Variational_Autoencoder _base_with_base_distributions..ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\ps1109\\AppData\\Local\\Continuum\\Anaconda3\\envs\\torch\\lib\\site-packages\\requests\\__init__.py:102: RequestsDependencyWarning: urllib3 (1.26.14) or chardet (5.1.0)/charset_normalizer (2.0.12) doesn't match a supported version!\n",
      "  warnings.warn(\"urllib3 ({}) or chardet ({})/charset_normalizer ({}) doesn't match a supported \"\n"
     ]
    }
   ],
   "source": [
    "# Source : https://towardsdatascience.com/variational-autoencoder-demystified-with-pytorch-implementation-3a06bee395ed\n",
    "\n",
    "# Adapted version for MNIST and Pytorch\n",
    "\n",
    "import torch\n",
    "import torch.nn.functional as F\n",
    "from torch import nn \n",
    "\n",
    "import torchvision.datasets as datasets\n",
    "from tqdm import tqdm \n",
    "from torchvision import transforms \n",
    "from torchvision.utils import save_image\n",
    "from torch.utils.data import DataLoader\n",
    "from torch import optim"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "DEVICE = torch.device(\"cpu\")#\"cuda\" if torch.cuda.is_available() else \"cpu\")\n",
    "INPUT_DIM = 784\n",
    "H_DIM = 200\n",
    "Z_DIM = 2\n",
    "NUM_EPOCHS  = 500\n",
    "BATCH_SIZE = 32\n",
    "LR_RATE = 1e-4"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "def gaussian_likelihood(x_hat, logscale, x):\n",
    "        scale = torch.exp(logscale)\n",
    "        mean = x_hat\n",
    "        dist = torch.distributions.Normal(mean, scale)\n",
    "\n",
    "        # measure prob of seeing image under p(x|z)\n",
    "        log_pxz = dist.log_prob(x)\n",
    "        return log_pxz.sum(dim=(1, 2, 3))\n",
    "\n",
    "def kl_divergence(z, mu, std):\n",
    "    # --------------------------\n",
    "    # Monte carlo KL divergence\n",
    "    # --------------------------\n",
    "    # 1. define the first two probabilities (in this case Normal for both)\n",
    "    p = torch.distributions.Normal(torch.zeros_like(mu), torch.ones_like(std))\n",
    "    q = torch.distributions.Normal(mu, std)\n",
    "\n",
    "    # 2. get the probabilities from the equation\n",
    "    log_qzx = q.log_prob(z)\n",
    "    log_pz = p.log_prob(z)\n",
    "\n",
    "    # kl\n",
    "    kl = (log_qzx - log_pz)\n",
    "    kl = kl.sum(-1)\n",
    "    return kl"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "class VariationalAutoencoder(nn.Module):\n",
    "    def __init__(self, input_dim= 1,  latent_dim = 20) -> None:\n",
    "        super().__init__()\n",
    "        self.latent_dim = latent_dim\n",
    "        self.activation = nn.ReLU()\n",
    "        self.log_scale = nn.Parameter(torch.Tensor([0.0]))\n",
    "        \n",
    "        self.conv1 = nn.Conv2d(input_dim, 32, kernel_size=3, stride=2, padding = 0)\n",
    "        self.conv2 = nn.Conv2d(32, 64, kernel_size=3, stride=1, padding = 1)\n",
    "        self.conv3 = nn.Conv2d(64, 64, kernel_size=2, stride=2, padding = 1)\n",
    "        self.linear_mu = nn.Linear(3136, self.latent_dim)\n",
    "        self.linear_sigma = nn.Linear(3136, self.latent_dim)\n",
    "        \n",
    "        self.Linear1 = nn.Linear(self.latent_dim, 32* 7* 7)\n",
    "        self.deconv1 = nn.ConvTranspose2d(32, 64, kernel_size=2, stride=2, padding=0)\n",
    "        self.deconv2 = nn.ConvTranspose2d(64, 32, kernel_size=2, stride=2, padding=0)\n",
    "        self.deconv3 = nn.ConvTranspose2d(32, 1, kernel_size=1, stride=1, padding=0)\n",
    "        # Decoder Layesr \n",
    "        \n",
    "    def forward(self, x):\n",
    "        x = self.activation(self.conv1(x))\n",
    "        x = self.activation(self.conv2(x))\n",
    "        x = self.activation(self.conv3(x))\n",
    "        x = x.view(x.shape[0], -1)\n",
    "        mu = self.linear_mu(x)\n",
    "        x_sigma = self.linear_sigma(x)\n",
    "\n",
    "        # sample z from q\n",
    "        std = torch.exp(x_sigma / 2)\n",
    "        q = torch.distributions.Normal(mu, std)\n",
    "        z= q.rsample()\n",
    "        \n",
    "        x = self.Linear1(z)\n",
    "        x = x.view(x.shape[0], 32, 7, 7)\n",
    "        x = self.activation(self.deconv1(x))\n",
    "        x = self.activation(self.deconv2(x))\n",
    "        x_hat = self.activation(self.deconv3(x))\n",
    "        return x_hat, z, mu, std, self.log_scale"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\Users\\ps1109\\AppData\\Local\\Continuum\\Anaconda3\\envs\\torch\\lib\\site-packages\\torch\\nn\\_reduction.py:42: UserWarning: size_average and reduce args will be deprecated, please use reduction='mean' instead.\n",
      "  warnings.warn(warning.format(ret))\n"
     ]
    }
   ],
   "source": [
    "dataset = datasets.MNIST(root = \"dataset/\", train = True, transform=transforms.ToTensor(), download=True)\n",
    "train_loader = DataLoader(dataset = dataset, batch_size = BATCH_SIZE, shuffle=True)\n",
