14-pytorch-autoencoder/Pytorch MNIST Autoencoder.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torchvision as tv\n",
    "import torchvision.transforms as transforms\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "from torch.autograd import Variable\n",
    "from torchvision.utils import save_image\n",
    "\n",
    "num_epochs = 100\n",
    "batch_size = 64\n",
    "\n",
    "transform = transforms.Compose([\n",
    "    transforms.ToTensor(),\n",
    "    transforms.Normalize((0.1307,), (0.3081,))\n",
    "])\n",
    "trainset = tv.datasets.MNIST(root='./data',  train=True, download=True, transform=transform)\n",
    "dataloader = torch.utils.data.DataLoader(trainset, batch_size=batch_size, shuffle=True, num_workers=4)\n",
    "\n",
    "class Autoencoder(nn.Module):\n",
    "    def __init__(self):\n",
    "        super(Autoencoder,self).__init__()\n",
    "        self.encoder = nn.Sequential(\n",
    "            # 28 x 28\n",
    "            nn.Conv2d(1, 4, kernel_size=5),\n",
    "            # 4 x 24 x 24\n",
    "            nn.ReLU(True),\n",
    "            nn.Conv2d(4, 8, kernel_size=5),\n",
    "            nn.ReLU(True),\n",
    "            # 8 x 20 x 20 = 3200\n",
    "            nn.Flatten(),\n",
    "            nn.Linear(3200, 10),\n",
    "            # 10\n",
    "            nn.Softmax(),\n",
    "            )\n",
    "        self.decoder = nn.Sequential(\n",
    "            # 10\n",
    "            nn.Linear(10, 400),\n",
    "            # 400\n",
    "            nn.ReLU(True),\n",
    "            nn.Linear(400, 4000),\n",
    "            # 4000\n",
    "            nn.ReLU(True),\n",
    "            nn.Unflatten(1, (10, 20, 20)),\n",
    "            # 10 x 20 x 20\n",
    "            nn.ConvTranspose2d(10, 10, kernel_size=5),\n",
    "            # 24 x 24\n",
    "            nn.ConvTranspose2d(10, 1, kernel_size=5),\n",
    "            # 28 x 28\n",
    "            nn.Sigmoid(),\n",
    "            )\n",
    "    def forward(self, x):\n",
    "        x = self.encoder(x)\n",
    "        x = self.decoder(x)\n",
    "        return x\n",
    "\n",
    "model = Autoencoder().cpu()\n",
    "distance = nn.MSELoss()\n",
    "#distance = nn.BCELoss()\n",
    "#distance = nn.BCEWithLogitsLoss()\n",
    "#optimizer = torch.optim.Adam(model.parameters(), weight_decay=1e-5)\n",
    "optimizer = torch.optim.SGD(model.parameters(), lr=0.01, momentum=0.5)\n",
    "\n",
    "for epoch in range(num_epochs):\n",
    "    for data in dataloader:\n",
    "        img, _ = data\n",
    "        img = Variable(img).cpu()\n",
    "        output = model(img)\n",
    "        loss = distance(output, img)\n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "    print('epoch [{}/{}], loss: {:.4f}'.format(epoch+1, num_epochs, loss.item()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[383. 186.  22. 326.  31.  21.   3.   0.   7.   1.]\n",
      " [  0.   0.  39.   3.   0.   0.  73.   4. 197. 819.]\n",
      " [  2. 203.  88. 171.  50.   4.  13. 256. 156.  89.]\n",
      " [ 27.   0.  20. 882.   3.   0.  13.  21.  20.  24.]\n",
      " [ 61.  79.  22.   0.   0.  29. 528.  47.  11. 205.]\n",
      " [ 60.   0.  74. 494.   0. 108.  12.   4.  28. 112.]\n",
      " [ 16.  16.  87.   1.   6. 103.  23.  50. 655.   1.]\n",
      " [148.   0.   0.   0.   1.   0. 366.   7.   2. 504.]\n",
      " [  4.   5.  56. 594.   6.   7.  18.  16.   3. 265.]\n",
      " [ 59.   8.   3.  29.   2.   7. 609.   1.   0. 291.]]\n",
      "10000\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "np.set_printoptions(suppress=True)\n",
    "\n",
    "confusion_matrix = np.zeros((10, 10))\n",
    "\n",
    "batch_size = 20*1000\n",
    "\n",
    "testset = tv.datasets.MNIST(root='./data',  train=False, download=True, transform=transform)\n",
    "dataloader = torch.utils.data.DataLoader(testset, batch_size=batch_size, shuffle=True, num_workers=4)\n",
    "\n",
    "for data in dataloader:\n",
    "    imgs, labels = data\n",
