h+RTzEU2hGFGUcm5S71DEL2+EVCk5DTHY94fOcUwYzTk3xuPlfxpKK5RySn1Rd+nihmL7J/xCDUJk0BVmM/EivGOT/mYcbKPCIsmKWlwGgutRKWA43xhhvdblEfDnKMnVlh7Z3sVYoIif24PvOOjECN1CoFTnmOJx0RKMZaBGOdeSZaXcqI4jKKRSISuR5pJERH1RaEuGK9EFOX5sYcjNGsZi/laUPKruZDfRwLrwuRQFsZTTtE0jEV/5hUtUSTXwwifZcHxMVosoAHc2nWurekMcK05ssi/V1KMiQYtbp41VNaaUDzLQRd/Y2pTXAPzwDsfgJikz4+TKgZw4YAfZ7+N56JA9E8JNQBRxOe+pyHEeC/Xa1xUpP5Kfud2n90p9gbAMAzDMDKIbQAMwzAMI4PYBsAwDMMwMohtAAzDMAwjgzjN5MUwDMMwjG9u7A2AYRiGYWQQ2wAYhmEYRgaxDYBhGIZhZBDbABiGYRhGBrENgGEYhmFkENsAGIZhGEYGsQ2AYRiGYWQQ2wAYhmEYRgaxDYBhGIZhZBDbABiGYRhGBrENgGEYhmFkENsAGIZhGEYGsQ2AYRiGYWSQTG8AnHOXnXMff7uvwzAmGeecd87d9XZfh2E45/6Vc+4fvN3X8c1CpjcAhvHNgm1mDcP4WrENwH8izrno7b4GwzgIG6OGYWjYBoDoEefc8865pnPul5xzRSIi59xfcs6dd87tOef+g3Nu6atfuPlK9Medc+eI6Jy7wT9zzm3dPM7zzrkHb8YWnHP/1Dl31Tm36Zz7Wedc6W26V+ObEOfczxHRMSL6dedcxzn312+O0b/gnLtKRJ9yzn3EOXddfO//emvgnAudc3/LOXfBOdd2zj3tnDuqnOsDzrlrzrlvfUtuzsg0zrl3OueeuTkmf4mIirf82UFr9Lc55167uR7/T865P3TO/cW35SYmGNsAEP0gEX07EZ0koncQ0Y855z5KRP/45p8dIaIrRPSL4nvfQ0TvIaL7iejbiOhDRHSGiOpE9ENEtHsz7p/c/PwRIrqLiJaJ6O++SfdiZBDv/Y8S0VUi+m7vfZWI/s3NP/owEd1HRJ+4g8P8l0T0w0T0x4hoioj+PBH1bg1wzn2CiH6BiL7fe//pN+bqDUPHOZcnon9HRD9HRLNE9MtE9P03/+y2a7Rzbp6IfoWI/iYRzRHRa0T0/rf26r8xsA0A0U9579e893tE9Ot044f6R4jof/PeP+O9H9KNgfQ+59yJW773j733e977PhGNiahGRPcSkfPev+K9X3fOOSL6S0T0127GtonoHxHRn3rL7s7IMn/Pe9+9OUZfj79IRH/He/+av8Fz3vvdW/78TxLRvyCiP+a9//KbcrWGwXkvEeWI6P/jvR9773+FiJ68+WcHrdF/jIhe8t7/qvc+JqKfIqKNt/7yJx/bAPCB0SOiKhEt0Y0dJRERee87dONv9Mu3xF675c8/RUQ/TUT/nIg2nXP/wjk3RUQLRFQmoqedcw3nXIOIfufm54bxZnPt9UP+L44S0YUD/vyvEtG/8d6/8J90RYZx5ywR0ar33t/y2ZVb/ux2a/QS8fXZExFLfxk3sA2AzhoRHf9qwzlXoRuvklZvibl1UJL3/qe8948S0QN045X/f0VEO0TUJ6IHvPf1m/9N33xNaxhvJP51PuvSjc0oEd3I+RPfiF4jotMHHP9PEtH3OOf+6n/CNRrG18I6ES3ffJP6VY7d/N+D1uh1Ilq55c/crW3jP2IbAJ2fJ6I/55x7xDlXoBuv7Z/w3l/Wgp1zjznn3uOcy9GNhXZARIn3PiWi/4WI/plzbvFm7PLNXKphvJFsEtGpA/78LBEVnXPfeXOc/h0iKtzy5/8/IvrvnHN33xS1vsM5N3fLn68R0ceI6K845/7zN/riDUPhi0QU040xFznnvo+IHr/5Zwet0b9JRA85577n5r+A+XEiOvzWX/7kYxsABe/9J4noJ4no39KN3eRpOjhvP0U3fuj36cZrqV0i+qc3/+xvENF5IvqSc65FRH9ARPe8OVduZJh/TER/52aa6QfkH3rvm0T0n9ONH/pVurFRvfW16P9AN8SDv0dELSL6X4moJI5xlW5sAv6GKaqNNxvv/YiIvo+IfoxurK0/RES/evPPbrtGe+936MYbq/8X3ViL7yeip4ho+JbewDcAjqdXDMMwDOObB+dcQDc2uz9i/3qFY28ADMMwjG8qnHOfcM7Vb6YH/hYROSL60tt8WROHbQAMwzCMbzbeRzf+VcsOEX03EX3PHf5z2ExhKQDDMAzDyCD2BsAwDMMwMsiBRUJ+7Affy14PTC+dgJgz7/4e1i5WZyDGU8raSTzCk8Uxa05VahBy78ll1r525TLEjBPH2mGItzga8/P3+22IiYKEtXMugZiA+DVH1IAYN9oTMQ5ixgMhTg3wrUw87IpPQojJFYusHRBes3Pi2OkYYuRLofEIr2c45H02bGMf/thP/ju82bcYYSJCaZpCDP9nxv9J53rdGHkq7Sux50Ff/zs6/k2nnGw84G9Fuz18S5rE/HvDYQwxlSq3tojyOO8GQynCxn5vtzusvbW1DTHy/I3mOsSsbb7E2peuon9Rt7vP2qMhzpd/9TO//7aPYSKin/yZ72QPoVYrQky1UmHtIMZnMG7w2xlv4Foci34I5/IQk5vhf3fM1fBcgePf89rPjePH8SPl76Rd/r1qgvdeKfDxF2II9YiP7TTENXTYE78fKR4oX+DXWMgp9+V5zPzMPITsb3JzwrOvPgsxmwM+tndTnJ/thD/DSLn5X/ynn1THsb0BMAzDMIwMYhsAwzAMw8ggtgEwDMMwjAxyoAYg7jdZ++qrz0FMaeoYa5++73GIiWOet0iUpGYU8L1I2B9AzGiH56t9bxdiKCiwZjFfwJAhzzOmfgevR+oUYsVEKunxEI/5dEr49zqDHoT0Wjx/XijivqxYnmJt7zFf6Xu8z8IAc1xByFNBWv47FSne5t4exIxjrknQtBaTgLy/UMn7TRryibxR/05HSwKuXuM59o2NTYg5cphrb0aKBkAe3UU4hl0g8s9jnC+dTkvE4DrQ7fL50m43ICaK5DqAmqJmkx+nWChBzKRQnyqzdqmE47hUEmtomoOYgVjWkjw+y7jHn0syQN2MF48lLEMI5aQOxCtzT64/yrjJV3lOe6GyDDHliD/f8RjvayrPNRKdPsY8/dLzrL167UWI2W02WHt5Ga+nWOT6h0KEz6IkNBtHFrEUR2eLP4veGLVW41A8DH/nshV7A2AYhmEYGcQ2AIZhGIaRQWwDYBiGYRgZxDYAhmEYhpFBDhYBCtOUfhsFOWtneX2FQzNoLNFsXmbtygxWwz106J2svXf+jyBm9dNcsOSLCxBz+lse4udevwYxvR0pdEIxSJjjog2fohTLed4/XjHeSaR4UFF0hREXbQQBPpZUnItSPFccC1FVDg08hiN+nDRBgU+/zcWfSYz9EwnzC9meVN4o62vtOJrJkOROTIfkk9UuWQrqnCLxSxJ+pFSODyLaFGYkmniq1xcGKjj0qD/k8z5RLrpa40qxS5cvQcylS+dZe2ERDVS2t/n8LZVQYFWfnmPt4fAExJRL3LSs3UKB1aRQqfD5lc/j884JAZ3T1NahGKOhIgIWzzcYYYxLpckPiq3h9JpIWhiTBYTPslbiJj/jPs6zVy+/ytrdAY7j+iIXrHeFmI+I6IXPf4q1z1++CDEbbS6AfuSBByFm+cQSaw9bLYh54vNfZu3v/Pb/DGKO3rfI2n3Nc8jzeTVS1uvbYW8ADMMwDCOD2AbAMAzDMDKIbQAMwzAMI4McmLiVabxiCXPK01We4x53X4KYvfUrrN3ZR+OdcMjzzoM2xkzn+X7l81/6HMTsxDw/ePcKXnO+wPNVYYgGILHIjXtZRIcITCwCLUYUhdAywIHjea9iEV01woibaIxGWBQiFQKDOMFk7VjkgQc9WWSIaDwSJj8O83IjkfMdj5QCT98g3EleXub3ZX6diGgk+kDTBEQRn3JBgHvwkRh7LVEgh4ioXOJjNoxwKu/vN/j1KGZWco6Xyzj22qLQk1P+3hCEfIwMlPEg6r7QtWtXIOba9cusnUrHGSJaXeMxx44dgxgibrJCKeaoHfF7XV+9rhxnMkiE8VesmOp4UeTMxTj+ErEeBYoxVhjyNTOvrEdTlTprRznsXwr54PLK+BsnfD2qFlHzkezzsfXq81jYaVMYWJVn8ThnX73M2muXML+/s8GL78wuLkLM3Q8dYe1QWQvSmH92/MRJiDl8hOsE8or2a+UwH9vRaB9irjS5Ns4FikjnNtgbAMMwDMPIILYBMAzDMIwMYhsAwzAMw8ggtgEwDMMwjAxyoAiwVuZCo6lpFJUszHHxx0ip4jdb42KHSr4IMUXPBSy/+kmswnT6yApr92vTEPOvf+WzrP33/uu/iOcqc1FTMkLxnixkNRqj6C5I+f4ppwhhpIeGFIoREcXCeaOgiNJk5bR+B40lEnFs71DgI0VoXjECkuZAKaGBR05UWwvuvADVxHMnJj9aJbtul4snY8WQoyTEe0GoVM0TgjqtgqG8Rt2ESIiwlPuanp4Sn+D17O1yge5wiPdem+JzcTBAwVeU4/cRKiY0NWEWpBWZ7PUbrH3hIs5NSvlxdnfQ5Gd7m4un9naV6qITQqvL19X8GIW5UcSfb1ERN+dCbqoTljGmcoRX1tNEycWIr+FpF8dfoczPRRGu+4GokhcoRlTnLnLR387OBsT0u1woq4m2Q6FCrS/OQkx9jptDpcrC1uvysb2/j8K8nFjT9xVvnmP3383a1ekZiMnn+PO5a6oKMZTj13it1cCY22BvAAzDMAwjg9gGwDAMwzAyiG0ADMMwDCODHKgBeOShu1hby1eWyjyvl89hnifveA4pUaqJnHzHt7P2r/zWBYj59BPPs/apu05DjMyVRVWMCXP82D5pQEwqrjEe9iAmErna0UjJw4o8vKYT6Pd57ngwQOOXWJh8DHsYQ8KQIknxeY0Dfj3SGOjGYfhn5VIFYvJCx/FGFdmZVOT9aUZAwyHPDWp6j5woMpUPcDykovCUC1ADMBTHThQthywqFaInFkViTkdKUFkUY7l2bRVidnZ4/ny/2YSYtQ0+rtIUE6Nxwu+r38fcfb7Ac56r17Hg15WLPC876ClFb2DITq6ZVb/Ln5PWd1MV/pyC0RzE+D7PM4cec8q9ToO1X37maYjZ3d9j7bwi1qjP1Fl7buk4xLz3wx9i7Reefwpirly6zM+9sw0x3RZfn4tKHrwofptyhPOq3+PH2dpCQ7rumM/96Sr2YXOdm0oN2ng90uSqoBRUS+7hBkIrK2h6dbzCNXZhUep6bo+9ATAMwzCMDGIbAMMwDMPIILYBMAzDMIwMYhsAwzAMw8ggB4oAjyzUWTseo9Boa5MLgkplNOcpL3DhyfTsYYhZOHoPa+cKKI66vsnFH7uK0EjqesqKWVA5zw2FLlzG6lK9NjfaqdexKlRY5KKSWBHU5SO+xxoP0SipICrCDbsofJJ1BDXjl0SYaISK4NB7/r1AqRy1vs4rYh0+tAIxhQIXvozHd16B6q3ECxMjzS7HCTGQV/bFwxG/v24PBWPtDhcQDQYoHC2XyiIGxVz9mD/rdk8xfUr5OCopQk1KhaDP4TWHwjAlr5lHiY+0wm/tDT4XO22cm93enmhvQcz2Ll9P1lZxvlRKfN4VFDHXQFQwvOeeRyDm8BIXT1WrKGCeFCKxVJc9mtg0znKB6c41FApHnvfL3hYK6i5evMza2ztokJQT69rdR49CzNwiX+eXjy5DzIWzL7P2pfOvQcz1K1dZu9PDeTXo83leqyhrX8Rj4iHOh6H4LFLEljPC0GquhsLZUcDXx9EQzarmpvicHSrrxf5Og7WLeWXyiTk7fxeKLW+HvQEwDMMwjAxiGwDDMAzDyCC2ATAMwzCMDHKgBmDtOs+9TNXnIWZqhn9WLqMJQeq5QUqpUoOYVovHrK2uQ4zM72uahFqV51VGikFKs82Nd0pFzKtMzTzE2j2l+MZ+g+fG5mfx3n3M8zoyJ02ExXdipZCKEzoBUkx+CgWeX/ZKIYtAFAMKPeav5mYXWDuKMGY44rlZWcBmUvCyIA5pVYv4Z7Fi8rPb4LnTwQCfY6fPn9tQeY5DUSBoqBzn3PnLrP3s81+GGBfwXOWRI5iDzYV8PObLeO/VKp+L9fIhJYa3Oz3M719b5cYn3TbqWFKRf97auQgx7S7PSVcqWIhmZ7PB2nmHRiw/8L1/grX/xPf+MMTUZ7k2SY6VSWIm4nndzhoW8fn0b32GtdMhPoNArCMbm2h0U5vi4+bhdz8KMaN9/r3Tx1An9NgHPszaWnGsZ7/8JdbeW0eTqVjk/OMezitp1DUaY35/JJ5vlFfWLFHkq1jEtU8aLk1N4e/ZuFpn7Z1VvK/mHtfEVKZRryb91XoD1MQ4UTyu2lFM4m6DvQEwDMMwjAxiGwDDMAzDyCC2ATAMwzCMDGIbAMMwDMPIIAeKAGvTvJpUdaoOMRUh6NOET419LuyZj1GYF4+4QEQTjMB3lKqCj9zNqyeNrz4HMUmJiyQefP8PQEyU58Y/L375MxBzeZUbCM3OnoIYJ6qt9YQAkYioGHExSqBUfxuL6m9hhI8uEpXmYo/9HOaFqCVBYdjMPDfwGPbQxCIVQjkpLpwceF86RQToPP8sdErlOGG81B+heC/2vA+GKR6nKYyg4jH27Wc+8xus/fu/85sQc/+9d7P2E83PQMzlS1yY5woocrvrvgdY+zu/409BzL3383G9uroGMb/yy/+Gtev1OsQsHuHiqb3mdYip1vgY9opmc2+PC9A+/pHHIebP/Nm/zNr5PAp088LISzPXmhSuvsrX1Qsvo2HOxhrvz8XZOsSUK7wf3vkurC5XzIs1q4ViwrGYRzPHTkKMFC6/+BWsKjjsCUF2Gc2Y+gMuePRd/I2RvxfjGP9uW1nk5kmRUsE0L9ZH5/A4RSFMjaVTFhE1hMGSU2LCgrhXLE9J+bxYU2L8zcuL349UMTi6HfYGwDAMwzAyiG0ADMMwDCOD2AbAMAzDMDLIgYnb6vSC+ARzFEOR0xwp+QdpPrG/fQViTj/8n7H2X/iLfx5i/v7f/29Zu9tHU4Rhg+cHz37pUxDzoT/1/axdP3IfxGyK/OnyCub3A1GYYX/7LMTUcjxflY5R25CGPD+UKCVrxvIzxawmFuYXmkYikZ8pegwpQcgrRkle5Ks085xJwHlxM4ouwolxHUrTJSKaqXOtS0sZeyORh0yVwjqygFWzjcVY1pp8DPsSmuGceeBdrL11HY1GXniea1TGLSwqVJviz9aTor0ROpbhCOf41uYGP47H8TC3yM9VLmG+N0l4v7Y7aDqUL/DrKRZxGauIYjBK6lRZzjSTqMnguSdfYu2BYsb04P1nWHt+bg5immL8rV7FtXgoxnZL0QA89K5HWPvoKdQAPPMlbvKzuYljtJjnfT7uobam2+E6gbZSDEg+yqn6DMTEwtAsUYrv5HM8JojQLGi7wedRcxeLJbVa/NinlrGYXDoW418xvYrE2psroDHRzgYvqjV3CO/9dtgbAMMwDMPIILYBMAzDMIwMYhsAwzAMw8ggtgEwDMMwjAxyoAiwVOaV9YaKaELuIbSKWi7gp2k3sQLV2ef+gLUfP43Vpd5z5i7WfuXsOYgpFrnwqpmigG3lzGOsXVs4ATFrl66x9uVraFoyEMYhFRBNEiX7F1g7UKoTjkS/Xt9FYVguzwUjsyWsgCYFb7JK1I3zczVUohRAG414TC5E4Yk0Kxp10eBoErh0kVecm51Dgcy0qMI1HqHRyGjIhT+jgSb84QKr6hSeK/VcQNcd47lOvotXXyvMH4cYX+Oiok6KAqugyufmfSuHIWbpCBeKFfL4d4JUGKYMFLOvqhAq+jGKuYZdbsCVL+Lg29nla0Muh0JKJxR9/QGK1JJUGGcFKDiURjWBYgwzKTSbfLytLOGzrIj1+tz5CxBz4SIX/aXKArA4z8ftvUJcSET0joe4cPriOTQm2tvmwtDOHs6Zvvi96Cvi2v6AP8ucUsUvEutRr42C11QIUw/NYfW9bpevxWtrWxCzur3P2vUI54wXYuvtTUWomuPrc5DD36pmm99r2MY1PRaVWX3tziuz2hsAwzAMw8ggtgEwDMMwjAxiGwDDMAzDyCAHagBikceThj5ERMMhzweOlRw3hbLAAuadNq6+ytpPvYQ6gXbKTUpqSoGFQ0tLrP0j//e/BjG1uWXWTsd4zXOHjrB2S9EtlOo8D9vfuwwx19ZfZG2v5BkHoiBGd68BMfNz/FyurOT3CzyHFATK4xXnj5QCKOOBMLRRYlwgCuiE+CwmgU998vdZ+4EH0PTpPe99N2vv7W1CzDPPfp61n3vhWYiZnuFmKA898i0Q0+jy4j9a4ad3PM41KotL+xCzfpZrG9p9zK8eO8Hz8o8/grqa0PN8bxAohbrE89dMn4qiiEpB0Z+sX77M2jVMwVIhz8enU4qC7QqNTLOFc1PmYNMAr1nm/CdYAkBFYQATK2ZML7zA8/trGziO50WBoKPLhyDm+FG+Pq4cw4JBY6Fbauxirlya/BQDHBMDUWgsGaAGoCh+d7zy+zEc8eN4xfBrZZ4XQgoT1Kns7PK5dm0D51V3yMfkfL0EMfJ59RWDPDmPuk2c56nj5xopGom7HryftbcVk6jbYW8ADMMwDCOD2AbAMAzDMDKIbQAMwzAMI4PYBsAwDMMwMsiBIkBpiJIoKplEVKAjRXiWFyY2pTpW1nvqs8+y9lcuorBns8OvJ6eYDn3rd/NKf+/6wLdDzN7WOmu3t9BEJe5zc5Gp+izEzC2dYO1dh4KlxMvKZRWIyRV4n505VYOYMOQxQYSCGidEkbLfiYgSIarq7Dcgplzl5y9WUa0lhYGaOcck8MzTT7D2xvpliBmPuWhmVxE0/dFnf52119YuQcz73v99rJ2OlIqBQsTT7qJhSae1x9rr1xSRm6iSV406EPPAO7lwdHkWBYeXr/Dza1X8pJxKEy7ub3NhXinE8Vmf4WPED3C+5HN8nA8V06Fhn4tm4xgFVmnKBV5pgjFEwnjFaX8fmgxx6+Isn4Mba+sQ0xeCunfcexfELB3m61ilgmvE4sI8a496aPLV3OfiuPYujtFWo8HahRyuEQUhXNZMppKEjxOoaEpEeXHsqSpW1puu87W33cT72twQlTiV558XgtL1Nl7zoTo//9IsCgUjsWamKf6eba9xIefcTB1iZg9xA7qLYv04CHsDYBiGYRgZxDYAhmEYhpFBbANgGIZhGBnkQA1Ap8HzeolippGOeX6mXMJ8cVTnRhJfeeIaxLzy7FOsPQjxOO09bpTwxz/2rRDzfT/yY6w9xkumqTovgLJ+7TLEfPJX/jlr1+p1iHn0W/80a5en5yDGhzzHNhxjrrYsinjk85gLGsc8z1QqYY4rFiZMiWJ0MRbPq1BGTUKUL4g2FgMi4jne0QhzvpPAhVd5wajN1asQs7PJx2NXMdLY2OSFVTrdBsTs7/C8aDrC/g9Cvuf+8hNPQczGKi+iMjOD+pMpkb6uKvnrw2mdtQeYNqYw5DqBYhULGCVemD4Rjodxn99rqYq50zDH86CjEWoABkN+rjhWcvCe33y/h/086HOjGi2/muT4fFldQ+Ochx5+CM//NhAIvVM8xvs5c4qvs8dXFiGmIfLypRLO/1jqhNo4H3a3uE6m3dDMZ/g1hiGa83hpMhXjgh0LDQApJlN5MUwiRR+2tcP1LmurqPUJhWZrOsTrOSTMlLbbqCUYjeS94/xc3WmwdrGExYAGHX7sRaWAUU9ojVod0wAYhmEYhnEAtgEwDMMwjAxiGwDDMAzDyCC2ATAMwzCMDHKgCPDpzz3N2kdOzkNMr8tFbXff806IWTrJq5tdf+HXIGZ+mQudLq6jYKRS4EqPP/6DfxpipMCv1elDTBhxsYUPUIw07nJB19CheO/KK59j7XaMIo6L57no7PSJKYhptPi5SiU053Bir9bv4315IcwLNfGeqKyVL6CYMBGVtDo9NKuRQqFapYrnmgCuXOQV0mZmUfTkEy6i2VSMoVpdLnLqDtH84/wl/r13dXHMRGU+RuaUkngzR3hfTs3gmFnf4tUAW/s4hiNRnbB+CKu61Q8fZ+2CYjTSiXn/DBWTn7TCTU3CCor3EiHeck4RPY35+MzlMcYFXBgVBjhfvKxYmKJQcG9vjbU/9fu/CzGTIgJsNfg916ZwHD/88BnWvnr+IsRIAV2kVPGMhbnb5iqqR/eFgVi314OY6SluKDZQRJ/tNp8jA6VqXipEgCHh+OuLr+23FJOfBr/GSKn+eFKM/9kZ7OcZYVZVy+Hfo9c7/IK6PTQF6wnB4WCAa3pBrNexsu4UpUnc1/DXensDYBiGYRgZxDYAhmEYhpFBbANgGIZhGBnkQA3A/JE6a9em0SQkX+DmHvMr90DMkRWeiwyHaBqRxDyvMxhjzu6uo4dZ++FHHoEY6f8QBpgvGo94fvDa5bMQ0xV5sLmqkrvffoW1n3sBDY6qwjwplrlJIjq3yr93cn4JYmrTddYOFDOMYpE/Cxfh402EdmA0VvJO4pm+evEcxBQLXF9QPXEcYiaBrQY3xejFmBscpDw/12huQ0y3y3Nv+RIWbHLisz5hzvNwnuf3v+9jH4WYYo0f59Xz2P+FCj/2hRHqDcJ772Xt5ZN3Q0yvx8d5eYDHGYlyQKPRPsSkJT6ueyMc577Nj1OpYYGUWoXPM++xD4cjYXiVYh57dY2bvLSamKNeX+OFX7QCO5NCKJxujh5GPdaWyNXv7qIhzMrRFdaWBd+IiJrCLKjRakBMu8OLpRUUrUZT6ASG0tCHiFLh1KatayTWTKfk7ttCO7DXxnWNhKHVyizqlu49xDVkpSLqS/aFbqEUotYqdPya20qRo9RJzRYWS5J6h2IFNVt9oXc4c/w+iLkd9gbAMAzDMDKIbQAMwzAMI4PYBsAwDMMwMohtAAzDMAwjgxwoAjx29wnWfvFlFCPJqnRzS0chJkm5IGJrgCqOaI6LyPJXX4CYR+7lx547hEIYWfUrUkxLVq9fYu0rr+K5+mP+vW2l2lXJcbHY4WkUekyVhTCsjzEBcfGHU4wu5H0oWkLqCyOJRAmKhTjFKUYgxZRfz/Is9nNZiFECxYxlElg5dYK1czns21A4gsxX8F7mE2H+MXUIYt793vex9mFFqBXu83GENiNEjfXrrD06dwFijgVceLRy76MQExS5oK51GUVu/XPcLKY4VYcYL5S1/QsvQUypw4VRHdTwUkeInhKlYltRrCexUtEyEBU2z567AjH/9td+nbXTBJ97rVpn7WYHBZCTQhDwvkqGaCyzv9dg7VJRMQITJl/bG1gBcX+fizxbLeyX/oCL7uIxGrelMX92Y+VZelnVUBEBhqJCIJg8EdEYYhSloBhvbcWY6NwmX9NzyrlkZb+yUsVPmvG02vi8ZmtinVEuWQ7b6TkU4jebDdZ+z/H344Fug70BMAzDMIwMYhsAwzAMw8ggtgEwDMMwjAxyoAbAi/RHr4EmKlVRTCbwWMyhIfIqpx56EGLm8/w4X37yaYjJC1OQxjbmNAtlbuQQx5hY2Vo/z9pXz7+I5xJGN3ubaA4TjrkZxuEjR/B6hDlPquTcazl+7zLHRIRFlzzhcaKc+CzF/NW2KGqj7QBLC1wDsHhoEWKKJZ69TtPJ3Et++KMfYe1AMYYKRLc5xWkkFIVrfIJT5/jCAmsvKbnBF778Gmu/cuEyxFx9hmtSgh7mTktizBy/516Iyc3zAluNAepYrrzCx/5FJZcbCU3EWhPzxuOuMJgizD+PRJ62UMIc7ECY/GiZ3EKRF1Da3sFiVV/44lOs3Wnj2uWFMUy7hzGTgh/xMdBp4rP0jvfdVK0OMd02X7PWN7Ygptfn+eqeYmKTpLzvYsVQqCTmUbeLvw2dAR8DqeIDNBOJAlGKTqA15Mfpa/l0L65HMYdaE4ZfJWUEHhIFvQ4r64UXn+0PcQ7Pi/W6WEJjrFjce0mJSSv8syTUZo3OZK7ahmEYhmG8qdgGwDAMwzAyiG0ADMMwDCOD2AbAMAzDMDLIgSLAKMfFDvfff1qJ4YfY2tyAmJPveJy1H/3AByHmySe/zNqd3V2IuXaJCz1e+PzvQszUoVOs3e2h0KO1w0VMoz5WjoqE8UagGG+AiU4eDWQCUVmvEGLFp0OHuGCqUsWYRIizkjEK/AJxPYFSyaoojIiKBRSquRzfF3aGKLIiEdNvtTFmApiuiopfioAoEK4dWkEyqRQMFIOQRBj4NPrYJ+efepK11zZ3IOae9z/G2ktLWB2yIK55fxOFeXsNYWoygxUMH/3h72XtvjJfOiM+9jXjrNYrz7P2WDH7Is/7UBOyesfXk0QxdBkIA5ekh3Oz1ebj3Hs8jjTckuNgkiiV+FwuRng/acrXCM3k6+oqX59bbXzekkAxJgsDfq5citczEoK6liIC7IrvBcpzykk3HGWCDoQIeaiI9wL3+s83TlAEK2l3+H1FOUW0HfO+HylryrDPhYH1eWXdr/LfBqdUHiQhWN/poED0dkzuiDcMwzAM403DNgCGYRiGkUFsA2AYhmEYGeRADcBoxHM2MsdERDQUMe0m5u5rs9y449Unn4SYl574Imt3Gnt4PYf4fmVr4xLEbK6J4iZVLJ7Qb/A8WC5FE4u9TX4fDSWneegQz+9HeexOR7zP8gEepyCKQiQeDVLawsCnGJUhJh/xfH6Yw5zSyvIya6dKzmsYc9OhnX00XJpu8Gdaqy9AzCSQCvMZp+UBhTmQk85ARJSIIkpV9OOg9sWzrP2Vq2sQ0xUGLvc8ikV8Hv6u7+KXV8OTRSnP3e5/7o8gZizG8FS+DjFL9z3A2hc2rkNMXuRTKzuoW5DGSKni6OJE7nYwQHOUXpfn81PpRqZ8L1HMvqThE5hkEVEU8c8KwYHL4dtKp8tNijSzl5zQY40U85n9Jp/bQ8X4KSfMZ/IO+27Yl4XHIIT2+vy3IVGOkxJf6ypFHOv1Gl/rogjXtbYwPVv0OM/jlJ9rsYRr6EyNF9A6u4lzuN/kz2J/XykiFfJnMaPosXYGfKzXlHF8ZI6beY0UjcTcAi9MVq+gcdvtsDcAhmEYhpFBbANgGIZhGBnENgCGYRiGkUFsA2AYhmEYGeRA1Uu7vS8+UUwSREWtUQ8FQp0Grzh1/oXPQ0xv/xxrf/zj74GYI3UugOgP5fURNTd4pb/gKFYevHSWixBbDTRRmVm5j7UPT89BzGyJi0EqiggwElWhEqUaXbvFjXbaTbyvUpELX2YWD0FMLJ5FrIg2vRDdSGHWjevhopa0pxxHmIzki2goNAn4WAgqFRGgF48tVUxNUjH2+x77JF/ngp1TyychZrbPxXuLD+L4rBXrrD2W90BERWGO0r90DWJ2vsTNtaKBYrzzQdEfSpXDeMgFX+MmimbreW4y1A1QFTYcc1HYYIAGPk0xF3KKkHU0EsfWPIfE320cdiFJf5vI4bkmha4w8CrUcL5N1bnguaWYc+WLvIrnfrcBMWWh6OslaOAzEAK/kmI6NHZyjVBEgDF/TicX5yHmlKh25xyO0XaZi+wWq7MQ04/5NQe7KDR/92PvZO2PFd4HMS++/BJr//ZzX4GYPVFpcEERAbaFmPHiDgro736U/w6FRRQuHj7Ehd3Tyr3fDnsDYBiGYRgZxDYAhmEYhpFBbANgGIZhGBnkQA3AYMDzGBVZWIWIUpEvimPMF3X2ufFORFh8554zPMc+t3gGYqYOnWDtl5/5DMR88tf/d9b++J/8LyGmIExBZg5hPv3D3/1j/IMB5lg3z32OtfMFLNRQKvOczU4D83JrWzznvzQ/DTGzopBLkMNH56WRhJLLlju+eIjPqxjxnNvCCTSWCEWhqOFwMosBJcIgJFCMmGS/abWAUlHIpjnAHHf5PTyff//73g8xwVluFuTyOGbSSBgTyWIoRBSJqZsM8XpaIp++qJjhrO5yfc7zzz4HMc8/zwv9XL6IBlwk9C/VMuaoC2N+r8MhagnGI25eE4U4zmtVmQfF/pEmZl7TbIgcbLWGxZImBWlgtbCAc7Ja4v27oRSICsR80P4K2BvwvnOK9Gu6xNeIVh/1HGOx/rzjEP5+BAX+DDbWcB25KMbx0ZkpiHn80GHWvvv++yGmtLLC2s98/osQs3WeG8k98q73QszD3/eD/NzveRxiXlzjfX/9/CsQ85w41/w09s/SYW6wlivhGA3F2B70X7/A01exNwCGYRiGkUFsA2AYhmEYGcQ2AIZhGIaRQWwDYBiGYRgZ5EARYLHAxTaBUs2JhNFNXqlSt7t2gbWHioFPJIxuuvtXIOaed387a/dHuH/54md+jrW1amKVKW6UsHQcRYDHTp1g7esXuxATC3FMQa00x7t4dgHPNT3DhR5OMd5IRlwwVSxjlUM/5EKcftyCmEQIcyJFTFgrcpGNU6pveXGNYTqZe8lEiFS9Uk3LCZWTZj4jxWmpUjkuCfn3+oo4bSwqDUZKTBLw63GKycrGNhf69CsoHH3Xn/gB1p695z6I+cMnnmDt3/nt34aYq1evsrYU2BERheIagwDHQ0kIxwqKaDYWpke9HgqapqdFJUpFvCevsdPBim1dUWFPMyaaFIZj3i+tNq5HoedjdK5WgZhtIU4rKYrXjjAH647xec8Jc55uiM+7L9as9T285j/xCS7WWzuMz/szX+IVKuMOxuQvX2bt7/7YJyCmKoSBR5eOQcwv/6//gl/PuXMQI39SHjxyFGI+dO+7Wbv54Y9BzL97ihvivfDkFyBmfo6v8/uKmHXQ4uZ7O70mxBB9l/KZvQEwDMMwjExiGwDDMAzDyCC2ATAMwzCMDPI6GgCe12s3GxATFvkeoh1jnqcqcm3DIRoBlSpcO5AroZFI4njyJV/C3OiZ49wQwsWY+yuXec6wsYUxrz3/adZeWLwXYiKRK9fqjcjcnZbzrVT49Xhfgpg05Xm48Qj7MBnzHGagFB5y4vytJuoE+n3+vEpFzCV6kVsfjbS7f/uROX+ZYybC/PX6+jrEXLgs8uCKydLsLNeWvPvd74aYUObGlRysT3kuNyjgeFjb54YpGylO5Qe/5SOs/dJZzGd+6pOfZO3V69chplDgc3FmBvUnUmsh8+tEaEyk5e7Lwjir3UZjmP19riGSxyUiGo95H2qmQ/0+n0OpUjxrUmh2+LWubWHhmH6PP6eyoh2pVnlRGmUYU0GsG3GC/TIzz8fAsscxcb3Fx8BWA3PT11b5c7nvriWIuXKVH2evgWvfBz/8cdauVlETM1rfZu2Z5RWI+fgP/BBrP/+pT0HM/hbXUQQx6sz2rnPjuEMB/p795Y9+J2v/rKL9SovcHGg6xKJCQ2Hm1mxuQcztsDcAhmEYhpFBbANgGIZhGBnENgCGYRiGkUFsA2AYhmEYGeRAEWBOVJeK+igqSVIuyHCy2hQRjWIe02ijiCPIc3FDsYSXJo2IAsV8ghz/Xqu5jSEhF2SUylhdygkxY2v/KsRIw5hIqbbmhKmGVmmuD8YNKLppdblQcW+vATGRF2ZKXezndouLqtauoWAkEdXnymUUoSVCGDQSRjnfSEjTmE2litr2Nu+n6xs4rt7x0MOsHUWKWZAQJUaK8KwoxrXsayKiQ0e4gOnhD3wEYr509lXW/t3f/wOIuXKFG26VKyj4rIoqoJpYTn6mmSlJsaVmvCP7TM4f7XuaWZAUAcpzE+E1a89rUsjl+Focj/EZXLzCq65q97w8y4WXs1Xs32bAhXmVEa7pG02+jjx4DM1wHjvCBX1eqey4ucOP88RTaxBTEfd+3/13Q8z3fuBbWXurgeta6zo3zFlYRFO2qRUuIl9YOAIxPuZja0pUIiQiqta5IDhUfs9K5XnWfugd74SY603+THPKT15RCuaTOxdk2xsAwzAMw8ggtgEwDMMwjAxiGwDDMAzDyCAHJr16A2GUoeTc52d4HqM+Mwcx3RE36lg+imYPtSlRgEYpetBv8dxsLo95xnyJawni8R7EkLiPUgUNSYo5nhvrKBqAeMTz8kFYhRhRK4lipZDKoM9zbttbmF9eFSYWezuY92zv8+c1VIp4JEKjMVRy97JfxyPsw0gMhbm5OsRMAjJfrOVFJSsraBDSaPNnPcS0KM3O8bxfmmCQ/Mwre/C1cxdZu9lCLcexux9k7fn6PMR8fo8/t0tXr0GMzHtLIx4i7EOtoJI0AlJS93BsLb8vjy1NiLTjaMWJ7sTUR55fu69JYSAKfS3NL0LMSBgbre+jQRKJwl4Pn8A8+MZrl1jbKwXD9vu8z78oxiwR0Qfe8QBrf8fHvxViUjGR5O8AEVGtyDVIlVgp6FXh36vNzEJMQRSEKpWwGFWe+HjbPnkcYtrn+TyaUs6VF/N6fhmP0+zyNX1a/lgQUSoK1z318isYQ7yYXGkKNVu3w94AGIZhGEYGsQ2AYRiGYWQQ2wAYhmEYRgaxDYBhGIZhZJADRYBlUX2vVEQB1aDFhRWtZBVigiIXrOQTFDXFLS4qGfSwCtimqKi0cNd7IWbY5VWydq89DzGp44KRkTBbICLqT/Nrdh4FS8MWF1l1mlgBrdHg/bO9uQ8xqxtcrLOzi9UJ+10hdFL0SrLSnFdEJV7ch5NqPiKKxNdm6yhuvOsUF3uurNTxgiYAKWDL51H4I41kSiUU0UwJcdLLZy9ATKPRYO1EEaIFQoQ4UsxwXv7yM/y4OzgXDk1zg5JejEK48y+9yNr9Lh6nXuHPVhPUSSHewsICxEgB3c7ODsTIZ1EsKhU/hemR9iwkmrCzWORiYO2+5DWH0esLRN8uOkKQXS1j3215bgDjExT4Xt9tsPapwyjaPnOSi2C7A+y7OOXrxt4OVvo7e5ULp1fOvQYx62tcCPfoo49BzAfew6tqHlaqUbaFUDa5juOvWOVjqUgo0s15fl+n7r8LYq4IEWKng79nsrrt2Ssokjz3Mp/npx99GGIeOrHM2q+8ihU9L1/iosTH3/cgxNwOewNgGIZhGBnENgCGYRiGkUFsA2AYhmEYGeR1NAA816EWfJG5tRzmi6Iaz4d0WmhQUSqK3OwA8yrjJjcCKhUxP1gp8lxt2sKcO1X59aR9zI0W8zzPtLl9GWJ219ZZe3Ud87l7Tf7ZcIh9uL0t8lcp7ssKwowjVgo+BI7nTwtFfLzT0/zeFxfReGNphecF5+dRAzBV4c/LKYU+JgGZH76T4jK7u7sQs7ffYO2hMj63NrmWpNdFLUcozr+1g+eaEWZaU2XMeW6tc63N/hCNodYucZ2CHEM34HM8r5hrPfquR1h7egqNs0ai+M61ShFizp0/z9pazRJpTDQc4noiC/2Uy1jASOoL2m2c41IXcCcmUW8XdVGQ6eoVLJrTbHENkleKio17fNx+/qXzEPNtj3EDn7pSgWavy83LGlI4RERdoa357B9+HmIub/H1+ZNf+DLE/PKvcr3LBx59BGI+9P73sPY9yychZuowz6cP9vG3odFs8Ou7fgVirrb4up8q+rAk4WOyvIBGXdHdp1j71YuXIWaU8jH68fd+C8S8snWdtcPgztdiewNgGIZhGBnENgCGYRiGkUFsA2AYhmEYGcQ2AIZhGIaRQQ5UC3hhyqGZz+QCLoBIYzRXmJ7lpjqDLTSx8GMuYMnn8dKGLW7u0Nm4DjELR+5j7a0rL0DMoUUuvmhffhFigpCLmDYuvAwx6YirmOan0WRmaZGLd3J5FObsCBFgt4fqqHyBH3ugiNDyBS5iOrKEwpPZWS6crFZRSJnP8+czUgxFKOHCOedRPDYJBNIcSan4VhUCK00w1tjnpk/lEt5vv8vFreMhikITIaRttVEo2BX93VMEfufPbbF2s4PX3BZi26JigpQIJd5RpVLn0hFu/HP1EpogBaLC5sLCNMSsb/A5tbeL5jFBwPtVKahIkahopwk7pXjQOfy7Ti7H+yOd4GqA8YiPpb2xUhHP8TUzH2DnJRFff9pdHKOfepKvh+86eQRiikJQWlZ+SUbCZGyric87FDGnjy1DTL3KBXUdZV7tNPix8z00zFnd4OK9/U2suhoK8eje1ibEDNtcPOgVcXynxcfWXheF74tCGJiMhxDjxI9usYqC1/sWuPHP2QvPQMztsDcAhmEYhpFBbANgGIZhGBnENgCGYRiGkUEO1ADI1Fp9DvODPsfzetvnn4KYuMVNK9pNzHv6ociRBGjK4ROeI7n20mcgpj57mLXXr1yFmMP3C5OSChqtfPGTv8k/6KJpxHSN5yILSg4xn+OdmCSYrzx+jOdLI8WwRRq0eI/5vSjH+6xYQjMWmT/V9oBxzPt51FP0BiJ/GuUncy+ZaEnk10EWvyEiqtW4TmBmrg4xuRzv28999rMQs7HB844XLmCREFlUqKsYCkktQ7eLhaj6ff7cpqbR9GmU8v6p1zF3v7fHzYouX8ZrrtfrrL1y6hTEyMI+SboHMeMxz1HHMWpmpFmQKk4Sn2naj4O/MVnsi6JrM4r50VA8y4RwrZFmR2WlX3aFLuULZ3ENPT5fZ+2qYhYUD/k60uujqROJ6zmkaJJOHudreqGA+pvGJv+NWR/jvHeiPyoBHkeOk0Dpn5xYC4aKZigninM5pbjdy09yLc2pFdQ/5ISpjy/iNXdaXP/Q2MV5dTsmc9U2DMMwDONNxTYAhmEYhpFBbANgGIZhGBnENgCGYRiGkUEOFAEWSlxoUqwuQMxYGIkEIQpPdq88z9ojxRQhX+B7EaeI3AplIc7qo0nD3nVuFhTFKI7avfQ51h50sCLbaJubDNWn0EQlCrkgI8rhvVeq/Jo10xLhoQLmD0REgajMGCvePIHjgppYMWXyQlMVaGLLVJhPFFFMKMUyqWJiMQnEsVJyTpAK8VReMcypVHgVxVARakqx3q/+6q9CTK/H+6nZxLkgjyPFhURoXrSxsQExkiDAsQeiJ8UwJxVCyloNqwHKzwKH/TMe82eRpijwS8XYk89Gi5FmTzc+k/f6jf13nV6fj5uqMiakYddYGrkRUSj6IU0V0zEhzBuMMea1dS40K0bYvzUh1ss5vJ6REEU/f/YsxAz7fI7Up1CoOljgZllhqQwxp09wYer83BzEtHf4fWkVFaXor5+guHHY4deTKyu/Z+L3YtjBtWBvlxt+lVYOYUyTX7NVAzQMwzAM40BsA2AYhmEYGcQ2AIZhGIaRQQ5MFnTaPLfQ6aIhjEhNU76IeZV+l+fYp6Ywpyxzqj7BvJMXCWyteIcL+GeLSkGc8T43YCgTnuueM/x7QaDk5UXeS0lFUpLIHCvmYWUBJc28JhRCAafs3QbCsEd+h4jIC31BqGgApAMUXjHm1kPlviaBoTAj0TQYsr/1/ufjUysYRMT7st9HXUQu4pqQU6dOQ4zUG5TLaI6yucn1L2traxAjTV8SJScs8/DNFhZsmZvlOdelJTQEq9W4yVC7jyKVZlP0mTJ/U6H9iRSDmVDojDRtg7x3aQKjMplDmIiInLjHTh8L4hREES8vF2ciGokiSXGKMbFYZ3WbJf5pc6A87wE/V157TkJzMogxn77/6mXWrhXw92NhkRtsHT2CuXJpRBWneK6S0HWFJdQD9Zvi9yKPeoziQp21naIZqtf59aSK/mZnyPVA3XNYlO6JT/8Ray8dx/l5O+wNgGEYhmFkENsAGIZhGEYGsQ2AYRiGYWQQ2wAYhmEYRgZxd1IlyzAMwzCMby7sDYBhGIZhZBDbABiGYRhGBrENgGEYhmFkENsAGIZhGEYGsQ2AYRiGYWQQ2wAYhmEYRgb5/wPojER28NnY9QAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 648x648 with 9 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "dls.show_batch()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn = cnn_learner(dls, resnet18, metrics=error_rate)\n",
