nb/gumbel/LabeledGSVAE.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Gumbel Softmax VAE"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This notebook is intended for loading and exploring arbitrary pre-trained models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import torch\n",
    "from torch.utils.data import DataLoader, SubsetRandomSampler\n",
    "from torchvision.transforms.functional import to_pil_image\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load pallets library from project root\n",
    "import sys, os\n",
    "sys.path.append(os.path.abspath(os.path.join(os.getcwd(), '..', '..')))\n",
    "\n",
    "from pallets import (\n",
    "    images as I,\n",
    "    datasets as DS,\n",
    "    models as M,\n",
    "    logging as L\n",
    ")\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Settings"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "SAVE_NAME = 'gsvae.labeled'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "USE_GPU = True\n",
    "LOG_LEVEL = 'INFO'\n",
    "LABELS_FILE = 'pallets_labels.json'\n",
    "\n",
    "\n",
    "TEST_SIZE = 1000\n",
    "EPOCHS = 50\n",
    "LR = 1e-03\n",
    "BATCH_SIZE = 32\n",
    "\n",
    "TEMPERATURE = 1\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Logger pallets (INFO)>"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# To GPU, or not to GPU\n",
    "device = M.get_device(require_gpu=USE_GPU)\n",
    "\n",
    "# Logging\n",
    "L.init_logger(level=LOG_LEVEL, notebook=True)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO | loading punk images\n",
      "INFO | - image 0\n",
      "INFO | - image 1000\n",
      "INFO | - image 2000\n",
      "INFO | - image 3000\n",
      "INFO | - image 4000\n",
      "INFO | - image 5000\n",
      "INFO | - image 6000\n",
      "INFO | - image 7000\n",
      "INFO | - image 8000\n",
      "INFO | - image 9000\n",
      "INFO | punk images complete\n",
      "INFO | loading punk labels: pallets_labels.json\n"
     ]
    }
   ],
   "source": [
    "all_colors = I.get_punk_colors()\n",
    "mapper = DS.ColorOneHotMapper(all_colors)\n",
    "dataset = DS.OneHotCPunksDataset(\n",
    "    mapper, labels_file=LABELS_FILE, test_size=TEST_SIZE\n",
    ")\n",
    "\n",
    "# Uncomment to put entire dataset in GPU RAM\n",
    "# dataset = DS.FastOneHotCPunksDataset(\n",
    "#     device, mapper, labels_file=LABELS_FILE, test_size=TEST_SIZE\n",
    "# )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_sampler = SubsetRandomSampler(dataset.train_idx)\n",
    "test_sampler = SubsetRandomSampler(dataset.test_idx)\n",
    "\n",
    "train_loader = DataLoader(dataset, batch_size=BATCH_SIZE, sampler=train_sampler)\n",
    "test_loader = DataLoader(dataset, batch_size=BATCH_SIZE, sampler=test_sampler)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_dim = 222 * 24 * 24\n",
    "hidden_dim = 128\n",
    "latent_dim = 32\n",
    "classes_dim = len(dataset._label_keys)  # 92\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "INFO | model: pallets.models.gumbel.LabeledGSVAE\n",
      "INFO | criterion: pallets.models.gumbel.Loss\n",
      "INFO | learn rate: 0.001\n",
      "INFO | epochs: 50\n",
      "INFO | epoch 1 (  0%) loss: 1419241.875000\n",
      "INFO | epoch 1 ( 35%) loss: 258467.555422\n",
      "INFO | epoch 1 ( 70%) loss: 135705.091204\n",
      "INFO | epoch 1 (100%) loss: 99979.637431\n",
      "INFO | epoch 1 (test) loss: 11156.620026\n",
      "INFO | epoch 2 (  0%) loss: 10660.712891\n",
      "INFO | epoch 2 ( 35%) loss: 11339.612517\n",
      "INFO | epoch 2 ( 70%) loss: 11307.118805\n",
