nb_model_task2.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "gradient": {
     "editing": false
    }
   },
   "source": [
    "# Task 2 Models\n",
    "\n",
    "1. Train the PCA models\n",
    "\n",
    "        This results in pickled PCA models.\n",
    "        \n",
    "2. Train autoencoder models (as replacement for PCA)\n",
    "\n",
    "        This results in saved autoencoder models.\n",
    "        \n",
    "3. Train a k-means cluster that sits on top of PCA & autoencoder\n",
    "\n",
    "        This results in pickled k-means models."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "gradient": {
     "editing": false
    }
   },
   "source": [
    "## LOAD"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "task_name = 'task2_bart_openface'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "model_path = './models/{0}/'.format(task_name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "import os\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import warnings\n",
    "import pickle\n",
    "\n",
    "from tqdm import tqdm\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "save_path = './data/tmp_analysis/{0}'.format(task_name)\n",
    "save_train_path = save_path + '_train.csv'\n",
    "save_test_path = save_path + '_test.csv'\n",
    "save_valid_path = save_path + '_valid.csv'\n",
    "\n",
    "\n",
    "Train = pd.read_csv(save_train_path)\n",
    "Test = pd.read_csv(save_test_path)\n",
    "Valid = pd.read_csv(save_valid_path)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "X = np.asarray(Train.to_numpy()[:,3:]).astype('float32')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "del Train"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "Xtest = np.asarray(Test.to_numpy()[:,3:]).astype('float32')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "del Test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "Xvalid = np.asarray(Valid.to_numpy()[:,3:]).astype('float32')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "del Valid"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## PCA Fit\n",
    "I found 32 components to be roughly the inflection curve.\n",
    "I am also creating an 18  component variant just in case we want a direct comparison with action units"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "from sklearn.decomposition import PCA"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[2.12116360e+06 9.41878768e+05 6.51112188e+04 1.46127087e+04\n",
      " 4.28960478e+03 3.52037785e+03 2.22753141e+03 9.14173594e+02\n",
      " 7.60174472e+02 5.90052640e+02 3.97382387e+02 3.87775970e+02\n",
      " 2.66193515e+02 2.20283752e+02 1.81378183e+02 1.53110760e+02\n",
      " 1.02648015e+02 8.98712482e+01 7.70245594e+01 6.65274985e+01\n",
      " 4.22923168e+01 3.59366880e+01 3.38732051e+01 3.20646285e+01\n",
      " 2.58124777e+01 2.33485284e+01 2.18275103e+01 1.61892713e+01\n",
