Books/DeepLearningwithPython - Jason Brownlee/Chapter 8. Evaluate The Performance of Deep Learning Models.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "### 8.1 Empirically Evaluate Network Configurations\n",
    "\n",
    "Many of these decisions can be resolved by copying the structure of other people’s networks and using heuristics.\n",
    "\n",
    "the best technique is to actually design small experiments and empirically evaluate options using real data. It also includes the lower level decisions like the choice of loss function, activation functions, optimization procedure and number of epochs.\n",
    "\n",
    "### 8.2 Data Splitting\n",
    "\n",
    "Keras provides two convenient ways of evaluating your deep learning algorithms this way:\n",
    "1. Use an automatic verification dataset.\n",
    "2. Use a manual verification dataset.\n",
    "\n",
    "#### 8.2.1 Use a Automatic Verification Dataset\n",
    "Keras can separate a portion of your training data into a validation dataset and evaluate the performance of your model on that validation dataset each epoch. \n",
    "\n",
    "You can do this by setting the validation split argument on the fit() function to a percentage of the size of your training dataset. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n",
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:15: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(12, input_dim=8, activation=\"relu\", kernel_initializer=\"uniform\")`\n",
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:16: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(8, activation=\"relu\", kernel_initializer=\"uniform\")`\n",
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:17: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(1, activation=\"sigmoid\", kernel_initializer=\"uniform\")`\n",
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:20: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 514 samples, validate on 254 samples\n",
      "Epoch 1/150\n",
      "514/514 [==============================] - 1s 1ms/step - loss: 0.6796 - acc: 0.6401 - val_loss: 0.6586 - val_acc: 0.6732\n",
      "Epoch 2/150\n",
      "514/514 [==============================] - 0s 167us/step - loss: 0.6686 - acc: 0.6401 - val_loss: 0.6535 - val_acc: 0.6732\n",
      "Epoch 3/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.6614 - acc: 0.6401 - val_loss: 0.6508 - val_acc: 0.6732\n",
      "Epoch 4/150\n",
      "514/514 [==============================] - 0s 190us/step - loss: 0.6540 - acc: 0.6401 - val_loss: 0.6452 - val_acc: 0.6732\n",
      "Epoch 5/150\n",
      "514/514 [==============================] - 0s 166us/step - loss: 0.6448 - acc: 0.6440 - val_loss: 0.6347 - val_acc: 0.7008\n",
      "Epoch 6/150\n",
      "514/514 [==============================] - 0s 166us/step - loss: 0.6371 - acc: 0.6479 - val_loss: 0.6228 - val_acc: 0.6811\n",
      "Epoch 7/150\n",
      "514/514 [==============================] - 0s 180us/step - loss: 0.6297 - acc: 0.6615 - val_loss: 0.6145 - val_acc: 0.6929\n",
      "Epoch 8/150\n",
      "514/514 [==============================] - 0s 180us/step - loss: 0.6220 - acc: 0.6654 - val_loss: 0.6047 - val_acc: 0.6929\n",
      "Epoch 9/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.6145 - acc: 0.6615 - val_loss: 0.6034 - val_acc: 0.6850\n",
      "Epoch 10/150\n",
      "514/514 [==============================] - 0s 179us/step - loss: 0.6096 - acc: 0.6887 - val_loss: 0.5964 - val_acc: 0.6732\n",
      "Epoch 11/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.6024 - acc: 0.6848 - val_loss: 0.6020 - val_acc: 0.6890\n",
      "Epoch 12/150\n",
      "514/514 [==============================] - 0s 170us/step - loss: 0.6069 - acc: 0.6907 - val_loss: 0.5821 - val_acc: 0.7047\n",
      "Epoch 13/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.6027 - acc: 0.6984 - val_loss: 0.5907 - val_acc: 0.6575\n",
      "Epoch 14/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.5928 - acc: 0.7062 - val_loss: 0.5820 - val_acc: 0.7165\n",
      "Epoch 15/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.5892 - acc: 0.7101 - val_loss: 0.5784 - val_acc: 0.6890\n",
      "Epoch 16/150\n",
      "514/514 [==============================] - 0s 170us/step - loss: 0.5980 - acc: 0.6887 - val_loss: 0.5771 - val_acc: 0.6969\n",
      "Epoch 17/150\n",
      "514/514 [==============================] - 0s 169us/step - loss: 0.5871 - acc: 0.6984 - val_loss: 0.5711 - val_acc: 0.7087\n",
      "Epoch 18/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.5829 - acc: 0.7023 - val_loss: 0.5796 - val_acc: 0.7047\n",
      "Epoch 19/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.5831 - acc: 0.7062 - val_loss: 0.5751 - val_acc: 0.7047\n",
      "Epoch 20/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.5804 - acc: 0.7101 - val_loss: 0.5729 - val_acc: 0.7047\n",
      "Epoch 21/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.5790 - acc: 0.7160 - val_loss: 0.5794 - val_acc: 0.6929\n",
      "Epoch 22/150\n",
      "514/514 [==============================] - 0s 166us/step - loss: 0.5737 - acc: 0.7393 - val_loss: 0.5685 - val_acc: 0.7087\n",
      "Epoch 23/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.5760 - acc: 0.6984 - val_loss: 0.5665 - val_acc: 0.7047\n",
      "Epoch 24/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.5613 - acc: 0.7218 - val_loss: 0.5634 - val_acc: 0.7047\n",
      "Epoch 25/150\n",
      "514/514 [==============================] - 0s 170us/step - loss: 0.5674 - acc: 0.7218 - val_loss: 0.5628 - val_acc: 0.7126\n",
      "Epoch 26/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.5650 - acc: 0.7412 - val_loss: 0.5631 - val_acc: 0.7244\n",
      "Epoch 27/150\n",
      "514/514 [==============================] - 0s 184us/step - loss: 0.5627 - acc: 0.7335 - val_loss: 0.5620 - val_acc: 0.7126\n",
      "Epoch 28/150\n",
      "514/514 [==============================] - 0s 170us/step - loss: 0.5613 - acc: 0.7237 - val_loss: 0.6415 - val_acc: 0.5748\n",