      "INFO | epoch 2 (100%) loss: 11252.291447\n",
      "INFO | epoch 2 (test) loss: 11047.788788\n",
      "INFO | epoch 3 (  0%) loss: 10925.098633\n",
      "INFO | epoch 3 ( 35%) loss: 11179.087871\n",
      "INFO | epoch 3 ( 70%) loss: 11190.389221\n",
      "INFO | epoch 3 (100%) loss: 11158.073227\n",
      "INFO | epoch 3 (test) loss: 10988.958115\n",
      "INFO | epoch 4 (  0%) loss: 11361.206055\n",
      "INFO | epoch 4 ( 35%) loss: 11156.276648\n",
      "INFO | epoch 4 ( 70%) loss: 11128.152528\n",
      "INFO | epoch 4 (100%) loss: 11070.894806\n",
      "INFO | epoch 4 (test) loss: 10817.397125\n",
      "INFO | epoch 5 (  0%) loss: 11084.065430\n",
      "INFO | epoch 5 ( 35%) loss: 10984.323445\n",
      "INFO | epoch 5 ( 70%) loss: 10953.445793\n",
      "INFO | epoch 5 (100%) loss: 10830.800323\n",
      "INFO | epoch 5 (test) loss: 10232.962639\n",
      "INFO | epoch 6 (  0%) loss: 10544.832031\n",
      "INFO | epoch 6 ( 35%) loss: 10140.533164\n",
      "INFO | epoch 6 ( 70%) loss: 10035.263881\n",
      "INFO | epoch 6 (100%) loss: 9931.645786\n",
      "INFO | epoch 6 (test) loss: 9534.939636\n",
      "INFO | epoch 7 (  0%) loss: 9215.046875\n",
      "INFO | epoch 7 ( 35%) loss: 9658.183671\n",
      "INFO | epoch 7 ( 70%) loss: 9594.923367\n",
      "INFO | epoch 7 (100%) loss: 9478.526946\n",
      "INFO | epoch 7 (test) loss: 9118.928741\n",
      "INFO | epoch 8 (  0%) loss: 9548.783203\n",
      "INFO | epoch 8 ( 35%) loss: 9152.980527\n",
      "INFO | epoch 8 ( 70%) loss: 9075.224175\n",
      "INFO | epoch 8 (100%) loss: 9017.774533\n",
      "INFO | epoch 8 (test) loss: 8725.387615\n",
      "INFO | epoch 9 (  0%) loss: 8574.556641\n",
      "INFO | epoch 9 ( 35%) loss: 8791.288439\n",
      "INFO | epoch 9 ( 70%) loss: 8715.977862\n",
      "INFO | epoch 9 (100%) loss: 8631.130575\n",
      "INFO | epoch 9 (test) loss: 8170.964226\n",
      "INFO | epoch 10 (  0%) loss: 8194.153320\n",
      "INFO | epoch 10 ( 35%) loss: 8181.579372\n",
      "INFO | epoch 10 ( 70%) loss: 7933.657708\n",
      "INFO | epoch 10 (100%) loss: 7726.208230\n",
      "INFO | epoch 10 (test) loss: 7005.164768\n",
      "INFO | epoch 11 (  0%) loss: 7142.008789\n",
      "INFO | epoch 11 ( 35%) loss: 6933.703096\n",
      "INFO | epoch 11 ( 70%) loss: 6740.247291\n",
      "INFO | epoch 11 (100%) loss: 6600.191117\n",
      "INFO | epoch 11 (test) loss: 6048.384838\n",
      "INFO | epoch 12 (  0%) loss: 6830.671387\n",
      "INFO | epoch 12 ( 35%) loss: 6011.165319\n",
      "INFO | epoch 12 ( 70%) loss: 5918.926607\n",
      "INFO | epoch 12 (100%) loss: 5804.299326\n",
      "INFO | epoch 12 (test) loss: 5424.441635\n",
      "INFO | epoch 13 (  0%) loss: 5327.593262\n",
      "INFO | epoch 13 ( 35%) loss: 5433.940343\n",
      "INFO | epoch 13 ( 70%) loss: 5307.998613\n",
      "INFO | epoch 13 (100%) loss: 5220.344547\n",
      "INFO | epoch 13 (test) loss: 4821.918690\n",
      "INFO | epoch 14 (  0%) loss: 4338.987305\n",
      "INFO | epoch 14 ( 35%) loss: 4831.636893\n",
      "INFO | epoch 14 ( 70%) loss: 4755.021616\n",
      "INFO | epoch 14 (100%) loss: 4657.056971\n",
      "INFO | epoch 14 (test) loss: 4190.395153\n",
      "INFO | epoch 15 (  0%) loss: 4112.315918\n",
      "INFO | epoch 15 ( 35%) loss: 4267.988129\n",
      "INFO | epoch 15 ( 70%) loss: 4161.600049\n",
      "INFO | epoch 15 (100%) loss: 4083.018158\n",
      "INFO | epoch 15 (test) loss: 3720.223080\n",
      "INFO | epoch 16 (  0%) loss: 3100.393799\n",
      "INFO | epoch 16 ( 35%) loss: 3766.022659\n",