    "model = VariationalAutoencoder(input_dim= 1,  latent_dim = Z_DIM).to(device=DEVICE)\n",
    "optimizer = optim.Adam(model.parameters(), lr =  LR_RATE)\n",
    "loss_fn = nn.BCELoss(reduce=\"sum\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "1875it [01:32, 20.29it/s, loss=612]t/s]\n",
      "1875it [01:25, 21.88it/s, loss=485]:31, 92.41s/it]\n",
      "1875it [01:25, 22.01it/s, loss=368]:12, 88.46s/it]\n",
      "1875it [01:22, 22.72it/s, loss=255]:22, 86.97s/it]\n",
      "1875it [01:30, 20.65it/s, loss=179]:27, 85.22s/it]\n",
      "1875it [01:22, 22.69it/s, loss=101]:37, 87.23s/it]\n",
      "1875it [01:20, 23.41it/s, loss=46.6]20, 85.67s/it]\n",
      "1875it [01:31, 20.46it/s, loss=46.8]58, 83.85s/it]\n",
      "1875it [01:28, 21.24it/s, loss=0.375]2, 86.33s/it]\n",
      "1875it [01:35, 19.59it/s, loss=60.5]24, 86.93s/it]\n",
      "1875it [01:26, 21.63it/s, loss=51.1]:08, 89.65s/it]\n",
      "1875it [01:30, 20.61it/s, loss=8.88]:15, 88.74s/it]\n",
      "1875it [01:31, 20.43it/s, loss=-23.1]20, 89.43s/it]\n",
      "1875it [01:28, 21.23it/s, loss=15.6]:36, 90.14s/it]\n",
      "1875it [01:33, 20.13it/s, loss=27.2]:41, 89.59s/it]\n",
      "1875it [01:32, 20.19it/s, loss=18.1]:54, 90.67s/it]\n",
      "1875it [01:25, 21.90it/s, loss=21.5]:41, 91.33s/it]\n",
      "1875it [01:29, 21.02it/s, loss=10.9]:22, 89.61s/it]\n",
      "1875it [01:26, 21.78it/s, loss=3.57]:56, 89.50s/it]\n",
      "1875it [01:29, 20.88it/s, loss=71.6]:16, 88.48s/it]\n",
      "1875it [01:26, 21.62it/s, loss=59.4]:57, 88.87s/it]\n",
      "1875it [01:27, 21.32it/s, loss=-9.7]:19, 88.22s/it]\n",
      "1875it [01:26, 21.72it/s, loss=-30.2]11, 88.14s/it]\n",
      "1875it [01:28, 21.19it/s, loss=6.79]:25, 87.60s/it]\n",
      "1875it [01:26, 21.66it/s, loss=51.8]:06, 87.87s/it]\n",
      "1875it [01:27, 21.50it/s, loss=7.99]:34, 87.48s/it]\n",
      "1875it [01:29, 20.89it/s, loss=35.5]:28, 87.40s/it]\n",
      "1875it [01:30, 20.76it/s, loss=-19.4]32, 88.10s/it]\n",
      "1875it [01:33, 20.04it/s, loss=25.6]:18, 88.77s/it]\n",
      "1875it [01:27, 21.42it/s, loss=-2.65]09, 90.21s/it]\n",
      "1875it [01:27, 21.41it/s, loss=19.7]:20, 89.40s/it]\n",
      "1875it [01:25, 21.89it/s, loss=-26.4]31, 88.85s/it]\n",
      "1875it [01:25, 21.81it/s, loss=-17.4]36, 87.90s/it]\n",
      "1875it [01:27, 21.45it/s, loss=-16.2]37, 87.32s/it]\n",
      "1875it [01:34, 19.82it/s, loss=32.9]:22, 87.35s/it]\n",
      "1875it [01:32, 20.32it/s, loss=9.32]:49, 89.53s/it]\n",
      "1875it [01:30, 20.81it/s, loss=18.1]:44, 90.36s/it]\n",
      "1875it [01:28, 21.23it/s, loss=14.6]:42, 90.29s/it]\n",
      "1875it [01:29, 20.86it/s, loss=11.9]:40, 89.70s/it]\n",
      "1875it [01:29, 20.85it/s, loss=4.79]:39, 89.76s/it]\n",
      "1875it [01:30, 20.70it/s, loss=-5.77]35, 89.82s/it]\n",
      "1875it [01:36, 19.45it/s, loss=-19.4]8:53, 90.05s/it]\n",
      "1875it [01:31, 20.57it/s, loss=-21.2]1:56, 91.96s/it]\n",
      "1875it [01:22, 22.75it/s, loss=-16.6]8:32, 91.71s/it]\n",
      "1875it [01:21, 22.92it/s, loss=38]1:15:50, 88.93s/it]\n",
      "1875it [01:23, 22.49it/s, loss=-3.14]8:10, 86.79s/it]\n",
      "1875it [01:22, 22.80it/s, loss=43.8]48:59, 85.77s/it]\n",
      "1875it [01:27, 21.38it/s, loss=21.6]39:35, 84.72s/it]\n",
      "1875it [01:21, 22.93it/s, loss=7.17]44:59, 85.62s/it]\n",
      "1875it [01:20, 23.40it/s, loss=50.3]34:56, 84.47s/it]\n",
      "1875it [01:23, 22.36it/s, loss=30.4]23:45, 83.17s/it]\n",
      "1875it [01:26, 21.78it/s, loss=36.8]23:56, 83.38s/it]\n",
      "1875it [01:27, 21.32it/s, loss=-16.6]8:36, 84.19s/it]\n",
      "1875it [01:27, 21.55it/s, loss=4.43]35:37, 85.32s/it]\n",
      "1875it [01:26, 21.59it/s, loss=51]0:37:58, 85.83s/it]\n",
      "1875it [01:29, 20.88it/s, loss=64.1]38:46, 86.13s/it]\n",
      "1875it [01:27, 21.51it/s, loss=25.5]45:29, 87.23s/it]\n",
      "1875it [01:31, 20.51it/s, loss=30.2]43:57, 87.22s/it]\n",
      "1875it [01:27, 21.48it/s, loss=-2.88]1:45, 88.47s/it]\n",
      "1875it [01:27, 21.50it/s, loss=-.331]7:39, 88.12s/it]\n",
      "1875it [01:27, 21.54it/s, loss=40.5]44:11, 87.84s/it]\n",
      "1875it [01:27, 21.38it/s, loss=19.5]41:01, 87.61s/it]\n",
      "1875it [01:28, 21.12it/s, loss=8.24]39:44, 87.64s/it]\n",
      "1875it [01:27, 21.52it/s, loss=46.4]40:47, 87.98s/it]\n",
      "1875it [01:26, 21.55it/s, loss=-5.15]7:28, 87.73s/it]\n",
      "1875it [01:27, 21.50it/s, loss=3.19]34:26, 87.51s/it]\n",
      "1875it [01:27, 21.49it/s, loss=-56]:32:20, 87.42s/it]\n",
      "1875it [01:27, 21.52it/s, loss=33.6]30:33, 87.38s/it]\n",
      "1875it [01:28, 21.18it/s, loss=-19.4]8:36, 87.31s/it]\n",
      "1875it [01:28, 21.20it/s, loss=23.7]29:47, 87.67s/it]\n",
      "1875it [01:27, 21.50it/s, loss=-5.13]0:00, 87.91s/it]\n",
      "1875it [01:27, 21.51it/s, loss=32.7]27:05, 87.70s/it]\n",
      "1875it [01:27, 21.51it/s, loss=62.9]24:31, 87.55s/it]\n",
      "1875it [01:27, 21.45it/s, loss=-29.3]2:13, 87.43s/it]\n",