      "Epoch 29/150\n",
      "514/514 [==============================] - 0s 166us/step - loss: 0.5610 - acc: 0.7140 - val_loss: 0.5647 - val_acc: 0.7205\n",
      "Epoch 30/150\n",
      "514/514 [==============================] - 0s 170us/step - loss: 0.5553 - acc: 0.7276 - val_loss: 0.5652 - val_acc: 0.7205\n",
      "Epoch 31/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.5495 - acc: 0.7218 - val_loss: 0.5808 - val_acc: 0.7126\n",
      "Epoch 32/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.5594 - acc: 0.7335 - val_loss: 0.5894 - val_acc: 0.6890\n",
      "Epoch 33/150\n",
      "514/514 [==============================] - 0s 171us/step - loss: 0.5724 - acc: 0.7257 - val_loss: 0.5861 - val_acc: 0.7244\n",
      "Epoch 34/150\n",
      "514/514 [==============================] - 0s 190us/step - loss: 0.5530 - acc: 0.7237 - val_loss: 0.5601 - val_acc: 0.7165\n",
      "Epoch 35/150\n",
      "514/514 [==============================] - 0s 165us/step - loss: 0.5500 - acc: 0.7257 - val_loss: 0.5518 - val_acc: 0.7244\n",
      "Epoch 36/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.5549 - acc: 0.7237 - val_loss: 0.5642 - val_acc: 0.7126\n",
      "Epoch 37/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.5463 - acc: 0.7315 - val_loss: 0.5621 - val_acc: 0.7126\n",
      "Epoch 38/150\n",
      "514/514 [==============================] - 0s 169us/step - loss: 0.5450 - acc: 0.7471 - val_loss: 0.5459 - val_acc: 0.7244\n",
      "Epoch 39/150\n",
      "514/514 [==============================] - 0s 178us/step - loss: 0.5540 - acc: 0.7082 - val_loss: 0.5596 - val_acc: 0.7283\n",
      "Epoch 40/150\n",
      "514/514 [==============================] - 0s 169us/step - loss: 0.5442 - acc: 0.7490 - val_loss: 0.5536 - val_acc: 0.7165\n",
      "Epoch 41/150\n",
      "514/514 [==============================] - 0s 169us/step - loss: 0.5422 - acc: 0.7412 - val_loss: 0.5512 - val_acc: 0.7283\n",
      "Epoch 42/150\n",
      "514/514 [==============================] - 0s 180us/step - loss: 0.5484 - acc: 0.7432 - val_loss: 0.5797 - val_acc: 0.7165\n",
      "Epoch 43/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.5450 - acc: 0.7354 - val_loss: 0.5558 - val_acc: 0.7244\n",
      "Epoch 44/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.5433 - acc: 0.7315 - val_loss: 0.5494 - val_acc: 0.7205\n",
      "Epoch 45/150\n",
      "514/514 [==============================] - 0s 168us/step - loss: 0.5358 - acc: 0.7412 - val_loss: 0.5847 - val_acc: 0.6654\n",
      "Epoch 46/150\n",
      "514/514 [==============================] - 0s 184us/step - loss: 0.5481 - acc: 0.7237 - val_loss: 0.5590 - val_acc: 0.7126\n",
      "Epoch 47/150\n",
      "514/514 [==============================] - 0s 186us/step - loss: 0.5507 - acc: 0.7218 - val_loss: 0.5568 - val_acc: 0.7441\n",
      "Epoch 48/150\n",
      "514/514 [==============================] - 0s 224us/step - loss: 0.5406 - acc: 0.7257 - val_loss: 0.5518 - val_acc: 0.7283\n",
      "Epoch 49/150\n",
      "514/514 [==============================] - 0s 217us/step - loss: 0.5397 - acc: 0.7451 - val_loss: 0.5472 - val_acc: 0.7244\n",
      "Epoch 50/150\n",
      "514/514 [==============================] - 0s 184us/step - loss: 0.5374 - acc: 0.7393 - val_loss: 0.5591 - val_acc: 0.7047\n",
      "Epoch 51/150\n",
      "514/514 [==============================] - 0s 199us/step - loss: 0.5321 - acc: 0.7412 - val_loss: 0.5445 - val_acc: 0.7402\n",
      "Epoch 52/150\n",
      "514/514 [==============================] - 0s 187us/step - loss: 0.5351 - acc: 0.7471 - val_loss: 0.5514 - val_acc: 0.7323\n",
      "Epoch 53/150\n",
      "514/514 [==============================] - 0s 206us/step - loss: 0.5337 - acc: 0.7510 - val_loss: 0.5471 - val_acc: 0.7244\n",
      "Epoch 54/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.5324 - acc: 0.7237 - val_loss: 0.5470 - val_acc: 0.7323\n",
      "Epoch 55/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.5289 - acc: 0.7432 - val_loss: 0.5456 - val_acc: 0.7362\n",
      "Epoch 56/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.5353 - acc: 0.7393 - val_loss: 0.5601 - val_acc: 0.7047\n",
      "Epoch 57/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.5376 - acc: 0.7529 - val_loss: 0.5459 - val_acc: 0.7323\n",
      "Epoch 58/150\n",
      "514/514 [==============================] - 0s 188us/step - loss: 0.5292 - acc: 0.7374 - val_loss: 0.5650 - val_acc: 0.6811\n",
      "Epoch 59/150\n",
      "514/514 [==============================] - 0s 170us/step - loss: 0.5416 - acc: 0.7276 - val_loss: 0.5426 - val_acc: 0.7520\n",
      "Epoch 60/150\n",
      "514/514 [==============================] - 0s 170us/step - loss: 0.5316 - acc: 0.7296 - val_loss: 0.5458 - val_acc: 0.7402\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 61/150\n",
      "514/514 [==============================] - 0s 166us/step - loss: 0.5342 - acc: 0.7354 - val_loss: 0.5433 - val_acc: 0.7402\n",
      "Epoch 62/150\n",
      "514/514 [==============================] - 0s 203us/step - loss: 0.5338 - acc: 0.7276 - val_loss: 0.5447 - val_acc: 0.7362\n",
      "Epoch 63/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.5241 - acc: 0.7432 - val_loss: 0.5409 - val_acc: 0.7283\n",
      "Epoch 64/150\n",
      "514/514 [==============================] - 0s 204us/step - loss: 0.5274 - acc: 0.7354 - val_loss: 0.5439 - val_acc: 0.7480\n",
      "Epoch 65/150\n",
      "514/514 [==============================] - 0s 199us/step - loss: 0.5332 - acc: 0.7296 - val_loss: 0.5501 - val_acc: 0.7165\n",
      "Epoch 66/150\n",
      "514/514 [==============================] - 0s 182us/step - loss: 0.5319 - acc: 0.7354 - val_loss: 0.5366 - val_acc: 0.7441\n",
      "Epoch 67/150\n",
      "514/514 [==============================] - 0s 180us/step - loss: 0.5255 - acc: 0.7374 - val_loss: 0.5412 - val_acc: 0.7441\n",
      "Epoch 68/150\n",