      " 1.53576457e+01 1.22650058e+01 1.04890955e+01 7.44619758e+00]\n"
     ]
    }
   ],
   "source": [
    "pca_n_32 = PCA(n_components=32)\n",
    "face_pca_n_32 = pca_n_32.fit(X)\n",
    "\n",
    "print(face_pca_n_32.explained_variance_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "pickle.dump(face_pca_n_32,open(model_path+'pca_n_32.p','wb'))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[2.12116510e+06 9.41878601e+05 6.51112188e+04 1.46127152e+04\n",
      " 4.28960250e+03 3.52038274e+03 2.22753255e+03 9.14173431e+02\n",
      " 7.60173411e+02 5.90051986e+02 3.97382265e+02 3.87775644e+02\n",
      " 2.66193413e+02 2.20283854e+02 1.81378101e+02 1.53110851e+02\n",
      " 1.02648137e+02 8.98712482e+01]\n"
     ]
    }
   ],
   "source": [
    "pca_n_18 = PCA(n_components=18)\n",
    "face_pca_n_18 = pca_n_18.fit(X)\n",
    "\n",
    "print(face_pca_n_18.explained_variance_)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "pickle.dump(face_pca_n_18,open(model_path+'pca_n_18.p','wb'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Autoencoder\n",
    "Right now just doing a single layer \n",
    "Perhaps we can try a two layer encoder / two layer decoder next?"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "32 Wide encoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "import keras\n",
    "from keras import layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(391949, 136)\n"
     ]
    }
   ],
   "source": [
    "print(X.shape)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "input_dim = X.shape[1]\n",
    "encoding_dim = 32\n",
    "\n",
    "input_layer = keras.Input(shape=(input_dim,))\n",
    "encoded_layer = layers.Dense(encoding_dim, activation='relu')(input_layer)\n",
    "decoded_layer = layers.Dense(input_dim, activation='relu')(encoded_layer)\n",
    "\n",
    "autoencoder_32 = keras.Model(input_layer,decoded_layer)\n",
    "\n",
    "encoder_32 = keras.Model(input_layer, encoded_layer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "autoencoder_32.compile(optimizer='adam',\n",
    "                    loss='binary_crossentropy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_1 (InputLayer)         [(None, 136)]             0         \n",
      "_________________________________________________________________\n",
      "dense (Dense)                (None, 32)                4384      \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 136)               4488      \n",
      "=================================================================\n",
      "Total params: 8,872\n",
      "Trainable params: 8,872\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "autoencoder_32.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      "6125/6125 [==============================] - 16s 2ms/step - loss: -6071.5892 - val_loss: -7445.5142\n",
      "Epoch 2/2\n",
      "6125/6125 [==============================] - 14s 2ms/step - loss: -7617.7796 - val_loss: -8263.9326\n"