      "INFO | epoch 16 ( 70%) loss: 3682.615718\n",
      "INFO | epoch 16 (100%) loss: 3614.384502\n",
      "INFO | epoch 16 (test) loss: 3285.094469\n",
      "INFO | epoch 17 (  0%) loss: 3567.601807\n",
      "INFO | epoch 17 ( 35%) loss: 3333.295241\n",
      "INFO | epoch 17 ( 70%) loss: 3299.077312\n",
      "INFO | epoch 17 (100%) loss: 3240.112328\n",
      "INFO | epoch 17 (test) loss: 2958.190662\n",
      "INFO | epoch 18 (  0%) loss: 2911.560547\n",
      "INFO | epoch 18 ( 35%) loss: 3004.278526\n",
      "INFO | epoch 18 ( 70%) loss: 2949.617690\n",
      "INFO | epoch 18 (100%) loss: 2901.565355\n",
      "INFO | epoch 18 (test) loss: 2703.374800\n",
      "INFO | epoch 19 (  0%) loss: 2803.573975\n",
      "INFO | epoch 19 ( 35%) loss: 2710.611055\n",
      "INFO | epoch 19 ( 70%) loss: 2666.335311\n",
      "INFO | epoch 19 (100%) loss: 2624.146039\n",
      "INFO | epoch 19 (test) loss: 2383.458006\n",
      "INFO | epoch 20 (  0%) loss: 2153.345703\n",
      "INFO | epoch 20 ( 35%) loss: 2382.345816\n",
      "INFO | epoch 20 ( 70%) loss: 2345.608871\n",
      "INFO | epoch 20 (100%) loss: 2325.922932\n",
      "INFO | epoch 20 (test) loss: 2157.000683\n",
      "INFO | epoch 21 (  0%) loss: 1862.597046\n",
      "INFO | epoch 21 ( 35%) loss: 2131.852231\n",
      "INFO | epoch 21 ( 70%) loss: 2098.383714\n",
      "INFO | epoch 21 (100%) loss: 2059.570679\n",
      "INFO | epoch 21 (test) loss: 1899.526346\n",
      "INFO | epoch 22 (  0%) loss: 1703.308594\n",
      "INFO | epoch 22 ( 35%) loss: 1881.223240\n",
      "INFO | epoch 22 ( 70%) loss: 1858.638136\n",
      "INFO | epoch 22 (100%) loss: 1836.720278\n",
      "INFO | epoch 22 (test) loss: 1710.061032\n",
      "INFO | epoch 23 (  0%) loss: 1916.642822\n",
      "INFO | epoch 23 ( 35%) loss: 1706.065871\n",
      "INFO | epoch 23 ( 70%) loss: 1707.138738\n",
      "INFO | epoch 23 (100%) loss: 1674.147783\n",
      "INFO | epoch 23 (test) loss: 1594.366585\n",
      "INFO | epoch 24 (  0%) loss: 1520.897095\n",
      "INFO | epoch 24 ( 35%) loss: 1520.024827\n",
      "INFO | epoch 24 ( 70%) loss: 1507.063297\n",
      "INFO | epoch 24 (100%) loss: 1484.403279\n",
      "INFO | epoch 24 (test) loss: 1422.071381\n",
      "INFO | epoch 25 (  0%) loss: 1459.372559\n",
      "INFO | epoch 25 ( 35%) loss: 1401.920046\n",
      "INFO | epoch 25 ( 70%) loss: 1375.029695\n",
      "INFO | epoch 25 (100%) loss: 1361.430252\n",
      "INFO | epoch 25 (test) loss: 1331.200106\n",
      "INFO | epoch 26 (  0%) loss: 1254.322388\n",
      "INFO | epoch 26 ( 35%) loss: 1262.761559\n",
      "INFO | epoch 26 ( 70%) loss: 1253.012518\n",
      "INFO | epoch 26 (100%) loss: 1237.412423\n",
      "INFO | epoch 26 (test) loss: 1240.796654\n",
      "INFO | epoch 27 (  0%) loss: 1139.271484\n",
      "INFO | epoch 27 ( 35%) loss: 1147.494309\n",
      "INFO | epoch 27 ( 70%) loss: 1139.690446\n",
      "INFO | epoch 27 (100%) loss: 1123.410379\n",
      "INFO | epoch 27 (test) loss: 1157.701841\n",
      "INFO | epoch 28 (  0%) loss: 1000.067932\n",
      "INFO | epoch 28 ( 35%) loss: 1070.047457\n",
      "INFO | epoch 28 ( 70%) loss: 1059.696016\n",
      "INFO | epoch 28 (100%) loss: 1046.389656\n",
      "INFO | epoch 28 (test) loss: 1074.886754\n",
      "INFO | epoch 29 (  0%) loss: 1063.481445\n",
      "INFO | epoch 29 ( 35%) loss: 962.788458\n",
      "INFO | epoch 29 ( 70%) loss: 963.336694\n",
      "INFO | epoch 29 (100%) loss: 954.612798\n",
      "INFO | epoch 29 (test) loss: 992.169963\n",
      "INFO | epoch 30 (  0%) loss: 956.468750\n",