    "#learn.fine_tune(1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Add Transforms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "from train_classification_lightning import choose_dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "mean, std, traindir, valdir, num_c = choose_dataset('cifar10')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "#cifar_stats = ([*mean],[*std])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "#import torchvision.transforms as transforms\n",
    "#data_transform_normal = transforms.Compose([\n",
    "#            transforms.Resize((300,300)),\n",
    "#            #transforms.CenterCrop((100, 100)),\n",
    "#            transforms.RandomCrop((250, 250), padding=4),\n",
    "#            transforms.RandomHorizontalFlip(p=0.5),\n",
    "#            #transforms.RandomRotation(degrees=(-90, 90)),\n",
    "#            #transforms.RandomVerticalFlip(p=0.5),\n",
    "#            transforms.ToTensor(),\n",
    "#            transforms.Normalize(mean, std)\n",
    "#            ])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "item_tfms = [ToTensor, RandomResizedCrop(128, min_scale=0.35)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "item_tfms2 = [ToTensor, Resize(size=(300,300)), \n",
    "              RandomCrop(size=(250,250))#, \n",
    "              #Dihedral()\n",
    "             ]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_tfms = [Dihedral()]\n",
    "#batch_tfms = [Dihedral(),Normalize()]\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "dls = ImageDataLoaders.from_folder(cifar10, train='train', \n",
    "                                   valid='test', device=1, \n",
    "                                   item_tfms=item_tfms2,\n",
    "                                  batch_tfms=batch_tfms,\n",
    "                                  bs=164)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[Path('../cv_data/cifar10/train/bird/20189_bird.png'),\n",
       " Path('../cv_data/cifar10/train/bird/31230_bird.png'),\n",
       " Path('../cv_data/cifar10/train/bird/13226_bird.png'),\n",
       " Path('../cv_data/cifar10/train/bird/8314_bird.png'),\n",
       " Path('../cv_data/cifar10/train/bird/12191_bird.png')]"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dls.train_ds.items[5:10]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "fastai.data.core.DataLoaders"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "type(dls)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "# call to resnet18 goes to the torchvision library so it is same as the one we have in \n",
    "# \n",
    "learn = cnn_learner(dls, resnet18, pretrained=False, \n",
    "                    metrics=[accuracy, top_k_accuracy])\n",
    "#learn.fine_tune(1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>top_k_accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <td>0</td>\n",
       "      <td>1.835039</td>\n",
       "      <td>1.938373</td>\n",
       "      <td>0.367400</td>\n",
       "      <td>0.868000</td>\n",
       "      <td>01:42</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>1</td>\n",
       "      <td>1.513667</td>\n",
       "      <td>1.272368</td>\n",
       "      <td>0.544600</td>\n",
       "      <td>0.944800</td>\n",
       "      <td>01:53</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>2</td>\n",
       "      <td>1.243599</td>\n",
       "      <td>1.508175</td>\n",
       "      <td>0.606100</td>\n",
       "      <td>0.953300</td>\n",
       "      <td>01:57</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>3</td>\n",
       "      <td>1.053734</td>\n",
       "      <td>1.016378</td>\n",
       "      <td>0.705100</td>\n",
       "      <td>0.976600</td>\n",
       "      <td>02:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>4</td>\n",
       "      <td>0.938526</td>\n",
       "      <td>0.732988</td>\n",
       "      <td>0.744500</td>\n",
       "      <td>0.981900</td>\n",
       "      <td>02:08</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>5</td>\n",
       "      <td>0.849213</td>\n",
       "      <td>0.642656</td>\n",
       "      <td>0.776800</td>\n",
       "      <td>0.985100</td>\n",
       "      <td>02:02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>6</td>\n",
       "      <td>0.765467</td>\n",
       "      <td>0.584992</td>\n",
       "      <td>0.800100</td>\n",
       "      <td>0.989000</td>\n",
       "      <td>02:07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>7</td>\n",
       "      <td>0.720816</td>\n",
       "      <td>0.608127</td>\n",
       "      <td>0.780000</td>\n",
       "      <td>0.991400</td>\n",
       "      <td>02:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>8</td>\n",
       "      <td>0.659474</td>\n",
       "      <td>0.523544</td>\n",
       "      <td>0.815900</td>\n",
       "      <td>0.990400</td>\n",
       "      <td>02:07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>9</td>\n",
       "      <td>0.634465</td>\n",
       "      <td>0.528767</td>\n",
       "      <td>0.816700</td>\n",
       "      <td>0.991600</td>\n",
       "      <td>02:01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>10</td>\n",
       "      <td>0.602564</td>\n",
       "      <td>0.459842</td>\n",
       "      <td>0.840200</td>\n",
       "      <td>0.993100</td>\n",
       "      <td>02:03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>11</td>\n",
       "      <td>0.581518</td>\n",
       "      <td>0.474212</td>\n",
       "      <td>0.840700</td>\n",
       "      <td>0.993500</td>\n",
       "      <td>02:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>12</td>\n",
       "      <td>0.545345</td>\n",
       "      <td>0.431500</td>\n",
       "      <td>0.851200</td>\n",
       "      <td>0.994200</td>\n",
       "      <td>02:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>13</td>\n",
       "      <td>0.518050</td>\n",
       "      <td>0.433864</td>\n",
       "      <td>0.847500</td>\n",
       "      <td>0.994700</td>\n",
       "      <td>02:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>14</td>\n",
       "      <td>0.504430</td>\n",
       "      <td>0.406744</td>\n",
       "      <td>0.861700</td>\n",
       "      <td>0.994900</td>\n",
       "      <td>02:03</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>15</td>\n",
       "      <td>0.496834</td>\n",
       "      <td>0.394793</td>\n",
       "      <td>0.866100</td>\n",
       "      <td>0.993500</td>\n",
       "      <td>02:08</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>16</td>\n",
       "      <td>0.457680</td>\n",
       "      <td>0.366922</td>\n",
       "      <td>0.873500</td>\n",
       "      <td>0.994600</td>\n",
       "      <td>02:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>17</td>\n",
       "      <td>0.448991</td>\n",
       "      <td>0.365247</td>\n",
       "      <td>0.873800</td>\n",
       "      <td>0.995700</td>\n",
       "      <td>02:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>18</td>\n",
       "      <td>0.438356</td>\n",
       "      <td>0.354962</td>\n",
       "      <td>0.874900</td>\n",
       "      <td>0.995900</td>\n",
       "      <td>02:02</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>19</td>\n",
       "      <td>0.413812</td>\n",
       "      <td>0.386303</td>\n",
       "      <td>0.867600</td>\n",
       "      <td>0.996500</td>\n",
       "      <td>02:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>20</td>\n",
       "      <td>0.415469</td>\n",
       "      <td>0.381111</td>\n",
       "      <td>0.869800</td>\n",
       "      <td>0.995600</td>\n",
       "      <td>02:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>21</td>\n",
       "      <td>0.386615</td>\n",
       "      <td>0.339867</td>\n",
       "      <td>0.883900</td>\n",
       "      <td>0.996900</td>\n",
       "      <td>02:13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>22</td>\n",
       "      <td>0.370650</td>\n",
       "      <td>0.335333</td>\n",
       "      <td>0.888500</td>\n",
       "      <td>0.995900</td>\n",
       "      <td>02:04</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>23</td>\n",
       "      <td>0.350464</td>\n",
       "      <td>0.325240</td>\n",
       "      <td>0.890000</td>\n",
       "      <td>0.996700</td>\n",
       "      <td>02:07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>24</td>\n",
       "      <td>0.356943</td>\n",
       "      <td>0.337944</td>\n",
       "      <td>0.887900</td>\n",
       "      <td>0.995900</td>\n",
       "      <td>02:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>25</td>\n",
       "      <td>0.351880</td>\n",
       "      <td>0.305306</td>\n",
       "      <td>0.893600</td>\n",
       "      <td>0.996600</td>\n",
       "      <td>02:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>26</td>\n",
       "      <td>0.331332</td>\n",
       "      <td>0.308509</td>\n",
       "      <td>0.898500</td>\n",
       "      <td>0.997700</td>\n",
       "      <td>02:07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>27</td>\n",
       "      <td>0.318552</td>\n",
       "      <td>0.296424</td>\n",
       "      <td>0.900900</td>\n",
       "      <td>0.997200</td>\n",
