Autoencoder.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "canadian-canyon",
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "import torch.optim as optim\n",
    "from torchvision import datasets, transforms\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "gentle-metallic",
   "metadata": {},
   "outputs": [],
   "source": [
    "transform = transforms.Compose(\n",
    "    [transforms.ToTensor(),\n",
    "     transforms.Normalize((0.5), (0.5))\n",
    "    ])\n",
    "\n",
    "transform = transforms.ToTensor()\n",
    "\n",
    "mnist_data = datasets.MNIST(root='./data', train=True, download=True, transform=transform)\n",
    "\n",
    "data_loader = torch.utils.data.DataLoader(dataset=mnist_data,\n",
    "                                          batch_size=64,\n",
    "                                          shuffle=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "id": "purple-birth",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "tensor(0.) tensor(1.)\n"
     ]
    }
   ],
   "source": [
    "dataiter = iter(data_loader)\n",
    "images, labels = dataiter.next()\n",
    "print(torch.min(images), torch.max(images))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "id": "endless-charlotte",
   "metadata": {},
   "outputs": [],
   "source": [
    "# repeatedly reduce the size\n",
    "class Autoencoder_Linear(nn.Module):\n",
    "    def __init__(self):\n",
    "        super().__init__()        \n",
    "        self.encoder = nn.Sequential(\n",
    "            nn.Linear(28 * 28, 128), # (N, 784) -> (N, 128)\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(128, 64),\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(64, 12),\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(12, 3) # -> N, 3\n",
    "        )\n",
    "        \n",
    "        self.decoder = nn.Sequential(\n",
    "            nn.Linear(3, 12),\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(12, 64),\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(64, 128),\n",
    "            nn.ReLU(),\n",
    "            nn.Linear(128, 28 * 28),\n",
    "            nn.Sigmoid()\n",
    "        )\n",
    "\n",
    "    def forward(self, x):\n",
    "        encoded = self.encoder(x)\n",
    "        decoded = self.decoder(encoded)\n",
    "        return decoded\n",
    "    \n",
    "# Input [-1, +1] -> use nn.Tanh"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 99,
   "id": "swedish-robin",
   "metadata": {},
   "outputs": [],
   "source": [
    "class Autoencoder(nn.Module):\n",
    "    def __init__(self):\n",
    "        super().__init__()        \n",
    "        # N, 1, 28, 28\n",
    "        self.encoder = nn.Sequential(\n",
    "            nn.Conv2d(1, 16, 3, stride=2, padding=1), # -> N, 16, 14, 14\n",
    "            nn.ReLU(),\n",
    "            nn.Conv2d(16, 32, 3, stride=2, padding=1), # -> N, 32, 7, 7\n",
    "            nn.ReLU(),\n",
    "            nn.Conv2d(32, 64, 7) # -> N, 64, 1, 1\n",
    "        )\n",
    "        \n",
    "        # N , 64, 1, 1\n",
    "        self.decoder = nn.Sequential(\n",
    "            nn.ConvTranspose2d(64, 32, 7), # -> N, 32, 7, 7\n",
    "            nn.ReLU(),\n",
    "            nn.ConvTranspose2d(32, 16, 3, stride=2, padding=1, output_padding=1), # N, 16, 14, 14 (N,16,13,13 without output_padding)\n",
    "            nn.ReLU(),\n",
    "            nn.ConvTranspose2d(16, 1, 3, stride=2, padding=1, output_padding=1), # N, 1, 28, 28  (N,1,27,27)\n",
    "            nn.Sigmoid()\n",
    "        )\n",
    "\n",
    "    def forward(self, x):\n",
    "        encoded = self.encoder(x)\n",
    "        decoded = self.decoder(encoded)\n",
    "        return decoded\n",
    "    \n",
    " \n",
    "# Note: nn.MaxPool2d -> use nn.MaxUnpool2d, or use different kernelsize, stride etc to compensate...\n",
    "# Input [-1, +1] -> use nn.Tanh"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 122,
   "id": "educated-meditation",
   "metadata": {},
   "outputs": [],
   "source": [
    "model = Autoencoder()\n",
    "\n",
    "criterion = nn.MSELoss()\n",
    "optimizer = torch.optim.Adam(model.parameters(),\n",
    "                             lr=1e-3, \n",
    "                             weight_decay=1e-5)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 123,
   "id": "steady-ceremony",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([64, 3])\n",
      "torch.Size([32, 3])\n",
      "Epoch:1, Loss:0.0484\n"
     ]
    }
   ],
   "source": [
    "# Point to training loop video\n",
    "num_epochs = 1\n",
    "outputs = []\n",
    "for epoch in range(num_epochs):\n",
    "    for (img, _) in data_loader:\n",
    "        # img = img.reshape(-1, 28*28) # -> use for Autoencoder_Linear\n",
    "        recon = model(img)\n",
    "        loss = criterion(recon, img)\n",
    "        \n",
    "        optimizer.zero_grad()\n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "\n",
    "    print(f'Epoch:{epoch+1}, Loss:{loss.item():.4f}')\n",
    "    outputs.append((epoch, img, recon))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 124,
   "id": "numerous-greensboro",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 648x144 with 18 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "for k in range(0, num_epochs, 4):\n",
    "    plt.figure(figsize=(9, 2))\n",
    "    plt.gray()\n",
    "    imgs = outputs[k][1].detach().numpy()\n",
    "    recon = outputs[k][2].detach().numpy()\n",
    "    for i, item in enumerate(imgs):\n",
    "        if i >= 9: break\n",
    "        plt.subplot(2, 9, i+1)\n",
    "        # item = item.reshape(-1, 28,28) # -> use for Autoencoder_Linear\n",
    "        # item: 1, 28, 28\n",
    "        plt.imshow(item[0])\n",
    "            \n",
    "    for i, item in enumerate(recon):\n",
    "        if i >= 9: break\n",
    "        plt.subplot(2, 9, 9+i+1) # row_length + i + 1\n",
    "        # item = item.reshape(-1, 28,28) # -> use for Autoencoder_Linear\n",
    "        # item: 1, 28, 28\n",
    "        plt.imshow(item[0])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "coastal-cement",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Homework: Use MaxPool2d, inspect the encoded data (can it be plotted as img?)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "diverse-hours",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "pytorch",
   "language": "python",
   "name": "pytorch"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}