      "INFO | epoch 30 ( 35%) loss: 889.922601\n",
      "INFO | epoch 30 ( 70%) loss: 877.243893\n",
      "INFO | epoch 30 (100%) loss: 869.755130\n",
      "INFO | epoch 30 (test) loss: 907.147220\n",
      "INFO | epoch 31 (  0%) loss: 812.599243\n",
      "INFO | epoch 31 ( 35%) loss: 826.604822\n",
      "INFO | epoch 31 ( 70%) loss: 822.895671\n",
      "INFO | epoch 31 (100%) loss: 811.536114\n",
      "INFO | epoch 31 (test) loss: 857.800843\n",
      "INFO | epoch 32 (  0%) loss: 760.481323\n",
      "INFO | epoch 32 ( 35%) loss: 748.917892\n",
      "INFO | epoch 32 ( 70%) loss: 738.108244\n",
      "INFO | epoch 32 (100%) loss: 729.291030\n",
      "INFO | epoch 32 (test) loss: 795.057493\n",
      "INFO | epoch 33 (  0%) loss: 625.461304\n",
      "INFO | epoch 33 ( 35%) loss: 677.995295\n",
      "INFO | epoch 33 ( 70%) loss: 674.814822\n",
      "INFO | epoch 33 (100%) loss: 675.568121\n",
      "INFO | epoch 33 (test) loss: 767.456862\n",
      "INFO | epoch 34 (  0%) loss: 687.284729\n",
      "INFO | epoch 34 ( 35%) loss: 674.909027\n",
      "INFO | epoch 34 ( 70%) loss: 654.972706\n",
      "INFO | epoch 34 (100%) loss: 638.287897\n",
      "INFO | epoch 34 (test) loss: 670.661122\n",
      "INFO | epoch 35 (  0%) loss: 606.744385\n",
      "INFO | epoch 35 ( 35%) loss: 575.211113\n",
      "INFO | epoch 35 ( 70%) loss: 585.386367\n",
      "INFO | epoch 35 (100%) loss: 580.171013\n",
      "INFO | epoch 35 (test) loss: 637.549601\n",
      "INFO | epoch 36 (  0%) loss: 629.574890\n",
      "INFO | epoch 36 ( 35%) loss: 557.244791\n",
      "INFO | epoch 36 ( 70%) loss: 555.558649\n",
      "INFO | epoch 36 (100%) loss: 548.219218\n",
      "INFO | epoch 36 (test) loss: 603.337631\n",
      "INFO | epoch 37 (  0%) loss: 478.806641\n",
      "INFO | epoch 37 ( 35%) loss: 504.412323\n",
      "INFO | epoch 37 ( 70%) loss: 505.711864\n",
      "INFO | epoch 37 (100%) loss: 508.164620\n",
      "INFO | epoch 37 (test) loss: 596.471814\n",
      "INFO | epoch 38 (  0%) loss: 466.213043\n",
      "INFO | epoch 38 ( 35%) loss: 485.524195\n",
      "INFO | epoch 38 ( 70%) loss: 489.665717\n",
      "INFO | epoch 38 (100%) loss: 477.712813\n",
      "INFO | epoch 38 (test) loss: 533.815420\n",
      "INFO | epoch 39 (  0%) loss: 382.447510\n",
      "INFO | epoch 39 ( 35%) loss: 440.846630\n",
      "INFO | epoch 39 ( 70%) loss: 435.666735\n",
      "INFO | epoch 39 (100%) loss: 431.610706\n",
      "INFO | epoch 39 (test) loss: 508.732305\n",
      "INFO | epoch 40 (  0%) loss: 410.924561\n",
      "INFO | epoch 40 ( 35%) loss: 409.432149\n",
      "INFO | epoch 40 ( 70%) loss: 407.274257\n",
      "INFO | epoch 40 (100%) loss: 402.854984\n",
      "INFO | epoch 40 (test) loss: 446.947064\n",
      "INFO | epoch 41 (  0%) loss: 397.842957\n",
      "INFO | epoch 41 ( 35%) loss: 375.967817\n",
      "INFO | epoch 41 ( 70%) loss: 372.095979\n",
      "INFO | epoch 41 (100%) loss: 373.823165\n",
      "INFO | epoch 41 (test) loss: 448.201968\n",
      "INFO | epoch 42 (  0%) loss: 437.194916\n",
      "INFO | epoch 42 ( 35%) loss: 379.820484\n",
      "INFO | epoch 42 ( 70%) loss: 369.796447\n",
      "INFO | epoch 42 (100%) loss: 364.083393\n",
      "INFO | epoch 42 (test) loss: 414.242637\n",
      "INFO | epoch 43 (  0%) loss: 346.781067\n",
      "INFO | epoch 43 ( 35%) loss: 330.710379\n",
      "INFO | epoch 43 ( 70%) loss: 331.364750\n",
      "INFO | epoch 43 (100%) loss: 344.351258\n",
      "INFO | epoch 43 (test) loss: 457.001387\n",
      "INFO | epoch 44 (  0%) loss: 311.488251\n",
      "INFO | epoch 44 ( 35%) loss: 337.059692\n",