       "      <td>02:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>28</td>\n",
       "      <td>0.306777</td>\n",
       "      <td>0.322276</td>\n",
       "      <td>0.891000</td>\n",
       "      <td>0.996600</td>\n",
       "      <td>02:13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>29</td>\n",
       "      <td>0.303445</td>\n",
       "      <td>0.304375</td>\n",
       "      <td>0.898900</td>\n",
       "      <td>0.997200</td>\n",
       "      <td>02:13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>30</td>\n",
       "      <td>0.291861</td>\n",
       "      <td>0.300158</td>\n",
       "      <td>0.901500</td>\n",
       "      <td>0.995700</td>\n",
       "      <td>02:07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>31</td>\n",
       "      <td>0.289277</td>\n",
       "      <td>0.308696</td>\n",
       "      <td>0.897800</td>\n",
       "      <td>0.997100</td>\n",
       "      <td>02:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>32</td>\n",
       "      <td>0.277173</td>\n",
       "      <td>0.298511</td>\n",
       "      <td>0.901200</td>\n",
       "      <td>0.997100</td>\n",
       "      <td>02:17</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>33</td>\n",
       "      <td>0.272114</td>\n",
       "      <td>0.282092</td>\n",
       "      <td>0.907000</td>\n",
       "      <td>0.997500</td>\n",
       "      <td>02:23</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>34</td>\n",
       "      <td>0.260101</td>\n",
       "      <td>0.298253</td>\n",
       "      <td>0.903600</td>\n",
       "      <td>0.997500</td>\n",
       "      <td>02:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>35</td>\n",
       "      <td>0.261253</td>\n",
       "      <td>0.309236</td>\n",
       "      <td>0.897500</td>\n",
       "      <td>0.996600</td>\n",
       "      <td>02:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>36</td>\n",
       "      <td>0.260535</td>\n",
       "      <td>0.293916</td>\n",
       "      <td>0.905600</td>\n",
       "      <td>0.997200</td>\n",
       "      <td>02:07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>37</td>\n",
       "      <td>0.252058</td>\n",
       "      <td>0.290914</td>\n",
       "      <td>0.904000</td>\n",
       "      <td>0.996500</td>\n",
       "      <td>02:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>38</td>\n",
       "      <td>0.251916</td>\n",
       "      <td>0.290786</td>\n",
       "      <td>0.909000</td>\n",
       "      <td>0.998000</td>\n",
       "      <td>02:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>39</td>\n",
       "      <td>0.223932</td>\n",
       "      <td>0.292896</td>\n",
       "      <td>0.905600</td>\n",
       "      <td>0.997000</td>\n",
       "      <td>02:07</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>40</td>\n",
       "      <td>0.226793</td>\n",
       "      <td>0.287885</td>\n",
       "      <td>0.908200</td>\n",
       "      <td>0.997500</td>\n",
       "      <td>02:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>41</td>\n",
       "      <td>0.229174</td>\n",
       "      <td>0.287210</td>\n",
       "      <td>0.909600</td>\n",
       "      <td>0.997900</td>\n",
       "      <td>02:09</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>42</td>\n",
       "      <td>0.215113</td>\n",
       "      <td>0.276295</td>\n",
       "      <td>0.910900</td>\n",
       "      <td>0.998100</td>\n",
       "      <td>02:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>43</td>\n",
       "      <td>0.218968</td>\n",
       "      <td>0.300341</td>\n",
       "      <td>0.907400</td>\n",
       "      <td>0.996800</td>\n",
       "      <td>02:05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>44</td>\n",
       "      <td>0.198462</td>\n",
       "      <td>0.281393</td>\n",
       "      <td>0.910600</td>\n",
       "      <td>0.997200</td>\n",
       "      <td>02:10</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>45</td>\n",
       "      <td>0.207056</td>\n",
       "      <td>0.280196</td>\n",
       "      <td>0.911500</td>\n",
       "      <td>0.997300</td>\n",
       "      <td>02:08</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>46</td>\n",
       "      <td>0.215518</td>\n",
       "      <td>0.269899</td>\n",
       "      <td>0.911400</td>\n",
       "      <td>0.997200</td>\n",
       "      <td>02:12</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>47</td>\n",
       "      <td>0.196897</td>\n",
       "      <td>0.284737</td>\n",
       "      <td>0.912100</td>\n",
       "      <td>0.997200</td>\n",
       "      <td>02:08</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>48</td>\n",
       "      <td>0.194964</td>\n",
       "      <td>0.295184</td>\n",
       "      <td>0.909300</td>\n",
       "      <td>0.997000</td>\n",
       "      <td>02:06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <td>49</td>\n",
       "      <td>0.202749</td>\n",
       "      <td>0.283486</td>\n",
       "      <td>0.909900</td>\n",
       "      <td>0.997700</td>\n",
       "      <td>02:09</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn.fit(n_epoch=50)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Test Timm Model\n",
    "\n",
    "https://rwightman.github.io/pytorch-image-models/models/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "['adv_inception_v3',\n",
      " 'cspdarknet53',\n",
      " 'cspdarknet53_iabn',\n",
      " 'cspresnet50',\n",
      " 'cspresnet50d',\n",
      " 'cspresnet50w',\n",
      " 'cspresnext50',\n",
      " 'cspresnext50_iabn',\n",
      " 'darknet53',\n",
      " 'densenet121',\n",
      " 'densenet121d',\n",
      " 'densenet161',\n",
      " 'densenet169',\n",
      " 'densenet201',\n",
      " 'densenet264',\n",
      " 'densenet264d_iabn',\n",
      " 'densenetblur121d',\n",
      " 'dla34',\n",
      " 'dla46_c',\n",
      " 'dla46x_c',\n",
      " 'dla60',\n",
      " 'dla60_res2net',\n",
      " 'dla60_res2next',\n",
      " 'dla60x',\n",
      " 'dla60x_c',\n",
      " 'dla102',\n",
      " 'dla102x',\n",
      " 'dla102x2',\n",
      " 'dla169',\n",
      " 'dpn68',\n",
      " 'dpn68b',\n",
      " 'dpn92',\n",
      " 'dpn98',\n",
      " 'dpn107',\n",
      " 'dpn131',\n",
      " 'eca_vovnet39b',\n",
      " 'ecaresnet18',\n",
      " 'ecaresnet50',\n",
      " 'ecaresnet50d',\n",
      " 'ecaresnet50d_pruned',\n",
      " 'ecaresnet101d',\n",
      " 'ecaresnet101d_pruned',\n",
      " 'ecaresnetlight',\n",
      " 'ecaresnext26tn_32x4d',\n",
      " 'efficientnet_b0',\n",
      " 'efficientnet_b1',\n",
      " 'efficientnet_b1_pruned',\n",
      " 'efficientnet_b2',\n",
      " 'efficientnet_b2_pruned',\n",
      " 'efficientnet_b2a',\n",
      " 'efficientnet_b3',\n",
      " 'efficientnet_b3_pruned',\n",
      " 'efficientnet_b3a',\n",
      " 'efficientnet_b4',\n",
      " 'efficientnet_b5',\n",
      " 'efficientnet_b6',\n",
      " 'efficientnet_b7',\n",
      " 'efficientnet_b8',\n",
      " 'efficientnet_cc_b0_4e',\n",
      " 'efficientnet_cc_b0_8e',\n",
      " 'efficientnet_cc_b1_8e',\n",
      " 'efficientnet_el',\n",
      " 'efficientnet_em',\n",
      " 'efficientnet_es',\n",
      " 'efficientnet_l2',\n",
      " 'efficientnet_lite0',\n",
      " 'efficientnet_lite1',\n",
      " 'efficientnet_lite2',\n",
      " 'efficientnet_lite3',\n",
      " 'efficientnet_lite4',\n",
      " 'ens_adv_inception_resnet_v2',\n",
      " 'ese_vovnet19b_dw',\n",
      " 'ese_vovnet19b_slim',\n",
      " 'ese_vovnet19b_slim_dw',\n",
      " 'ese_vovnet39b',\n",
      " 'ese_vovnet39b_evos',\n",
      " 'ese_vovnet57b',\n",
      " 'ese_vovnet99b',\n",
      " 'ese_vovnet99b_iabn',\n",
      " 'fbnetc_100',\n",
      " 'gluon_inception_v3',\n",
      " 'gluon_resnet18_v1b',\n",
      " 'gluon_resnet34_v1b',\n",
      " 'gluon_resnet50_v1b',\n",
      " 'gluon_resnet50_v1c',\n",
      " 'gluon_resnet50_v1d',\n",
      " 'gluon_resnet50_v1s',\n",
      " 'gluon_resnet101_v1b',\n",
      " 'gluon_resnet101_v1c',\n",
      " 'gluon_resnet101_v1d',\n",
      " 'gluon_resnet101_v1s',\n",
      " 'gluon_resnet152_v1b',\n",
      " 'gluon_resnet152_v1c',\n",
      " 'gluon_resnet152_v1d',\n",
      " 'gluon_resnet152_v1s',\n",
      " 'gluon_resnext50_32x4d',\n",
      " 'gluon_resnext101_32x4d',\n",
      " 'gluon_resnext101_64x4d',\n",
      " 'gluon_senet154',\n",
      " 'gluon_seresnext50_32x4d',\n",
      " 'gluon_seresnext101_32x4d',\n",
      " 'gluon_seresnext101_64x4d',\n",
      " 'gluon_xception65',\n",
      " 'hrnet_w18',\n",
      " 'hrnet_w18_small',\n",
      " 'hrnet_w18_small_v2',\n",
      " 'hrnet_w30',\n",
      " 'hrnet_w32',\n",
      " 'hrnet_w40',\n",
      " 'hrnet_w44',\n",
      " 'hrnet_w48',\n",
      " 'hrnet_w64',\n",
      " 'ig_resnext101_32x8d',\n",
      " 'ig_resnext101_32x16d',\n",
      " 'ig_resnext101_32x32d',\n",
      " 'ig_resnext101_32x48d',\n",
      " 'inception_resnet_v2',\n",
      " 'inception_v3',\n",
      " 'inception_v4',\n",
      " 'legacy_senet154',\n",
      " 'legacy_seresnet18',\n",
      " 'legacy_seresnet34',\n",
      " 'legacy_seresnet50',\n",
      " 'legacy_seresnet101',\n",
      " 'legacy_seresnet152',\n",
      " 'legacy_seresnext26_32x4d',\n",
      " 'legacy_seresnext50_32x4d',\n",
      " 'legacy_seresnext101_32x4d',\n",
      " 'mixnet_l',\n",
      " 'mixnet_m',\n",
      " 'mixnet_s',\n",
      " 'mixnet_xl',\n",
      " 'mixnet_xxl',\n",
      " 'mnasnet_050',\n",
      " 'mnasnet_075',\n",
      " 'mnasnet_100',\n",
      " 'mnasnet_140',\n",
      " 'mnasnet_a1',\n",
      " 'mnasnet_b1',\n",
      " 'mnasnet_small',\n",
      " 'mobilenetv2_100',\n",
      " 'mobilenetv2_110d',\n",
      " 'mobilenetv2_120d',\n",
