lec_7_v4.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "3d15b389",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import gzip, pickle\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib as mpl\n",
    "from pathlib import Path\n",
    "import pandas as pd\n",
    "\n",
    "mp.rcParams['image.cmap'] = 'gray'\n",
    "mpl.rcParams['figure.dpi'] = 30"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "e6e99cd2",
   "metadata": {},
   "outputs": [],
   "source": [
    "dpth = Path('./data/mnist.pkl.gz')\n",
    "with gzip.open(dpth, 'rb') as f:\n",
    "    data = pickle.load(f, encoding='latin')\n",
    "((x_train, y_train), (x_val, y_val), _) = data\n",
    "(x_train, y_train, x_val, y_val) = map(torch.tensor, (x_train, y_train, x_val, y_val))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "id": "5ca77b78",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[[(0, 0), (0, 1), (0, 2), (0, 3), (0, 4)],\n",
       " [(1, 0), (1, 1), (1, 2), (1, 3), (1, 4)],\n",
       " [(2, 0), (2, 1), (2, 2), (2, 3), (2, 4)],\n",
       " [(3, 0), (3, 1), (3, 2), (3, 3), (3, 4)],\n",
       " [(4, 0), (4, 1), (4, 2), (4, 3), (4, 4)]]"
      ]
     },
     "execution_count": 46,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x_imgs = x_train.view(-1, 28, 28)\n",
    "xv_imgs = x_val.view(-1, 28, 28)\n",
    "\n",
    "im3 = x_imgs[7]\n",
    "\n",
    "top_edge = torch.tensor([\n",
    "    [-1, -1, -1],\n",
    "    [0, 0, 0],\n",
    "    [1, 1, 1]  \n",
    "]).float()\n",
    "\n",
    "df = pd.DataFrame(im3[:13, :23])\n",
    "df.style.format(precision=2).set_properties(**{'font-size':'7pt'}).background_gradient('Greys')\n",
    "\n",
    "\n",
    "(im3[1:4, 14:17] * top_edge).sum()\n",
    "(im3[7:10, 14:17] * top_edge).sum()\n",
    "\n",
    "def apply_kernel(row, col, kernel):\n",
    "    return (im3[row-1:row+2, col-1:col+2]*kernel).sum()\n",
    "\n",
    "apply_kernel(8, 15, top_edge)\n",
    "\n",
    "[[(i, j) for j in range(5)] for i in range(5)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "id": "c027e5d5",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([26, 26])"
      ]
     },