     ]
    }
   ],
   "source": [
    "history = autoencoder_32.fit(X, X, epochs=2, batch_size=64, verbose=1, validation_data=(Xtest, Xtest))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "gradient": {}
   },
   "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": [
    "fig, ax = plt.subplots()\n",
    "\n",
    "plt.plot([0]+history.history['loss'], label='train')\n",
    "plt.plot([0]+history.history['val_loss'], label='test')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "predictions = autoencoder_32.predict(Xvalid)\n",
    "# TODO I should probably do something with the validation set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "autoencoder_32.save(model_path + 'autoencoder_32.h5', save_format='h5')\n",
    "encoder_32.save(model_path + 'encoder_32.h5', save_format='h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "5 wide encoder"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "import keras\n",
    "from keras import layers"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "input_dim = X.shape[1]\n",
    "encoding_dim = 18\n",
    "\n",
    "input_layer = keras.Input(shape=(input_dim,))\n",
    "encoded_layer = layers.Dense(encoding_dim, activation='relu')(input_layer)\n",
    "decoded_layer = layers.Dense(input_dim, activation='relu')(encoded_layer)\n",
    "\n",
    "autoencoder_18 = keras.Model(input_layer,decoded_layer)\n",
    "\n",
    "encoder_18 = keras.Model(input_layer, encoded_layer)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "autoencoder_18.compile(optimizer='adam',\n",
    "                    loss='binary_crossentropy')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"model_2\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "input_2 (InputLayer)         [(None, 136)]             0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 18)                2466      \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 136)               2584      \n",
      "=================================================================\n",
      "Total params: 5,050\n",
      "Trainable params: 5,050\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "autoencoder_18.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/2\n",
      "6125/6125 [==============================] - 15s 2ms/step - loss: -4731.3882 - val_loss: -5230.1035\n",
      "Epoch 2/2\n",
      "6125/6125 [==============================] - 15s 2ms/step - loss: -5229.2962 - val_loss: -5230.1787\n"
     ]
    }
   ],
   "source": [
    "history = autoencoder_18.fit(X, X, epochs=2, batch_size=64, verbose=1, validation_data=(Xtest, Xtest))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "gradient": {}
   },
   "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": [
    "fig, ax = plt.subplots()\n",
    "\n",
    "plt.plot([0]+history.history['loss'], label='train')\n",
    "plt.plot([0]+history.history['val_loss'], label='test')\n",