      "1875it [01:27, 21.49it/s, loss=16.5]20:44, 87.43s/it]\n",
      "1875it [01:30, 20.81it/s, loss=6.83]18:52, 87.37s/it]\n",
      "1875it [01:27, 21.43it/s, loss=45.5]23:13, 88.19s/it]\n",
      "1875it [01:27, 21.53it/s, loss=-41.7]0:17, 87.98s/it]\n",
      "1875it [01:27, 21.49it/s, loss=13.4]16:57, 87.72s/it]\n",
      "1875it [01:27, 21.46it/s, loss=18]0:14:29, 87.58s/it]\n",
      "1875it [01:27, 21.55it/s, loss=18.4]12:35, 87.51s/it]\n",
      "1875it [01:27, 21.50it/s, loss=14.8]10:07, 87.37s/it]\n",
      "1875it [01:29, 20.93it/s, loss=-4.78]8:18, 87.32s/it]\n",
      "1875it [01:27, 21.45it/s, loss=6.29]11:34, 88.00s/it]\n",
      "1875it [01:27, 21.52it/s, loss=-3]0:08:51, 87.82s/it]\n",
      "1875it [01:27, 21.49it/s, loss=-19.5]5:59, 87.61s/it]\n",
      "1875it [01:27, 21.52it/s, loss=38.2]03:47, 87.51s/it]\n",
      "1875it [01:27, 21.49it/s, loss=-9.26]1:34, 87.40s/it]\n",
      "1875it [01:27, 21.49it/s, loss=2.7]59:50, 87.36s/it] \n",
      "1875it [01:29, 20.88it/s, loss=-15.1]:12, 87.33s/it]\n",
      "1875it [01:27, 21.50it/s, loss=11.7]01:47, 88.07s/it]\n",
      "1875it [01:27, 21.50it/s, loss=-16.3]:31, 87.80s/it] \n",
      "1875it [01:27, 21.49it/s, loss=-77.7]:52, 87.63s/it]\n",
      "1875it [01:27, 21.44it/s, loss=-17.7]:38, 87.51s/it]\n",
      "1875it [01:27, 21.51it/s, loss=17.4]2:04, 87.50s/it]\n",
      "1875it [01:27, 21.44it/s, loss=18.6]9:58, 87.40s/it]\n",
      "1875it [01:30, 20.70it/s, loss=-26.3]:38, 87.42s/it]\n",
      "1875it [01:27, 21.41it/s, loss=0.88]3:31, 88.37s/it]\n",
      "1875it [01:27, 21.48it/s, loss=12.5]0:28, 88.13s/it]\n",
      "1875it [01:31, 20.39it/s, loss=-11.7]:18, 87.88s/it]\n",
      "1875it [01:27, 21.46it/s, loss=10.5]54:02, 89.11s/it]\n",
      "1875it [01:27, 21.45it/s, loss=9.14]49:06, 88.59s/it]\n",
      "1875it [01:27, 21.31it/s, loss=28.7]45:19, 88.24s/it]\n",
      "1875it [01:29, 21.03it/s, loss=13.8]43:18, 88.16s/it]\n",
      "1875it [01:27, 21.49it/s, loss=3.29]43:48, 88.46s/it]\n",
      "1875it [01:27, 21.46it/s, loss=-34.9]9:59, 88.10s/it]\n",
      "1875it [01:27, 21.42it/s, loss=-24.9]7:06, 87.89s/it]\n",
      "1875it [01:27, 21.43it/s, loss=-33.5]4:59, 87.78s/it]\n",
      "1875it [01:27, 21.47it/s, loss=26.7]32:56, 87.70s/it]\n",
      "1875it [01:29, 20.93it/s, loss=38.7]30:47, 87.59s/it]\n",
      "1875it [01:28, 21.26it/s, loss=22.6]33:16, 88.20s/it]\n",
      "1875it [01:27, 21.45it/s, loss=-46.8]1:48, 88.20s/it]\n",
      "1875it [01:27, 21.41it/s, loss=12.5]28:49, 87.96s/it]\n",
      "1875it [01:27, 21.49it/s, loss=-33.5]6:36, 87.85s/it]\n",
      "1875it [01:27, 21.46it/s, loss=-15.5]3:57, 87.66s/it]\n",
      "1875it [01:27, 21.37it/s, loss=20.6]21:58, 87.58s/it]\n",
      "1875it [01:30, 20.63it/s, loss=-17.5]0:50, 87.63s/it]\n",
      "1875it [01:27, 21.41it/s, loss=-7.41]5:35, 88.61s/it]\n",
      "1875it [01:27, 21.41it/s, loss=21.6]22:08, 88.30s/it]\n",
      "1875it [01:27, 21.47it/s, loss=6.12]19:16, 88.08s/it]\n",
      "1875it [01:27, 21.48it/s, loss=28.6]16:21, 87.85s/it]\n",
      "1875it [01:27, 21.44it/s, loss=19.1]13:51, 87.68s/it]\n",
      "1875it [01:27, 21.46it/s, loss=27.3]11:56, 87.61s/it]\n",
      "1875it [01:29, 20.87it/s, loss=-14.3]0:03, 87.54s/it]\n",
      "1875it [01:27, 21.44it/s, loss=-25.1]2:54, 88.23s/it]\n",
      "1875it [01:27, 21.46it/s, loss=-4.76]9:57, 87.99s/it]\n",
      "1875it [01:27, 21.44it/s, loss=-5.87]7:19, 87.81s/it]\n",
      "1875it [01:27, 21.43it/s, loss=27.5]05:14, 87.71s/it]\n",
      "1875it [01:27, 21.44it/s, loss=-10.7]3:22, 87.64s/it]\n",
      "1875it [01:27, 21.42it/s, loss=60.4]01:36, 87.59s/it]\n",
      "1875it [01:30, 20.74it/s, loss=-26]:00:03, 87.58s/it]\n",
      "1875it [01:27, 21.43it/s, loss=-11.9]3:49, 88.43s/it]\n",
      "1875it [01:27, 21.40it/s, loss=-2.02]0:38, 88.15s/it]\n",
      "1875it [01:27, 21.44it/s, loss=18.7]58:13, 87.99s/it]\n",
      "1875it [01:27, 21.42it/s, loss=-7.18]5:48, 87.84s/it]\n",
      "1875it [01:27, 21.43it/s, loss=-6.77]3:49, 87.75s/it]\n",
      "1875it [01:27, 21.39it/s, loss=-29.3]1:51, 87.67s/it]\n",
      "1875it [01:30, 20.62it/s, loss=-5.17]0:24, 87.67s/it]\n",
      "1875it [01:27, 21.32it/s, loss=35.3]54:50, 88.65s/it]\n",
      "1875it [01:27, 21.32it/s, loss=22.1]52:07, 88.44s/it]\n",
      "1875it [01:27, 21.32it/s, loss=-7.64]9:43, 88.29s/it]\n",
      "1875it [01:31, 20.45it/s, loss=-26]:47:38, 88.18s/it]\n",
      "1875it [01:27, 21.41it/s, loss=28.9]52:25, 89.23s/it]\n",
      "1875it [01:28, 21.17it/s, loss=-10.5]7:57, 88.73s/it]\n",
      "1875it [01:29, 21.05it/s, loss=-5.09]6:10, 88.68s/it]\n",
      "1875it [01:27, 21.43it/s, loss=11.6]45:26, 88.81s/it]\n",
      "1875it [01:27, 21.38it/s, loss=-18.8]1:37, 88.41s/it]\n",
      "1875it [01:27, 21.44it/s, loss=0.841]8:55, 88.20s/it]\n",
      "1875it [01:27, 21.42it/s, loss=9.67]36:07, 87.98s/it]\n",
      "1875it [01:27, 21.40it/s, loss=29.7]33:54, 87.85s/it]\n",
      "1875it [01:29, 20.96it/s, loss=-10.3]2:00, 87.77s/it]\n",