      "514/514 [==============================] - 0s 184us/step - loss: 0.5301 - acc: 0.7393 - val_loss: 0.5521 - val_acc: 0.7323\n",
      "Epoch 69/150\n",
      "514/514 [==============================] - 0s 207us/step - loss: 0.5298 - acc: 0.7335 - val_loss: 0.5401 - val_acc: 0.7402\n",
      "Epoch 70/150\n",
      "514/514 [==============================] - 0s 166us/step - loss: 0.5287 - acc: 0.7412 - val_loss: 0.5350 - val_acc: 0.7520\n",
      "Epoch 71/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.5224 - acc: 0.7549 - val_loss: 0.5372 - val_acc: 0.7480\n",
      "Epoch 72/150\n",
      "514/514 [==============================] - 0s 180us/step - loss: 0.5239 - acc: 0.7451 - val_loss: 0.5303 - val_acc: 0.7717\n",
      "Epoch 73/150\n",
      "514/514 [==============================] - 0s 190us/step - loss: 0.5248 - acc: 0.7335 - val_loss: 0.5334 - val_acc: 0.7559\n",
      "Epoch 74/150\n",
      "514/514 [==============================] - 0s 171us/step - loss: 0.5195 - acc: 0.7393 - val_loss: 0.5347 - val_acc: 0.7559\n",
      "Epoch 75/150\n",
      "514/514 [==============================] - 0s 166us/step - loss: 0.5245 - acc: 0.7393 - val_loss: 0.5303 - val_acc: 0.7520\n",
      "Epoch 76/150\n",
      "514/514 [==============================] - 0s 164us/step - loss: 0.5193 - acc: 0.7393 - val_loss: 0.5306 - val_acc: 0.7441\n",
      "Epoch 77/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.5192 - acc: 0.7412 - val_loss: 0.5339 - val_acc: 0.7559\n",
      "Epoch 78/150\n",
      "514/514 [==============================] - 0s 159us/step - loss: 0.5181 - acc: 0.7471 - val_loss: 0.5282 - val_acc: 0.7677\n",
      "Epoch 79/150\n",
      "514/514 [==============================] - 0s 169us/step - loss: 0.5241 - acc: 0.7432 - val_loss: 0.5334 - val_acc: 0.7480\n",
      "Epoch 80/150\n",
      "514/514 [==============================] - 0s 167us/step - loss: 0.5191 - acc: 0.7412 - val_loss: 0.5273 - val_acc: 0.7638\n",
      "Epoch 81/150\n",
      "514/514 [==============================] - 0s 173us/step - loss: 0.5273 - acc: 0.7490 - val_loss: 0.5342 - val_acc: 0.7362\n",
      "Epoch 82/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.5185 - acc: 0.7296 - val_loss: 0.5241 - val_acc: 0.7677\n",
      "Epoch 83/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.5124 - acc: 0.7412 - val_loss: 0.5360 - val_acc: 0.7283\n",
      "Epoch 84/150\n",
      "514/514 [==============================] - 0s 170us/step - loss: 0.5238 - acc: 0.7393 - val_loss: 0.5304 - val_acc: 0.7598\n",
      "Epoch 85/150\n",
      "514/514 [==============================] - 0s 190us/step - loss: 0.5181 - acc: 0.7432 - val_loss: 0.5271 - val_acc: 0.7559\n",
      "Epoch 86/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.5154 - acc: 0.7412 - val_loss: 0.5299 - val_acc: 0.7441\n",
      "Epoch 87/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.5119 - acc: 0.7451 - val_loss: 0.5235 - val_acc: 0.7520\n",
      "Epoch 88/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.5115 - acc: 0.7490 - val_loss: 0.5481 - val_acc: 0.7441\n",
      "Epoch 89/150\n",
      "514/514 [==============================] - 0s 171us/step - loss: 0.5146 - acc: 0.7412 - val_loss: 0.5240 - val_acc: 0.7717\n",
      "Epoch 90/150\n",
      "514/514 [==============================] - 0s 169us/step - loss: 0.5131 - acc: 0.7432 - val_loss: 0.5304 - val_acc: 0.7480\n",
      "Epoch 91/150\n",
      "514/514 [==============================] - 0s 180us/step - loss: 0.5174 - acc: 0.7374 - val_loss: 0.5395 - val_acc: 0.7520\n",
      "Epoch 92/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.5246 - acc: 0.7412 - val_loss: 0.5242 - val_acc: 0.7520\n",
      "Epoch 93/150\n",
      "514/514 [==============================] - 0s 215us/step - loss: 0.5096 - acc: 0.7607 - val_loss: 0.5191 - val_acc: 0.7795\n",
      "Epoch 94/150\n",
      "514/514 [==============================] - 0s 194us/step - loss: 0.5157 - acc: 0.7490 - val_loss: 0.5193 - val_acc: 0.7756\n",
      "Epoch 95/150\n",
      "514/514 [==============================] - 0s 212us/step - loss: 0.5114 - acc: 0.7549 - val_loss: 0.5306 - val_acc: 0.7283\n",
      "Epoch 96/150\n",
      "514/514 [==============================] - 0s 192us/step - loss: 0.5159 - acc: 0.7393 - val_loss: 0.5233 - val_acc: 0.7598\n",
      "Epoch 97/150\n",
      "514/514 [==============================] - 0s 171us/step - loss: 0.5104 - acc: 0.7529 - val_loss: 0.5156 - val_acc: 0.7677\n",
      "Epoch 98/150\n",
      "514/514 [==============================] - 0s 199us/step - loss: 0.5169 - acc: 0.7393 - val_loss: 0.5393 - val_acc: 0.7441\n",
      "Epoch 99/150\n",
      "514/514 [==============================] - 0s 206us/step - loss: 0.5180 - acc: 0.7354 - val_loss: 0.5247 - val_acc: 0.7598\n",
      "Epoch 100/150\n",
      "514/514 [==============================] - 0s 180us/step - loss: 0.5111 - acc: 0.7432 - val_loss: 0.5185 - val_acc: 0.7913\n",
      "Epoch 101/150\n",
      "514/514 [==============================] - 0s 185us/step - loss: 0.5062 - acc: 0.7471 - val_loss: 0.5197 - val_acc: 0.7402\n",
      "Epoch 102/150\n",
      "514/514 [==============================] - 0s 165us/step - loss: 0.5022 - acc: 0.7510 - val_loss: 0.5182 - val_acc: 0.7598\n",
      "Epoch 103/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.5142 - acc: 0.7490 - val_loss: 0.5319 - val_acc: 0.7402\n",
      "Epoch 104/150\n",
      "514/514 [==============================] - 0s 167us/step - loss: 0.5113 - acc: 0.7354 - val_loss: 0.5225 - val_acc: 0.7638\n",
      "Epoch 105/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.5011 - acc: 0.7471 - val_loss: 0.5184 - val_acc: 0.7717\n",
      "Epoch 106/150\n",
      "514/514 [==============================] - 0s 204us/step - loss: 0.5134 - acc: 0.7393 - val_loss: 0.5156 - val_acc: 0.7677\n",
      "Epoch 107/150\n",
      "514/514 [==============================] - 0s 176us/step - loss: 0.5046 - acc: 0.7335 - val_loss: 0.5365 - val_acc: 0.7323\n",