      " 'mobilenetv2_140',\n",
      " 'mobilenetv3_large_075',\n",
      " 'mobilenetv3_large_100',\n",
      " 'mobilenetv3_rw',\n",
      " 'mobilenetv3_small_075',\n",
      " 'mobilenetv3_small_100',\n",
      " 'nasnetalarge',\n",
      " 'pnasnet5large',\n",
      " 'regnetx_002',\n",
      " 'regnetx_004',\n",
      " 'regnetx_006',\n",
      " 'regnetx_008',\n",
      " 'regnetx_016',\n",
      " 'regnetx_032',\n",
      " 'regnetx_040',\n",
      " 'regnetx_064',\n",
      " 'regnetx_080',\n",
      " 'regnetx_120',\n",
      " 'regnetx_160',\n",
      " 'regnetx_320',\n",
      " 'regnety_002',\n",
      " 'regnety_004',\n",
      " 'regnety_006',\n",
      " 'regnety_008',\n",
      " 'regnety_016',\n",
      " 'regnety_032',\n",
      " 'regnety_040',\n",
      " 'regnety_064',\n",
      " 'regnety_080',\n",
      " 'regnety_120',\n",
      " 'regnety_160',\n",
      " 'regnety_320',\n",
      " 'res2net50_14w_8s',\n",
      " 'res2net50_26w_4s',\n",
      " 'res2net50_26w_6s',\n",
      " 'res2net50_26w_8s',\n",
      " 'res2net50_48w_2s',\n",
      " 'res2net101_26w_4s',\n",
      " 'res2next50',\n",
      " 'resnest14d',\n",
      " 'resnest26d',\n",
      " 'resnest50d',\n",
      " 'resnest50d_1s4x24d',\n",
      " 'resnest50d_4s2x40d',\n",
      " 'resnest101e',\n",
      " 'resnest200e',\n",
      " 'resnest269e',\n",
      " 'resnet18',\n",
      " 'resnet18d',\n",
      " 'resnet26',\n",
      " 'resnet26d',\n",
      " 'resnet34',\n",
      " 'resnet34d',\n",
      " 'resnet50',\n",
      " 'resnet50d',\n",
      " 'resnet101',\n",
      " 'resnet101d',\n",
      " 'resnet101d_320',\n",
      " 'resnet152',\n",
      " 'resnet152d',\n",
      " 'resnet152d_320',\n",
      " 'resnet200',\n",
      " 'resnet200d',\n",
      " 'resnet200d_320',\n",
      " 'resnetblur18',\n",
      " 'resnetblur50',\n",
      " 'resnext50_32x4d',\n",
      " 'resnext50d_32x4d',\n",
      " 'resnext101_32x4d',\n",
      " 'resnext101_32x8d',\n",
      " 'resnext101_64x4d',\n",
      " 'rexnet_100',\n",
      " 'rexnet_130',\n",
      " 'rexnet_150',\n",
      " 'rexnet_200',\n",
      " 'rexnetr_100',\n",
      " 'rexnetr_130',\n",
      " 'rexnetr_150',\n",
      " 'rexnetr_200',\n",
      " 'selecsls42',\n",
      " 'selecsls42b',\n",
      " 'selecsls60',\n",
      " 'selecsls60b',\n",
      " 'selecsls84',\n",
      " 'semnasnet_050',\n",
      " 'semnasnet_075',\n",
      " 'semnasnet_100',\n",
      " 'semnasnet_140',\n",
      " 'senet154',\n",
      " 'seresnet18',\n",
      " 'seresnet34',\n",
      " 'seresnet50',\n",
      " 'seresnet50tn',\n",
      " 'seresnet101',\n",
      " 'seresnet152',\n",
      " 'seresnet152d',\n",
      " 'seresnet152d_320',\n",
      " 'seresnext26_32x4d',\n",
      " 'seresnext26d_32x4d',\n",
      " 'seresnext26t_32x4d',\n",
      " 'seresnext26tn_32x4d',\n",
      " 'seresnext50_32x4d',\n",
      " 'seresnext101_32x4d',\n",
      " 'seresnext101_32x8d',\n",
      " 'skresnet18',\n",
      " 'skresnet34',\n",
      " 'skresnet50',\n",
      " 'skresnet50d',\n",
      " 'skresnext50_32x4d',\n",
      " 'spnasnet_100',\n",
      " 'ssl_resnet18',\n",
      " 'ssl_resnet50',\n",
      " 'ssl_resnext50_32x4d',\n",
      " 'ssl_resnext101_32x4d',\n",
      " 'ssl_resnext101_32x8d',\n",
      " 'ssl_resnext101_32x16d',\n",
      " 'swsl_resnet18',\n",
      " 'swsl_resnet50',\n",
      " 'swsl_resnext50_32x4d',\n",
      " 'swsl_resnext101_32x4d',\n",
      " 'swsl_resnext101_32x8d',\n",
      " 'swsl_resnext101_32x16d',\n",
      " 'tf_efficientnet_b0',\n",
      " 'tf_efficientnet_b0_ap',\n",
      " 'tf_efficientnet_b0_ns',\n",
      " 'tf_efficientnet_b1',\n",
      " 'tf_efficientnet_b1_ap',\n",
      " 'tf_efficientnet_b1_ns',\n",
      " 'tf_efficientnet_b2',\n",
      " 'tf_efficientnet_b2_ap',\n",
      " 'tf_efficientnet_b2_ns',\n",
      " 'tf_efficientnet_b3',\n",
      " 'tf_efficientnet_b3_ap',\n",
      " 'tf_efficientnet_b3_ns',\n",
      " 'tf_efficientnet_b4',\n",
      " 'tf_efficientnet_b4_ap',\n",
      " 'tf_efficientnet_b4_ns',\n",
      " 'tf_efficientnet_b5',\n",
      " 'tf_efficientnet_b5_ap',\n",
      " 'tf_efficientnet_b5_ns',\n",
      " 'tf_efficientnet_b6',\n",
      " 'tf_efficientnet_b6_ap',\n",
      " 'tf_efficientnet_b6_ns',\n",
      " 'tf_efficientnet_b7',\n",
      " 'tf_efficientnet_b7_ap',\n",
      " 'tf_efficientnet_b7_ns',\n",
      " 'tf_efficientnet_b8',\n",
      " 'tf_efficientnet_b8_ap',\n",
      " 'tf_efficientnet_cc_b0_4e',\n",
      " 'tf_efficientnet_cc_b0_8e',\n",
      " 'tf_efficientnet_cc_b1_8e',\n",
      " 'tf_efficientnet_el',\n",
      " 'tf_efficientnet_em',\n",
      " 'tf_efficientnet_es',\n",
      " 'tf_efficientnet_l2_ns',\n",
      " 'tf_efficientnet_l2_ns_475',\n",
      " 'tf_efficientnet_lite0',\n",
      " 'tf_efficientnet_lite1',\n",
      " 'tf_efficientnet_lite2',\n",
      " 'tf_efficientnet_lite3',\n",
      " 'tf_efficientnet_lite4',\n",
      " 'tf_inception_v3',\n",
      " 'tf_mixnet_l',\n",
      " 'tf_mixnet_m',\n",
      " 'tf_mixnet_s',\n",
      " 'tf_mobilenetv3_large_075',\n",
      " 'tf_mobilenetv3_large_100',\n",
      " 'tf_mobilenetv3_large_minimal_100',\n",
      " 'tf_mobilenetv3_small_075',\n",
      " 'tf_mobilenetv3_small_100',\n",
      " 'tf_mobilenetv3_small_minimal_100',\n",
      " 'tresnet_l',\n",
      " 'tresnet_l_448',\n",
      " 'tresnet_m',\n",
      " 'tresnet_m_448',\n",
      " 'tresnet_xl',\n",
      " 'tresnet_xl_448',\n",
      " 'tv_densenet121',\n",
      " 'tv_resnet34',\n",
      " 'tv_resnet50',\n",
      " 'tv_resnet101',\n",
      " 'tv_resnet152',\n",
      " 'tv_resnext50_32x4d',\n",
      " 'vit_base_patch16_224',\n",
      " 'vit_base_patch16_384',\n",
      " 'vit_base_patch32_384',\n",
      " 'vit_base_resnet26d_224',\n",
      " 'vit_base_resnet50d_224',\n",
      " 'vit_huge_patch16_224',\n",
      " 'vit_huge_patch32_384',\n",
      " 'vit_large_patch16_224',\n",
      " 'vit_large_patch16_384',\n",
      " 'vit_large_patch32_384',\n",
      " 'vit_small_patch16_224',\n",
      " 'vit_small_resnet26d_224',\n",
      " 'vit_small_resnet50d_s3_224',\n",
      " 'vovnet39a',\n",
      " 'vovnet57a',\n",
      " 'wide_resnet50_2',\n",
      " 'wide_resnet101_2',\n",
      " 'xception',\n",
      " 'xception41',\n",
      " 'xception65',\n",
      " 'xception71']\n"
     ]
    }
   ],
   "source": [
    "import timm\n",
    "from pprint import pprint\n",
    "model_names = timm.list_models(pretrained=False)\n",
    "pprint(model_names)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "efficient_net_compare = timm.create_model('efficientnet_b0', pretrained=False, num_classes=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true
    }
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "EfficientNet(\n",
      "  (conv_stem): Conv2d(3, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
      "  (bn1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  (act1): SiLU(inplace=True)\n",
      "  (blocks): Sequential(\n",
      "    (0): Sequential(\n",
      "      (0): DepthwiseSeparableConv(\n",
      "        (conv_dw): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)\n",
      "        (bn1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(32, 8, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(8, 32, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pw): Conv2d(32, 16, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): Identity()\n",
      "      )\n",
      "    )\n",
      "    (1): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(16, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(96, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=96, bias=False)\n",
      "        (bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(96, 4, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(4, 96, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(96, 24, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (1): InvertedResidual(\n",
      "        (conv_pw): Conv2d(24, 144, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(144, 144, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=144, bias=False)\n",
      "        (bn2): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(144, 6, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(6, 144, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(144, 24, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "    (2): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(24, 144, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(144, 144, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=144, bias=False)\n",
      "        (bn2): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(144, 6, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(6, 144, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(144, 40, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(40, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (1): InvertedResidual(\n",
      "        (conv_pw): Conv2d(40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(240, 240, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=240, bias=False)\n",
      "        (bn2): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(240, 10, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(10, 240, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(240, 40, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(40, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "    (3): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(240, 240, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=240, bias=False)\n",
      "        (bn2): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(240, 10, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(10, 240, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(240, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (1): InvertedResidual(\n",