      "1875it [01:28, 21.29it/s, loss=-37.1]3:30, 88.28s/it]\n",
      "1875it [01:27, 21.42it/s, loss=-9.52]1:40, 88.22s/it]\n",
      "1875it [01:27, 21.42it/s, loss=8.82]29:00, 88.01s/it]\n",
      "1875it [01:27, 21.46it/s, loss=-26.3]6:42, 87.87s/it]\n",
      "1875it [01:27, 21.38it/s, loss=-13]:24:24, 87.72s/it]\n",
      "1875it [01:27, 21.44it/s, loss=-14.8]2:53, 87.71s/it]\n",
      "1875it [01:30, 20.67it/s, loss=3.02]21:00, 87.64s/it]\n",
      "1875it [01:27, 21.37it/s, loss=9.63]24:46, 88.56s/it]\n",
      "1875it [01:27, 21.33it/s, loss=-47.5]1:55, 88.32s/it]\n",
      "1875it [01:27, 21.41it/s, loss=6.32]19:46, 88.20s/it]\n",
      "1875it [01:20, 23.33it/s, loss=33]8:17:14, 88.01s/it]\n",
      "1875it [01:07, 27.60it/s, loss=-25.2]2:51, 85.71s/it]\n",
      "1875it [01:07, 27.70it/s, loss=-20.1]1:28, 80.38s/it]\n",
      "1875it [01:07, 27.58it/s, loss=8.19]08:49, 76.58s/it]\n",
      "1875it [01:08, 27.23it/s, loss=-34.2]3:09, 74.00s/it]\n",
      "1875it [01:07, 27.58it/s, loss=-23.5]3:20, 72.46s/it]\n",
      "1875it [01:08, 27.49it/s, loss=-34.5]4:42, 71.12s/it]\n",
      "1875it [01:08, 27.52it/s, loss=-32]:28:41, 70.24s/it]\n",
      "1875it [01:07, 27.66it/s, loss=2.57]24:01, 69.61s/it]\n",
      "1875it [01:07, 27.69it/s, loss=-12.6]9:53, 69.07s/it]\n",
      "1875it [01:07, 27.64it/s, loss=-24.6]6:29, 68.66s/it]\n",
      "1875it [01:07, 27.72it/s, loss=3.86]13:59, 68.41s/it]\n",
      "1875it [01:08, 27.43it/s, loss=-29.6]1:34, 68.18s/it]\n",
      "1875it [01:08, 27.39it/s, loss=-15.6]0:44, 68.24s/it]\n",
      "1875it [01:07, 27.73it/s, loss=36]6:09:58, 68.30s/it]\n",
      "1875it [01:08, 27.49it/s, loss=23.1]07:44, 68.10s/it]\n",
      "1875it [01:07, 27.68it/s, loss=19.5]06:47, 68.13s/it]\n",
      "1875it [01:07, 27.66it/s, loss=-3.76]5:01, 68.02s/it]\n",
      "1875it [01:07, 27.68it/s, loss=-20.4]3:31, 67.95s/it]\n",
      "1875it [01:07, 27.69it/s, loss=4.39]02:01, 67.88s/it]\n",
      "1875it [01:07, 27.59it/s, loss=51.7]00:37, 67.83s/it]\n",
      "1875it [01:09, 27.07it/s, loss=0.0186]:42, 67.87s/it]\n",
      "1875it [01:08, 27.50it/s, loss=34.6]00:46, 68.29s/it]\n",
      "1875it [01:07, 27.64it/s, loss=-37.3]9:29, 68.26s/it]\n",
      "1875it [01:07, 27.67it/s, loss=9.41]57:42, 68.13s/it]\n",
      "1875it [01:07, 27.74it/s, loss=-27.5]5:59, 68.02s/it]\n",
      "1875it [01:07, 27.69it/s, loss=30.2]54:11, 67.90s/it]\n",
      "1875it [01:08, 27.39it/s, loss=56.5]52:45, 67.84s/it]\n",
      "1875it [01:08, 27.23it/s, loss=16.5]52:34, 68.02s/it]\n",
      "1875it [01:07, 27.61it/s, loss=-3.62]2:45, 68.27s/it]\n",
      "1875it [01:09, 27.11it/s, loss=33.5]51:04, 68.17s/it]\n",
      "1875it [01:07, 27.67it/s, loss=31.2]51:27, 68.47s/it]\n",
      "1875it [01:08, 27.57it/s, loss=49.9]49:13, 68.25s/it]\n",
      "1875it [01:07, 27.61it/s, loss=-24.8]7:44, 68.18s/it]\n",
      "1875it [01:07, 27.64it/s, loss=-66]:46:12, 68.11s/it]\n",
      "1875it [01:07, 27.64it/s, loss=-17.7]4:40, 68.03s/it]\n",
      "1875it [01:07, 27.59it/s, loss=19.3]43:16, 67.97s/it]\n",
      "1875it [01:08, 27.56it/s, loss=18.1]42:06, 67.97s/it]\n",
      "1875it [01:07, 27.68it/s, loss=38]5:41:04, 67.99s/it]\n",
      "1875it [01:09, 27.16it/s, loss=11]5:39:33, 67.91s/it]\n",
      "1875it [01:07, 27.73it/s, loss=-43]:40:06, 68.25s/it]\n",
      "1875it [01:07, 27.65it/s, loss=44.5]38:02, 68.06s/it]\n",
      "1875it [01:07, 27.61it/s, loss=34.6]36:31, 67.99s/it]\n",
      "1875it [01:07, 27.69it/s, loss=12.5]35:18, 67.97s/it]\n",
      "1875it [01:07, 27.60it/s, loss=-10.5]3:47, 67.89s/it]\n",
      "1875it [01:08, 27.57it/s, loss=-10.6]2:43, 67.90s/it]\n",
      "1875it [01:07, 27.61it/s, loss=-25.9]1:46, 67.94s/it]\n",
      "1875it [01:07, 27.60it/s, loss=-24.5]0:34, 67.93s/it]\n",
      "1875it [01:09, 27.02it/s, loss=25.9]29:28, 67.93s/it]\n",
      "1875it [01:07, 27.63it/s, loss=35.4]30:27, 68.37s/it]\n",
      "1875it [01:07, 27.66it/s, loss=0.857]8:35, 68.22s/it]\n",
      "1875it [01:07, 27.66it/s, loss=-12.1]6:51, 68.09s/it]\n",
      "1875it [01:07, 27.66it/s, loss=-15.3]5:16, 68.00s/it]\n",
      "1875it [01:08, 27.56it/s, loss=-7.51]3:50, 67.94s/it]\n",
      "1875it [01:07, 27.73it/s, loss=3.05]22:51, 67.97s/it]\n",
      "1875it [01:07, 27.64it/s, loss=20.3]21:12, 67.86s/it]\n",
      "1875it [01:08, 27.55it/s, loss=6.68]20:04, 67.86s/it]\n",
      "1875it [01:08, 27.30it/s, loss=12.2]19:12, 67.92s/it]\n",
      "1875it [01:07, 27.61it/s, loss=16.1]19:10, 68.15s/it]\n",
      "1875it [01:07, 27.71it/s, loss=-17.8]7:43, 68.08s/it]\n",
      "1875it [01:07, 27.67it/s, loss=-38.7]6:00, 67.96s/it]\n",
      "1875it [01:07, 27.71it/s, loss=-38.2]4:35, 67.90s/it]\n",
      "1875it [01:08, 27.53it/s, loss=28.6]13:08, 67.83s/it]\n",
      "1875it [01:07, 27.65it/s, loss=1.12]12:23, 67.91s/it]\n",
      "1875it [01:07, 27.57it/s, loss=18.9]11:08, 67.88s/it]\n",
      "1875it [01:08, 27.40it/s, loss=-13.8]0:10, 67.92s/it]\n",