      "Epoch 108/150\n",
      "514/514 [==============================] - 0s 180us/step - loss: 0.5129 - acc: 0.7549 - val_loss: 0.5236 - val_acc: 0.7362\n",
      "Epoch 109/150\n",
      "514/514 [==============================] - 0s 206us/step - loss: 0.5094 - acc: 0.7451 - val_loss: 0.5133 - val_acc: 0.7598\n",
      "Epoch 110/150\n",
      "514/514 [==============================] - 0s 173us/step - loss: 0.4981 - acc: 0.7568 - val_loss: 0.5330 - val_acc: 0.7362\n",
      "Epoch 111/150\n",
      "514/514 [==============================] - 0s 187us/step - loss: 0.5042 - acc: 0.7588 - val_loss: 0.5143 - val_acc: 0.7559\n",
      "Epoch 112/150\n",
      "514/514 [==============================] - 0s 187us/step - loss: 0.4963 - acc: 0.7529 - val_loss: 0.5205 - val_acc: 0.7362\n",
      "Epoch 113/150\n",
      "514/514 [==============================] - 0s 195us/step - loss: 0.4994 - acc: 0.7549 - val_loss: 0.5165 - val_acc: 0.7559\n",
      "Epoch 114/150\n",
      "514/514 [==============================] - 0s 179us/step - loss: 0.5001 - acc: 0.7374 - val_loss: 0.5120 - val_acc: 0.7441\n",
      "Epoch 115/150\n",
      "514/514 [==============================] - 0s 183us/step - loss: 0.4984 - acc: 0.7549 - val_loss: 0.5101 - val_acc: 0.7756\n",
      "Epoch 116/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.4957 - acc: 0.7412 - val_loss: 0.5102 - val_acc: 0.7559\n",
      "Epoch 117/150\n",
      "514/514 [==============================] - 0s 168us/step - loss: 0.5036 - acc: 0.7471 - val_loss: 0.5252 - val_acc: 0.7520\n",
      "Epoch 118/150\n",
      "514/514 [==============================] - 0s 171us/step - loss: 0.4945 - acc: 0.7568 - val_loss: 0.5187 - val_acc: 0.7520\n",
      "Epoch 119/150\n",
      "514/514 [==============================] - 0s 209us/step - loss: 0.5043 - acc: 0.7529 - val_loss: 0.5189 - val_acc: 0.7402\n",
      "Epoch 120/150\n",
      "514/514 [==============================] - 0s 211us/step - loss: 0.4961 - acc: 0.7568 - val_loss: 0.5087 - val_acc: 0.7638\n",
      "Epoch 121/150\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "514/514 [==============================] - 0s 204us/step - loss: 0.4931 - acc: 0.7568 - val_loss: 0.5191 - val_acc: 0.7520\n",
      "Epoch 122/150\n",
      "514/514 [==============================] - 0s 197us/step - loss: 0.4975 - acc: 0.7412 - val_loss: 0.5088 - val_acc: 0.7638\n",
      "Epoch 123/150\n",
      "514/514 [==============================] - 0s 214us/step - loss: 0.4902 - acc: 0.7451 - val_loss: 0.5243 - val_acc: 0.7520\n",
      "Epoch 124/150\n",
      "514/514 [==============================] - 0s 180us/step - loss: 0.4941 - acc: 0.7646 - val_loss: 0.5134 - val_acc: 0.7441\n",
      "Epoch 125/150\n",
      "514/514 [==============================] - 0s 176us/step - loss: 0.4957 - acc: 0.7568 - val_loss: 0.5165 - val_acc: 0.7520\n",
      "Epoch 126/150\n",
      "514/514 [==============================] - 0s 184us/step - loss: 0.4852 - acc: 0.7451 - val_loss: 0.5179 - val_acc: 0.7638\n",
      "Epoch 127/150\n",
      "514/514 [==============================] - 0s 167us/step - loss: 0.4915 - acc: 0.7607 - val_loss: 0.5126 - val_acc: 0.7559\n",
      "Epoch 128/150\n",
      "514/514 [==============================] - 0s 185us/step - loss: 0.4904 - acc: 0.7568 - val_loss: 0.5112 - val_acc: 0.7480\n",
      "Epoch 129/150\n",
      "514/514 [==============================] - 0s 180us/step - loss: 0.4921 - acc: 0.7412 - val_loss: 0.5073 - val_acc: 0.7598\n",
      "Epoch 130/150\n",
      "514/514 [==============================] - 0s 176us/step - loss: 0.4851 - acc: 0.7646 - val_loss: 0.5061 - val_acc: 0.7638\n",
      "Epoch 131/150\n",
      "514/514 [==============================] - 0s 195us/step - loss: 0.4833 - acc: 0.7549 - val_loss: 0.5115 - val_acc: 0.7520\n",
      "Epoch 132/150\n",
      "514/514 [==============================] - 0s 198us/step - loss: 0.4890 - acc: 0.7607 - val_loss: 0.5164 - val_acc: 0.7559\n",
      "Epoch 133/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.4928 - acc: 0.7646 - val_loss: 0.5118 - val_acc: 0.7717\n",
      "Epoch 134/150\n",
      "514/514 [==============================] - 0s 162us/step - loss: 0.4792 - acc: 0.7646 - val_loss: 0.5070 - val_acc: 0.7638\n",
      "Epoch 135/150\n",
      "514/514 [==============================] - 0s 176us/step - loss: 0.4843 - acc: 0.7412 - val_loss: 0.5047 - val_acc: 0.7756\n",
      "Epoch 136/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.4821 - acc: 0.7471 - val_loss: 0.5095 - val_acc: 0.7717\n",
      "Epoch 137/150\n",
      "514/514 [==============================] - 0s 194us/step - loss: 0.4930 - acc: 0.7451 - val_loss: 0.5121 - val_acc: 0.7559\n",
      "Epoch 138/150\n",
      "514/514 [==============================] - 0s 219us/step - loss: 0.4814 - acc: 0.7685 - val_loss: 0.5074 - val_acc: 0.7598\n",
      "Epoch 139/150\n",
      "514/514 [==============================] - 0s 200us/step - loss: 0.4862 - acc: 0.7626 - val_loss: 0.5049 - val_acc: 0.7677\n",
      "Epoch 140/150\n",
      "514/514 [==============================] - 0s 210us/step - loss: 0.4765 - acc: 0.7665 - val_loss: 0.5088 - val_acc: 0.7677\n",
      "Epoch 141/150\n",
      "514/514 [==============================] - 0s 195us/step - loss: 0.4785 - acc: 0.7549 - val_loss: 0.5039 - val_acc: 0.7717\n",
      "Epoch 142/150\n",
      "514/514 [==============================] - 0s 184us/step - loss: 0.4788 - acc: 0.7607 - val_loss: 0.5076 - val_acc: 0.7756\n",
      "Epoch 143/150\n",
      "514/514 [==============================] - 0s 197us/step - loss: 0.4850 - acc: 0.7510 - val_loss: 0.5006 - val_acc: 0.7559\n",
      "Epoch 144/150\n",
      "514/514 [==============================] - 0s 200us/step - loss: 0.4937 - acc: 0.7665 - val_loss: 0.5255 - val_acc: 0.7638\n",
      "Epoch 145/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.4847 - acc: 0.7490 - val_loss: 0.5091 - val_acc: 0.7756\n",