      "INFO | epoch 44 ( 70%) loss: 374.315165\n",
      "INFO | epoch 44 (100%) loss: 379.800330\n",
      "INFO | epoch 44 (test) loss: 412.440963\n",
      "INFO | epoch 45 (  0%) loss: 302.687744\n",
      "INFO | epoch 45 ( 35%) loss: 355.851597\n",
      "INFO | epoch 45 ( 70%) loss: 352.225682\n",
      "INFO | epoch 45 (100%) loss: 337.899366\n",
      "INFO | epoch 45 (test) loss: 380.886229\n",
      "INFO | epoch 46 (  0%) loss: 327.723907\n",
      "INFO | epoch 46 ( 35%) loss: 285.938289\n",
      "INFO | epoch 46 ( 70%) loss: 289.584123\n",
      "INFO | epoch 46 (100%) loss: 296.310273\n",
      "INFO | epoch 46 (test) loss: 489.300241\n",
      "INFO | epoch 47 (  0%) loss: 410.508362\n",
      "INFO | epoch 47 ( 35%) loss: 312.088402\n",
      "INFO | epoch 47 ( 70%) loss: 293.947879\n",
      "INFO | epoch 47 (100%) loss: 287.281642\n",
      "INFO | epoch 47 (test) loss: 336.888661\n",
      "INFO | epoch 48 (  0%) loss: 322.537811\n",
      "INFO | epoch 48 ( 35%) loss: 255.423993\n",
      "INFO | epoch 48 ( 70%) loss: 253.283199\n",
      "INFO | epoch 48 (100%) loss: 253.270398\n",
      "INFO | epoch 48 (test) loss: 311.696913\n",
      "INFO | epoch 49 (  0%) loss: 229.307999\n",
      "INFO | epoch 49 ( 35%) loss: 251.983548\n",
      "INFO | epoch 49 ( 70%) loss: 257.109513\n",
      "INFO | epoch 49 (100%) loss: 253.610413\n",
      "INFO | epoch 49 (test) loss: 300.304478\n",
      "INFO | epoch 50 (  0%) loss: 232.249512\n",
      "INFO | epoch 50 ( 35%) loss: 245.557051\n",
      "INFO | epoch 50 ( 70%) loss: 258.118448\n",
      "INFO | epoch 50 (100%) loss: 270.159137\n",
      "INFO | epoch 50 (test) loss: 318.765559\n",
      "INFO | model blob saved to /home/jmsdnns/ML/pallets/saved/gsvae.labeled.pkl\n",
      "INFO | training losses saved to /home/jmsdnns/ML/pallets/saved/gsvae.labeled.json\n"
     ]
    }
   ],
   "source": [
    "model = M.gumbel.LabeledGSVAE(\n",
    "    input_dim, hidden_dim, latent_dim, classes_dim, TEMPERATURE\n",
    ")\n",
    "criterion = M.gumbel.Loss()\n",
    "\n",
    "train_losses, test_losses = M.gumbel.train(\n",
    "    device, model, criterion, train_loader, test_loader,\n",
    "    learn_rate=LR, epochs=EPOCHS, with_labels=True\n",
    ")\n",
    "\n",
    "M.save(SAVE_NAME, model, train_losses, test_losses)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Results"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random Inputs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "import random\n",
    "\n",
    "def rand_label():\n",
    "    label_idx = int(random.random() * len(dataset._labels))\n",
    "    features = dataset._labels[label_idx]\n",
    "    enabled_names = [k for k,v in zip(dataset._label_keys, features) if v.item() == 1]\n",
    "    return features.to(device), enabled_names\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def rand_punk():\n",
    "    z = torch.randn(1, latent_dim*classes_dim).to(device)\n",
    "    # print(z.shape)\n",
    "\n",
    "    with torch.no_grad():\n",
    "        model.eval()\n",
    "        generated_image = model.decoder(z)\n",
    "        # print(generated_image.shape)\n",
    "\n",
    "    decoded_one_hot = generated_image[0].view((222, 24, 24))\n",
    "    # print(decoded_one_hot.shape)\n",
    "\n",
    "    decoded = DS.one_hot_to_rgba(decoded_one_hot, dataset.mapper)\n",
    "    # print(decoded.shape)\n",
    "    return decoded\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 5 Randoms"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "decoded = rand_punk()\n",