      "1875it [01:08, 27.41it/s, loss=33.3]09:43, 68.07s/it]\n",
      "1875it [01:07, 27.60it/s, loss=-9.17]9:04, 68.18s/it]\n",
      "1875it [01:07, 27.60it/s, loss=-39]:07:35, 68.10s/it]\n",
      "1875it [01:07, 27.62it/s, loss=5.34]06:15, 68.06s/it]\n",
      "1875it [01:08, 27.52it/s, loss=-3.95]4:53, 68.01s/it]\n",
      "1875it [01:07, 27.60it/s, loss=-11.7]3:55, 68.04s/it]\n",
      "1875it [01:08, 27.53it/s, loss=-10.3]2:38, 68.01s/it]\n",
      "1875it [01:07, 27.65it/s, loss=16]5:01:39, 68.04s/it]\n",
      "1875it [01:09, 27.08it/s, loss=22]5:00:12, 67.97s/it]\n",
      "1875it [01:08, 27.57it/s, loss=-17.5]0:46, 68.36s/it]\n",
      "1875it [01:07, 27.64it/s, loss=-29.7]9:10, 68.25s/it]\n",
      "1875it [01:07, 27.63it/s, loss=-9.9]57:29, 68.13s/it]\n",
      "1875it [01:07, 27.65it/s, loss=35.3]56:00, 68.05s/it]\n",
      "1875it [01:07, 27.67it/s, loss=-8.96]4:35, 67.98s/it]\n",
      "1875it [01:07, 27.65it/s, loss=-8.78]3:12, 67.92s/it]\n",
      "1875it [01:09, 27.08it/s, loss=-19.3]1:55, 67.89s/it]\n",
      "1875it [01:07, 27.62it/s, loss=-38.3]2:32, 68.30s/it]\n",
      "1875it [01:09, 27.09it/s, loss=-7.28]0:52, 68.17s/it]\n",
      "1875it [01:07, 27.63it/s, loss=-32.4]1:03, 68.48s/it]\n",
      "1875it [01:08, 27.56it/s, loss=-10.3]9:08, 68.30s/it]\n",
      "1875it [01:07, 27.58it/s, loss=44.4]47:39, 68.22s/it]\n",
      "1875it [01:07, 27.61it/s, loss=5.06]46:13, 68.15s/it]\n",
      "1875it [01:07, 27.68it/s, loss=0.352]4:47, 68.08s/it]\n",
      "1875it [01:07, 27.68it/s, loss=-11.7]3:14, 67.98s/it]\n",
      "1875it [01:07, 27.66it/s, loss=22.7]41:48, 67.91s/it]\n",
      "1875it [01:07, 27.59it/s, loss=7.53]40:31, 67.87s/it]\n",
      "1875it [01:08, 27.19it/s, loss=-21.9]9:30, 67.90s/it]\n",
      "1875it [01:07, 27.64it/s, loss=22.3]39:40, 68.21s/it]\n",
      "1875it [01:07, 27.62it/s, loss=-15.1]8:04, 68.10s/it]\n",
      "1875it [01:08, 27.55it/s, loss=-28.2]6:40, 68.03s/it]\n",
      "1875it [01:07, 27.62it/s, loss=-.534]5:33, 68.04s/it]\n",
      "1875it [01:07, 27.63it/s, loss=-24.7]4:14, 67.99s/it]\n",
      "1875it [01:08, 27.44it/s, loss=-5.74]2:56, 67.95s/it]\n",
      "1875it [01:07, 27.63it/s, loss=-3.18]2:16, 68.07s/it]\n",
      "1875it [01:08, 27.56it/s, loss=25.3]30:54, 68.01s/it]\n",
      "1875it [01:09, 27.03it/s, loss=1.04]29:48, 68.02s/it]\n",
      "1875it [01:07, 27.65it/s, loss=2.51]30:17, 68.43s/it]\n",
      "1875it [01:08, 27.56it/s, loss=-5.39]8:25, 68.24s/it]\n",
      "1875it [01:08, 27.53it/s, loss=-53]:27:03, 68.18s/it]\n",
      "1875it [01:07, 27.60it/s, loss=-12]:25:49, 68.16s/it]\n",
      "1875it [01:07, 27.59it/s, loss=37.4]24:24, 68.09s/it]\n",
      "1875it [01:07, 27.66it/s, loss=28.2]23:08, 68.05s/it]\n",
      "1875it [01:07, 27.64it/s, loss=-19.3]1:41, 67.97s/it]\n",
      "1875it [01:08, 27.28it/s, loss=24.6]20:23, 67.93s/it]\n",
      "1875it [01:08, 27.31it/s, loss=-34.1]0:10, 68.17s/it]\n",
      "1875it [01:07, 27.66it/s, loss=-12.2]9:35, 68.32s/it]\n",
      "1875it [01:08, 27.53it/s, loss=-33.7]7:51, 68.16s/it]\n",
      "1875it [01:08, 27.38it/s, loss=-10.9]6:39, 68.14s/it]\n",
      "1875it [01:08, 27.42it/s, loss=-35]:15:55, 68.25s/it]\n",
      "1875it [01:08, 27.50it/s, loss=6.01]14:57, 68.29s/it]\n",
      "1875it [01:08, 27.52it/s, loss=-9.67]3:42, 68.26s/it]\n",
      "1875it [01:08, 27.52it/s, loss=1.97]12:25, 68.22s/it]\n",
      "1875it [01:09, 27.04it/s, loss=4.5]:11:10, 68.19s/it]\n",
      "1875it [01:08, 27.42it/s, loss=8.88]11:18, 68.54s/it]\n",
      "1875it [01:09, 26.84it/s, loss=17.4]09:59, 68.49s/it]\n",
      "1875it [01:08, 27.45it/s, loss=32]4:10:19, 68.90s/it]\n",
      "1875it [01:08, 27.55it/s, loss=-33.4]8:33, 68.72s/it]\n",
      "1875it [01:08, 27.48it/s, loss=35.8]06:41, 68.52s/it]\n",
      "1875it [01:08, 27.41it/s, loss=-26.1]5:13, 68.44s/it]\n",
      "1875it [01:08, 27.54it/s, loss=29.2]04:04, 68.43s/it]\n",
      "1875it [01:08, 27.52it/s, loss=49]4:02:33, 68.33s/it]\n",
      "1875it [01:09, 26.88it/s, loss=-55]:01:13, 68.27s/it]\n",
      "1875it [01:07, 27.60it/s, loss=-44.1]1:39, 68.72s/it]\n",
      "1875it [01:08, 27.50it/s, loss=-28.4]9:40, 68.48s/it]\n",
      "1875it [01:08, 27.46it/s, loss=-1.51]8:13, 68.39s/it]\n",
      "1875it [01:08, 27.47it/s, loss=10.3]56:59, 68.36s/it]\n",
      "1875it [01:08, 27.50it/s, loss=42.5]55:45, 68.34s/it]\n",
      "1875it [01:08, 27.48it/s, loss=-53.8]4:27, 68.29s/it]\n",
      "1875it [01:09, 27.16it/s, loss=-19.5]3:15, 68.27s/it]\n",
      "1875it [01:08, 27.25it/s, loss=25.7]52:54, 68.50s/it]\n",
      "1875it [01:09, 27.09it/s, loss=48.7]52:04, 68.59s/it]\n",
      "1875it [01:07, 27.60it/s, loss=-5.56]1:33, 68.78s/it]\n",
      "1875it [01:08, 27.49it/s, loss=12.6]49:34, 68.53s/it]\n",
      "1875it [01:08, 27.55it/s, loss=-12]:48:06, 68.43s/it]\n",
      "1875it [01:08, 27.55it/s, loss=-24.5]6:35, 68.32s/it]\n",
      "1875it [01:08, 27.47it/s, loss=-3.58]5:12, 68.24s/it]\n",
      "1875it [01:08, 27.51it/s, loss=17.8]44:05, 68.25s/it]\n",
