01_Variational_AutoEncoder.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Implementation of Variational AutoEncoder (VAE)\n",
    "\n",
    "     VAE from \"Auto-Encoding Variational Bayes\" (2014, D.P. Kingma et. al.)\n",
    "    \n",
    "     Kernel-author: Jackson Kang @ Deep-learning Lab. (Handong Global University, S.Korea)\n",
    "     \n",
    "     Author-email:  mskang@handong.edu\n",
    "\n",
    "     python and pytorch version: python=3.7, pytorch=1.3.1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "\n",
    "import numpy as np\n",
    "\n",
    "from tqdm import tqdm\n",
    "from torchvision.utils import save_image, make_grid"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Model Hyperparameters\n",
    "\n",
    "dataset_path = '~/datasets'\n",
    "\n",
    "cuda = True\n",
    "DEVICE = torch.device(\"cuda\" if cuda else \"cpu\")\n",
    "\n",
    "\n",
    "batch_size = 100\n",
    "\n",
    "x_dim  = 784\n",
    "hidden_dim = 400\n",
    "latent_dim = 200\n",
    "\n",
    "lr = 1e-3\n",
    "\n",
    "epochs = 30"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "###    Step 1. Load (or download) Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torchvision.datasets import MNIST\n",
    "import torchvision.transforms as transforms\n",
    "from torch.utils.data import DataLoader\n",
    "\n",
    "\n",
    "mnist_transform = transforms.Compose([\n",
    "        transforms.ToTensor(),\n",
    "])\n",
    "\n",
    "kwargs = {'num_workers': 1, 'pin_memory': True} \n",
    "\n",
    "train_dataset = MNIST(dataset_path, transform=mnist_transform, train=True, download=True)\n",
    "test_dataset  = MNIST(dataset_path, transform=mnist_transform, train=False, download=True)\n",
    "\n",
    "train_loader = DataLoader(dataset=train_dataset, batch_size=batch_size, shuffle=True, **kwargs)\n",
    "test_loader  = DataLoader(dataset=test_dataset,  batch_size=batch_size, shuffle=False, **kwargs)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Step 2. Define our model: Variational AutoEncoder (VAE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "    A simple implementation of Gaussian MLP Encoder and Decoder\n",
    "\"\"\"\n",
    "\n",
    "class Encoder(nn.Module):\n",
    "    \n",
    "    def __init__(self, input_dim, hidden_dim, latent_dim):\n",
    "        super(Encoder, self).__init__()\n",
    "\n",
    "        self.FC_input = nn.Linear(input_dim, hidden_dim)\n",
    "        self.FC_input2 = nn.Linear(hidden_dim, hidden_dim)\n",
    "        self.FC_mean  = nn.Linear(hidden_dim, latent_dim)\n",
    "        self.FC_var   = nn.Linear (hidden_dim, latent_dim)\n",
    "        \n",
    "        self.LeakyReLU = nn.LeakyReLU(0.2)\n",
    "        \n",
    "        self.training = True\n",
    "        \n",
    "    def forward(self, x):\n",
    "        h_       = self.LeakyReLU(self.FC_input(x))\n",
    "        h_       = self.LeakyReLU(self.FC_input2(h_))\n",
    "        mean     = self.FC_mean(h_)\n",
    "        log_var  = self.FC_var(h_)                     # encoder produces mean and log of variance \n",
    "                                                       #             (i.e., parateters of simple tractable normal distribution \"q\"\n",
