mnist_class_demo.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Using device cuda\n",
      "UNET_Classifier(\n",
      "  (0): Conv2d(1, 32, kernel_size=(3, 3), stride=(1, 1), padding=(1, 1))\n",
      "  (1): ReLU()\n",
      "  (2): UNETAttentionBlock(\n",
      "    (groupnorm): GroupNorm(32, 32, eps=1e-05, affine=True)\n",
      "    (conv_input): Conv2d(32, 32, kernel_size=(1, 1), stride=(1, 1))\n",
      "    (layernorm1): LayerNorm((32,), eps=1e-05, elementwise_affine=True)\n",
      "    (self_attention): MultiheadAttention(\n",
      "      (out_proj): NonDynamicallyQuantizableLinear(in_features=32, out_features=32, bias=True)\n",
      "    )\n",
      "    (conv_output): Conv2d(32, 32, kernel_size=(1, 1), stride=(1, 1))\n",
      "  )\n",
      "  (3): MaxPool2d(kernel_size=2, stride=2, padding=0, dilation=1, ceil_mode=False)\n",
      "  (4): Flatten(start_dim=1, end_dim=-1)\n",
      "  (5): Linear(in_features=6272, out_features=128, bias=True)\n",
      "  (6): ReLU()\n",
      "  (7): Linear(in_features=128, out_features=10, bias=True)\n",
      "  (8): LogSoftmax(dim=1)\n",
      ")\n"
     ]
    }
   ],
   "source": [
    "from data_loader import load_classifier_data\n",
    "from classifier_models import CNN_Classifier\n",
    "from classifier_models import UNET_Classifier\n",
    "import torch\n",
    "\n",
    "DEVICE = \"cpu\"\n",
    "\n",
    "ALLOW_CUDA = True\n",
    "ALLOW_MPS = False # Metal API for MacOS\n",
    "\n",
    "if torch.cuda.is_available() and ALLOW_CUDA:\n",
    "    DEVICE = \"cuda\"\n",
    "elif (torch.has_mps or torch.backends.mps.is_available()) and ALLOW_MPS:\n",
    "    DEVICE = \"mps\"\n",
    "print(f\"Using device {DEVICE}\")\n",
    "\n",
    "# load dataset\n",
    "(trainX, trainy), (testX, testy) = load_classifier_data(DEVICE)\n",
    "\n",
    "model = UNET_Classifier().to(DEVICE)\n",
    "model.load_state_dict(torch.load(\"./data/state_dict_model.pt\"), strict=True)\n",
    "model.eval()\n",
    "\n",
    "print(model)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Accuracy: 94.90%\n",
      "Accuracy 0: 96 / 100 = 96.00%\n",
      "Accuracy 1: 94 / 96 = 97.92%\n",
      "Accuracy 2: 123 / 126 = 97.62%\n",
      "Accuracy 3: 90 / 100 = 90.00%\n",
      "Accuracy 4: 88 / 93 = 94.62%\n",
      "Accuracy 5: 87 / 91 = 95.60%\n",
      "Accuracy 6: 87 / 87 = 100.00%\n",
      "Accuracy 7: 90 / 93 = 96.77%\n",
      "Accuracy 8: 98 / 109 = 89.91%\n",
      "Accuracy 9: 96 / 105 = 91.43%\n",
      "Predicted: 3, target: 3 True\n",
      "Predicted: 4, target: 4 True\n",
      "Predicted: 4, target: 4 True\n",
      "Predicted: 2, target: 2 True\n",
      "Predicted: 9, target: 9 True\n",
      "Predicted: 0, target: 0 True\n",
      "Predicted: 6, target: 6 True\n",
      "Predicted: 1, target: 1 True\n",