      "1875it [01:08, 27.53it/s, loss=-25.2]2:51, 68.22s/it]\n",
      "1875it [01:08, 27.32it/s, loss=-5.62]1:36, 68.19s/it]\n",
      "1875it [01:09, 27.16it/s, loss=6.19]40:53, 68.32s/it]\n",
      "1875it [01:08, 27.53it/s, loss=-19.3]0:27, 68.54s/it]\n",
      "1875it [01:08, 27.49it/s, loss=-56.6]8:53, 68.40s/it]\n",
      "1875it [01:08, 27.48it/s, loss=20]3:37:34, 68.35s/it]\n",
      "1875it [01:08, 27.51it/s, loss=-14.2]6:19, 68.31s/it]\n",
      "1875it [01:08, 27.56it/s, loss=42.7]35:01, 68.26s/it]\n",
      "1875it [01:08, 27.46it/s, loss=-33.2]3:41, 68.20s/it]\n",
      "1875it [01:08, 27.53it/s, loss=14.4]32:38, 68.23s/it]\n",
      "1875it [01:09, 26.88it/s, loss=-10.8]1:23, 68.19s/it]\n",
      "1875it [01:08, 27.37it/s, loss=6.58]31:42, 68.66s/it]\n",
      "1875it [01:08, 27.49it/s, loss=9.98]30:25, 68.62s/it]\n",
      "1875it [01:07, 27.58it/s, loss=31.3]28:54, 68.49s/it]\n",
      "1875it [01:08, 27.51it/s, loss=11.1]27:18, 68.34s/it]\n",
      "1875it [01:08, 27.52it/s, loss=23]3:25:59, 68.29s/it]\n",
      "1875it [01:08, 27.56it/s, loss=-6.64]4:43, 68.24s/it]\n",
      "1875it [01:08, 27.53it/s, loss=-27.1]3:23, 68.18s/it]\n",
      "1875it [01:08, 27.56it/s, loss=-13]:22:11, 68.16s/it]\n",
      "1875it [01:09, 26.99it/s, loss=8.65]20:56, 68.12s/it]\n",
      "1875it [01:08, 27.48it/s, loss=-18]:21:00, 68.52s/it]\n",
      "1875it [01:08, 27.49it/s, loss=-28.1]9:37, 68.44s/it]\n",
      "1875it [01:08, 27.36it/s, loss=-1.71]8:17, 68.38s/it]\n",
      "1875it [01:08, 27.44it/s, loss=22.3]17:17, 68.42s/it]\n",
      "1875it [01:08, 27.52it/s, loss=-.672]6:04, 68.40s/it]\n",
      "1875it [01:08, 27.43it/s, loss=7.05]14:42, 68.32s/it]\n",
      "1875it [01:08, 27.57it/s, loss=16.4]13:36, 68.33s/it]\n",
      "1875it [01:08, 27.34it/s, loss=18.5]12:11, 68.24s/it]\n",
      "1875it [01:09, 26.97it/s, loss=-2.73]1:20, 68.34s/it]\n",
      "1875it [01:08, 27.40it/s, loss=25.3]11:12, 68.70s/it]\n",
      "1875it [01:08, 27.31it/s, loss=-25.9]9:50, 68.62s/it]\n",
      "1875it [01:08, 27.30it/s, loss=-6.15]8:43, 68.63s/it]\n",
      "1875it [01:08, 27.42it/s, loss=-18.7]7:37, 68.64s/it]\n",
      "1875it [01:08, 27.44it/s, loss=-21.7]6:16, 68.57s/it]\n",
      "1875it [01:08, 27.42it/s, loss=-5.95]4:56, 68.49s/it]\n",
      "1875it [01:08, 27.53it/s, loss=-52.8]3:42, 68.46s/it]\n",
      "1875it [01:08, 27.25it/s, loss=4.09]02:17, 68.36s/it]\n",
      "1875it [01:09, 26.92it/s, loss=50.2]01:31, 68.50s/it]\n",
      "1875it [01:09, 27.14it/s, loss=-54.9]1:17, 68.85s/it]\n",
      "1875it [01:08, 27.46it/s, loss=-55]:00:20, 68.92s/it]\n",
      "1875it [01:08, 27.48it/s, loss=-8.45]8:41, 68.73s/it]\n",
      "1875it [01:08, 27.41it/s, loss=0.663]7:10, 68.58s/it]\n",
      "1875it [01:08, 27.41it/s, loss=-17.6]5:53, 68.53s/it]\n",
      "1875it [01:08, 27.37it/s, loss=-6.9]54:39, 68.50s/it]\n",
      "1875it [01:08, 27.46it/s, loss=-16.6]3:32, 68.50s/it]\n",
      "1875it [01:10, 26.47it/s, loss=-6.93]2:13, 68.43s/it]\n",
      "1875it [01:08, 27.36it/s, loss=-8.5]52:52, 69.15s/it]\n",
      "1875it [01:08, 27.46it/s, loss=52.6]51:15, 68.96s/it]\n",
      "1875it [01:08, 27.39it/s, loss=-6.39]9:36, 68.76s/it]\n",
      "1875it [01:08, 27.47it/s, loss=-10.1]8:14, 68.67s/it]\n",
      "1875it [01:08, 27.30it/s, loss=-63.3]6:48, 68.55s/it]\n",
      "1875it [01:08, 27.52it/s, loss=-12.8]5:45, 68.59s/it]\n",
      "1875it [01:08, 27.50it/s, loss=17.6]44:16, 68.45s/it]\n",
      "1875it [01:08, 27.48it/s, loss=43.4]42:57, 68.37s/it]\n",
      "1875it [01:09, 26.93it/s, loss=20.9]41:43, 68.33s/it]\n",
      "1875it [01:08, 27.43it/s, loss=-13.4]1:29, 68.72s/it]\n",
      "1875it [01:08, 27.41it/s, loss=-57.4]0:05, 68.61s/it]\n",
      "1875it [01:08, 27.46it/s, loss=41.2]38:48, 68.55s/it]\n",
      "1875it [01:08, 27.49it/s, loss=2.47]37:29, 68.47s/it]\n",
      "1875it [01:08, 27.40it/s, loss=5.57]36:09, 68.39s/it]\n",
      "1875it [01:08, 27.42it/s, loss=4.71]35:03, 68.41s/it]\n",
      "1875it [01:08, 27.25it/s, loss=-37.6]3:54, 68.40s/it]\n",
      "1875it [01:08, 27.42it/s, loss=-10.8]3:02, 68.53s/it]\n",
      "1875it [01:09, 26.80it/s, loss=-4.6]31:48, 68.48s/it]\n",
      "1875it [01:08, 27.37it/s, loss=-21.6]1:38, 68.93s/it]\n",
      "1875it [01:08, 27.42it/s, loss=-35.7]0:12, 68.80s/it]\n",
      "1875it [01:08, 27.42it/s, loss=-24.5]8:47, 68.67s/it]\n",
      "1875it [01:08, 27.42it/s, loss=-16.4]7:27, 68.59s/it]\n",
      "1875it [01:08, 27.41it/s, loss=-25.2]6:10, 68.52s/it]\n",
      "1875it [01:08, 27.42it/s, loss=24.3]24:58, 68.49s/it]\n",
      "1875it [01:08, 27.41it/s, loss=6.87]23:45, 68.46s/it]\n",
      "1875it [01:08, 27.26it/s, loss=-50.3]2:35, 68.44s/it]\n",
      "1875it [01:09, 27.08it/s, loss=-1.14]1:39, 68.55s/it]\n",
      "1875it [01:08, 27.47it/s, loss=8.17]20:56, 68.76s/it]\n",
      "1875it [01:08, 27.34it/s, loss=8.29]19:29, 68.60s/it]\n",
      "1875it [01:08, 27.30it/s, loss=-3.16]8:20, 68.60s/it]\n",