    "plt.legend()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "predictions = autoencoder_18.predict(Xvalid)\n",
    "# TODO I should probably do something with the validation set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "autoencoder_18.save(model_path + 'autoencoder_18.h5', save_format='h5')\n",
    "encoder_18.save(model_path + 'encoder_18.h5', save_format='h5')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## K-means\n",
    "Not sure what this is going to do \n",
    "(my guess is it is going to cluster similar participants)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "from sklearn.cluster import KMeans"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "def visualize_labels(y):\n",
    "    idx = np.arange(len(y))\n",
    "\n",
    "    fig, (ax1,ax2) = plt.subplots(nrows=2, sharex=True)\n",
    "\n",
    "    ax1.imshow(y[np.newaxis,:], cmap=\"plasma\", aspect=\"auto\")\n",
    "    ax1.set_yticks([])\n",
    "\n",
    "    ax2.scatter(idx,y)\n",
    "\n",
    "    plt.tight_layout()\n",
    "    plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "X_pca = face_pca_n_32.transform(X)\n",
    "kmeans_5 = KMeans(n_clusters=5, random_state=0).fit(X_pca)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y = kmeans_5.predict(face_pca_n_32.transform(Xtest))\n",
    "visualize_labels(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAagAAAEYCAYAAAAJeGK1AAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/Il7ecAAAACXBIWXMAAAsTAAALEwEAmpwYAAATEklEQVR4nO3db6xkdX3H8c/3nDNzdy9/dkEIWJdyUUwTYhqBiaHRNAZbRCTSBz7AtJG2aUjaJ9o2MUt9YNqkSds0BkxN7QZstbWiRdIawBBaSdQm0M6iZS24uopFNigicvlzd++dOefbB+c3Z+bevbt3djnDfrnzfiU3d+Y355z5ne/5nfOZOffcGXN3AQAQTXa6OwAAwGYIKABASAQUACAkAgoAEBIBBQAIqZjFQl+32PWLzj5TWV7J3eQuWZauFnST5HLPJJcsq6cxG19NaKZmHq+y8fyuNG16PP2WrF5Omcnyqn6aqp5OqpehtPzxPBo/7qNpNdEHV1VZ3V9zSaYsLyVJVZmtm3c0fZqzeS5tcoFk81xWr49scjpbt16jZY76b+ZNrdY/r03cryaWafLKmnllo7qMbzd1MK/vp+XIXFleNetfL0eytM7H1Gy0PpMrNHHTq6zZ1vLxekpSVpSpfWM91KxXs66juxPTTPbDldYxc5lUb8PJbmbe1KCuqzaMw7r2WTaqsY/7ZJKX9bbIsqoewxrXv94WvmE8W9Mmt3p8jurU1Gy0rqPlHdvXeptnx4zdUY1G69u0Tm7/tH2a+qWx5elxs3r9qzJL2zstwEdjZzweR9tx1P/ReFCabdS/cS3r/carbP1YnDBZr3U1cEvPtd6o783KWtpfy/Hr7VHtJreBlGrZjMPx803uU1La/hN1XLe/alTD8fo2+1PaFqN1GO9bE/tw2peafXFygG3YCdYfu8bbb2N7U8MNyx0fR6w5VkreHE/Xlae5o2aaUQ29SrVtxsNof51Yz2ZfmdhvUvMxx73RQ2k8Pfn8Ef1sZXXDUXhGAfWLuxd1/weu0c5zX9LwaEfVMFdncVXupmqQy7JK5WpX1TBTZ3FVVZUpy8umIHm3VLlaqNi5psHKgtxNeXcouVQOimMO1D7MVSyuavDygjpnrEqShke6yopSZq7Bka7yhaFU1QcId2tu10GUHbPTFDsGGry8UIdRVu+sC7teluWVVl9YlJkrK0qVqx0pqw9Yo4GYd4eSpGqYrTvgy03VMJfllbJOKR9msqKSD7Nm3mqY1wfsMlO+MJAkDVYWlOWV8oWBykGhvDMcP2/a+atBvSnz7kBW1AfBaphreLSjfGGoLKtkRaXyaEdZZyjLXcMj3XqHTf0pj3aanTzrlOqecbQeQHlaT6vDpDzaUTkolBVlU6+sUzbPOTqoWO7yst5pBysL6pxxVFWZq1wtVJV53ae80o7dL6scFCkoXNUgbwZ+udapt1mZKe8OlBXpRc8wG7cvDJt+VGWm8mhH+Y6BsrzS4EhXkpoXL3l3qHKtruFgZaE5yHQWVzU82m3CsrNzTcOjnXoMpT5ZUWntxR3Kikrds47U