      "Predicted: 2, target: 2 True\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAfoAAAGgCAYAAABCAKXYAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjguNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8fJSN1AAAACXBIWXMAAA9hAAAPYQGoP6dpAAA6TUlEQVR4nO3dfXQUVZ7/8W+CpHlKOgRMQhZa4g4KioNrTEIAGZGsiMqIoK677qDjA4IJOxHn5wyKqBw0R2ZmZUUQd+RBdkQcdAHBEWcMTwcNMMRBjEDElZGskBb0pDs8JIGkfn/MoTf3Ap2u9FNV9ft1Tp3Tn67q7pvuL9yuvlW3kgzDMAQAADhScrwbAAAAooeOHgAAB6OjBwDAwejoAQBwMDp6AAAcjI4eAAAHo6MHAMDB6OgBAHAwOnoAAByMjh4AAAeLWke/YMEC6d+/v3Tp0kUKCwtl586d0XopIKKoXdgVtYvzSYrGXPdvvfWWTJo0SRYtWiSFhYUyb948WbVqldTU1EhmZmbQx7a2tsrhw4clNTVVkpKSIt00RIFhGNLQ0CA5OTmSnGzvH4mo3cRC7f4NtWs/pmrXiIKCggKjpKQkkFtaWoycnByjvLy83cfW1tYaIsJiw6W2tjYa5RRT1G5iLtQutWvXJZTajfhX2ObmZqmqqpLi4uLAfcnJyVJcXCyVlZXnbN/U1CR+vz+wGFxMz7ZSU1Pj3YSwULuJi9qldu0qlNqNeEd/7NgxaWlpkaysLOX+rKwsqaurO2f78vJycbvdgcXj8US6SYgRu//kR+0mLmqX2rWrUGo37oNSM2bMEJ/PF1hqa2vj3SQgJNQu7IraTSwXRfoJe/fuLZ06dRKv16vc7/V6JTs7+5ztXS6XuFyuSDcDMI3ahV1Ruwgm4nv0KSkpkpeXJxUVFYH7WltbpaKiQoqKiiL9ckDEULuwK2oXQXX8GM8LW7lypeFyuYxly5YZe/fuNSZPnmykp6cbdXV17T7W5/PF/ShGlo4tPp8vGuUUU9RuYi7ULrVr1yWU2o1KR28YhjF//nzD4/EYKSkpRkFBgbF9+/aQHkfB2Xdxwn+WhkHtJuJC7VK7dl1Cqd2oTJgTDr/fL263O97NQAf4fD5JS0uLdzPihtq1L2qX2rWrUGo37kfdAwCA6KGjBwDAwejoAQBwMDp6AAAcjI4eAAAHo6MHAMDBIj4FbqL5wx/+oOSxY8cq+YUXXgjcXrRokbLur3/9a9TaBQCACHv0AAA4Gh09AAAOxk/3YfL7/UrWJxp8/PHHA7cffvhhZV1VVZWS586dq+S2F6gQEWlpaelwO5F49GuM/+Uvf1FyRkaGkrt166bkU6dORadhAGKKPXoAAByMjh4AAAejowcAwMEYow/TrFmzlHzttdcq+dJLLw3c1q8OdcMNNwTNf/zjH5X8/PPPK3nr1q3mGouEoo+561k/nkS/AhZj9LCrrl27Bm4/9thjyrqZM2cq2eVyKXnv3r1KHjVqlJK//fbbSDQxptijBwDAwejoAQBwMDp6AAAcLMnQB+rizO/3nzOWbWd33nln4PakSZOUdbfccoup52pqalLy8OHDlfzJJ5+YbF1k+Xy+c8Z5E4nVard3795K1uujb9++Sh49erSSN23aFJ2GicjIkSOVXFNTo2Sv1xu11z4fatdatWtW586dlVxaWhq4/etf/zqs596+fbuS9f934y2U2mWPHgAAB6OjBwDAwejoAQBwMMboY6hTp05Kzs/PV/Irr7yi5CFDhgR9vt/97ndK1o8BiDXGOa1Vu7feequS33333aDbd+/eXcmRPI8+KytLyV988YWSf/vb3yr55z//ecReOxTUrrVq16zi4mIlf/DBBxfc9vjx40qeNm2akvVrjujXhFi1apWS77nnnpDbGQ2M0QMAkODo6AEAcDA6egAAHIy57uPowIEDSv7666+V3N4YvT5HM9DWxIkTg67fvHmzkpubm6PWljFjxig5NTVVyfX19VF7bTiPXk///d//fcFtfT6fkm+++WYlf/bZZ0r+5S9/qeSLL75Yyfo5+3bAHj0AAA5GRw8AgIOZ7ui3bt0q48aNk5ycHElKSpI1a9Yo6w3DkFmzZkmfPn2ka9euUlxcfM5P1EA8ULuwK2oX4TA9Rn/ixAkZMmSI3H///TJhwoRz1s+dO1deeuklef311yU3N1eeeuopGTNmjOzdu1e6dOkSkUbb1XPPPafkxx9/3NTj9bnu33jjjbDblEgSrXbbmyIjJydHycnJ6vf+lpaWDr92UlKSks/3fre1bt26Dr9WIki02tXl5eUp+cUXX1Sy/jfu378/cFs/VqXtOhGRESNGKPnyyy8P2pY333wzeGMtyHRHP3bsWBk7dux51xmGIfPmzZOZM2fKbbfdJiIiy5cvl6ysLFmzZo3cfffd5zymqalJ6cD8fr/ZJgEhoXZhV9QuwhHRMfqDBw9KXV2dMkuR2+2WwsJCqaysPO9jysvLxe12B5Z+/fpFsklASKhd2BW1i/ZEtKOvq6sTkXOnu8zKygqs082YMUN8Pl9gqa2tjWSTgJBQu7Arahftift59C6Xy9Lng/fq1UvJAwYMULJ+rvvVV199wecyOyfyiRMnlDxr1iwltzd3OaLL6rXb3tzln376qZJPnz4dsdfu2rWrkn/84x9H7LkRPqvXbu/evZWsz/nQrVs3Jevz1z/88MOB2/qYfCKK6B59dna2iIh4vV7lfq/XG1gHWBG1C7uidtGeiHb0ubm5kp2dLRUVFYH7/H6/7NixQ4qKiiL5UkBEUbuwK2oX7TH90/3x48flyy+/DOSDBw/K7t27JSMjQzwej5SVlcmcOXNkwIABgdM8cnJyZPz48ZFsN2AatQu7onYRDtMd/a5du2TUqFGBPH36dBERuffee2XZsmXy+OOPy4kTJ2Ty5MlSX18vI0aMkA0bNtjmXM7CwkIlv/TSS0rWryEfTW2/oYuILFiwIGav7UROr13dj370o3g34YL040/0DJXTa1c/T769MXndzJkzlbxt27aItMspTHf0119/fdCJOJKSkmT27Nkye/bssBoGRBq1C7uidhEO5roHAMDB6OgBAHCwuJ9HbzXDhg1TcizH5HX6ucf/8R//oeSpU6fGsjlwGP063JF0++23B13f2tqq5Pbm5Yez6HM8lJeXK7m9MflXX31VyYsWLYpMw87j2LFjSt66dWvUXita2KMHAMDB6OgBAHAwfrrXvPLKK0pOS0tT8kUXqW9Z23Nbz6ftXNMffPCBsu7KK69U8saNG5V88cUXK7nttI4iIuvXr1fye++9F7QtQFt/+ctfovbcqampQdf/z//8T9AMZ9EvW/z8888refTo0UEfr/80X1ZWpuRwpm9+4IEHgq7XLw/+3Xffdfi14oU9egAAHIyOHgAAB6OjBwDAwRij1zQ2Nir52Wefjdprff7550qeN2+ekp977rmgj584caKSGaNHW2+//baS9WM8+vTpE7HX0q+Sps/Qpo/R/uM//mPEXhvWd//99yt5ypQpQbf/+OOPlVxSUqLkK664QsnDhw9XclZWVuD2ZZddFvS1rrvuuqDre/bsqeR//ud/VvLatWuVfPLkyaDPFw/s0QMA4GB09AAAOBgdPQAADpZkWGzuSb/ff870iIlCHws6cOCAkjMyMpSsn8/Zt29fJevnf0abz+c7Z96BRGK12r366quV/Mknnyh53759Sv7Zz36m5A8//PCCz61PUaqPyZ+9jOqF9O7dW8nff/990O2jjdqNbO3q/xdVVVUpWf/8v/rqKyXr59mPHDlSyRMmTFByjx49OtTOSKiurlbyn/70JyW///77StYvPx6uUGqXPXoAAByMjh4AAAejowcAwMEYo7ew3/zmN0p+9NFHg26vny/a3jz8kcY4p7Vqt3v37kr+85//rOSBAwcquaGhQclvvvlm4Pbu3buVddOmTVPyoEGDTLWNMXpriUTttv1Mt2zZoqzTa82sM2fOKPk///M/lez1epXcdpy8pqYm6HMvXrxYyePHj1ey/u/i17/+tZL/5V/+Rcn9+vUL+npXXXVV4PZf//rXoNuGgjF6AAASHB09AAAORkcPAICDMde9heXn55va3krjw4i/EydOKFkfe9Sv6a2f+3zPPfcEbk+ePFlZp19Xoa6uTsmjRo1Ssn4e9fHjx8/faNjW4MGDA7fNjsnrc4K88cYbStbPRf/jH/9osnUX1t617P1+v5LnzJkTNOvzV+jj55EYlzeLPXoAAByMjh4AAAejowcAwMEcP0a/fPlyJbc9h1FEZOrUqUrevn171Nt0Ifo1mocOHRp0+y+++ELJe/bsiXib4Bx6vTzyyCNBtx8xYkTgtn6dhXfffVfJ+hwO+/fvV/KRI0eU3NzcHLyxsB0zcymsW7dOyfrxIvEYx76QlStXmtpen3PCCtijBwDAwejoAQBwMFMdfXl5ueTn50tqaqpkZmbK+PHjz5lesLGxUUpKSqRXr17So0cPmThx4jnTEwKxRu3CrqhdhMvUGP2WLVukpKRE8vPz5cyZM/LEE0/IjTfeKHv37g3Mq/3oo4/Ke++9J6tWrRK32y2lpaUyYcIE+eijj6LyB7TnBz/4gZKHDBmi5Llz5ypZn19+7dq10WmYiNx4441Kfuqpp5R80UXBPx69be2dD5rI7Fi78bZt27aIPZd+PXF9zgefzxex13Iau9TuW2+9Fbitnxd/6NAhJe/atUvJ+lz2VqL/LXZkqqPfsGGDkpctWyaZmZlSVVUlI0eOFJ/PJ4sXL5YVK1bIDTfcICIiS5culUGDBsn27dvPe3BZU1OTNDU1BbI+OQEQCdQu7IraRbjCGqM/+y387BG5VVVVcvr0aSkuLg5sM3DgQPF4PFJZWXne5ygvLxe32x1Y2rvyDxAJ1C7sitqFWR3u6FtbW6WsrEyGDx8emPqwrq5OUlJSJD09Xdk2KyvrnCkyz5oxY4b4fL7AUltb29EmASGhdmFX1C46osPn0ZeUlEh1dXXY43gul0tcLldYzxGMfq5w2+sUi6jnCouIDB8+XMmNjY1KPnr0qJL1OZiD+Yd/+AclX3vttUpOTg7+vevgwYNKXrhwYcivjf9jl9p1Ev2a3i0tLXFqib1ZuXa///77wO3f//73EX3ueNKv62BHHdqjLy0tlfXr18umTZuUC2FkZ2dLc3Oz1NfXK9t7vV7Jzs4Oq6FAJFC7sCtqFx1lqqM3DENKS0tl9erVsnHjRsnNzVXW5+XlSefOnaWioiJwX01NjRw6dEiKiooi02KgA6hd2BW1i3CZ+um+pKREVqxYIWvXrpXU1NTA+I/b7ZauXbuK2+2WBx54QKZPny4ZGRmSlpYm06ZNk6KionancwWiidqFXVG7CFeSYRhGyBsnJZ33/qVLl8p9990nIn8b037sscfkzTfflKamJhkzZowsXLgw5J+Q/H5/VK+r3qdPHyUvWrRIydddd52S9QNcounUqVNKXrFihZKfeOIJJevHC8Sbz+c759rLVuGE2rWy9ua618/nPnsa2FnxngOC2k3c2tXnsr/zzjuVnJqaquSTJ09GvU1mhFK7pvboQ/lO0KVLF1mwYIEsWLDAzFMDUUXtwq6oXYSLue4BAHAwOnoAABzM8dej1+nXxb7tttuU7PF4lPz3f//3Sr7nnnuUfMcddyg52FjJJ598ouR33nlHyfr5mlxfHnbR9hxqEZEvvvhCyddcc42Sz87RfpZ+ahiAyGGPHgAAB6OjBwDAwRLup/v26JdT1POmTZuU/OCDD0a9TYDVHTt2TMnTpk1T8htvvKHk1tbWqLcJ6IgPP/xQyW2v8mdX7NEDAOBgdPQAADgYHT0AAA7GGD2AiNMvB52ZmRmnlgDm1NbWKtkJl1Rmjx4AAAejowcAwMHo6AEAcDDG6AEACevuu++OdxOijj16AAAcjI4eAAAHo6MHAMDB6OgBAHAwOnoAAByMjh4AAAezXEdvGEa8m4AOSvTPLtH/fjtL9M8u0f9+Owvls7NcR9/Q0BDvJqCDEv2zS/S/384S/bNL9L/fzkL57JIMi32Va21tlcOHD4thGOLxeKS2tlbS0tLi3Szb8Pv90q9fv5i+b4ZhSENDg+Tk5EhysuW+O8YMtRseajd+qN3wWL12LTczXnJysvTt21f8fr+IiKSlpVFwHRDr983tdsfstayK2o0Majf2qN3IsGrtJu5XWAAAEgAdPQAADmbZjt7lcsnTTz8tLpcr3k2xFd63+OMz6Bjet/jjM+gYq79vljsYDwAARI5l9+gBAED46OgBAHAwOnoAAByMjh4AAAejowcAwMEs29EvWLBA+vfvL126dJHCwkLZuXNnvJtkGeXl5ZKfny+pqamSmZkp48ePl5qaGmWbxsZGKSkpkV69ekmPHj1k4sSJ4vV649TixELtXhi1a23U7oXZunYNC1q5cqWRkpJiLFmyxPj888+Nhx56yEhPTze8Xm+8m2YJY8aMMZYuXWpUV1cbu3fvNm6++WbD4/EYx48fD2wzZcoUo1+/fkZFRYWxa9cuY+jQocawYcPi2OrEQO0GR+1aF7UbnJ1r15IdfUFBgVFSUhLILS0tRk5OjlFeXh7HVlnXt99+a4iIsWXLFsMwDKO+vt7o3LmzsWrVqsA2+/btM0TEqKysjFczEwK1aw61ax3Urjl2ql3L/XTf3NwsVVVVUlxcHLgvOTlZiouLpbKyMo4tsy6fzyciIhkZGSIiUlVVJadPn1bew4EDB4rH4+E9jCJq1