    "        \n",
    "        return mean, log_var"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Decoder(nn.Module):\n",
    "    def __init__(self, latent_dim, hidden_dim, output_dim):\n",
    "        super(Decoder, self).__init__()\n",
    "        self.FC_hidden = nn.Linear(latent_dim, hidden_dim)\n",
    "        self.FC_hidden2 = nn.Linear(hidden_dim, hidden_dim)\n",
    "        self.FC_output = nn.Linear(hidden_dim, output_dim)\n",
    "        \n",
    "        self.LeakyReLU = nn.LeakyReLU(0.2)\n",
    "        \n",
    "    def forward(self, x):\n",
    "        h     = self.LeakyReLU(self.FC_hidden(x))\n",
    "        h     = self.LeakyReLU(self.FC_hidden2(h))\n",
    "        \n",
    "        x_hat = torch.sigmoid(self.FC_output(h))\n",
    "        return x_hat\n",
    "        "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Model(nn.Module):\n",
    "    def __init__(self, Encoder, Decoder):\n",
    "        super(Model, self).__init__()\n",
    "        self.Encoder = Encoder\n",
    "        self.Decoder = Decoder\n",
    "        \n",
    "    def reparameterization(self, mean, var):\n",
    "        epsilon = torch.randn_like(var).to(DEVICE)        # sampling epsilon        \n",
    "        z = mean + var*epsilon                          # reparameterization trick\n",
    "        return z\n",
    "        \n",
    "                \n",
    "    def forward(self, x):\n",
    "        mean, log_var = self.Encoder(x)\n",
    "        z = self.reparameterization(mean, torch.exp(0.5 * log_var)) # takes exponential function (log var -> var)\n",
    "        x_hat            = self.Decoder(z)\n",
    "        \n",
    "        return x_hat, mean, log_var"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "encoder = Encoder(input_dim=x_dim, hidden_dim=hidden_dim, latent_dim=latent_dim)\n",
    "decoder = Decoder(latent_dim=latent_dim, hidden_dim = hidden_dim, output_dim = x_dim)\n",
    "\n",
    "model = Model(Encoder=encoder, Decoder=decoder).to(DEVICE)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Step 3. Define Loss function (reprod. loss) and optimizer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torch.optim import Adam\n",
    "\n",
    "BCE_loss = nn.BCELoss()\n",
    "\n",
    "def loss_function(x, x_hat, mean, log_var):\n",
    "    reproduction_loss = nn.functional.binary_cross_entropy(x_hat, x, reduction='sum')\n",
    "    KLD      = - 0.5 * torch.sum(1+ log_var - mean.pow(2) - log_var.exp())\n",
    "\n",
    "    return reproduction_loss + KLD\n",
    "\n",
    "\n",
    "optimizer = Adam(model.parameters(), lr=lr)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Step 4. Train Variational AutoEncoder (VAE)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Start training VAE...\n",
      "\tEpoch 1 complete! \tAverage Loss:  173.12485116796222\n",
      "\tEpoch 2 complete! \tAverage Loss:  128.31116850023477\n",
      "\tEpoch 3 complete! \tAverage Loss:  117.0532701801179\n",
      "\tEpoch 4 complete! \tAverage Loss:  112.62500634195013\n",
      "\tEpoch 5 complete! \tAverage Loss:  109.84031165223288\n",
      "\tEpoch 6 complete! \tAverage Loss:  108.16002750352149\n",