9cuqpv5ZZ9iMp9GLmdEOOFrv7llH5aXJq6yp2Sj0ytXu+ICcuYqFgYarHWVZpXzHQMMj3bq+aZvVByRXNSia9a2fs97P3E3lWlHvU52yXpeikpemfGGoapirXKv3p87iqlZfWFQ1zJTl3rwoyfJ6G1XDXHl3oOHRroqFQR08aZn1Pl0f0MrVQpZXzX5aDQot7H5Jg5d3qFwrlOVVXaf0gtXLNOZT2/DIQn0QzCtVg3TcGL2QGh0fUt/r+U2Wu/LOUGsv7mzGwahek6Fa13yozuKqykEhH2b1+CrKep8aFFJ6rs4Zq+vqmHVKeWnKinGNs6JSuZZL6fHhaqc+bu0YjPeropJlVb0Pd+rnGR7p1tMfGY+3rCibfb85BlT1fpMVVXO8yToT7Wl7Z0VZj7msavqRZaM3Bza+XWZNsBU715p9c+MxSKpDrVhc1XBloT4GHek261lv86J+Idcpm+nNvD6OrBWyrBrvq6WpGhTN/lcN8nR8qOvSWVzVO//u65tmCaf4AAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkAgoAEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACAkc/f2F2r2oqSDrS94/pwn6dnT3YltgDq2gzq2h1qud7G7n7+xsZjRkx10996Mlj03zKxPHV856tgO6tgeajkdTvEBAEIioAAAIc0qoPbNaLnzhjq2gzq2gzq2h1pOYSYXSQAA8Epxig8AEBIBBQAIiYACAIREQAEAQiKgAAAhEVAAgJAIKABASAQUACCkmXxY7HnnnedLS0uzWDQAYJvZv3//s6/ap5kvLS2p3+/PYtEAgG3GzP5vs/apT/GZWW5m3zSze9rrFgAAmzuZd1AfkvS4pLO3mvDA4WUt7b33lDsFAJgf3QsvvXKz9qneQZnZHknvlXR7m50CAOB4pj3Fd6ukj0iqZtcVAADGtgwoM7te0jPuvn+L6W42s76Z9cuV5dY6CACYT9O8g3q7pPeZ2Q8l3SnpajP7p40Tufs+d++5ey9f3NVyNwEA82bLgHL3W9x9j7svSbpR0lfd/bdm3jMAwFzjkyQAACHN5Cvfe72e84+6AIBpmNl+d+9tbOcdFAAgJAIKABASAQUACImAAgCEREABAEIioAAAIRFQAICQCCgAQEgEFAAgJAIKABASAQUACImAAgCEREABAEIioAAAIRFQAICQCCgAQEgEFAAgJAIKABBSsdUEZrZD0tckLaTp73L3j51ongOHl7W09952eggA2Na6F1565WbtWwaUpFVJV7v7S2bWkfQNM/uKuz/Uag8BAJiwZUC5u0t6Kd3tpB+fZacAAJjqb1BmlpvZtyQ9I+kBd394k2luNrO+mfXLleWWuwkAmDdTBZS7l+7+Vkl7JL3NzN6yyTT73L3n7r18cVfL3QQAzJuTuorP3Z+X9KCka2fSGwAAki0DyszON7Pd6fZOSb8u6Tsz7hcAYM5ZfQ3ECSYw+2VJn5GUqw60L7r7n51onl6v5/1+v7VOAgC2LzPb7+69je3TXMX3qKTLZ9IrAACOg0+SAACEREABAEIioAAAIRFQAICQCCgAQEgEFAAgJAIKABASAQUACImAAgCEREABAEIioAAAIRFQAICQCCgAQEgEFAAgJAIKABASAQUACImAAgCEREABAELa8ivfzewiSZ+VdIEkl7TP3W870TwHDi9rae+97fQQALCtdS+89MrN2rcMKElDSX/s7o+Y2VmS9pvZA+7+WKs9BABgwpan+Nz9aXd/JN1+UdLjkt4w644BAObbSf0NysyWJF0u6eFNHrvZzPpm1i9XllvqHgBgXk0dUGZ2pqQvSfqwu7+w8XF33+fuPXfv5Yu72uwjAGAOTRVQZtZRHU6fc/e7Z9slAACmCCgzM0l3SHrc3T8++y4BACCZu594ArN3SPq6pAOSqtT8J+5+3/Hm6fV63u/3W+skAGD7MrP97t7b2L7lZebu/g1JNpNeAQBwHHySBAAgJAIKABASAQUACImAAgCEREABAEIioAAAIRFQAICQCCgAQEgEFAAgJAIKABASAQUACImAAgCEREABAEIioAAAIRFQAICQCCgAQEgEFAAgJAIKABDSll/5bmaflnS9pGfc/S3TLPTA4WUt7b33lfYNADAHuhdeeuVm7dO8g/oHSde22hsAALawZUC5+9ckPfcq9AUAgEZrf4Mys5vNrG9m/XJlua3FAgDmVGsB5e773L3n7r18cVdbiwUAzCmu4gMAhERAAQBCMnc/8QRmn5f0TknnSfqJpI+5+x0nmqfX63m/32+rjwCAbczM9rt7b2P7lv8H5e4fmE2XAAA4Pk7xAQBCIqAAACERUACAkAgoAEBIBBQAICQCCgAQEgEFAAiJgAIAhERAAQBCIqAAACERUACAkAgoAEBIBBQAICQCCgAQEgEFAAiJgAIAhERAAQBC2vIbdSXJzK6VdJukXNLt7v4XJ5r+wOFlLe29t4XuxbPYyTQoKw2q092T+BaKTL2Ld+s/v//c6e4KcIzcpLN3dvTzlcHp7src617wpis2a9/yHZSZ5ZI+Kek9ki6T9AEzu6zd7r12rAwIp2mtDivCCWGVLsIpCjPbrHmaU3xvk3TI3X/g7muS7pR0Q5t9AwBgo2kC6g2SfjRx/6nUto6Z3WxmfTPrlyvLbfUPADCnWrtIwt33uXvP3Xv54q62FgsAmFPTBNRhSRdN3N+T2gAAmJlpAuq/Jb3ZzC4xs66kGyV9ebbdimuxk6nDxflTWSgyvf1N557ubgCbyk06Z7FzursBSXL3zZrtOO3rJzK7TtKtqi8z/7S7//mJpu/1et7v90+hlwCAeWNm+929t7F9qv+Dcvf7JN3Xeq8AADiOqd5BnfRCzV6UdLD1