zxq1xqoXfPsVLuW6+iPHTsmLS0tkpWVpdyflZUldXV1cWqVdbW2tkpZWZkMHz5cBg8eLCIidXV1kpKSIunp6cq2vIfRRe2aQ+1aB7Vrjt1q13KXqYU5JSUlUl1dLdu2bYt3UwBTqF3Yld1q13J79L1795ZOnTqdc6Si1+uV7OzsOLXKmkpLS2X9+vWyadMm6du3b+D+7OxsaW5ulvr6emV73sPoonZDR+1aC7UbOjvWruU6+pSUFMnLy5OKiorAfa2trVJRUSFFRUVxbJl1GIYhpaWlsnr1atm4caPk5uYq6/Py8qRz587Ke1hTUyOHDh3iPYwiard91K41Ubvts3XtRusov5dfftm45JJLDJfLZRQUFBg7duwI+bErV640XC6XsWzZMmPv3r3G5MmTjfT0dKOuri5azbWVqVOnGm6329i8ebNx5MiRwHLy5MnANlOmTDE8Ho+xceNGY9euXUZRUZFRVFQUx1bbB7UbPdRudFG70WPn2o3KZWrfeustmTRpkixatEgKCwtl3rx5smrVKqmpqZHMzMygj21tbZXDhw/LihUrZP78+eL1euWHP/yhzJ07V6699tpIN9WW3G73ee9fuHCh3HPPPSLyt4kbnnzySXn77belqalJRo8eLf/+7/9+zsE2kWAYhjQ0NEhOTo4kJ1vuRyJTqN3oonajh9qNLlvXbjS+PYRzPmZtba0hIiw2XGpra6NRTjFF7SbmQu1Su3ZdQqndiH+FNXs+ZlNTk/j9/sBiRP4HBsRIampqvJsQFmo3cVG71K5dhVK7Ee/ozZ6PWV5eLm63O7B4PJ5INwkxkpSUFO8mhIXaTVzULrVrV6HUbtwHpWbMmCE+ny+w1NbWxrtJQEioXdgVtZtYIj5hjtnzMV0ul7hcrkg3AzCN2oVdUbsIJuJ79JyPCbuidmFX1C6C6vgxnhcWzvmYPp8v7kcxsnRs8fl80SinmKJ2E3Ohdqlduy6h1G7UJsyZP3++4fF4jJSUFKOgoMDYvn17SI+j4Oy7OOE/S8OgdhNxoXapXbsuodRuVCbMCYff77/gxASwNp/PJ2lpafFuRtxQu/ZF7VK7dhVK7cb9qHsAABA9dPQAADgYHT0AAA5GRw8AgIPR0QMA4GB09AAAOBgdPQAADhbxue4RPZmZmUp+4YUXlHzvvfcGffzo0aOVvGnTpsg0DAAsqr2pYkaNGqXkzZs3R7E18cEePQAADkZHDwCAg/HTvYU9+eSTSp4yZYqSc3JylKz/RPX1118r+bPPPotg6wDAmp555pmQtzU7hKn/tP/ss88GXW8F7NEDAOBgdPQAADgYHT0AAA7GZWrDlJGRoeQJEyYoeciQIYHb119/vbKuvbf+iiuuUHJSUlLQ7Xfv3q3k2bNnK3nt2rVBHx8uLvVprdr9u7/7OyWXlZUpefr06Uo+ePCgktetWxe47ff7lXX6qZ0nT57saDMtgdq1Vu2GK57dWqxP1+MytQAAJDg6egAAHIyOHgAAB+M8+nYMHTpUye+++66Su3fvruQuXbpc8Ln0MfZwx5EWLVqk5CeeeELJPp8vrOeHtaWkpCh5zpw5Sr7//vuVnJ6ermS9/vr376/kadOmBW7rtTto0CAlz5o1S8n79+8/f6OBKAhnOu/2xtD1Y6vMtqW9Y6tigT16AAAcjI4eAAAHo6MHAMDBGKPXjBgxQsmrV69Wsn7efDjq6uqUnJ2dberxy5cvVzJj8omlZ8+eStbPizfr/fffV3LbMf1hw4Yp6/T5IvRLKBcXFyv5zJkzYbUNaEsfNzczjq6PyevnvevCPZZKn3ffzDz8kcIePQAADkZHDwCAg9HRAwDgYIzRax555BElhzsm33a+8PLycmWdPkb/4IMPKlk/L1531113KXnHjh0daSJsaurUqaa21+eA+Nd//VclNzU1KTk5+f/2A9577z1l3Q033KBk/dgWvXb16y4AZuhj8OGcN9/emLxOH9M3e169FbBHDwCAg5nu6Ldu3Srjxo2TnJwcSUpKkjVr1ijrDcOQWbNmSZ8+faRr165SXFwsBw4ciFR7gQ6jdmFX1C7CYbqjP3HihAwZMkQWLFhw3vVz586Vl156SRYtWiQ7duyQ7t27y5gxY6SxsTHsxgLhoHZhV9QuwmF6jH7s2LEyduzY864zDEPmzZsnM2fOlNtuu01E/naud1ZWlqxZs0buvvvu8FobA1OmTFFyZWWlkvXzhXX6efeffvpp4HZLS0vQxz7//PNK1ucm148feOihh4I+/rvvvgv6eonGabXbqVMnU9vr5++2dw35tvWqnze/ceNGJV9zzTVKbjtPvojIwoULlXzs2LGgrw2V02q3Pfo4+NNPP93h5zI7Jq/bsmWLkhN+jP7gwYNSV1enTJbhdrulsLDwnA7zrKamJvH7/coCxBq1C7uidtGeiHb0Z48iz8rKUu7Pyso65wjzs8rLy8XtdgeWfv36RbJJQEioXdgVtYv2xP2o+xkzZojP5wsstbW18W4SEBJqF3ZF7SaWiJ5Hf3audq/XK3369Anc7/V65eqrrz7vY1wul7hcrkg2Iyz6T1jz58+P2Wvn5+cr+eGHHw66fdtz9EUYkw+HHWvX7FHVd955p5L37NkT8mOPHz+uZP0c/PXr1yv50ksvVbJ+fAnn1UeOHWtXF8nz5EXUcfn2rjffnh/96EdhPd4KIrpHn5ubK9nZ2VJRURG4z+/3y44dO6SoqCiSLwVEFLULu6J20R7Te/THjx+XL7/8MpAPHjwou3fvloyMDPF4PFJWViZz5syRAQMGSG5urjz11FOSk5Mj48ePj2S7AdOoXdgVtYtwmO7od+3apfwscvbSmPfee68sW7ZMHn/8cTlx4oRMnjxZ6uvrZcSIEbJhwwbp0qVL5FoNdAC1C7uidhGOJCPci+1GmN/vF7fbHe9mxIU+1/2rr74adPtf/vKXSv7Vr34V8TaZ4fP5JC0tLa5tiKdY165+lPU333wTdPsPP/xQyePGjVPy6dOnO9wW/bz5F198Ucn6dRiGDx/e4deKBmo3trUbzTF5kfDH5dvS558we07/s88+G/T5whVK7cb9qHsAABA9dPQAADgYHT0AAA7G9ejjyOPxKPmxxx4Luv3u3buVvGzZsgi3CHaiz/nw5z//Wcn6vAxtp0gV+dukKW29/vrrSv76669DbsupU6eCrtfPq9fzV199FfJrwf7CnS9eH/eO5Ji8E7FHDwCAg9HRAwDgYPx0H0P66SvPPfecki+//PKgj9dPUTp69GhkGgZb0n8u/81vfqPklStXBn38rFmzlKxPU/vZZ58Fbrc3FfSkSZOCrr/44ouDZn66d7ZwLzsb7VPUnI49egAAHIyOHgAAB6OjBwDAwRijj6Frr71WyfopRvpsxPqUpGfOnAn6/J06dVKyPkXq4cOHQ2on7Em/bPGTTz6p5JkzZyq5a9euSu7du7eS204resMNNyjrLDZzNizO7Ji8jtPnwsMePQAADkZHDwCAg9HRAwDgYIzRm6SPsy9ZsiTkxw4ePFjJ7Y1zJier38P0c1H37Nmj5KSkJCXPmzdPyYsXLw6hlbCrpqYmJb/wwgtK1i8Fessttyj5nnvuUXLbeR/08fvW1lZTbdPngNAznM3slLdWmuI23OMLrIA9egAAHIyOHgAAB6OjBwDAwRJ+jF4fB7/qqquUfNdddylZnw88LS0t5NfSx9DbG6PXz4u/8sorg26vP19jY2PIbYPz7dy5M2jWxyLbzvNQVFSkrNPn1dfH8HWcd59Ywp2LPp5z2Yd7CV2dFeblZ48eAAAHo6MHAMDB6OgBAHCwhB+jf+ONN5Ssj8lHkj5O6ff7lbxv376wnl+/Prn+twFmtL1GvH69+CuuuELJv/jFL2LSJiDS9DF5fb4Js/Q5AKyAPXoAAByMjh4AAAejowcAwMESboz+tttuU/Idd9xh6vH6OZGTJk1Ssn6N+bb0MXn9tSsqKky1BYiX999/X8k///nPlazPAQHEiz4Gr88XYfa8eX3efSvNy38h7NEDAOBgpjr68vJyyc/Pl9TUVMnMzJTx48dLTU2Nsk1jY6OUlJRIr169pEePHjJx4kTxer0RbTRgFrULu6J2ES5THf2WLVukpKREtm/fLn/605/k9OnTcuONN8qJEycC2zz66KOybt06WbVqlWzZskUOHz4sEyZMiHjDATOoXdgVtYtwJRlhTEJ99OhRyczMlC1btsjIkSPF5/PJxRdfLCtWrAiMP+/fv18GDRoklZWVMnTo0Haf0+/3K9fBDtfIkSOVvG7dOiX36NEj6PrVq1crefbs2UrW5/ju0qWLks+cORO4PW7cOGXdli1blKxfT9xufD6fqbn/48kOtWsnn376qZL16zJ89913Sh4+fLiSv/zyy+g0LETUbmRrVz+Wyew13cM9Fz2a15AfNWqUkuM9Jh9K7YY1Ru/z+UREJCMjQ0REqqqq5PTp01JcXBzYZuDAgeLxeKSysvK8z9HU1CR+v19ZgGijdmFX1C7M6nBH39raKmVlZTJ8+HAZPHiwiIjU1dVJSkqKpKenK9tmZWVJXV3deZ+nvLxc3G53YOnXr19HmwSEhNqFXVG76IgOd/QlJSVSXV0tK1euDKsBM2bMEJ/PF1hqa2vDej6gPdQu7IraRUd06Dz60tJSWb9+vWzdulX69u0buD87O1uam5ulvr5e+Xbp9XolOzv7vM/lcrnE5XJ1pBkh0ce9O3fuHHR7/Tz4JUuWmHo9fayy7fXrt2/fbuq5EHl2ql07+f3vf69kfYy1V69eStavbx/vMXo7sFPt6uPWZsfMoznGbpYdzpNvj6k9esMwpLS0VFavXi0bN26U3NxcZX1eXp507txZmfilpqZGDh06dM4/bCCWqF3YFbWLcJnaoy8pKZEVK1bI2rVrJTU1NTD+43a7pWvXruJ2u+WBBx6Q6dOnS0ZGhqSlpcm0adOkqKgopCM/gWihdmFX1C7CZaqjf+WVV0Tk3CkDly5dKvfdd5+IiLz44ouSnJwsEydOlKamJhkzZowsXLgwIo0FOorahV1RuwhXWOfRR0Okz+e8+OKLlbxz504lezweU8/X0tKi5OXLlyv5pZdeUvKePXtMPb+d2elc5GhI5PPo9eNPrr322qDb68fOzJkzR8nl5eWRaViIqN3o1q5+jXez88tHkx3nrm8r6ufRAwAAa6OjBwDAwejoAQBwMMdfj75nz55KPnDggJL1MXp9Jin9/OBXX31Vyfv37w+3iYDtffzxx0pub4xeP4e7W7duEW8TrEOfHz7aY/Ztx9X1a4ro8/AnAvboAQBwMDp6AAAczPE/3X/xxRdKvvHGG+PUEsC59u7da2r7b775Rsk1NTWRbA4sTv8pH9HFHj0AAA5GRw8AgIPR0QMA4GCOH6MHEH1vvfVW0PX6ZUdvueUWJVdXV0e8TQD+hj16AAAcjI4eAAAHo6MHAMDBHH+ZWsQOl/qkdu2K2qV27YrL1AIAkODo6AEAcDA6egAAHIyOHgAAB6OjBwDAwejoAQBwMMt19BY72w8mJPpnl+h/v50l+meX6H+/nYXy2Vmuo29oaIh3E9BBif7ZJfrfb2eJ/tkl+t9vZ6F8dpabMKe1tVUOHz4shmGIx+OR2trahJ7Iwiy/3y/9+vWL6ftmGIY0NDRITk6OJCdb7rtjzFC74aF244faDY/Va9dyV69LTk6Wvn37it/vFxGRtLQ0Cq4DYv2+MasWtRsp1G7sUbuRYdXaTdyvsAAAJAA6egAAHMyyHb3L5ZKnn35aXC5XvJtiK7xv8cdn0DG8b/HHZ9AxVn/fLHcwHgAAiBzL7tEDAIDw0dEDAOBgdPQAADgYHT0AAA5m2Y5+wYIF0r9/f+nSpYsUFhbKzp07490kyygvL5f8/HxJTU2VzMxMGT9+vNTU1CjbNDY2SklJifTq1Ut69OghEydOFK/XG6cWJxZq98KoXWujdi/M1rVrWNDKlSuNlJQUY8mSJcbnn39uPPTQQ0Z6errh9Xrj3TRLGDNmjLF06VKjurra2L17t3HzzTcbHo/HOH78eGCbKVOmGP369TMqKiqMXbt2GUOHDjWGDRsWx1YnBmo3OGrXuqjd4Oxcu5bs6AsKCoySkpJAbmlpMXJycozy8vI4tsq6vv32W0NEjC1bthiGYRj19fVG586djVWrVgW22bdvnyEiRmVlZbyamRCoXXOoXeugds2xU+1a7qf75uZmqaqqkuLi4sB9ycnJUlxcLJWVlXFsmXX5fD4REcnIyBARkaqqKjl9+rTyHg4cOFA8Hg/vYRRRu+ZRu9ZA7Zpnp9q1XEd/7NgxaWlpkaysLOX+rKwsqauri1OrrKu1tVXKyspk+PDhMnjwYBERqaurk5SUFElPT1e25T2MLmrXHGrXOqhdc+xWu5a7eh3MKSkpkerqatm2bVu8mwKYQu3CruxWu5bbo+/du7d06tTpnCMVvV6vZGdnx6lV1lRaWirr16+XTZs2Sd++fQP3Z2dnS3Nzs9TX1yvb8x5GF7UbOmrXWqjd0Nmxdi3X0aekpEheXp5UVFQE7mttbZWKigopKiqKY8uswzAMKS0tldWrV8vGjRslNzdXWZ+XlyedO3dW3sOamho5dOgQ72EUUbvto3atidptn61rN66HAl7AypUrDZfLZSxbtszYu3evMXnyZCM9Pd2oq6uLd9MsYerUqYbb7TY2b95sHDlyJLCcPHkysM2UKVMMj8djbNy40di1a5dRVFRkFBUVxbHViYHaDY7atS5qNzg7127UOvqXX37ZuOSSSwyXy2UUFBQYO3bsMPX4+fPnGx6Px0hJSTEKCgqM7du3R6ml9iMi512WLl0a2ObUqVPGI488YvTs2dPo1q2bcfvttxtHjhyJX6NthNqNHmo3uqjd6LFz7UblMrVvvfWWTJo0SRYtWiSFhYUyb948WbVqldTU1EhmZmbQx7a2tsrhw4clNTVVkpKSIt00RIFhGNLQ0CA5OTmSnGy50SBTqN3EQu3+DbVrP6ZqNxrfHsKZeKG2tvaC35xYrL3U1tZGo5xiitpNzIXapXbtuoRSuxH/Cmt24oWmpibx+/2BxYj8DwyIkdTU1Hg3ISzUbuKidqlduwqldiPe0ZudeKG8vFzcbndg8Xg8kW4SYsTuP/lRu4mL2qV27SqU2o37oNSMGTPE5/MFltra2ng3CQgJtQu7onYTS8RnxjM78YLL5RKXyxXpZgCmUbuwK2oXwUR8j56JF2BX1C7sitpFUB0/xvPCwpl4wefzxf0oRpaOLT6fLxrlFFPUbmIu1C61a9cllNqNykVt/umf/kmOHj0qs2bNkrq6Orn66qtlw4YN5xwoYkdXX321kjdu3Kjkb775RsmjR48O3P7222+j1i5EhpNrF85G7eJCojJhTjj8fr+43e54N+OC6OgvzOfzSVpaWrybETdWr11cGLVL7dpVKLUb96PuAQBA9NDRAwDgYFEZo3eyf/u3f1Oy/nOXnhcuXBi4fccdd0SvYQDgUK+99pqSBwwYoOQ9e/Zc8LFNTU1Knj17tpL9fn+YrbM+9ugBAHAwOnoAAByMjh4AAAdjjL4dixcvVvKkSZNMPf7HP/5x4LZ+at7u3bs72iwg6l5//fXA7VmzZinrvv7661g3Bwns0ksvVfJ1110XNAdz4403KvnIkSNKPnnypJKXLFmi5K1btyrZ5/OF/Nrxwh49AAAORkcPAICDMTNeO44eParkjIyMDj9XQUGBkquqqjr8XFbE7GLWqt1w1dTUBG7rM0BOnTo11s2JKmrX2rWrX4Fv+PDhStYv3NP29LubbrpJWXfRReqItd4FJicH3/996KGHlLxs2TIlt7S0BH18pDEzHgAACY6OHgAAB6OjBwDAwTi9TqOPv6Snp4f1fB988EHgdrBpGgGrefvttwO377//fmWd08boYW11dXVKfuedd4LmYEaNGqXk5uZmJefn5yt52rRpSv7tb3+r5N69eyv5hRdeCLktscIePQAADkZHDwCAg9HRAwDgYIzRa/TzMds7p7I9O3fuDNw+ffp0WM8FxNKOHTsCt/UxesCuNm3aFHT9Rx99pORdu3YpWZ8C1w7YowcAwMHo6AEAcDA6egAAHCzhx+j18+Z/8pOfhPV8N998s5I3b94c1vMB8dLQ0BC43alTJ2Vd165dlXzq1KmYtAmINn3O/7lz58apJZHDHj0AAA5GRw8AgIPR0QMA4GAJP0bfs2dPJZs9b/5///d/lVxdXa3kpqamjjUMiLO25xvr83lfddVVSm47XwRgJ7fddpuSn3nmGSUPGTIk6ON/+MMfRrpJEccePQAADkZHDwCAg5nu6Ldu3Srjxo2TnJwcSUpKkjVr1ijrDcOQWbNmSZ8+faRr165SXFwsBw4ciFR7gQ6jdmFX1C7CYXqM/sSJEzJkyBC5//77ZcKECeesnzt3rrz00kvy+uuvS25urjz11FMyZswY2bt3r3Tp0iUijY6kH/zgB2E9fvHixUr+5ptvwnq+YFJTU5XcvXt3JevXbB48eLCS/X6/kg8dOhTB1lmf02oXiYPajZzx48crefny5Uru0aOHqefbs2dPuE2KOtMd/dixY2Xs2LHnXWcYhsybN09mzpwZOMBh+fLlkpWVJWvWrJG77777nMc0NTUpB6zpnREQKdQu7IraRTgiOkZ/8OBBqaurk+Li4sB9brdbCgsLpbKy8ryPKS8vF7fbHVj69esXySYBIaF2YVfULtoT0Y7+7E/HWVlZyv1ZWVnn/Kx81owZM8Tn8wWW2traSDYJCAm1C7uidtGeuJ9H73K5xOVyxez1fvrTnyo53Ots6+M7ZuTl5SlZPzf5gQceULI+Jq/PN37s2DEl68cfHD9+XMltx+j19yXRxu87Ita1C0RKItVuenq6kvXz5Nsbk29paVHy9OnTlawfp2VFEd2jz87OFhERr9er3O/1egPrACuidmFX1C7aE9GOPjc3V7Kzs6WioiJwn9/vlx07dkhRUVEkXwqIKGoXdkXtoj2mf7o/fvy4fPnll4F88OBB2b17t2RkZIjH45GysjKZM2eODBgwIHCaR05OzjmnNACxRu3CrqhdhMN0R79r1y4ZNWpUIJ8dr7j33ntl2bJl8vjjj8uJEydk8uTJUl9fLyNGjJANGzZY5lzOmTNnKjkpKcnU4//whz8o+UIHu5zVq1evwO158+Yp62699VYlp6WlmWqL7rLLLgu6PjMzU8mXXnpp4Pb/+3//T1mnX4PZCQfr2L12reTyyy9XMnPdRxe123Hr1q1Tsj43/XfffafkZ599VsmGYSh5wYIFEWxdbJju6K+//vpz/vC2kpKSZPbs2TJ79uywGgZEGrULu6J2EQ7mugcAwMHo6AEAcLC4n0cfa7m5uUoO9nPY+ehj8o2NjUrOyMhQctvz7G+66SZTrxVLjzzyiJL1v/O5556LZXNgcadOnYp3E4AA/f/dn/3sZ4HbhYWFQR/7/vvvK/nll1+OXMMsgj16AAAcjI4eAAAHS7if7vXT6cz+dL9ixYqg6/XpEKP5c70+Te3JkyeVPHDgwKi9NpzvkksuCdzW/920vfIZEGtjxoxRsn62QX5+/gUf29DQoORt27ZFrmEWxR49AAAORkcPAICD0dEDAOBgCTdGb3ZMXrd///6g682Mi+tj7DU1NUp+7bXXgj5++/btStanu3zzzTeVfM0114TcNmDIkCGB2/q/G/1KaUA0DR48WMn/9V//peTevXuH/FwlJSVK/t3vftfxhtkEe/QAADgYHT0AAA5GRw8AgIMl3Bh9uO677z4l6+PgnTt3Dvm5SktLlfzee+8p2eVyKblPnz5K/slPfqLkBx98UMn9+/cPuS1AMEePHlXyp59+GqeWwIk6deqk5LOX4T3rqaeeUnKPHj1Cfu7Ro0crORFrlz16AAAcjI4eAAAHo6MHAMDBEm6MPty57ufMmRM0m7Fo0SIlf/DBB0rWzw0dN25ch18LMKvtvAutra3KOua6Rzi6du2q5LaX8xYRmThxYtDH6/9v6+Pud955Z+D2V199FfSxiYA9egAAHIyOHgAAB6OjBwDAwRJujH7lypVKvuuuu+LUEpGcnBwl//SnP43p67cdq9LnA9iwYUNM2wLrGTBgQLybAIe47LLLlKxfP/6OO+5Qcnvj6PoxIi+88IKSf/WrXwVuL168WFl38uTJoM/drVs3Jfft21fJK1asULJ+ffuePXsqubm5WcknTpwI+vrRwB49AAAORkcPAICD0dEDAOBgSYbFTir0+/3idrtj9no+n0/JZuZQthv9o257HWZ9Dv+O8Pl8kpaWFvbz2FWsazfajhw5Erh97NgxZd1VV10V6+ZEFbUb3dqdP3++kvVrwtuJfl7+qVOnlKz3Ia+//rqS2x6foM9P0RGh1C579AAAOBgdPQAADmaqoy8vL5f8/HxJTU2VzMxMGT9+vNTU1CjbNDY2SklJifTq1Ut69OghEydOFK/XG9FGA2ZRu7ArahfhMjVGf9NNN8ndd98t+fn5cubMGXniiSekurpa9u7dK927dxcRkalTp8p7770ny5YtE7fbLaWlpZKcnCwfffRRSK8R63HOgoICJf/iF79Q8q233qrkiy6yztQDZ86cUbI+3rN69Wolf/zxx0p++eWXI9oeK49zOrF2o63tGP2aNWuUdVOnTo1xa6KL2nXuGH19fX3QrI+pZ2RkKFn/wtSnT5+w2vP9998Hbo8cOVJZt3fvXtPPF0rtmuq19ElUli1bJpmZmVJVVSUjR44Un88nixcvlhUrVsgNN9wgIiJLly6VQYMGyfbt22Xo0KHnPGdTU5My+YHf7zfTJCAk1C7sitpFuMIaoz97xPrZb0BVVVVy+vRpKS4uDmwzcOBA8Xg8UllZed7nKC8vF7fbHVj69esXTpOAkFC7sCtqF2Z1uKNvbW2VsrIyGT58uAwePFhEROrq6iQlJUXS09OVbbOysqSuru68zzNjxgzx+XyBpba2tqNNAkJC7cKuqF10RIcHnEtKSqS6ulq2bdsWVgNcLpe4XK6wniMcO3fuVLJ+HeS8vDwlP/HEE0oeP358xNpy6NAhJb/22mtBt3/nnXeUvH///oi1xcmcUruRpv8tycn/tx/w/vvvx7o5OA+71q7+f5tOP1RMP55Iv+7C7t27lVxRUaHkffv2BW4fOHBAWacfyKjPZT9o0CAlV1VVKXnYsGESjP58+vEIV155ZeD2pZdeqqzryBh9KDq0R19aWirr16+XTZs2KX9Udna2NDc3n3Owg9frlezs7LAaCkQCtQu7onbRUaY6esMwpLS0VFavXi0bN26U3NxcZX1eXp507txZ+XZVU1Mjhw4dkqKiosi0GOgAahd2Re0iXKZ+ui8pKZEVK1bI2rVrJTU1NTD+43a7pWvXruJ2u+WBBx6Q6dOnS0ZGhqSlpcm0adOkqKjovEd+ArFC7cKuqF2Ey9R59ElJSee9f+nSpYG50hsbG+Wxxx6TN998U5qammTMmDGycOHCkH9Cctq5yInEyuciU7vt0+eUaHvEtj5u+cUXX8SkTbFC7Ua3dlNSUpTcv39/JetzgOgHB/bq1UvJhw8fjlzjokw/SPKZZ54J3NYPltTPktiyZUu7zx/x8+hD+U7QpUsXWbBggSxYsMDMUwNRRe3CrqhdhIu57gEAcDA6egAAHMw6E7cDiKshQ4ZccJ3TxuQRW83NzUo2W092GpPX6ac9lpWVxbwN7NEDAOBgdPQAADiYqdPrYsHupyglMiufohQL1K59UbvUrl2FUrvs0QMA4GB09AAAOBgdPQAADkZHDwCAg9HRAwDgYHT0AAA4GB09AAAORkcPAICD0dEDAOBgdPQAADgYHT0AAA5GRw8AgIPR0QMA4GB09AAAOJjlOnqLXTUXJiT6Z5fof7+dJfpnl+h/v52F8tlZrqNvaGiIdxPQQYn+2SX6329nif7ZJfrfb2ehfHZJhsW+yrW2tsrhw4fFMAzxeDxSW1sraWlp8W6Wbfj9funXr19M3zfDMKShoUFycnIkOdly3x1jhtoND7UbP9RueKxeuxfFpEUmJCcnS9++fcXv94uISFpaGgXXAbF+39xud8xey6qo3cigdmOP2o0Mq9Zu4n6FBQAgAdDRAwDgYJbt6F0ulzz99NPicrni3RRb4X2LPz6DjuF9iz8+g46x+vtmuYPxAABA5Fh2jx4AAISPjh4AAAejowcAwMHo6AEAcDA6egAAHMyyHf2CBQukf//+0qVLFyksLJSdO3fGu0mWUV5eLvn5+ZKamiqZmZkyfvx4qampUbZpbGyUkpIS6dWrl/To0UMmTpwoXq83Ti1OLNTuhVG71kbtXpita9ewoJUrVxopKSnGkiVLjM8//9x46KGHjPT0dMPr9ca7aZYwZswYY+nSpUZ1dbWxe/du4+abbzY8Ho9x/PjxwDZTpkwx+vXrZ1RUVBi7du0yhg4dagwbNiyOrU4M1G5w1K51UbvB2bl2LdnRFxQUGCUlJYHc0tJi5OTkGOXl5XFslXV9++23hogYW7ZsMQzDMOrr643OnTsbq1atCmyzb98+Q0SMysrKeDUzIVC75lC71kHtmmOn2rXcT/fNzc1SVVUlxcXFgfuSk5OluLhYKisr49gy6/L5fCIikpGRISIiVVVVcvr0aeU9HDhwoHg8Ht7DKKJ2zaN2rYHaNc9OtWu5jv7YsWPS0tIiWVlZyv1ZWVlSV1cXp1ZZV2trq5SVlcnw4cNl8ODBIiJSV1cnKSkpkp6ermzLexhd1K451K51ULvm2K12LXeZWphTUlIi1dXVsm3btng3BTCF2oVd2a12LbdH37t3b+nUqdM5Ryp6vV7Jzs6OU6usqbS0VNavXy+bNm2Svn37Bu7Pzs6W5uZmqa+vV7bnPYwuajd01K61ULuhs2PtWq6jT0lJkby8PKmoqAjc19raKhUVFVJUVBTHllmHYRhSWloqq1evlo0bN0pubq6yPi8vTzp37qy8hzU1NXLo0CHewyiidttH7VoTtds+W9duXA8FvICVK1caLpfLWLZsmbF3715j8uTJRnp6ulFXVxfvplnC1KlTDbfbbWzevNk4cuRIYDl58mRgmylTphgej8fYuHGjsWvXLqOoqMgoKiqKY6sTA7UbHLVrXdRucHauXUt29IZhGPPnzzc8Ho+RkpJiFBQUGNu3b493kyxDRM67LF26NLDNqVOnjEceecTo2bOn0a1bN+P22283jhw5Er9GJxBq98KoXWujdi/MzrXL9egBAHAwy43RAwCAyKGjBwDAwejoAQBwMDp6AAAcjI4eAAAHo6MHAMDB6OgBAHAwOnoAAByMjh4AAAejowcAwMHo6AEAcLD/D3ipKdHEn3OxAAAAAElFTkSuQmCC",