     "execution_count": 69,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAGkAAABnCAYAAAADiGxKAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAASdAAAEnQF8NGuhAAAHD0lEQVR4nO2d308VVxDHv0cFCaBAUVBT1FiMIsZGq03K39AXXzT+CT7Yf6YP7X9g41PfbNL0UWIaFDTRRkQUUIEqBCKCKHL6cPE4M93de++ye3cnzudpTmbde3Q8c87Oj13nvYdRbnbVeqFzrgPAMQDv85uOsUUzgEnv/TJQh5EAHLty5crd7u7ufKZlBBYWFnD9+vVzAEaBBCM55y4DaAFwy3s/AaCvu7sbvb29jZmpETzWjoSL2gDcAHAh9+kYiSQZaRXAJQAjW+Mn+U/HiCLW3Xnvf2vkRIx4klaSURLMSAowIynAjKQAM5IC6ok4FMru3bvZ2DkX5F27+F/j4MGDQX779i3TvXz5MtXvv3nzJsjT09NMt2fPniD39fXFzjMttpIUYEZSgBlJAaXak1paWti4q6sryHv37mW6paWlIK+vrzNdU1NT9pOL+W0AeP78eZB37OD/7w8dOhSrqxVbSQowIymgVO7uyJEjbHz48OEgLy4uMp10OUUyOTkZZFmO0NnZGeT29vZU97eVpIBYIznnrjrnfnTOfXqK/KZBczIESStpBkATKhlao0CS9qRZAN8C+BrAIhqQmZVFLoODg0F++PAh0z169CjI9Agsx0n7QGtrKxu3tX3+/7i6uhr75+i8AGB0dDTIDx48YLqzZ8/WNJckkjKzdwDcSXVXI1Ps4KCAUh3BZeRgYWEhyC9evGA6egSXrqm5uTnISS5m586dbHzy5Mkgv3v3junGxsaCLCMHm5ubkXLUtWmwlaQAM5ICzEgKKNWeNDU1xcYfP34MMs2MVqPWTpGOjg42PnXqVJBfvXrFdPfv3w+yDFHRPXH//v1MJ4/5abCVpAAzkgJK5e5ev37NxsvLy7HX0uIT6WKSij9oVFp2iMikYxwyqkDd3cDAANOZu/tCMCMpwIykgFLtSZIPHz7UdJ0sPKH7jtyvaMhIFlXSyLp8HKCPADJERQs38yiOZLN0zg0B+AHABoB17/2vRG1Jv4Jg7s57PwygFcA4rMu8NMiVNACgB5U+2X/Ftbkn/agrArjbkjoaBV9ZWWE6mryjMsBdqDzi02ShTDLS+m/pwk6cOBFkWoeeFcxI3vt/AFzL/FeMbWGnOwWYkRRQ+BGchmKOHz/OdLTvZ25uLpPfo3uZDEPRyPft27eZbmNjI8gy9LNv374gy2xvFthKUoAZSQGlcnf0KAvw4zMtSqkGdWmyZly2Z1JoslC2XNJI98WLF5mOtrdkEWGQ2EpSgBlJAWYkBRS+J1FkAQkdyyP47OxskGUBZNK+kxQhp5FvGYGnhZsyei7HWWMrSQFmJAUU7u5ozfX4+DjT0YjD+/fZZE6oa5ItLAcOHIj8bYC7Vxl1z5t6VpIl/QoiKjP7PSo5pd+9938XMiuDEZWZXQEwBeArca29g7UgojKzZwD8BWAQwB95T4DuSbQQBODH5VqLUqpB679Pnz7NdLTVv0xEZWZ/KmguRgx2BFdAw4/g8mhL69lkpJs+5cvosrxPHNRlArxQZG1tjelkS0tZsJWkADOSAsxICshlT5IFiTTaLFsgb968GeRnz54x3fnz54MsiyMpMgpNo+K07R/ge5IsgKRj2epP98Q8sq9J2EpSgBlJAbm4u/7+fjamLxecn59nOnrslUdw6XLikEk/6l5liyWt/x4ZGWG6e/fuBVkmIOnfodbjf1bYSlKAGUkBZiQF5LInyZpuukfNzMwwHfX90tdncdSV9d5Pnz4N8vDwMNPRAha6BwG81b/Re1JU0u8CgE4A89aOWQ6ikn7tAO7C2jFLg1xJ1wAsAegHsCaurTkzS9tE5Fi2htDEmzyCJ33OgBamyDo7Gj2XUQzq/uSfoy8QPHr0KNPR9hYZUckbmfT7uaG/btSEne4UYEZSQC5H8CdP+PZF9wj51qqhoaEgy7eQPH78OMiynZ8Wpsi9i/6GzL7SY70MJ9F9R7b6NzryTbGVpAAzkgJycXcTExNsTFsipRvJo1ubIhOCtDBFfnZBfoGzLNhKUoAZSQFmJAXksifJYy99iZ8MxfT09KT6DXrsltlXerSWxZEasZWkADOSAhpSC04LSmS9dVL9ddLLBb8kopJ+5wB8B+AX0elnSb+CiEr6eQBj+H+nn1EQUUm/TVQSf7LTb6aelzAZ6dn6dw511a7Wz9g45zoAHAPw6cXX2+mh/eQ6t9uHm8V9yjiXGQCT3vtloA4jGcVhR3AFmJEUUPdzknPuMoAWALe89xPVro+5x1VU/O6f3vv1atfH3CPpkw313GNbLxepUqtY730iH3/SrKQ2ADe2JpaWbX9XPYtPNlR5uUg999h2rWLS408aI60CuARgpNqFCcwC6ELlu+qpEJ9siG8DrH6PMwDmUHnkSHMPWquYah7kPsDnx5/POjvdlR87OCjAjKSA/wAlykxQFm5kNgAAAABJRU5ErkJggg==",
      "text/plain": [
       "<Figure size 180x120 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "rng = range(1, 27)\n",
    "top_edge3 = torch.tensor([[apply_kernel(i, j, top_edge) for j in rng] for i in rng])\n",
