mdn-rnn.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "import torch.nn as nn\n",
    "import torch.nn.functional as F\n",
    "from torch.nn.utils import clip_grad_norm_\n",
    "from torch.autograd import Variable\n",
    "from torchsummary import summary\n",
    "\n",
    "from torchvision.utils import save_image\n",
    "from IPython.core.display import Image, display\n",
    "\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "from vae import VAE\n",
    "\n",
    "%load_ext autoreload\n",
    "%autoreload 2"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "device(type='cuda')"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Device configuration\n",
    "device = torch.device('cuda' if torch.cuda.is_available() else 'cpu')\n",
    "device"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "bsz = 200\n",
    "epochs = 500\n",
    "seqlen = 16\n",
    "\n",
    "z_size = 32\n",
    "n_hidden = 256\n",
    "n_gaussians = 5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "torch.Size([200, 150, 32])"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "z = torch.from_numpy(np.load('z.npy'))\n",
    "z = z.view(bsz, -1, z.size(1)).to(device)\n",
    "z.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Truncated backpropagation\n",
    "def detach(states):\n",
    "    return [state.detach() for state in states] "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "class MDNRNN(nn.Module):\n",
    "    def __init__(self, z_size, n_hidden=256, n_gaussians=5, n_layers=1):\n",
    "        super(MDNRNN, self).__init__()\n",
    "\n",
    "        self.z_size = z_size\n",
    "        self.n_hidden = n_hidden\n",
    "        self.n_gaussians = n_gaussians\n",
    "        self.n_layers = n_layers\n",
    "        \n",
    "        self.lstm = nn.LSTM(z_size, n_hidden, n_layers, batch_first=True)\n",
    "        self.fc1 = nn.Linear(n_hidden, n_gaussians*z_size)\n",
    "        self.fc2 = nn.Linear(n_hidden, n_gaussians*z_size)\n",
    "        self.fc3 = nn.Linear(n_hidden, n_gaussians*z_size)\n",
    "        \n",
    "    def get_mixture_coef(self, y):\n",
    "        rollout_length = y.size(1)\n",
    "        pi, mu, sigma = self.fc1(y), self.fc2(y), self.fc3(y)\n",
    "        \n",
    "        pi = pi.view(-1, rollout_length, self.n_gaussians, self.z_size)\n",
    "        mu = mu.view(-1, rollout_length, self.n_gaussians, self.z_size)\n",
    "        sigma = sigma.view(-1, rollout_length, self.n_gaussians, self.z_size)\n",
    "        \n",
    "        pi = F.softmax(pi, 2)\n",
    "        sigma = torch.exp(sigma)\n",
    "        return pi, mu, sigma\n",
    "        \n",
    "        \n",
    "    def forward(self, x, h):\n",
    "        # Forward propagate LSTM\n",
    "        y, (h, c) = self.lstm(x, h)\n",
    "        pi, mu, sigma = self.get_mixture_coef(y)\n",
    "        return (pi, mu, sigma), (h, c)\n",
    "    \n",
    "    def init_hidden(self, bsz):\n",
    "        return (torch.zeros(self.n_layers, bsz, self.n_hidden).to(device),\n",
    "                torch.zeros(self.n_layers, bsz, self.n_hidden).to(device))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "model = MDNRNN(z_size, n_hidden).to(device)\n",
    "# model.load_state_dict(torch.load('rnn.torch', map_location='cpu'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# inputs = z[:bsz, :seqlen, :]\n",
    "# (pi, mu, sigma), (h, c) = model(inputs, model.init_hidden(bsz))\n",
    "# inputs.shape, pi.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "def mdn_loss_fn(y, pi, mu, sigma):\n",
    "    m = torch.distributions.Normal(loc=mu, scale=sigma)\n",
    "    loss = torch.exp(m.log_prob(y))\n",
    "    loss = torch.sum(loss * pi, dim=2)\n",
    "    loss = -torch.log(loss)\n",
    "    return loss.mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "def criterion(y, pi, mu, sigma):\n",
    "    y = y.unsqueeze(2)\n",
    "    return mdn_loss_fn(y, pi, mu, sigma)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "optimizer = torch.optim.Adam(model.parameters())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch [0/500], Loss: -1.2711\n",
      "Epoch [100/500], Loss: -1.3434\n",
      "Epoch [200/500], Loss: -1.3428\n",
      "Epoch [300/500], Loss: -1.3833\n",
      "Epoch [400/500], Loss: -1.4378\n"
     ]
    }
   ],
   "source": [
    "# Train the model\n",
    "for epoch in range(epochs):\n",
    "    # Set initial hidden and cell states\n",
    "    hidden = model.init_hidden(bsz)\n",
    "    \n",
    "    for i in range(0, z.size(1) - seqlen, seqlen):\n",
    "        # Get mini-batch inputs and targets\n",
    "        inputs = z[:, i:i+seqlen, :]\n",
    "        targets = z[:, (i+1):(i+1)+seqlen, :]\n",
    "        \n",
    "        # Forward pass\n",
    "        hidden = detach(hidden)\n",
    "        (pi, mu, sigma), hidden = model(inputs, hidden)\n",
    "        loss = criterion(targets, pi, mu, sigma)\n",
    "        \n",
    "        # Backward and optimize\n",
    "        model.zero_grad()\n",
    "        loss.backward()\n",
    "        # clip_grad_norm_(model.parameters(), 0.5)\n",
    "        optimizer.step()\n",
    "        \n",
    "    if epoch % 100 == 0:\n",
    "        print ('Epoch [{}/{}], Loss: {:.4f}'\n",
    "               .format(epoch, epochs, loss.item()))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "# np.linalg.norm(z[0, 0, :].cpu().data - z1[0, :, :].cpu().data, axis=1)\n",
    "# np.linalg.norm(inputs.cpu().data - targets.cpu().data)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### RNN prediction\n",
    "There are 5 modes of gaussians. Below is next true sample and 5 modes of predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "vae = VAE(image_channels=3).to(device)\n",
    "vae.load_state_dict(torch.load('vae.torch', map_location='cpu'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<IPython.core.display.Image object>"
      ]
     },
     "metadata": {
      "image/png": {
       "unconfined": true,
       "width": 1000
      }
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "zero = np.random.randint(z.size(0))\n",
    "one = np.random.randint(z.size(1))\n",
    "x = z[zero:zero+1, one:one+1, :]\n",
    "y = z[zero:zero+1, one+1:one+2, :]\n",
    "\n",
    "hidden = model.init_hidden(1)\n",
    "(pi, mu, sigma), _ = model(x, hidden)\n",
    "\n",
    "y_preds = [torch.normal(mu, sigma)[:, :, i, :] for i in range(n_gaussians)]\n",
    "\n",
    "compare_x = vae.decode(torch.cat([x, y] + y_preds))\n",
    "\n",
    "# compare_x = vae.decode(z_out)\n",
    "save_image(compare_x.data.cpu(), 'sample_image.png')\n",
    "display(Image('sample_image.png', width=1000, unconfined=True))"
   ]
  },
  {
   "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}