      "Epoch 146/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.4816 - acc: 0.7549 - val_loss: 0.5047 - val_acc: 0.7717\n",
      "Epoch 147/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.4724 - acc: 0.7529 - val_loss: 0.5096 - val_acc: 0.7559\n",
      "Epoch 148/150\n",
      "514/514 [==============================] - 0s 185us/step - loss: 0.4808 - acc: 0.7626 - val_loss: 0.5240 - val_acc: 0.7559\n",
      "Epoch 149/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.4718 - acc: 0.7724 - val_loss: 0.5134 - val_acc: 0.7638\n",
      "Epoch 150/150\n",
      "514/514 [==============================] - 0s 173us/step - loss: 0.4753 - acc: 0.7724 - val_loss: 0.5126 - val_acc: 0.7598\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f5eecc88cc0>"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# MLP with automatic validation set\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense\n",
    "import numpy\n",
    "# fix random seed for reproducibility\n",
    "seed = 7\n",
    "numpy.random.seed(seed)\n",
    "# load pima indians dataset\n",
    "dataset = numpy.loadtxt(\"pima-indians-diabetes.csv\", delimiter=\",\")\n",
    "# split into input (X) and output (Y) variables\n",
    "X = dataset[:,0:8]\n",
    "Y = dataset[:,8]\n",
    "# create model\n",
    "model = Sequential()\n",
    "model.add(Dense(12, input_dim=8, init='uniform', activation='relu')) \n",
    "model.add(Dense(8, init='uniform', activation='relu'))\n",
    "model.add(Dense(1, init='uniform', activation='sigmoid'))\n",
    "# Compile model\n",
    "model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy']) # Fit the model\n",
    "model.fit(X, Y, validation_split=0.33, nb_epoch=150, batch_size=10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### 8.2.2 Use a Manual Verification Dataset\n",
    "Keras also allows you to manually specify the dataset to use for validation during training.\n",
    "\n",
    "The validation dataset can be specified to the fit() function in Keras by the validation data argument."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/sklearn/cross_validation.py:41: DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. Also note that the interface of the new CV iterators are different from that of this module. This module will be removed in 0.20.\n",
      "  \"This module will be removed in 0.20.\", DeprecationWarning)\n",
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:17: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(12, input_dim=8, activation=\"relu\", kernel_initializer=\"uniform\")`\n",
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:18: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(8, activation=\"relu\", kernel_initializer=\"uniform\")`\n",
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:19: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(1, activation=\"sigmoid\", kernel_initializer=\"uniform\")`\n",
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:23: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 514 samples, validate on 254 samples\n",
      "Epoch 1/150\n",
      "514/514 [==============================] - 1s 1ms/step - loss: 0.6789 - acc: 0.6576 - val_loss: 0.6746 - val_acc: 0.6378\n",
      "Epoch 2/150\n",
      "514/514 [==============================] - 0s 182us/step - loss: 0.6642 - acc: 0.6576 - val_loss: 0.6674 - val_acc: 0.6378\n",
      "Epoch 3/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.6582 - acc: 0.6576 - val_loss: 0.6584 - val_acc: 0.6417\n",
      "Epoch 4/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.6538 - acc: 0.6576 - val_loss: 0.6530 - val_acc: 0.6378\n",
      "Epoch 5/150\n",
      "514/514 [==============================] - 0s 179us/step - loss: 0.6501 - acc: 0.6634 - val_loss: 0.6605 - val_acc: 0.6378\n",
      "Epoch 6/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.6446 - acc: 0.6537 - val_loss: 0.6385 - val_acc: 0.6535\n",
      "Epoch 7/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.6332 - acc: 0.6673 - val_loss: 0.6470 - val_acc: 0.6575\n",
      "Epoch 8/150\n",
      "514/514 [==============================] - 0s 190us/step - loss: 0.6254 - acc: 0.6751 - val_loss: 0.6258 - val_acc: 0.6772\n",
      "Epoch 9/150\n",
      "514/514 [==============================] - 0s 185us/step - loss: 0.6154 - acc: 0.6829 - val_loss: 0.6135 - val_acc: 0.6969\n",
      "Epoch 10/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.6064 - acc: 0.6965 - val_loss: 0.6332 - val_acc: 0.6417\n",
      "Epoch 11/150\n",
      "514/514 [==============================] - 0s 170us/step - loss: 0.6013 - acc: 0.6984 - val_loss: 0.5991 - val_acc: 0.7047\n",
      "Epoch 12/150\n",
      "514/514 [==============================] - 0s 178us/step - loss: 0.6009 - acc: 0.6868 - val_loss: 0.5979 - val_acc: 0.6890\n",
      "Epoch 13/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.5995 - acc: 0.7101 - val_loss: 0.6193 - val_acc: 0.6654\n",
      "Epoch 14/150\n",
      "514/514 [==============================] - 0s 209us/step - loss: 0.5848 - acc: 0.7023 - val_loss: 0.6106 - val_acc: 0.6732\n",
      "Epoch 15/150\n",
      "514/514 [==============================] - 0s 183us/step - loss: 0.5886 - acc: 0.7140 - val_loss: 0.5835 - val_acc: 0.7283\n",
      "Epoch 16/150\n",
      "514/514 [==============================] - 0s 187us/step - loss: 0.5875 - acc: 0.7121 - val_loss: 0.5874 - val_acc: 0.7008\n",
      "Epoch 17/150\n",
      "514/514 [==============================] - 0s 178us/step - loss: 0.5815 - acc: 0.7179 - val_loss: 0.5931 - val_acc: 0.6929\n",
      "Epoch 18/150\n",
      "514/514 [==============================] - 0s 191us/step - loss: 0.5751 - acc: 0.7160 - val_loss: 0.5762 - val_acc: 0.7165\n",
      "Epoch 19/150\n",
      "514/514 [==============================] - 0s 203us/step - loss: 0.5727 - acc: 0.7257 - val_loss: 0.5742 - val_acc: 0.7126\n",
      "Epoch 20/150\n",