    "plt.imshow(to_pil_image(decoded))\n",
    "plt.axis('off')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "decoded = rand_punk()\n",
    "plt.imshow(to_pil_image(decoded))\n",
    "plt.axis('off')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "decoded = rand_punk()\n",
    "plt.imshow(to_pil_image(decoded))\n",
    "plt.axis('off')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "decoded = rand_punk()\n",
    "plt.imshow(to_pil_image(decoded))\n",
    "plt.axis('off')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "decoded = rand_punk()\n",
    "plt.imshow(to_pil_image(decoded))\n",
    "plt.axis('off')\n",
    "plt.show()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Reconstruction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "def reconstruct_punk(idx):\n",
    "    punk = I.get_punk_tensor(idx)\n",
    "    p = DS.rgba_to_one_hot(punk, dataset.mapper)\n",
    "    p = p.unsqueeze(0)\n",
    "    p = p.to(device)\n",
    "\n",
    "    _, labels = dataset[idx]\n",
    "    l = labels.unsqueeze(0)\n",
    "    l = l.to(device)\n",
    "    enabled_features = [k for k,v in zip(dataset._label_keys, labels) if v.item() == 1]\n",
    "\n",
    "    model.eval()\n",
    "    with torch.no_grad():\n",
    "        reconstructed, x_y = model.forward(p, l)\n",
    "\n",
    "    recon_punk = reconstructed[0].cpu()\n",
    "    recon_punk = recon_punk.view((222, 24, 24))\n",
    "    recon_punk = DS.one_hot_to_rgba(recon_punk, dataset.mapper)\n",
    "\n",
    "    return punk, recon_punk, enabled_features\n",
    "\n",
    "\n",
    "def draw_two(img1, img2):\n",
    "    page_size = 2\n",
    "    view_x, view_y = 4*page_size, 2*page_size\n",
    "    fig = plt.figure(figsize=(view_x, view_y))\n",
    "    fig.add_subplot(1, 2, 1)\n",
    "    plt.imshow(to_pil_image(img1))\n",
    "    plt.axis('off')\n",
    "    fig.add_subplot(1, 2, 2)\n",
    "    plt.imshow(to_pil_image(img2))\n",
    "    plt.axis('off')\n",
    "    plt.show()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 5 Reconstructions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Features: female, rosyCheeks, darkHair, cigarette\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAn8AAAEtCAYAAAB54AaaAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjguMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/SrBM8AAAACXBIWXMAAA9hAAAPYQGoP6dpAAAHkUlEQVR4nO3dMW4kVRRA0S4wAanjMSkRMdIk3gpbMBtxxBJIWAD5JLMBIiY1iMhOCQAVsYUH+Y/9u6vrnhOXS78k99PVT96yrut6AAAg4bNTHwAAgOMRfwAAIeIPACBE/AEAhIg/AIAQ8QcAECL+AABCxB8AQIj4AwAIEX8AACHiDwAgRPwBAISIPwCAEPEHABAi/gAAQsQfAECI+AMACBF/AAAh4g8AIET8AQCEiD8AgBDxBwAQIv4AAELEHwBAiPgDAAgRfwAAIeIPACBE/AEAhIg/AIAQ8QcAECL+AABCxB8AQIj4AwAIEX8AACHiDwAgRPwBAISIPwCAEPEHABAi/gAAQsQfAECI+AMACBF/AAAh4g8AIET8AQCEiD8AgBDxBwAQIv4AAELEHwBAiPgDAAgRfwAAIeIPACBE/AEAhIg/AIAQ8QcAECL+AABCxB8AQIj4AwAIEX8AACHiDwAgRPwBAISIPwCAEPEHABAi/gAAQsQfAECI+AMACBF/AAAh4g8AIET8AQCEiD8AgBDxBwAQIv4AAELEHwBAiPgDAAgRfwAAIeIPACBE/AEAhIg/AIAQ8QcAEHJx6gPwupZlOfURNmld11MfATgT5ujTzNH9cPMHABAi/gAAQsQfAECI+AMACBF/AAAh4g8AIET8AQCEiD8AgBDxBwAQIv4AAELEHwBAyLJa1rdpdkyehp8F7Ic5ehrm6Ha5+QMACBF/AAAh4g8AIET8AQCEiD8AgBDxBwAQIv4AAELEHwBAiPgDAAgRfwAAIeIPACDEbt8js2Nyn/yM4HjM0X0yR4/HzR8AQIj4AwAIEX8AACHiDwAgRPwBAISIPwCAEPEHABAi/gAAQsQfAECI+AMACBF/AAAhdvu+kB2TjPKTg8fMUUaZoy/j5g8AIET8AQCEiD8AgBDxBwAQIv4AAELEHwBAiPgDAAgRfwAAIeIPACBE/AEAhFyc+gDA/3t4eBh6/vLyctJJAM6TOfqYmz8AgBDxBwAQIv4AAELEHwBAiPgDAAgRfwAAIeIPACBE/AEAhIg/AIAQ8QcAECL+AABC7PbdmZvrq2nvvn13N/T8zLPMNvqtI5ZlGXr+/v5+0kmAp5ijr8Mc3S43fwAAIeIPACBE/AEAhIg/AIAQ8QcAECL+AABCxB8AQIj4AwAIEX8AACHiDwAgRPwBAITY7