      "        (conv_pw): Conv2d(80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(480, 480, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=480, bias=False)\n",
      "        (bn2): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(480, 20, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(20, 480, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(480, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (2): InvertedResidual(\n",
      "        (conv_pw): Conv2d(80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(480, 480, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=480, bias=False)\n",
      "        (bn2): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(480, 20, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(20, 480, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(480, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "    (4): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(480, 480, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=480, bias=False)\n",
      "        (bn2): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(480, 20, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(20, 480, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(480, 112, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(112, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (1): InvertedResidual(\n",
      "        (conv_pw): Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(672, 672, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=672, bias=False)\n",
      "        (bn2): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(672, 28, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(28, 672, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(672, 112, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(112, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (2): InvertedResidual(\n",
      "        (conv_pw): Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(672, 672, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=672, bias=False)\n",
      "        (bn2): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(672, 28, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(28, 672, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(672, 112, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(112, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "    (5): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(672, 672, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=672, bias=False)\n",
      "        (bn2): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(672, 28, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(28, 672, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(672, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (1): InvertedResidual(\n",
      "        (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(1152, 1152, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=1152, bias=False)\n",
      "        (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (2): InvertedResidual(\n",
      "        (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(1152, 1152, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=1152, bias=False)\n",
      "        (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (3): InvertedResidual(\n",
      "        (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(1152, 1152, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=1152, bias=False)\n",
      "        (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "    (6): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(1152, 1152, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1152, bias=False)\n",
      "        (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(1152, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(320, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "  )\n",
      "  (conv_head): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "  (bn2): BatchNorm2d(1280, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  (act2): SiLU(inplace=True)\n",
      "  (global_pool): SelectAdaptivePool2d (pool_type=avg, flatten=True)\n",
      "  (classifier): Linear(in_features=1280, out_features=10, bias=True)\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "print(efficient_net_compare)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "# See first layer\n",
    "from fastai.vision.learner import _get_first_layer\n",
    "c,p,n = _get_first_layer(efficient_net_compare)\n",
    "# c is the layer\n",
    "# p is the whole network\n",
    "# n is the name of the layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Conv2d(3, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n"
     ]
    }
   ],
   "source": [
    "print(c)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "conv_stem\n"
     ]
    }
   ],
   "source": [
    "print(n)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "from fastai.vision.learner import _update_first_layer\n",
    "\n",
    "# from https://walkwithfastai.com/vision.external.timm\n",
    "def create_timm_body(arch:str, pretrained=True, cut=None, n_in=3):\n",
    "    \"Creates a body from any model in the `timm` library.\"\n",
    "    model = timm.create_model(arch, pretrained=pretrained, num_classes=0, global_pool='')\n",
    "    _update_first_layer(model, n_in, pretrained)\n",
    "    if cut is None:\n",
    "        ll = list(enumerate(model.children()))\n",
    "        cut = next(i for i,o in reversed(ll) if has_pool_type(o))\n",
    "    if isinstance(cut, int): return nn.Sequential(*list(model.children())[:cut])\n",
    "    elif callable(cut): return cut(model)\n",
    "    else: raise NamedError(\"cut must be either integer or function\")\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "body = create_timm_body('efficientnet_b0')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Sequential(\n",
      "  (0): Conv2d(3, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
      "  (1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  (2): SiLU(inplace=True)\n",
      "  (3): Sequential(\n",
      "    (0): Sequential(\n",
      "      (0): DepthwiseSeparableConv(\n",
      "        (conv_dw): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)\n",
      "        (bn1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(32, 8, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(8, 32, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pw): Conv2d(32, 16, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn2): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): Identity()\n",
      "      )\n",
      "    )\n",
      "    (1): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(16, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(96, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=96, bias=False)\n",
      "        (bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(96, 4, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(4, 96, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(96, 24, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (1): InvertedResidual(\n",
      "        (conv_pw): Conv2d(24, 144, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(144, 144, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=144, bias=False)\n",
      "        (bn2): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(144, 6, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(6, 144, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(144, 24, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "    (2): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(24, 144, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(144, 144, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=144, bias=False)\n",
      "        (bn2): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(144, 6, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(6, 144, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(144, 40, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(40, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (1): InvertedResidual(\n",
      "        (conv_pw): Conv2d(40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(240, 240, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=240, bias=False)\n",
      "        (bn2): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(240, 10, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(10, 240, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(240, 40, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(40, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "    (3): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(240, 240, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=240, bias=False)\n",
      "        (bn2): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(240, 10, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(10, 240, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(240, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (1): InvertedResidual(\n",
      "        (conv_pw): Conv2d(80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(480, 480, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=480, bias=False)\n",
      "        (bn2): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(480, 20, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(20, 480, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(480, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (2): InvertedResidual(\n",
      "        (conv_pw): Conv2d(80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(480, 480, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=480, bias=False)\n",