      "514/514 [==============================] - 0s 200us/step - loss: 0.5738 - acc: 0.7101 - val_loss: 0.5700 - val_acc: 0.7283\n",
      "Epoch 21/150\n",
      "514/514 [==============================] - 0s 182us/step - loss: 0.5661 - acc: 0.7257 - val_loss: 0.5758 - val_acc: 0.7205\n",
      "Epoch 22/150\n",
      "514/514 [==============================] - 0s 186us/step - loss: 0.5633 - acc: 0.7296 - val_loss: 0.5715 - val_acc: 0.7283\n",
      "Epoch 23/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.5713 - acc: 0.7121 - val_loss: 0.5699 - val_acc: 0.7283\n",
      "Epoch 24/150\n",
      "514/514 [==============================] - 0s 210us/step - loss: 0.5658 - acc: 0.7237 - val_loss: 0.5663 - val_acc: 0.7126\n",
      "Epoch 25/150\n",
      "514/514 [==============================] - 0s 180us/step - loss: 0.5628 - acc: 0.7140 - val_loss: 0.5894 - val_acc: 0.7165\n",
      "Epoch 26/150\n",
      "514/514 [==============================] - 0s 182us/step - loss: 0.5598 - acc: 0.7335 - val_loss: 0.5628 - val_acc: 0.7165\n",
      "Epoch 27/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.5564 - acc: 0.7198 - val_loss: 0.5691 - val_acc: 0.7323\n",
      "Epoch 28/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.5558 - acc: 0.7237 - val_loss: 0.5614 - val_acc: 0.7283\n",
      "Epoch 29/150\n",
      "514/514 [==============================] - 0s 182us/step - loss: 0.5564 - acc: 0.7140 - val_loss: 0.5640 - val_acc: 0.7283\n",
      "Epoch 30/150\n",
      "514/514 [==============================] - 0s 196us/step - loss: 0.5602 - acc: 0.7296 - val_loss: 0.5684 - val_acc: 0.7126\n",
      "Epoch 31/150\n",
      "514/514 [==============================] - 0s 189us/step - loss: 0.5516 - acc: 0.7198 - val_loss: 0.5618 - val_acc: 0.7244\n",
      "Epoch 32/150\n",
      "514/514 [==============================] - 0s 196us/step - loss: 0.5455 - acc: 0.7257 - val_loss: 0.5594 - val_acc: 0.7205\n",
      "Epoch 33/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.5497 - acc: 0.7315 - val_loss: 0.5678 - val_acc: 0.7126\n",
      "Epoch 34/150\n",
      "514/514 [==============================] - 0s 193us/step - loss: 0.5495 - acc: 0.7237 - val_loss: 0.5555 - val_acc: 0.7244\n",
      "Epoch 35/150\n",
      "514/514 [==============================] - 0s 179us/step - loss: 0.5449 - acc: 0.7296 - val_loss: 0.5580 - val_acc: 0.7126\n",
      "Epoch 36/150\n",
      "514/514 [==============================] - 0s 171us/step - loss: 0.5382 - acc: 0.7315 - val_loss: 0.5619 - val_acc: 0.7126\n",
      "Epoch 37/150\n",
      "514/514 [==============================] - 0s 189us/step - loss: 0.5378 - acc: 0.7315 - val_loss: 0.5582 - val_acc: 0.7126\n",
      "Epoch 38/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.5448 - acc: 0.7218 - val_loss: 0.5531 - val_acc: 0.7520\n",
      "Epoch 39/150\n",
      "514/514 [==============================] - 0s 178us/step - loss: 0.5416 - acc: 0.7393 - val_loss: 0.5509 - val_acc: 0.7441\n",
      "Epoch 40/150\n",
      "514/514 [==============================] - 0s 201us/step - loss: 0.5399 - acc: 0.7374 - val_loss: 0.5506 - val_acc: 0.7323\n",
      "Epoch 41/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.5372 - acc: 0.7354 - val_loss: 0.5542 - val_acc: 0.7323\n",
      "Epoch 42/150\n",
      "514/514 [==============================] - 0s 198us/step - loss: 0.5361 - acc: 0.7296 - val_loss: 0.5610 - val_acc: 0.7362\n",
      "Epoch 43/150\n",
      "514/514 [==============================] - 0s 179us/step - loss: 0.5366 - acc: 0.7374 - val_loss: 0.5710 - val_acc: 0.7205\n",
      "Epoch 44/150\n",
      "514/514 [==============================] - 0s 173us/step - loss: 0.5491 - acc: 0.7198 - val_loss: 0.5511 - val_acc: 0.7323\n",
      "Epoch 45/150\n",
      "514/514 [==============================] - 0s 188us/step - loss: 0.5358 - acc: 0.7393 - val_loss: 0.5668 - val_acc: 0.7205\n",
      "Epoch 46/150\n",
      "514/514 [==============================] - 0s 193us/step - loss: 0.5259 - acc: 0.7412 - val_loss: 0.5527 - val_acc: 0.7244\n",
      "Epoch 47/150\n",
      "514/514 [==============================] - 0s 178us/step - loss: 0.5301 - acc: 0.7432 - val_loss: 0.5761 - val_acc: 0.7126\n",
      "Epoch 48/150\n",
      "514/514 [==============================] - 0s 193us/step - loss: 0.5357 - acc: 0.7276 - val_loss: 0.5605 - val_acc: 0.7165\n",
      "Epoch 49/150\n",
      "514/514 [==============================] - 0s 186us/step - loss: 0.5267 - acc: 0.7354 - val_loss: 0.5453 - val_acc: 0.7283\n",
      "Epoch 50/150\n",
      "514/514 [==============================] - 0s 200us/step - loss: 0.5232 - acc: 0.7276 - val_loss: 0.5611 - val_acc: 0.7205\n",
      "Epoch 51/150\n",
      "514/514 [==============================] - 0s 185us/step - loss: 0.5236 - acc: 0.7335 - val_loss: 0.5509 - val_acc: 0.7441\n",
      "Epoch 52/150\n",
      "514/514 [==============================] - 0s 208us/step - loss: 0.5181 - acc: 0.7529 - val_loss: 0.5679 - val_acc: 0.7244\n",
      "Epoch 53/150\n",
      "514/514 [==============================] - 0s 193us/step - loss: 0.5280 - acc: 0.7490 - val_loss: 0.5530 - val_acc: 0.7362\n",
      "Epoch 54/150\n",
      "514/514 [==============================] - 0s 192us/step - loss: 0.5451 - acc: 0.7296 - val_loss: 0.5676 - val_acc: 0.7165\n",
      "Epoch 55/150\n",
      "514/514 [==============================] - 0s 195us/step - loss: 0.5246 - acc: 0.7374 - val_loss: 0.5488 - val_acc: 0.7362\n",
      "Epoch 56/150\n",
      "514/514 [==============================] - 0s 191us/step - loss: 0.5261 - acc: 0.7354 - val_loss: 0.5647 - val_acc: 0.7205\n",
      "Epoch 57/150\n",
      "514/514 [==============================] - 0s 208us/step - loss: 0.5297 - acc: 0.7218 - val_loss: 0.5709 - val_acc: 0.7165\n",
      "Epoch 58/150\n",
      "514/514 [==============================] - 0s 213us/step - loss: 0.5164 - acc: 0.7471 - val_loss: 0.5443 - val_acc: 0.7244\n",
      "Epoch 59/150\n",