      "\tEpoch 7 complete! \tAverage Loss:  107.07083751239044\n",
      "\tEpoch 8 complete! \tAverage Loss:  106.1414350740818\n",
      "\tEpoch 9 complete! \tAverage Loss:  105.50088035723601\n",
      "\tEpoch 10 complete! \tAverage Loss:  104.8424516609714\n",
      "\tEpoch 11 complete! \tAverage Loss:  104.30121563282555\n",
      "\tEpoch 12 complete! \tAverage Loss:  103.77578454324916\n",
      "\tEpoch 13 complete! \tAverage Loss:  103.46035459489775\n",
      "\tEpoch 14 complete! \tAverage Loss:  103.11576248500104\n",
      "\tEpoch 15 complete! \tAverage Loss:  102.79475204442299\n",
      "\tEpoch 16 complete! \tAverage Loss:  102.45230698625313\n",
      "\tEpoch 17 complete! \tAverage Loss:  102.15569195729863\n",
      "\tEpoch 18 complete! \tAverage Loss:  101.95428802235497\n",
      "\tEpoch 19 complete! \tAverage Loss:  101.75358199538293\n",
      "\tEpoch 20 complete! \tAverage Loss:  101.61855894263877\n",
      "\tEpoch 21 complete! \tAverage Loss:  101.41381663123435\n",
      "\tEpoch 22 complete! \tAverage Loss:  101.24662673466193\n",
      "\tEpoch 23 complete! \tAverage Loss:  101.1379146885434\n",
      "\tEpoch 24 complete! \tAverage Loss:  101.01240045257721\n",
      "\tEpoch 25 complete! \tAverage Loss:  100.85082658923727\n",
      "\tEpoch 26 complete! \tAverage Loss:  100.74188830342237\n",
      "\tEpoch 27 complete! \tAverage Loss:  100.65018820429883\n",
      "\tEpoch 28 complete! \tAverage Loss:  100.53461661362688\n",
      "\tEpoch 29 complete! \tAverage Loss:  100.39944208772434\n",
      "\tEpoch 30 complete! \tAverage Loss:  100.33088696003756\n",
      "Finish!!\n"
     ]
    }
   ],
   "source": [
    "print(\"Start training VAE...\")\n",
    "model.train()\n",
    "\n",
    "for epoch in range(epochs):\n",
    "    overall_loss = 0\n",
    "    for batch_idx, (x, _) in enumerate(train_loader):\n",
    "        x = x.view(batch_size, x_dim)\n",
    "        x = x.to(DEVICE)\n",
    "\n",
    "        optimizer.zero_grad()\n",
    "\n",
    "        x_hat, mean, log_var = model(x)\n",
    "        loss = loss_function(x, x_hat, mean, log_var)\n",
    "        \n",
    "        overall_loss += loss.item()\n",
    "        \n",
    "        loss.backward()\n",
    "        optimizer.step()\n",
    "        \n",
    "    print(\"\\tEpoch\", epoch + 1, \"complete!\", \"\\tAverage Loss: \", overall_loss / (batch_idx*batch_size))\n",
    "    \n",
    "print(\"Finish!!\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Step 5. Generate images from test dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "  0%|          | 0/100 [00:00<?, ?it/s]\n"
     ]
    }
   ],
   "source": [
    "model.eval()\n",
    "\n",
    "with torch.no_grad():\n",
    "    for batch_idx, (x, _) in enumerate(tqdm(test_loader)):\n",
    "        x = x.view(batch_size, x_dim)\n",
    "        x = x.to(DEVICE)\n",
    "        \n",
    "        x_hat, _, _ = model(x)\n",
    "\n",
    "\n",
    "        break"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "def show_image(x, idx):\n",
    "    x = x.view(batch_size, 28, 28)\n",
    "\n",
    "    fig = plt.figure()\n",