      "        (bn2): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(480, 20, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(20, 480, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(480, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "    (4): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(480, 480, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=480, bias=False)\n",
      "        (bn2): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(480, 20, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(20, 480, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(480, 112, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(112, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (1): InvertedResidual(\n",
      "        (conv_pw): Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(672, 672, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=672, bias=False)\n",
      "        (bn2): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(672, 28, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(28, 672, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(672, 112, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(112, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (2): InvertedResidual(\n",
      "        (conv_pw): Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(672, 672, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=672, bias=False)\n",
      "        (bn2): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(672, 28, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(28, 672, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(672, 112, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(112, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "    (5): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(672, 672, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=672, bias=False)\n",
      "        (bn2): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(672, 28, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(28, 672, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(672, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (1): InvertedResidual(\n",
      "        (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(1152, 1152, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=1152, bias=False)\n",
      "        (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (2): InvertedResidual(\n",
      "        (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(1152, 1152, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=1152, bias=False)\n",
      "        (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "      (3): InvertedResidual(\n",
      "        (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(1152, 1152, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=1152, bias=False)\n",
      "        (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "    (6): Sequential(\n",
      "      (0): InvertedResidual(\n",
      "        (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act1): SiLU(inplace=True)\n",
      "        (conv_dw): Conv2d(1152, 1152, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1152, bias=False)\n",
      "        (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "        (act2): SiLU(inplace=True)\n",
      "        (se): SqueezeExcite(\n",
      "          (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
      "          (act1): SiLU(inplace=True)\n",
      "          (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
      "        )\n",
      "        (conv_pwl): Conv2d(1152, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "        (bn3): BatchNorm2d(320, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "      )\n",
      "    )\n",
      "  )\n",
      "  (4): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
      "  (5): BatchNorm2d(1280, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
      "  (6): SiLU(inplace=True)\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "print(body)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "def create_timm_model(arch:str, n_out, cut=None, pretrained=False, n_in=3, init=nn.init.kaiming_normal_, custom_head=None,\n",
    "                     concat_pool=True, **kwargs):\n",
    "    \"Create custom architecture using `arch`, `n_in` and `n_out` from the `timm` library\"\n",
    "    body = create_timm_body(arch, pretrained, None, n_in)\n",
    "    if custom_head is None:\n",
    "        nf = num_features_model(nn.Sequential(*body.children())) * (2 if concat_pool else 1)\n",
    "        head = create_head(nf, n_out, concat_pool=concat_pool, **kwargs)\n",
    "    else: head = custom_head\n",
    "    model = nn.Sequential(body, head)\n",
    "    if init is not None: apply_init(model[1], init)\n",
    "    return model, body, head"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "10"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "get_c(dls)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create timm model is broken and not calculating right"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [],
   "source": [
    "validate_model_processing, body, head = create_timm_model('efficientnet_b0', 10, concat_pool=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "1280"
      ]
     },
     "execution_count": 59,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "num_features_model(nn.Sequential(*body.children()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Sequential(\n",
       "  (0): AdaptiveAvgPool2d(output_size=1)\n",
       "  (1): Flatten(full=False)\n",
       "  (2): BatchNorm1d(1280, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (3): Dropout(p=0.25, inplace=False)\n",
       "  (4): Linear(in_features=1280, out_features=512, bias=False)\n",
       "  (5): ReLU(inplace=True)\n",
       "  (6): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "  (7): Dropout(p=0.5, inplace=False)\n",
       "  (8): Linear(in_features=512, out_features=10, bias=False)\n",
       ")"
      ]
     },
     "execution_count": 60,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "head"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [],
   "source": [
    "def timm_learner(dls, arch:str, loss_func=None, pretrained=False, cut=None, splitter=None,\n",
    "                y_range=None, config=None, n_out=None, normalize=True, **kwargs):\n",
    "    \"Build a convnet style learner from `dls` and `arch` using the `timm` library\"\n",
    "    if config is None: config = {}\n",
    "    if n_out is None: n_out = get_c(dls)\n",
    "    assert n_out, \"`n_out` is not defined, and could not be inferred from data, set `dls.c` or pass `n_out`\"\n",
    "    if y_range is None and 'y_range' in config: y_range = config.pop('y_range')\n",
    "    model, body, head = create_timm_model(arch, n_out, default_split, pretrained, y_range=y_range, \n",
    "                                          concat_pool=False, **config)\n",
    "    learn = Learner(dls, model, loss_func=loss_func, splitter=default_split, **kwargs)\n",
    "    if pretrained: learn.freeze()\n",
    "    return learn"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "concat_pool is resulting in incorrect number of layers. Setting this to false lets the head be computed properly. We need to try and see what this was set the way that it was?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [],
   "source": [
    "dls = ImageDataLoaders.from_folder(cifar10, train='train', \n",
    "                                   valid='test', device=1, \n",
    "                                   item_tfms=item_tfms2,\n",
    "                                  batch_tfms=batch_tfms,\n",
    "                                  bs=32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 70,
   "metadata": {},
   "outputs": [],
   "source": [
    "learn2 = timm_learner(dls, 'efficientnet_b0', pretrained=False,\n",
    "                    metrics=[accuracy, top_k_accuracy])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true,
    "jupyter": {
     "outputs_hidden": true
    }
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Sequential(\n",
       "  (0): Sequential(\n",
       "    (0): Conv2d(3, 32, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), bias=False)\n",
       "    (1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (2): SiLU(inplace=True)\n",
       "    (3): Sequential(\n",
       "      (0): Sequential(\n",
       "        (0): DepthwiseSeparableConv(\n",
       "          (conv_dw): Conv2d(32, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=32, bias=False)\n",
       "          (bn1): BatchNorm2d(32, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(32, 8, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(8, 32, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pw): Conv2d(32, 16, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn2): BatchNorm2d(16, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): Identity()\n",
       "        )\n",
       "      )\n",
       "      (1): Sequential(\n",
       "        (0): InvertedResidual(\n",
       "          (conv_pw): Conv2d(16, 96, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(96, 96, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=96, bias=False)\n",
       "          (bn2): BatchNorm2d(96, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(96, 4, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(4, 96, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(96, 24, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (1): InvertedResidual(\n",
       "          (conv_pw): Conv2d(24, 144, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(144, 144, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=144, bias=False)\n",