    "plt.imshow(top_edge3)\n",
    "\n",
    "left_edge = torch.tensor([\n",
    "    [-1, 0, 1],\n",
    "    [-1, 0, 1],\n",
    "    [-1, 0, 1]\n",
    "]).float()\n",
    "\n",
    "left_edge3 = torch.tensor([[apply_kernel(i, j, left_edge) for j in rng] for i in rng])\n",
    "plt.imshow(left_edge3)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "id": "00c7cedb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x173cb6c10>"
      ]
     },
     "execution_count": 76,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 180x120 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import torch.nn.functional as F\n",
    "\n",
    "inp = im3[None,None,:,:].float()\n",
    "inp_unf = F.unfold(inp, (3, 3))[0]\n",
    "inp_unf.shape\n",
    "torch.Size([9, 676])\n",
    "\n",
    "w = left_edge.view(-1)\n",
    "out = (w @ inp_unf).view(26, 26)\n",
    "plt.imshow(out)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 86,
   "id": "81e7cb24",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "12.7 ms ± 5.51 ms per loop (mean ± std. dev. of 7 runs, 1 loop each)\n"
     ]
    }
   ],
   "source": [
    "%timeit -n 1 torch.tensor([[apply_kernel(i, j, left_edge) for j in rng] for i in rng])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 88,
   "id": "ad373e15",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "92.3 µs ± 20.1 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)\n"
     ]
    }
   ],
   "source": [
    "%timeit -n 100 (w@F.unfold(inp, (3,3))[0]).view(26, 26)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 94,
   "id": "41c96b86",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "67 µs ± 15.8 µs per loop (mean ± std. dev. of 7 runs, 100 loops each)\n"
     ]
    }
   ],
   "source": [
    "%timeit -n 100 F.conv2d(inp, left_edge[None, None]).shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 96,
   "id": "e533928f",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x173a32990>"
      ]
     },
     "execution_count": 96,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 180x120 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "diag1_edge = torch.tensor([\n",
    "    [0, -1, 1],\n",
    "    [-1, 1, 0],\n",
    "    [1, 0, 0]\n",
    "]).float()\n",
    "plt.imshow(diag1_edge)\n",
    "\n",
    "diag2_edge = torch.tensor([\n",
    "    [1, -1, 0],\n",
    "    [0, 1, -1],\n",
    "    [0, 0, 1]\n",
    "]).float()\n",