    "    plt.imshow(x[idx].cpu().numpy())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAPsAAAD4CAYAAAAq5pAIAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuNCwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy8QVMy6AAAACXBIWXMAAAsTAAALEwEAmpwYAAANMElEQVR4nO3dDYwU5R3H8R3gOLwDW07kSgHFFzRFq9gg1GAbDNWixqDxpZKU2MR6ptUEG9PW0BhJ2lhqfAmt1uSsVDSItaKRKLXSiy0xtSdgAHlrQYoVetwJZwu+3QtMfw8dzIo7z8Ht7Mzc/b+f5J9nZp7d24fhfjezO7v7BGEYFgD0fwOyHgCAdBB2wAjCDhhB2AEjCDtgxKA0H2xwUB0OKdSm+ZCAKR8XPih0hh1B4mEPgmCGmgWqgarf6DLefN/tXdCnBNPLeUgAHs1hU/Kn8Qq6C/hDqktVE1SztM21APrZc/bJqm06mm9XdWr5KdXMZIYFIE9hH616p2h9Z7TtU3S0b1CtdtVV6Cjj4QDk+tV4HfUbVZNcVRWqK/1wACoQ9l2qsUXrY6JtAPpZ2Fepxuv0/BTVYC1fr1qWzLAAJK3Xl950Wt6tkN+qxT+q3CvzC7VtY2IjA5Cosq6zK9zL1bgCkHO8XRYwgrADRhB2wAjCDhhB2AEjCDtgBGEHjCDsgBGEHTCCsANGEHbACMIOGEHYASMIO2AEYQeMIOyAEYQdMIKwA0YQdsAIwg4YQdgBIwg7YARhB4wg7IARhB0wgrADRhB2wAjCDhhB2AEjCDtgRFlTNgdBsEPNftUBVXcYhpMSGRWAfIU9cpFCvieBnwOggjiNB4woN+yh6mWdzq9RNZS6gduuWu2qq9BR5sMByOo0/kKdwu9SkEdqeYXaLVpfWXwDrTeqcVU4PqhzfxwA9LUjuwt61LapeU41OYlBAchR2HUUr1UNO7ys5hLVhqQGBiA/p/H1qucU9MM/50kd4V9KZFQA8hN2BXu7mnMTHAuACuLSG2AEYQeMIOyAEYQdMIKwA0Yk8UEYE/bedEFs30mzt3nvu6XNXaWM19lR5e0fvcTfX7Pz/di+g2s3ee8LOziyA0YQdsAIwg4YQdgBIwZkPQAA6SDsgBGEHTCC6+xH6Uc/fDK27+ra9/x3Pu2o/z9Km+bv3tH9YWzfgncvKvPB+67X206O7au973Pe+w5qWpP0cDLHkR0wgrADRhB2wAjCDhhB2AEjCDtgBGEHjAjCML1JWtyMMFOC6ak9XpI+uGZKbN+ec/x/M4dv9u/j97506Ou4Yw0+5z/e/nvOfja27+LjPvLe98UPh3r7L6+J/6x8uT4KO739zR1uOoJ404Z09fqxT3/xZm//GQ2rev2zs9QcNhX2he0lf6E4sgNGEHbACMIOGEHYASMIO2AEYQeMIOyAEXye/SjVPtPs6SvvP+H48u5e+NUX4j/w/rOp4/yP/Rf/d97fM+30Xo3paAz66KC3v3Z9i7f/hJVLvf1fHhz/ffs1O/zfxW/yyB4EwUJVm2pD0bY61QrV1qgdXtlhAkjjNP4x1Ywjtt2hatK778a7NloH0JfDrkCvVNN+xOaZqkXRsmuvTHhcAHLynL1efwQOP6Ha7dbjbqhT/AY1rgpDCjW9fDgAmb8ar9C7T3nEftJD3Y2qSa6qCtXlPhyAlMPeqiP2KLcQtW29/DkAch72ZaobomXXPp/McABk9pxdR+4latyF3BFa3qn2LtV81dNav1Ht26rrKjVA9Kx7d2tsX+3S+D7nQA8/u/aZvZn9F7R+9wJv/1mD/b++97afGds37rfbvfft9vb207DrufasmK6++S0UgFG8XRYwgrADRhB2wAjCDhhB2AEj+IgrsvvlO3mst//BuQ96+6uCgd7+3y/4RmzfCS2vee/bH3FkB4wg7IARhB0wgrADRhB2wAjCDhhB2AEjuM6OzGz5wWhv//nV/qmsN3b6p6Ou2/ThMY+pP+PIDhhB2AEjCDtgBGEHjCDsgBGEHTCCsANGcJ0dFdVx+fmxfW9c80AP9/bPIPS9OXO8/cf99fUefr4tHNkBIwg7YARhB4wg7IARhB0wgrADRhB2wAius6Oi/nVp/PFkaOC/jj7rnxd7+2teWuftD7299vR4ZA+CYKGqTbWhaNs81S7V2qguq+wwAaRxGv+YakaJ7Q+EYTgxquXlDgRAxmFXkFeqaa/sMADk+QW6W3X6vj46zR8edyP1NahWu+oqdJTxcACyCPvDqtNUE1Utqvs8ZwaNqkmuqnr4YAOAnIVdwW1VHVAd1OojqsnJDgtALsKuU/JRRatXqT55pR5AH73OrmAvUTNNNULLO9Xe5da1PDG6lLlDdXNFR4ncGjBsmLd/9tdeje3bd/Bj733b7j7V21/dscrbj2MMu07VZ5XY/GhP9wOQL7xdFjCCsANGEHbACMIOGEHYASP4iCvKsnXeWd7+F0b8OrZv5tarvfetXs6ltSRxZAeMIOyAEYQdMIKwA0YQdsAIwg4YQdgBI7jODq//fvur3v713/qlt/+t7q7Yvvd/McZ73+pDX4KEpHBkB4wg7IARhB0wgrADRhB2wAjCDhhB2AEjuM5u3KDRX/T233bn77z91YH/V+j6dbNj+078A59XTxNHdsAIwg4YQdgBIwg7YARhB4wg7IARhB0wguvs/VwwyP9ffO4LbhbueNcO3evtX7x/pLe//s7448lB7z2R+pE9CIKxqldUm1QbVXOi7XWqFaqtUTs86cEBSPc0vlt1exiGE9S6ry25RcF2y3eomrR9vGujdQB9NewKc4vqjWh5v5rNqtGqmapF0c1ce2WlBgkg5efsOqKPU3OeqllV7/4QRF273XrMfRrUuCoMKdT0fqQA0nk1XqEdqmap6jaFfF9xn9ZD15S6n7oaVZNcVRWqyxosgAqHXUGvioK+WKF9Ntrcqu2jon7XtvV+GAAyP41XkAM1j6o2K+j3F3UtU92gmh+1z1dkhCjPuWd6u3868omyfvxDd1/r7f/8utfK+vlI9zn7VJX7UPKbyv3aaNvcKORPa9uNat9WXZfcsACkHnYdzV9V447upUxPdjgAKoW3ywJGEHbACMIOGEHYASMIO2AEH3HtBwZOOCO2r+Gp8t7+MGHhLd7+cU/8rayfj/RwZAeMIOyAEYQdMIKwA0YQdsAIwg4YQdgBI7jO3g9s+X78F/teUfOpLxU6ZmP+3Om/waEvKUJfwJEdMIKwA0YQdsAIwg4YQdgBIwg7YARhB