      "1875it [01:08, 27.47it/s, loss=-40.7]7:14, 68.62s/it]\n",
      "1875it [01:08, 27.33it/s, loss=24.4]15:53, 68.51s/it]\n",
      "1875it [01:08, 27.40it/s, loss=-4.96]4:48, 68.55s/it]\n",
      "1875it [01:08, 27.29it/s, loss=-32.6]3:36, 68.51s/it]\n",
      "1875it [01:09, 26.97it/s, loss=-24.9]2:34, 68.58s/it]\n",
      "1875it [01:08, 27.27it/s, loss=1.58]11:58, 68.86s/it]\n",
      "1875it [01:08, 27.39it/s, loss=-21.8]0:46, 68.83s/it]\n",
      "1875it [01:08, 27.28it/s, loss=18]2:09:25, 68.72s/it]\n",
      "1875it [01:08, 27.34it/s, loss=-34.2]8:17, 68.73s/it]\n",
      "1875it [01:08, 27.37it/s, loss=-22.9]7:03, 68.68s/it]\n",
      "1875it [01:08, 27.31it/s, loss=2.05]05:49, 68.63s/it]\n",
      "1875it [01:08, 27.25it/s, loss=-12.3]4:41, 68.64s/it]\n",
      "1875it [01:08, 27.32it/s, loss=2.48]03:38, 68.69s/it]\n",
      "1875it [01:10, 26.72it/s, loss=35.5]02:28, 68.68s/it]\n",
      "1875it [01:08, 27.46it/s, loss=24.6]02:08, 69.13s/it]\n",
      "1875it [01:08, 27.32it/s, loss=-2.68]0:32, 68.88s/it]\n",
      "1875it [01:08, 27.32it/s, loss=-16.8]9:15, 68.81s/it]\n",
      "1875it [01:08, 27.41it/s, loss=-46.4]8:01, 68.75s/it]\n",
      "1875it [01:08, 27.28it/s, loss=-3.36]6:42, 68.65s/it]\n",
      "1875it [01:08, 27.34it/s, loss=-9.41]5:36, 68.68s/it]\n",
      "1875it [01:08, 27.35it/s, loss=21.6]54:25, 68.65s/it]\n",
      "1875it [01:08, 27.30it/s, loss=-44.4]3:14, 68.63s/it]\n",
      "1875it [01:11, 26.06it/s, loss=-14.6]2:07, 68.64s/it]\n",
      "1875it [01:08, 27.33it/s, loss=-5.81]2:35, 69.64s/it]\n",
      "1875it [01:08, 27.27it/s, loss=-17.8]0:55, 69.33s/it]\n",
      "1875it [01:08, 27.33it/s, loss=-31.3]9:30, 69.16s/it]\n",
      "1875it [01:08, 27.33it/s, loss=-45.9]8:05, 69.00s/it]\n",
      "1875it [01:08, 27.29it/s, loss=-11.9]6:46, 68.88s/it]\n",
      "1875it [01:08, 27.27it/s, loss=-25.1]5:32, 68.83s/it]\n",
      "1875it [01:08, 27.33it/s, loss=45.2]44:21, 68.81s/it]\n",
      "1875it [01:09, 27.11it/s, loss=49.4]43:07, 68.75s/it]\n",
      "1875it [01:09, 26.98it/s, loss=-15.1]2:09, 68.88s/it]\n",
      "1875it [01:08, 27.23it/s, loss=-14.9]1:17, 69.06s/it]\n",
      "1875it [01:08, 27.24it/s, loss=13.2]40:03, 69.00s/it]\n",
      "1875it [01:08, 27.33it/s, loss=37.8]38:50, 68.96s/it]\n",
      "1875it [01:08, 27.40it/s, loss=16.9]37:32, 68.85s/it]\n",
      "1875it [01:08, 27.27it/s, loss=14.2]36:13, 68.73s/it]\n",
      "1875it [01:08, 27.24it/s, loss=-22.4]5:05, 68.74s/it]\n",
      "1875it [01:08, 27.42it/s, loss=-16.3]3:58, 68.77s/it]\n",
      "1875it [01:10, 26.66it/s, loss=2.09]32:40, 68.65s/it]\n",
      "1875it [01:08, 27.36it/s, loss=23.3]32:12, 69.16s/it]\n",
      "1875it [01:08, 27.38it/s, loss=-11.1]0:48, 68.97s/it]\n",
      "1875it [01:08, 27.31it/s, loss=-12.7]9:28, 68.83s/it]\n",
      "1875it [01:08, 27.29it/s, loss=-14.5]8:15, 68.78s/it]\n",
      "1875it [01:09, 27.14it/s, loss=-.165]7:05, 68.75s/it]\n",
      "1875it [01:08, 27.38it/s, loss=-22.9]6:03, 68.85s/it]\n",
      "1875it [01:08, 27.31it/s, loss=-27.2]4:46, 68.74s/it]\n",
      "1875it [01:08, 27.31it/s, loss=27.9]23:36, 68.72s/it]\n",
      "1875it [01:09, 26.80it/s, loss=-32]:22:27, 68.71s/it]\n",
      "1875it [01:08, 27.33it/s, loss=-25.7]1:45, 69.09s/it]\n",
      "1875it [01:08, 27.24it/s, loss=1.4]:20:26, 68.94s/it]\n",
      "1875it [01:08, 27.36it/s, loss=18.4]19:15, 68.91s/it]\n",
      "1875it [01:08, 27.34it/s, loss=-13.4]7:58, 68.80s/it]\n",
      "1875it [01:08, 27.37it/s, loss=27.7]16:45, 68.73s/it]\n",
      "1875it [01:08, 27.30it/s, loss=-14]:15:32, 68.67s/it]\n",
      "1875it [01:08, 27.35it/s, loss=-18.3]4:23, 68.67s/it]\n",
      "1875it [01:08, 27.26it/s, loss=-17.5]3:12, 68.64s/it]\n",
      "1875it [01:09, 26.95it/s, loss=-2.17]2:07, 68.68s/it]\n",
      "1875it [01:08, 27.21it/s, loss=-13.4]1:14, 68.95s/it]\n",
      "1875it [01:08, 27.28it/s, loss=-8.96]0:05, 68.94s/it]\n",
      "1875it [01:08, 27.25it/s, loss=7.91]08:52, 68.88s/it]\n",
      "1875it [01:08, 27.35it/s, loss=-15.4]7:42, 68.86s/it]\n",
      "1875it [01:08, 27.27it/s, loss=-21.1]6:28, 68.76s/it]\n",
      "1875it [01:08, 27.35it/s, loss=-4.16]5:19, 68.76s/it]\n",
      "1875it [01:08, 27.33it/s, loss=-69]:04:07, 68.70s/it]\n",
      "1875it [01:09, 26.86it/s, loss=11.6]02:56, 68.67s/it]\n",
      "1875it [01:09, 27.04it/s, loss=-3]1:02:06, 69.01s/it]\n",
      "1875it [01:09, 27.16it/s, loss=-50.4]1:02, 69.11s/it]\n",
      "1875it [01:08, 27.29it/s, loss=-10.3]52, 69.09s/it]  \n",
      "1875it [01:09, 27.13it/s, loss=-28.5]37, 68.98s/it]\n",
      "1875it [01:08, 27.27it/s, loss=-2.44]30, 69.02s/it]\n",
      "1875it [01:08, 27.40it/s, loss=47.1]:17, 68.94s/it]\n",
      "1875it [01:08, 27.22it/s, loss=-5.15]01, 68.79s/it]\n",
      "1875it [01:08, 27.35it/s, loss=6.19]:54, 68.81s/it]\n",
      "1875it [01:10, 26.78it/s, loss=26.4]:41, 68.73s/it]\n",
      "1875it [01:10, 26.72it/s, loss=4.5]1:50, 69.12s/it]\n",
      "1875it [01:08, 27.37it/s, loss=-27.3]55, 69.44s/it]\n",