    "plt.imshow(diag2_edge)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 113,
   "id": "a98727bb",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x173ebfb10>"
      ]
     },
     "execution_count": 113,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAGkAAABnCAYAAAADiGxKAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjcuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/bCgiHAAAACXBIWXMAAASdAAAEnQF8NGuhAAAHqklEQVR4nO2dW09USRDH/61yEVEuKpoIxiBekMRVUBR8NNGXNfFJP8M+sCZ+BT/Cxuw++WTixhizj2vWjfrgegmy6yWiCCiiAl4QdERRpPdhsK0qzxxmzsycmTb1e6rZOvb0WnZ1narqHmOthVLcLEr3QWNMFYBGAB/zNx1ljlIAg9baSSADIwFo7Ojo6KmsrMzPtBRHIpHAlStXWgH8C4QYyRhzGEA5gMvW2n4ADZWVlaiqqopnporzWAtCHloC4DSAnXmfjhJKmJGmABwC0D33eSD/01GCSOnurLW/xzkRJTVhK0kpEtRIHpBJCO4l1dXV7HNjY6OTGxoamG54eNjJg4ODTDcxMZHzuaWLriQPUCN5wHfp7ugL95YtW5ju7du3Tn78+DHT7dmzx8mdnZ1Md/LkSSfH7fp0JXmAGskD1Ege8F3sSTU1Nezzpk2bnNzX18d0GzdudPLr16+Z7uLFi07eu3cv03V1dTn5+PHjTDc+Pp7ZhDNEV5IHqJE8wBt3t2gRnyoNs2kWAQCGhoac3NTUxHTU/bW0tDDdwMBA4HMAsG/fPidT1wcAx44dc/Ls7Gzw/0AW6ErygJRGMsb8ZIz50RhTNvef1sc0J0UQtpKGAZQgWaFVCkjYnjQC4AcA9QDGUeDKrNw/aHab7iUAUF9f72S6PwFAW1ubk1etWsV05eXlTqbhuOTgwYPs844dO5zc3d3NdLnYo8IqszcA3Mj6G5Ss0cDBA4o6BKcubWxsLOVz69fzmObatWtOXr58OdNRlybHpLrVq1cz3YULF5xcWlrKdDTDIbMYDx48SDnvdNGV5AFqJA9QI3lAUe1JsoWZVlWlbx8dHXXyq1evmI7uZStWrGC6M2fOpNTR75P7Dq3oyhTVggVf/63now1bV5IHqJE8oODujh6lqaurYzqaiV63bh3T3b5928kbNmxgOhpKywwAdUdPnz5NOS/ZwDIzM+Pkq1evMt2RI0fSGjMqupI8QI3kAWokD4h9T1q4cCH7TPeaT58+Md3SpUudfPbsWabbvHlzyjE/f/7sZJqhBoBz5845uaSkhOlo6N7T08N0ixcvdrLcd27evOlkma2ne2LUpkpmJGNMJ4AOADMApq21vxG1Fv0KBHN31tp/AFQA6IOeMi8a5EpqBlCH5DnZ5+LZnBT9qAsDeM9abW0t09Fs9u7du5mOurS7d+8yHQ2fZaZbZsUpskCYLolEInBeAG+Eka8D6cKMZK3tBdCV4lmlQGh05wFqJA+IPQSnTSIA8OTJEye/ePGC6Wj4TP0+AKxdu9bJHz/yGIeGz7m6wcUYk9Zz79+/Z59p6B4VXUkeoEbygFjcHW3qoCe8AX4CXL7l07442SN3//59J8vmD5qBkO4nKvTKOen6aCZBnnbPBbqSPECN5AFqJA+IZU+iobXcW/r7+53c3NzMdLQaKps/Xr58mcspAuB7GW14BHjq6flznjF79uxZzudC0ZXkAWokD4jF3dHMsMxKr1y50slr1qxhuqmpKSfn270BPGMts9n0yCd10QDP3suCYEVFRdbzzGQladGvQARVZtuRrCn9Ya29XpBZKYygymwCwBCAWvGs3sFaIIIqs1sB/A2gBcCfuf5C6etpeC6bRmj65c6dOzn5/rCrBWgKSe4ttBlz27ZtTEcz9LK6LFNdUQiqzP6c9ahKTtEQ3ANiCcHDCmbbt293snyTpxnlDx8+RPpueSMKzXjIPjh6hEa6OzpPOReaDXn48CHT0axJVHQleYAayQPUSB4Qy54U9kNatKp69OhRpjt16pSTZUMHDXXl/d7Lli1zssxQ08YXeowSAN69e+dkenkhwPdVeR7q+vX8vvPrSvIANZIHFPw45vT0tJPlPd30mOOJEyeYjr7lUzclx6RHMwHeCCOz2a2trU6W97rSYzL5dm8SXUkeoEbyADWSB8SyJ9EwW6aIaDX23r17TEdDd3r8Uj4re8Fv3brl5Ddv3jAdDeX379/PdI8ePXKyvBFFNmDGSVDRbyeAagBjehyzOAgq+lUC6IEexywa5ErqAjABoAmAbKKOXJmlLubSpUtMR0