      "text/plain": [
       "<Figure size 640x480 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from matplotlib import pyplot as plt\n",
    "import torchmetrics\n",
    "import numpy as np\n",
    "\n",
    "model.eval()\n",
    "\n",
    "# demo batch\n",
    "demo_batch_size = 1000\n",
    "ri = np.random.permutation(testX.shape[0])[:demo_batch_size]\n",
    "inputs, targets = testX[ri], testy[ri]\n",
    "\n",
    "predictions = model(inputs)\n",
    "\n",
    "# get accuracy\n",
    "label_predictions = torch.argmax(predictions, dim=1)\n",
    "accuracy = torchmetrics.Accuracy(task='multiclass', num_classes=10).to(DEVICE)\n",
    "acc = accuracy(predictions, targets)\n",
    "print(f\"Accuracy: {acc * 100:.2f}%\")\n",
    "\n",
    "# get per-digit accuracy\n",
    "correct = np.zeros(10)\n",
    "total = np.zeros(10)\n",
    "for i in range(demo_batch_size):\n",
    "    #print(f\"Predicted: {label_predictions[i]}, target: {targets[i]} {label_predictions[i] == targets[i]}\")\n",
    "    total[targets[i]] += 1\n",
    "    correct[targets[i]] += (label_predictions[i] == targets[i])\n",
    "\n",
    "digit_acc = correct / total * 100\n",
    "for i in range(10):\n",
    "    print(f\"Accuracy {i}: {correct[i]:.0f} / {total[i]:.0f} = {digit_acc[i]:.2f}%\")\n",
    "\n",
    "# plot up to 9 images\n",
    "for i in range(min(demo_batch_size, 9)):\n",
    "    print(f\"Predicted: {label_predictions[i]}, target: {targets[i]} {label_predictions[i] == targets[i]}\")\n",
    "    # define subplot\n",
    "    plt.subplot(330 + 1 + i)\n",
    "    # plot raw pixel data\n",
    "    plt.imshow(inputs[i].squeeze().cpu(), cmap=plt.get_cmap('gray'))\n",
    "# show the figure\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Input: torch.Size([1, 1, 28, 28])\n",
      "Pred: torch.Size([1, 10])\n",
      "Label: tensor([3], device='cuda:0')\n",
      "Label[0]: 3\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x1588e01c110>"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from data_loader import load_classifier_image\n",
    "\n",
    "input_image_tensor = load_classifier_image(\"./data/digit.jpg\", DEVICE)\n",
    "\n",
    "pred = model(input_image_tensor)\n",
    "label_pred = torch.argmax(pred, dim=1)\n",
    "\n",
    "print(f\"Input: {input_image_tensor.shape}\")\n",
    "print(f\"Pred: {pred.shape}\")\n",
    "print(f\"Label: {label_pred}\")\n",
    "print(f\"Label[0]: {label_pred[0]}\")\n",
    "\n",
    "# define subplot\n",
    "plt.subplot(330 + 1)\n",
    "# plot raw pixel data\n",
    "plt.imshow(input_image_tensor.squeeze().cpu(), cmap=plt.get_cmap('gray'))"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.11.0"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}