      "514/514 [==============================] - 0s 204us/step - loss: 0.5123 - acc: 0.7490 - val_loss: 0.5495 - val_acc: 0.7126\n",
      "Epoch 60/150\n",
      "514/514 [==============================] - 0s 190us/step - loss: 0.5151 - acc: 0.7568 - val_loss: 0.5659 - val_acc: 0.7362\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 61/150\n",
      "514/514 [==============================] - 0s 190us/step - loss: 0.5160 - acc: 0.7374 - val_loss: 0.5447 - val_acc: 0.7323\n",
      "Epoch 62/150\n",
      "514/514 [==============================] - 0s 294us/step - loss: 0.5077 - acc: 0.7529 - val_loss: 0.5514 - val_acc: 0.7283\n",
      "Epoch 63/150\n",
      "514/514 [==============================] - 0s 225us/step - loss: 0.5203 - acc: 0.7529 - val_loss: 0.5426 - val_acc: 0.7362\n",
      "Epoch 64/150\n",
      "514/514 [==============================] - 0s 207us/step - loss: 0.5089 - acc: 0.7432 - val_loss: 0.5482 - val_acc: 0.7244\n",
      "Epoch 65/150\n",
      "514/514 [==============================] - 0s 208us/step - loss: 0.5104 - acc: 0.7529 - val_loss: 0.5517 - val_acc: 0.7362\n",
      "Epoch 66/150\n",
      "514/514 [==============================] - 0s 183us/step - loss: 0.5251 - acc: 0.7471 - val_loss: 0.5441 - val_acc: 0.7441\n",
      "Epoch 67/150\n",
      "514/514 [==============================] - 0s 209us/step - loss: 0.5100 - acc: 0.7568 - val_loss: 0.5456 - val_acc: 0.7480\n",
      "Epoch 68/150\n",
      "514/514 [==============================] - 0s 189us/step - loss: 0.5059 - acc: 0.7451 - val_loss: 0.5441 - val_acc: 0.7402\n",
      "Epoch 69/150\n",
      "514/514 [==============================] - 0s 186us/step - loss: 0.5057 - acc: 0.7568 - val_loss: 0.5437 - val_acc: 0.7362\n",
      "Epoch 70/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.5075 - acc: 0.7412 - val_loss: 0.5418 - val_acc: 0.7323\n",
      "Epoch 71/150\n",
      "514/514 [==============================] - 0s 187us/step - loss: 0.4988 - acc: 0.7607 - val_loss: 0.5403 - val_acc: 0.7283\n",
      "Epoch 72/150\n",
      "514/514 [==============================] - 0s 185us/step - loss: 0.5086 - acc: 0.7529 - val_loss: 0.5493 - val_acc: 0.7480\n",
      "Epoch 73/150\n",
      "514/514 [==============================] - 0s 189us/step - loss: 0.4971 - acc: 0.7607 - val_loss: 0.5406 - val_acc: 0.7362\n",
      "Epoch 74/150\n",
      "514/514 [==============================] - 0s 176us/step - loss: 0.5020 - acc: 0.7490 - val_loss: 0.5559 - val_acc: 0.7283\n",
      "Epoch 75/150\n",
      "514/514 [==============================] - 0s 197us/step - loss: 0.5106 - acc: 0.7588 - val_loss: 0.5496 - val_acc: 0.7323\n",
      "Epoch 76/150\n",
      "514/514 [==============================] - 0s 187us/step - loss: 0.5020 - acc: 0.7568 - val_loss: 0.5463 - val_acc: 0.7441\n",
      "Epoch 77/150\n",
      "514/514 [==============================] - 0s 196us/step - loss: 0.5028 - acc: 0.7607 - val_loss: 0.5485 - val_acc: 0.7362\n",
      "Epoch 78/150\n",
      "514/514 [==============================] - 0s 190us/step - loss: 0.5060 - acc: 0.7471 - val_loss: 0.5355 - val_acc: 0.7441\n",
      "Epoch 79/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.4937 - acc: 0.7626 - val_loss: 0.5473 - val_acc: 0.7638\n",
      "Epoch 80/150\n",
      "514/514 [==============================] - 0s 189us/step - loss: 0.4975 - acc: 0.7665 - val_loss: 0.5375 - val_acc: 0.7441\n",
      "Epoch 81/150\n",
      "514/514 [==============================] - 0s 196us/step - loss: 0.4944 - acc: 0.7529 - val_loss: 0.5373 - val_acc: 0.7441\n",
      "Epoch 82/150\n",
      "514/514 [==============================] - 0s 176us/step - loss: 0.4986 - acc: 0.7529 - val_loss: 0.5475 - val_acc: 0.7480\n",
      "Epoch 83/150\n",
      "514/514 [==============================] - 0s 190us/step - loss: 0.5111 - acc: 0.7510 - val_loss: 0.5500 - val_acc: 0.7283\n",
      "Epoch 84/150\n",
      "514/514 [==============================] - 0s 173us/step - loss: 0.5116 - acc: 0.7471 - val_loss: 0.5477 - val_acc: 0.7283\n",
      "Epoch 85/150\n",
      "514/514 [==============================] - 0s 178us/step - loss: 0.4985 - acc: 0.7549 - val_loss: 0.5363 - val_acc: 0.7362\n",
      "Epoch 86/150\n",
      "514/514 [==============================] - 0s 172us/step - loss: 0.4926 - acc: 0.7763 - val_loss: 0.5460 - val_acc: 0.7323\n",
      "Epoch 87/150\n",
      "514/514 [==============================] - 0s 201us/step - loss: 0.4898 - acc: 0.7665 - val_loss: 0.5349 - val_acc: 0.7520\n",
      "Epoch 88/150\n",
      "514/514 [==============================] - 0s 169us/step - loss: 0.4903 - acc: 0.7724 - val_loss: 0.5375 - val_acc: 0.7362\n",
      "Epoch 89/150\n",
      "514/514 [==============================] - 0s 188us/step - loss: 0.4879 - acc: 0.7704 - val_loss: 0.5556 - val_acc: 0.7244\n",
      "Epoch 90/150\n",
      "514/514 [==============================] - 0s 193us/step - loss: 0.4853 - acc: 0.7646 - val_loss: 0.5371 - val_acc: 0.7362\n",
      "Epoch 91/150\n",
      "514/514 [==============================] - 0s 195us/step - loss: 0.4898 - acc: 0.7588 - val_loss: 0.5365 - val_acc: 0.7244\n",
      "Epoch 92/150\n",
      "514/514 [==============================] - 0s 169us/step - loss: 0.4869 - acc: 0.7685 - val_loss: 0.5358 - val_acc: 0.7480\n",
      "Epoch 93/150\n",
      "514/514 [==============================] - 0s 171us/step - loss: 0.4941 - acc: 0.7646 - val_loss: 0.5528 - val_acc: 0.7402\n",
      "Epoch 94/150\n",
      "514/514 [==============================] - 0s 179us/step - loss: 0.4870 - acc: 0.7568 - val_loss: 0.5323 - val_acc: 0.7520\n",
      "Epoch 95/150\n",
      "514/514 [==============================] - 0s 194us/step - loss: 0.4814 - acc: 0.7646 - val_loss: 0.5346 - val_acc: 0.7441\n",
      "Epoch 96/150\n",
      "514/514 [==============================] - 0s 179us/step - loss: 0.4877 - acc: 0.7568 - val_loss: 0.5355 - val_acc: 0.7520\n",
      "Epoch 97/150\n",