4zgOnsf8PEVk739TVfc5+llyi38H0d2wAjCDhhB2AEjCDtgBGEHjCDsgBGEHTDiaOZnH6vmcVW9yn14uTEMwwXaPk/LN6nejW46V9uXV2ykhv176kBv/0mDen8tffH+kd7+qn3+z7Pzafb+9aaabtXtCvIbCvgwLa9RuyLqe0Db763c8ACkOT97i5qWaHm/gr5Zi6OTGgCAHD5nV9DHqTlP1RxtulXb1qsWqkp+N5K2N6hWu+oqdJQ5XAAVD7vCOlTNUtVtOsK7LzZ7WHWaamJ05C/5Bm3d1j3Hn+SqqlDd23ECSCPsCnpVFPTFCu2zbpvaVtUB1UGtPqLyf1oDQL7DrqAHah5VbVaw7y/aPqroZlepNiQ/PABpvho/VTVb9aYCvjbaNlc1S+sTo6svO1Q3JzUoJOfneyd4+1/7pnsZJl7Y8ib/HYZejX9VjTu6H4lr6kAfwjvoACMIO2AEYQeMIOyAEYQdMIKwA0YEurSW2oMdH9SFU4LpqT0eYE1z2FTYF7aXulTOkR2wgtN4wAjCDhhB2AEjCDtgBGEHjCDsgBGpXmcPgsB97fTbRZtGqPakNoBjk9ex5XVcDmPLfr+drEyfmHnYP/PgQbDafTddZgPwyOvY8jouh7Hle79xGg8YQdgBIwZk/PiNGT++T17HltdxOYwtx/st0+fsAOwc2QGkhLADRgzI6FLDDNXfVdtUd2Qxhjgazw7Voe/Id5dEMh6Lm0OvTfXJBBxarnOz6Kq2Ru3wHI1tnmpXtO9cXZbR2MaqXlFtUm1UzcnDvvOMK5X9lvpzdv1D3GTj/1BdrNqpWqWapXFsSnUgnrCrcXPT7cnBWL6u5n3V4xrP2dG2e9S0a31+9IdyuJZ/nJOxzXPbsp7GO5qtaFTxNOOFQuFK1Xey3HeecV2Xxn7L4sju5oTbpn/YdlWnlp9SzcxgHLmn/bNSTfsRm92+WhQtL4p+WfIytlxw04y7QEXL+9VsjqYZz3TfecaViizC7v5x7xSt78zZfO/uVOdl/eVd46abznowJdS7X5poebdbz3IwJfQ4jXeajphmvD4v+64305+XixfoPutC/UJ8Re2lqlui09Vc0jjdH6Y8XTs9qmm801JimvFc7LveTn/eF8O+SzW2aH1MtC0XtPMPjUVtm5rncjgVdevhGXSj1o0zF/I0jXepacbzsO+ynP48i7C7F+TG6x99imqwlq9XLctgHJ+h8dRGL5wcWlZzSQ6nonb76oZo2bXPZziWT8nLNN5x04xnve8yn/7cnc2kXXJZ9Ir8W6qfZDGGmHGdqloX1casxyZLotO6rui1jRtVJ6iaVFtVf1LV5WhsT6jcHM/ro2CNymhsF7pf7WgcbprxtdHvXKb7zjOuVPYbb5cFjOAFOsAIwg4YQdgBIwg7YARhB4wg7IARhB0w4n9BCfQT/O8q4AAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_image(x, idx=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_image(x_hat, idx=0)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Step 6. Generate image from noise vector"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Please note that this is not the correct generative process.**\n",
    "\n",
    "* Even if we don't know exact p(z|x), we can generate images from noise, since the loss function of training VAE regulates the q(z|x) (simple and tractable posteriors) must close enough to N(0, I). If q(z|x) is close to N(0, I) \"enough\"(but not tightly close due to posterior collapse problem), N(0, I) may replace the encoder of VAE.\n",
    "\n",
    "* To show this, I just tested with a noise vector sampled from N(0, I) similar with Generative Adversarial Network."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "with torch.no_grad():\n",
    "    noise = torch.randn(batch_size, latent_dim).to(DEVICE)\n",
    "    generated_images = decoder(noise)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "save_image(generated_images.view(batch_size, 1, 28, 28), 'generated_sample.png')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_image(generated_images, idx=12)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_image(generated_images, idx=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_image(generated_images, idx=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_image(generated_images, idx=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_image(generated_images, idx=20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "show_image(generated_images, idx=50)"
   ]
  },
  {
   "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": "3.7.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}