NpWfSj/d5lZWVMRzMJ8jgmzXTLozY0uy370umf6+3tRbEgi36/FGoiSmo0uvMANZIHxBKC09TMrl27mO78+fNOlhnr9vb2lGPSy3HlvjM5OZnyz9GQnzaXAN8eBy0WdCV5gBrJA2LPgsvjigcOHHDyyMgI09FLCWURjv5Yr+yfo7110oXRop8v6EryADWSB6iRPKDglVma7pGX4YZBM+Typ3m+N3QleYAayQPUSB4QVPRrBdAG4Fdx0k+LfgUiqOhnAfyHb0/6KQUiqOg3i2ThT570G5Z3zin5Ye7v2TVZmLA+bYoxpgpAI4AvjdfZxL1fXGe2sXMuxinGuQwDGLTWTgIZGEkpHBrdeYAayQMyTgsZYw4DKAdw2VrbP9/zKcb4CUm/+5e1dnq+51OMEfaTDZmMkdXlIvP0KmY6TuDrT5SVtATA6bmJRSXr31XPxU82zHO5SCZjZN2rGPb6E8VIUwAOAYh2p3+SEQA1SP6ueiTETzaUzvN42BhbAYwi+coRZQzaqxhpHmQc4Ovrz1edRnfFjwYOHqBG8oD/AUalfkutmbeGAAAAAElFTkSuQmCC",
      "text/plain": [
       "<Figure size 180x120 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "xb = x_imgs[:16][:,None]\n",
    "torch.Size([16, 1, 28, 28])\n",
    "\n",
    "edge_kernels = torch.stack([left_edge, top_edge, diag1_edge, diag2_edge])[:,None]\n",
    "torch.Size([4, 1, 3, 3])\n",
    "\n",
    "batch_features = F.conv2d(xb, edge_kernels)\n",
    "torch.Size([16, 4, 26, 26])\n",
    "\n",
    "img0 = xb[1,0]\n",
    "plt.imshow(img0)\n",
    "\n",
    "plt.imshow(batch_features[1][3])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 174,
   "id": "ca41ff8b",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([16, 10])"
      ]
     },
     "execution_count": 174,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "n,m = x_train.shape\n",
    "nh = 50\n",
    "c = y_train.max()+1\n",
    "\n",
    "model = nn.Sequential(*[nn.Linear(m, nh), nn.ReLU(), nn.Linear(nh, 10)])\n",
    "\n",
    "broken_cnn = nn.Sequential(*[\n",
    "   nn.Conv2d(1, 30, kernel_size=3, padding=1),\n",
    "   nn.ReLU(),\n",
    "   nn.Conv2d(30, 10, kernel_size=3, padding=1)\n",
    "])\n",
    "t = broken_cnn(xb)\n",
    "torch.Size([16, 10, 28, 28])\n",
    "\n",
    "def conv(ni, nf, ks=3, stride=2, act=True):\n",
    "    res = nn.Conv2d(ni, nf, stride=stride, kernel_size=ks, padding=ks//2)\n",
    "    if act:\n",
    "        res = nn.Sequential(res, nn.ReLU())\n",
    "    return res\n",
    "\n",
    "simple_cnn = nn.Sequential(\n",
    "    conv(1, 4), #14x14\n",
    "    conv(4, 8), #7x7\n",
    "    conv(8, 16), #2x2\n",
    "    conv(16, 16), #1x1\n",
    "    conv(16, 10, act=False),\n",
    "    nn.Flatten()\n",
    ")\n",
    "t = simple_cnn(xb)\n",
    "t.shape"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "python_main",
   "language": "python",
   "name": "python_main"
  },
  "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": 5
}