fuE0Z2BM43udfz9j3m7FLe0Y3LmdwIvZ44+zRxlC9z8AQCEiD8AgBDxBwAQIv4AAELEHwBAiPgDAAgRfwAAIeIPACBE/AEAhIg/AIAQ8QcAEGK378789OH5z37/7ZdD714HzzLTyHceDuPfejj8Ofg8sBfm6NPM0f1w8wcAECL+AABCxB8AQIj4AwAIEX8AACHiDwAgRPwBAISIPwCAEPEHABAi/gAAQqx325mb66tnP7ulNUOjRr7zcPiUb70b/gtgH8zRp5mj++HmDwAgRPwBAISIPwCAEPEHABAi/gAAQsQfAECI+AMACBF/AAAh4g8AIET8AQCEiD8AgJDEbt9lWU59hKP54u9/pr37r4vPp7171MzvBP6rNEdh79z8AQCEiD8AgBDxBwAQIv4AAELEHwBAiPgDAAgRfwAAIeIPACBE/AEAhIg/AIAQ8QcAEJLY7Vuypf27t+/uhp6/ub569rNb+s7ZLi8vT30E4ERmztESc/QxN38AACHiDwAgRPwBAISIPwCAEPEHABAi/gAAQsQfAECI+AMACBF/AAAh4g8AIET8AQCE2O3LNDN3TI7uuzxny7IMPb+u66STAMdmjr4Oc/QxN38AACHiDwAgRPwBAISIPwCAEPEHABAi/gAAQsQfAECI+AMACBF/AAAh4g8AIGRZz3CHyeialpKZq4B+/eW3ae8+HA6Hnx+e/6/4/v37iSfZljc/fDf0/Fc/fhh6/gxHAK/AHP04c3R/zNHH3PwBAISIPwCAEPEHABAi/gAAQsQfAECI+AMACBF/AAAh4g8AIET8AQCEiD8AgBDxBwAQcnHqA3yKLe3QK+3H/PqbN3PfP/Ds27dvp53j3G3p98F2ben/xBx9xfcPPGuOftyWfh8zuPkDAAgRfwAAIeIPACBE/AEAhIg/AIAQ8QcAECL+AABCxB8AQIj4AwAIEX8AACHiDwAg5Cx3+27J7P1/M3de3r67G3r+5vpq0knm2/ueRjhn5uh5MEf3w80fAECI+AMACBF/AAAh4g8AIET8AQCEiD8AgBDxBwAQIv4AAELEHwBAiPgDAAgRfwAAIXb7ho3umBzdYTmTHZPAFpijnCM3fwAAIeIPACBE/AEAhIg/AIAQ8QcAECL+AABCxB8AQIj4AwAIEX8AACHiDwAgxHo3nm32GiOrhoC9M0fZAjd/AAAh4g8AIET8AQCEiD8AgBDxBwAQIv4AAELEHwBAiPgDAAgRfwAAIeIPACBE/AEAhIg/AIAQ8QcAECL+AABCxB8AQIj4AwAIEX8AACHiDwAgRPwBAISIPwCAEPEHABAi/gAAQsQfAEDIsq7reupDAABwHG7+AABCxB8AQIj4AwAIEX8AACHiDwAgRPwBAISIPwCAEPEHABAi/gAAQv4F5LLbgdQmMcsAAAAASUVORK5CYII=",
      "text/plain": [
       "<Figure size 800x400 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "punk, recon_punk, features = reconstruct_punk(1000)\n",
    "\n",
    "print(f\"Features: {', '.join(features)}\")\n",
    "draw_two(punk, recon_punk)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Features: female, stringyHair, purpleLipstick, earring\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x400 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "punk, recon_punk, features = reconstruct_punk(2001)\n",
    "\n",
    "print(f\"Features: {', '.join(features)}\")\n",
    "draw_two(punk, recon_punk)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Features: female, wildWhiteHair, greenEyeShadow, hotLipstick\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x400 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "punk, recon_punk, features = reconstruct_punk(5000)\n",
    "\n",
    "print(f\"Features: {', '.join(features)}\")\n",
    "draw_two(punk, recon_punk)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Features: male, rosyCheeks, mustache, vr, cap, cigarette\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x400 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "punk, recon_punk, features = reconstruct_punk(8000)\n",
    "\n",
    "print(f\"Features: {', '.join(features)}\")\n",
    "draw_two(punk, recon_punk)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Features: male, handlebars, bandana\n"
     ]
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 800x400 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "punk, recon_punk, features = reconstruct_punk(1337)\n",
    "\n",
    "print(f\"Features: {', '.join(features)}\")\n",
    "draw_two(punk, recon_punk)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Losses"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<function matplotlib.pyplot.show(close=None, block=None)>"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.title(\"Train & Test loss\")\n",
    "plt.plot(train_losses, label='train loss')\n",
    "plt.plot(test_losses, label='test loss')\n",
    "plt.legend()\n",
    "plt.show\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[199959.2748625194,\n",