Bc+f8yQ9e7o7sQ1Qx3ZQx/ZQy/UudvfzNzZO9Q7qFBzc7O0aTo6Z9anjK0cd20Ed20Mtp8Of+wEAIRFQAICQZhVQ+2a03HlDHdtBHdtBHdtDLacwk4skAAB4pTjFBwAIiYACAITUakCZ2bVmdtDMDpnZ3jaX/VplZheZ2YNm9piZ/a+ZfSi1n2tmD5jZ99Lvc1K7mdknUg0fNbMrJpZ1U5r+e2Z200T7lWZ2IM3zCTvOd6tsB2aWm9k3zeyedP8SM3s4rfsX0sdxycwW0v1D6fGliWXcktoPmtm7J9rnYvya2W4zu8vMvmNmj5vZrzAeT42Z/WHar79tZp83sx2MyRa5eys/qj8G6fuS3iipK+l/JF3W1vJfqz+SXi/pinT7LEnfVf3Fj38laW9q3yvpL9Pt6yR9RZJJukrSw6n9XEk/SL/PSbfPSY/9V5rW0rzvOd3rPcN6/pGkf5Z0T7r/RUk3ptufkvT76fYfSPpUun2jpC+k25elsbkg6ZI0ZvN5Gr+SPiPp99LtrqTdjMdTquMbJD0haefEWPxtxmR7P22+g+KLDTfh7k+7+yPp9ouSHlc9sG9QfaBQ+v0b6fYNkj7rtYck7Taz10t6t6QH3P05d/+5pAckXZseO9vdH/J6tH92YlnbipntkfReSben+ybpakl3pUk21nFU37skvStNf4OkO9191d2fkHRI9didi/FrZrsk/aqkOyTJ3dfc/XkxHk9VIWmnmRWSFiU9LcZka9oMqKm+2HCepbf0l0t6WNIF7v50eujHki5It49XxxO1P7VJ+3Z0q6SPSKrS/ddJet7dh+n+5Lo39UqPL6fpT7a+280lkn4q6e/TqdLbzewMMR5PmrsflvTXkp5UHUzLkvaLMdkaLpJ4lZjZmZK+JOnD7v7C5GPplSbX+5+AmV0v6Rl333+6+/IaV0i6QtLfuvvlkl5WfUqvwXicTvo73Q2qQ/8XJJ0h6drT2qltps2A4osNj8PMOqrD6XPufndq/kk6HaL0+5nUfrw6nqh9zybt283bJb3PzH6o+lTH1ZJuU33KafSZkpPr3tQrPb5L0s908vXdbp6S9JS7P5zu36U6sBiPJ+/XJD3h7j9194Gku1WPU8ZkS9oMKL7YcBPpHPMdkh53949PPPRlSaMrn26S9G8T7R9MV09dJWk5nXq5X9I1ZnZOeuV2jaT702MvmNlV6bk+OLGsbcPdb3H3Pe6+pHpsfdXdf1PSg5LenybbWMdRfd+fpvfUfmO6ouoSSW9W/Uf9uRi/7v5jST8ys19KTe+S9JgYj6fiSUlXmdliWtdRLRmTbWnzigvVV/x8V/WVJx893VeARPiR9A7Vp0selfSt9HOd6nPP/yHpe5L+XdK5aXqT9MlUwwOSehPL+l3Vf0A9JOl3Jtp7kr6d5vkbpU8I2a4/kt6p8VV8b1S9Mx+S9C+SFlL7jnT/UHr8jRPzfzTV6qAmrjCbl/Er6a2S+mlM/qvqq/AYj6dWyz+V9J20vv+o+ko8xmRLP3zUEQAgJC6SAACEREABAEIioAAAIRFQAICQCCgAQEgEFAAgJAIKABDS/wN4dtL9jylmbgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y = kmeans_5.predict(face_pca_n_32.transform(Xvalid))\n",
    "visualize_labels(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "pickle.dump(kmeans_5,open(model_path+'kmeans_5_from_pca.p','wb'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "   "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "WARNING:tensorflow:No training configuration found in the save file, so the model was *not* compiled. Compile it manually.\n"
     ]
    }
   ],
   "source": [
    "encoder_32 = keras.models.load_model(model_path + 'encoder_32.h5')\n",
    "\n",
    "X_auto = encoder_32.predict(X)\n",
    "kmeans_5 = KMeans(n_clusters=5, random_state=0).fit(X_auto)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y = kmeans_5.predict(encoder_32.predict(Xtest))\n",
    "visualize_labels(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "gradient": {}
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "y = kmeans_5.predict(encoder_32.predict(Xvalid))\n",
    "visualize_labels(y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "gradient": {}
   },
   "outputs": [],
   "source": [
    "pickle.dump(kmeans_5,open(model_path+'kmeans_5_from_auto.p','wb'))"
   ]
  }
 ],
 "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": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}