    "    imgs = Variable(imgs).cpu()\n",
    "    encs = model.encoder(imgs).detach().numpy()\n",
    "    for i in range(len(encs)):\n",
    "        predicted = np.argmax(encs[i])\n",
    "        actual = labels[i]\n",
    "        confusion_matrix[actual][predicted] += 1\n",
    "print(confusion_matrix)\n",
    "print(int(confusion_matrix.sum()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Actual label 0 most likely activates on neuron 0 (39%)\n",
      "Actual label 1 most likely activates on neuron 9 (72%)\n",
      "Actual label 2 most likely activates on neuron 7 (24%)\n",
      "Actual label 3 most likely activates on neuron 3 (87%)\n",
      "Actual label 4 most likely activates on neuron 6 (53%)\n",
      "Actual label 5 most likely activates on neuron 3 (55%)\n",
      "Actual label 6 most likely activates on neuron 8 (68%)\n",
      "Actual label 7 most likely activates on neuron 9 (49%)\n",
      "Actual label 8 most likely activates on neuron 3 (60%)\n",
      "Actual label 9 most likely activates on neuron 6 (60%)\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "({0, 3, 6, 7, 8, 9}, 6)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "active_neurons = set()\n",
    "for actual in range(10):\n",
    "    predicted = np.argmax(confusion_matrix[actual])\n",
    "    pct = 100 * confusion_matrix[actual][predicted] / np.sum(confusion_matrix[actual])\n",
    "    print('Actual label %d most likely activates on neuron %d (%d%%)' % (actual, predicted, pct))\n",
    "    active_neurons.add(predicted)\n",
    "active_neurons, len(active_neurons)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "import numpy as np\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "fig, ax = plt.subplots(figsize=(10, 10))\n",
    "im = ax.imshow(confusion_matrix, cmap=\"YlGn\")\n",
    "ax.set_xticks(np.arange(10))\n",
    "ax.set_yticks(np.arange(10))\n",
    "ax.set_title(\"Confusion matrix\")\n",
    "for i in range(10):\n",
    "    for j in range(10):\n",
    "        text = ax.text(j, i, int(confusion_matrix[i, j]),\n",
    "                       ha=\"center\", va=\"center\")\n",
    "fig.tight_layout()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[[383. 186.  22. 326.  31.  21.   3.   0.   7.   1.]\n",
      " [  0.   0.  39.   3.   0.   0.  73.   4. 197. 819.]\n",
      " [  2. 203.  88. 171.  50.   4.  13. 256. 156.  89.]\n",
      " [ 27.   0.  20. 882.   3.   0.  13.  21.  20.  24.]\n",
      " [ 61.  79.  22.   0.   0.  29. 528.  47.  11. 205.]\n",
      " [ 60.   0.  74. 494.   0. 108.  12.   4.  28. 112.]\n",
      " [ 16.  16.  87.   1.   6. 103.  23.  50. 655.   1.]\n",
      " [148.   0.   0.   0.   1.   0. 366.   7.   2. 504.]\n",
      " [  4.   5.  56. 594.   6.   7.  18.  16.   3. 265.]\n",
      " [ 59.   8.   3.  29.   2.   7. 609.   1.   0. 291.]]\n",
      "10000\n",
      "0.4053\n"
     ]
    }
   ],
   "source": [
    "classifier = {\n",
    "    # enc -> predict\n",
    "    0: 0,\n",
    "    1: 2,\n",
    "    2: 2,\n",
    "    3: 3,\n",
    "    4: 2,\n",
    "    5: 5,\n",
    "    6: 9,\n",
    "    7: 2,\n",
    "    8: 6,\n",
    "    9: 1,\n",
    "}\n",
    "\n",
    "testset = tv.datasets.MNIST(root='./data',  train=False, download=True, transform=transform)\n",
    "dataloader = torch.utils.data.DataLoader(testset, batch_size=batch_size, shuffle=True, num_workers=4)\n",
    "total = 0\n",
    "correct = 0\n",
    "for data in dataloader:\n",
    "    imgs, labels = data\n",
    "    imgs = Variable(imgs).cpu()\n",
    "    encs = model.encoder(imgs).detach().numpy()\n",
    "    for i in range(len(encs)):\n",
    "        predicted = classifier[np.argmax(encs[i])]\n",
    "        actual = labels[i]\n",
    "        if predicted == actual:\n",
    "            correct += 1\n",
    "        total += 1\n",
    "print(confusion_matrix)\n",
    "print(int(confusion_matrix.sum()))\n",
    "\n",
    "print(correct/total)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}