       " 22504.58289353391,\n",
       " 22316.14645320811,\n",
       " 22141.789611383534,\n",
       " 21661.600646539784,\n",
       " 19863.291571434507,\n",
       " 18957.053891012856,\n",
       " 18035.54906533965,\n",
       " 17262.261149954287,\n",
       " 15452.416459753158,\n",
       " 13200.382234181072,\n",
       " 11608.598651858932,\n",
       " 10440.689094705785,\n",
       " 9314.113941814883,\n",
       " 8166.036316350842,\n",
       " 7228.769003577267,\n",
       " 6480.224655692459,\n",
       " 5803.130710466534,\n",
       " 5248.2920774635695,\n",
       " 4651.845864153923,\n",
       " 4119.141358287622,\n",
       " 3673.4405550043634,\n",
       " 3348.2955657742546,\n",
       " 2968.806557513298,\n",
       " 2722.860504366827,\n",
       " 2474.824845983627,\n",
       " 2246.8207583596522,\n",
       " 2092.779311809134,\n",
       " 1909.2255963264627,\n",
       " 1739.510260724007,\n",
       " 1623.0722281814467,\n",
       " 1458.5820598196476,\n",
       " 1351.136242697425,\n",
       " 1276.575794734008,\n",
       " 1160.3420262438185,\n",
       " 1096.4384358723958,\n",
       " 1016.3292398114577,\n",
       " 955.4256267141789,\n",
       " 863.2214119011629,\n",
       " 805.7099677552568,\n",
       " 747.6463303802707,\n",
       " 728.1667864102844,\n",
       " 688.7025161634946,\n",
       " 759.6006595259862,\n",
       " 675.7987315401118,\n",
       " 592.6205452993406,\n",
       " 574.5632843666888,\n",
       " 506.54079523830546,\n",
       " 507.2208254658584,\n",
       " 540.3182739366031]"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_losses"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[22313.24005126953,\n",
       " 22095.577575683594,\n",
       " 21977.916229248047,\n",
       " 21634.79425048828,\n",
       " 20465.92527770996,\n",
       " 19069.879272460938,\n",
       " 18237.857482910156,\n",
       " 17450.775230407715,\n",
       " 16341.928451538086,\n",
       " 14010.329536437988,\n",
       " 12096.769676208496,\n",
       " 10848.883270263672,\n",
       " 9643.837379455566,\n",
       " 8380.790306091309,\n",
       " 7440.446159362793,\n",
       " 6570.188938140869,\n",
       " 5916.381324768066,\n",
       " 5406.749599456787,\n",
       " 4766.916011810303,\n",
       " 4314.001365661621,\n",
       " 3799.05269241333,\n",
       " 3420.122064590454,\n",
       " 3188.733169555664,\n",
       " 2844.1427612304688,\n",
       " 2662.4002113342285,\n",
       " 2481.593307495117,\n",
       " 2315.4036827087402,\n",
       " 2149.7735080718994,\n",
       " 1984.3399257659912,\n",
       " 1814.294439315796,\n",
       " 1715.6016855239868,\n",
       " 1590.1149854660034,\n",
       " 1534.913724899292,\n",
       " 1341.322244644165,\n",
       " 1275.099202156067,\n",
       " 1206.6752610206604,\n",
       " 1192.9436283111572,\n",
       " 1067.6308403015137,\n",
       " 1017.4646096229553,\n",
       " 893.8941287994385,\n",
       " 896.4039359092712,\n",
       " 828.4852743148804,\n",
       " 914.0027747154236,\n",
       " 824.8819265365601,\n",
       " 761.7724571228027,\n",
       " 978.6004819869995,\n",
       " 673.7773218154907,\n",
       " 623.3938264846802,\n",
       " 600.6089553833008,\n",
       " 637.5311179161072]"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_losses"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "pallets",
   "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": "3.11.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}