       "          (bn2): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(144, 6, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(6, 144, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(144, 24, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(24, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (2): Sequential(\n",
       "        (0): InvertedResidual(\n",
       "          (conv_pw): Conv2d(24, 144, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(144, 144, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=144, bias=False)\n",
       "          (bn2): BatchNorm2d(144, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(144, 6, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(6, 144, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(144, 40, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(40, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (1): InvertedResidual(\n",
       "          (conv_pw): Conv2d(40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(240, 240, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=240, bias=False)\n",
       "          (bn2): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(240, 10, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(10, 240, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(240, 40, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(40, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (3): Sequential(\n",
       "        (0): InvertedResidual(\n",
       "          (conv_pw): Conv2d(40, 240, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(240, 240, kernel_size=(3, 3), stride=(2, 2), padding=(1, 1), groups=240, bias=False)\n",
       "          (bn2): BatchNorm2d(240, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(240, 10, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(10, 240, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(240, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (1): InvertedResidual(\n",
       "          (conv_pw): Conv2d(80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(480, 480, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=480, bias=False)\n",
       "          (bn2): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(480, 20, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(20, 480, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(480, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (2): InvertedResidual(\n",
       "          (conv_pw): Conv2d(80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(480, 480, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=480, bias=False)\n",
       "          (bn2): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(480, 20, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(20, 480, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(480, 80, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(80, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (4): Sequential(\n",
       "        (0): InvertedResidual(\n",
       "          (conv_pw): Conv2d(80, 480, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(480, 480, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=480, bias=False)\n",
       "          (bn2): BatchNorm2d(480, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(480, 20, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(20, 480, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(480, 112, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(112, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (1): InvertedResidual(\n",
       "          (conv_pw): Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(672, 672, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=672, bias=False)\n",
       "          (bn2): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(672, 28, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(28, 672, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(672, 112, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(112, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (2): InvertedResidual(\n",
       "          (conv_pw): Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(672, 672, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=672, bias=False)\n",
       "          (bn2): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(672, 28, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(28, 672, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(672, 112, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(112, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (5): Sequential(\n",
       "        (0): InvertedResidual(\n",
       "          (conv_pw): Conv2d(112, 672, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(672, 672, kernel_size=(5, 5), stride=(2, 2), padding=(2, 2), groups=672, bias=False)\n",
       "          (bn2): BatchNorm2d(672, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(672, 28, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(28, 672, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(672, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (1): InvertedResidual(\n",
       "          (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(1152, 1152, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=1152, bias=False)\n",
       "          (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (2): InvertedResidual(\n",
       "          (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(1152, 1152, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=1152, bias=False)\n",
       "          (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "        (3): InvertedResidual(\n",
       "          (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(1152, 1152, kernel_size=(5, 5), stride=(1, 1), padding=(2, 2), groups=1152, bias=False)\n",
       "          (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(1152, 192, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(192, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "      (6): Sequential(\n",
       "        (0): InvertedResidual(\n",
       "          (conv_pw): Conv2d(192, 1152, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn1): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act1): SiLU(inplace=True)\n",
       "          (conv_dw): Conv2d(1152, 1152, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1), groups=1152, bias=False)\n",
       "          (bn2): BatchNorm2d(1152, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "          (act2): SiLU(inplace=True)\n",
       "          (se): SqueezeExcite(\n",
       "            (conv_reduce): Conv2d(1152, 48, kernel_size=(1, 1), stride=(1, 1))\n",
       "            (act1): SiLU(inplace=True)\n",
       "            (conv_expand): Conv2d(48, 1152, kernel_size=(1, 1), stride=(1, 1))\n",
       "          )\n",
       "          (conv_pwl): Conv2d(1152, 320, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "          (bn3): BatchNorm2d(320, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "        )\n",
       "      )\n",
       "    )\n",
       "    (4): Conv2d(320, 1280, kernel_size=(1, 1), stride=(1, 1), bias=False)\n",
       "    (5): BatchNorm2d(1280, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (6): SiLU(inplace=True)\n",
       "  )\n",
       "  (1): Sequential(\n",
       "    (0): AdaptiveConcatPool2d(\n",
       "      (ap): AdaptiveAvgPool2d(output_size=1)\n",
       "      (mp): AdaptiveMaxPool2d(output_size=1)\n",
       "    )\n",
       "    (1): Flatten(full=False)\n",
       "    (2): BatchNorm1d(5120, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (3): Dropout(p=0.25, inplace=False)\n",
       "    (4): Linear(in_features=5120, out_features=512, bias=False)\n",
       "    (5): ReLU(inplace=True)\n",
       "    (6): BatchNorm1d(512, eps=1e-05, momentum=0.1, affine=True, track_running_stats=True)\n",
       "    (7): Dropout(p=0.5, inplace=False)\n",
       "    (8): Linear(in_features=512, out_features=10, bias=False)\n",
       "  )\n",
       ")"
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "learn2.model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='0' class='' max='50' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      0.00% [0/50 00:00<00:00]\n",
       "    </div>\n",
       "    \n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: left;\">\n",
       "      <th>epoch</th>\n",
       "      <th>train_loss</th>\n",
       "      <th>valid_loss</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>top_k_accuracy</th>\n",
       "      <th>time</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "  </tbody>\n",
       "</table><p>\n",
       "\n",
       "    <div>\n",
       "        <style>\n",
       "            /* Turns off some styling */\n",
       "            progress {\n",
       "                /* gets rid of default border in Firefox and Opera. */\n",
       "                border: none;\n",
       "                /* Needs to be in here for Safari polyfill so background images work as expected. */\n",
       "                background-size: auto;\n",
       "            }\n",
       "            .progress-bar-interrupted, .progress-bar-interrupted::-webkit-progress-bar {\n",
       "                background: #F44336;\n",
       "            }\n",
       "        </style>\n",
       "      <progress value='802' class='' max='1562' style='width:300px; height:20px; vertical-align: middle;'></progress>\n",
       "      51.34% [802/1562 02:02<01:56 2.0890]\n",
       "    </div>\n",
       "    "
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "learn2.fit(n_epoch=50)"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python [conda env:computer_vision]",
   "language": "python",
   "name": "conda-env-computer_vision-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}