      "1875it [01:08, 27.18it/s, loss=-13.7]33, 69.16s/it]\n",
      "1875it [01:08, 27.31it/s, loss=-11.9]22, 69.11s/it]\n",
      "1875it [01:08, 27.30it/s, loss=-24.2]07, 68.97s/it]\n",
      "1875it [01:08, 27.27it/s, loss=-31.7]55, 68.88s/it]\n",
      "1875it [01:08, 27.22it/s, loss=-9.17]45, 68.85s/it]\n",
      "1875it [01:08, 27.24it/s, loss=-61.5]36, 68.86s/it]\n",
      "1875it [01:09, 26.82it/s, loss=-4.02]27, 68.85s/it]\n",
      "1875it [01:08, 27.22it/s, loss=-28.5]30, 69.17s/it]\n",
      "1875it [01:09, 27.10it/s, loss=-9.18]17, 69.08s/it]\n",
      "1875it [01:08, 27.19it/s, loss=-4.54]10, 69.12s/it]\n",
      "1875it [01:08, 27.29it/s, loss=-8.34]59, 69.08s/it]\n",
      "1875it [01:08, 27.30it/s, loss=-9.49]47, 68.97s/it]\n",
      "1875it [01:08, 27.37it/s, loss=18.9]:35, 68.88s/it]\n",
      "1875it [01:08, 27.24it/s, loss=-11.1]23, 68.77s/it]\n",
      "1875it [01:09, 26.85it/s, loss=-57.8]14, 68.79s/it]\n",
      "1875it [01:09, 27.03it/s, loss=-24.7]14, 69.10s/it]\n",
      "1875it [01:08, 27.31it/s, loss=-5.93]07, 69.18s/it]\n",
      "1875it [01:08, 27.19it/s, loss=-17.1]54, 69.03s/it]\n",
      "1875it [01:08, 27.21it/s, loss=-16.6]45, 69.01s/it]\n",
      "1875it [01:08, 27.24it/s, loss=8.83]:35, 68.98s/it]\n",
      "1875it [01:08, 27.21it/s, loss=-14.1]25, 68.94s/it]\n",
      "1875it [01:08, 27.21it/s, loss=-3.99]16, 68.93s/it]\n",
      "1875it [01:08, 27.26it/s, loss=4.77]:07, 68.92s/it]\n",
      "1875it [01:10, 26.76it/s, loss=-3.66]:57, 68.88s/it]\n",
      "1875it [01:08, 27.20it/s, loss=45.3]1:55, 69.24s/it]\n",
      "1875it [01:08, 27.25it/s, loss=-28.4]:44, 69.15s/it]\n",
      "1875it [01:09, 27.16it/s, loss=-4.48]:33, 69.05s/it]\n",
      "1875it [01:08, 27.21it/s, loss=-17.5]:24, 69.05s/it]\n",
      "1875it [01:08, 27.33it/s, loss=43]<17:15, 69.01s/it]\n",
      "1875it [01:08, 27.20it/s, loss=-6.09]:04, 68.89s/it]\n",
      "1875it [01:08, 27.29it/s, loss=16]<14:55, 68.90s/it]\n",
      "1875it [01:08, 27.30it/s, loss=8.35]3:46, 68.84s/it]\n",
      "1875it [01:10, 26.67it/s, loss=-8.2]2:36, 68.80s/it]\n",
      "1875it [01:08, 27.33it/s, loss=-8.99]:32, 69.25s/it]\n",
      "1875it [01:08, 27.18it/s, loss=-23.9]:21, 69.06s/it]\n",
      "1875it [01:09, 27.17it/s, loss=-20.6]:12, 69.04s/it]\n",
      "1875it [01:08, 27.29it/s, loss=-54.8]:03, 69.03s/it]\n",
      "1875it [01:09, 27.09it/s, loss=22.8]6:53, 68.94s/it]\n",
      "1875it [01:08, 27.27it/s, loss=-67.7]:45, 69.02s/it]\n",
      "1875it [01:08, 27.21it/s, loss=-46]04:35, 68.94s/it]\n",
      "1875it [01:09, 26.97it/s, loss=17.2]3:26, 68.93s/it]\n",
      "1875it [01:09, 26.93it/s, loss=10.5]2:18, 69.11s/it]\n",
      "1875it [01:08, 27.26it/s, loss=-6.19]:09, 69.26s/it]\n",
      "100%|██████████| 500/500 [10:23:07<00:00, 74.78s/it]\n"
     ]
    }
   ],
   "source": [
    "\n",
    "\n",
    "for epoch in tqdm(range(NUM_EPOCHS)):\n",
    "    loop = tqdm(enumerate(train_loader))\n",
    "    for i, (x, _) in loop:\n",
    "        x = x.to(DEVICE)\n",
    "        x_hat, z, mu, std, log_scale = model(x)\n",
    "\n",
    "        # Compute Loss \n",
    "        reconstruction_loss = gaussian_likelihood(x_hat,log_scale, x)\n",
    "        kl_div =  kl_divergence(z, mu, std)\n",
    "        \n",
    "        # Backprop\n",
    "        loss = kl_div - reconstruction_loss \n",
    "        elbo = loss.mean()\n",
    "        optimizer.zero_grad()\n",
    "        elbo.backward()\n",
    "        optimizer.step()\n",
    "        loop.set_postfix(loss = elbo.item())"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "from torchvision.utils import make_grid, save_image\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "valid_dataset = datasets.MNIST(root = \"dataset/\", train = False, transform=transforms.ToTensor(), download=True)\n",
    "valid_loader = DataLoader(dataset = valid_dataset, batch_size = BATCH_SIZE, shuffle=True)\n",
    "\n",
    "def show(imgs):\n",
    "    if not isinstance(imgs, list):\n",
    "        imgs = [imgs]\n",
    "    fig, axs = plt.subplots(ncols=len(imgs), squeeze=False)\n",
    "    for i, img in enumerate(imgs):\n",
    "        img = img.detach()\n",
    "\n",
    "        img = transforms.functional.to_pil_image(img)\n",
    "        axs[0, i].imshow(np.asarray(img))\n",
    "        axs[0, i].set(xticklabels=[], yticklabels=[], xticks=[], yticks=[])\n",
    "        \n",
    "def infer_batch(valid_loader):\n",
    "    valid_loader_iter = iter(valid_loader)\n",
    "    (x, _) = next(valid_loader_iter)\n",
    "    x = x.to(DEVICE)\n",
    "    x_hat, z, mu, std, log_scale = model(x)\n",
    "\n",
    "    grid = make_grid(x_hat)\n",
    "    show(grid)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "infer_batch(valid_loader)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "torch",
   "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": "3.10.9"
  },
  "orig_nbformat": 4
 },
 "nbformat": 4,
 "nbformat_minor": 2
}