      "514/514 [==============================] - 0s 196us/step - loss: 0.4785 - acc: 0.7763 - val_loss: 0.5314 - val_acc: 0.7362\n",
      "Epoch 98/150\n",
      "514/514 [==============================] - 0s 196us/step - loss: 0.4794 - acc: 0.7685 - val_loss: 0.5395 - val_acc: 0.7520\n",
      "Epoch 99/150\n",
      "514/514 [==============================] - 0s 197us/step - loss: 0.4787 - acc: 0.7685 - val_loss: 0.5794 - val_acc: 0.7087\n",
      "Epoch 100/150\n",
      "514/514 [==============================] - 0s 185us/step - loss: 0.4859 - acc: 0.7646 - val_loss: 0.5373 - val_acc: 0.7323\n",
      "Epoch 101/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.4828 - acc: 0.7665 - val_loss: 0.5371 - val_acc: 0.7480\n",
      "Epoch 102/150\n",
      "514/514 [==============================] - 0s 178us/step - loss: 0.4901 - acc: 0.7724 - val_loss: 0.5379 - val_acc: 0.7520\n",
      "Epoch 103/150\n",
      "514/514 [==============================] - 0s 189us/step - loss: 0.4775 - acc: 0.7704 - val_loss: 0.5360 - val_acc: 0.7205\n",
      "Epoch 104/150\n",
      "514/514 [==============================] - 0s 226us/step - loss: 0.4783 - acc: 0.7665 - val_loss: 0.5326 - val_acc: 0.7559\n",
      "Epoch 105/150\n",
      "514/514 [==============================] - 0s 167us/step - loss: 0.4744 - acc: 0.7646 - val_loss: 0.5423 - val_acc: 0.7559\n",
      "Epoch 106/150\n",
      "514/514 [==============================] - 0s 178us/step - loss: 0.4750 - acc: 0.7607 - val_loss: 0.5404 - val_acc: 0.7402\n",
      "Epoch 107/150\n",
      "514/514 [==============================] - 0s 184us/step - loss: 0.4759 - acc: 0.7685 - val_loss: 0.5413 - val_acc: 0.7362\n",
      "Epoch 108/150\n",
      "514/514 [==============================] - 0s 171us/step - loss: 0.4856 - acc: 0.7646 - val_loss: 0.5491 - val_acc: 0.7441\n",
      "Epoch 109/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.4852 - acc: 0.7626 - val_loss: 0.5369 - val_acc: 0.7441\n",
      "Epoch 110/150\n",
      "514/514 [==============================] - 0s 192us/step - loss: 0.4739 - acc: 0.7763 - val_loss: 0.5440 - val_acc: 0.7323\n",
      "Epoch 111/150\n",
      "514/514 [==============================] - 0s 186us/step - loss: 0.4729 - acc: 0.7743 - val_loss: 0.5375 - val_acc: 0.7441\n",
      "Epoch 112/150\n",
      "514/514 [==============================] - 0s 174us/step - loss: 0.4717 - acc: 0.7840 - val_loss: 0.5359 - val_acc: 0.7480\n",
      "Epoch 113/150\n",
      "514/514 [==============================] - 0s 171us/step - loss: 0.4717 - acc: 0.7704 - val_loss: 0.5370 - val_acc: 0.7480\n",
      "Epoch 114/150\n",
      "514/514 [==============================] - 0s 178us/step - loss: 0.4803 - acc: 0.7704 - val_loss: 0.5460 - val_acc: 0.7677\n",
      "Epoch 115/150\n",
      "514/514 [==============================] - 0s 166us/step - loss: 0.4746 - acc: 0.7782 - val_loss: 0.5340 - val_acc: 0.7677\n",
      "Epoch 116/150\n",
      "514/514 [==============================] - 0s 179us/step - loss: 0.4821 - acc: 0.7588 - val_loss: 0.5370 - val_acc: 0.7559\n",
      "Epoch 117/150\n",
      "514/514 [==============================] - 0s 179us/step - loss: 0.4709 - acc: 0.7743 - val_loss: 0.5399 - val_acc: 0.7362\n",
      "Epoch 118/150\n",
      "514/514 [==============================] - 0s 182us/step - loss: 0.4832 - acc: 0.7665 - val_loss: 0.5314 - val_acc: 0.7638\n",
      "Epoch 119/150\n",
      "514/514 [==============================] - 0s 168us/step - loss: 0.4774 - acc: 0.7724 - val_loss: 0.5393 - val_acc: 0.7559\n",
      "Epoch 120/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.4723 - acc: 0.7626 - val_loss: 0.5346 - val_acc: 0.7441\n",
      "Epoch 121/150\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "514/514 [==============================] - 0s 181us/step - loss: 0.4677 - acc: 0.7685 - val_loss: 0.5327 - val_acc: 0.7677\n",
      "Epoch 122/150\n",
      "514/514 [==============================] - 0s 178us/step - loss: 0.4693 - acc: 0.7802 - val_loss: 0.5492 - val_acc: 0.7362\n",
      "Epoch 123/150\n",
      "514/514 [==============================] - 0s 178us/step - loss: 0.4733 - acc: 0.7646 - val_loss: 0.5346 - val_acc: 0.7402\n",
      "Epoch 124/150\n",
      "514/514 [==============================] - 0s 190us/step - loss: 0.4656 - acc: 0.7724 - val_loss: 0.5307 - val_acc: 0.7559\n",
      "Epoch 125/150\n",
      "514/514 [==============================] - 0s 187us/step - loss: 0.4736 - acc: 0.7763 - val_loss: 0.5358 - val_acc: 0.7402\n",
      "Epoch 126/150\n",
      "514/514 [==============================] - 0s 224us/step - loss: 0.4681 - acc: 0.7724 - val_loss: 0.5337 - val_acc: 0.7559\n",
      "Epoch 127/150\n",
      "514/514 [==============================] - 0s 214us/step - loss: 0.4673 - acc: 0.7743 - val_loss: 0.5312 - val_acc: 0.7638\n",
      "Epoch 128/150\n",
      "514/514 [==============================] - 0s 191us/step - loss: 0.4654 - acc: 0.7802 - val_loss: 0.5339 - val_acc: 0.7283\n",
      "Epoch 129/150\n",
      "514/514 [==============================] - 0s 210us/step - loss: 0.4596 - acc: 0.7685 - val_loss: 0.5364 - val_acc: 0.7638\n",
      "Epoch 130/150\n",
      "514/514 [==============================] - 0s 200us/step - loss: 0.4711 - acc: 0.7685 - val_loss: 0.5334 - val_acc: 0.7402\n",
      "Epoch 131/150\n",
      "514/514 [==============================] - 0s 215us/step - loss: 0.4617 - acc: 0.7665 - val_loss: 0.5329 - val_acc: 0.7520\n",
      "Epoch 132/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.4657 - acc: 0.7704 - val_loss: 0.5628 - val_acc: 0.7283\n",
      "Epoch 133/150\n",
      "514/514 [==============================] - 0s 193us/step - loss: 0.4688 - acc: 0.7840 - val_loss: 0.5359 - val_acc: 0.7598\n",
      "Epoch 134/150\n",
      "514/514 [==============================] - 0s 191us/step - loss: 0.4721 - acc: 0.7802 - val_loss: 0.5330 - val_acc: 0.7520\n",
      "Epoch 135/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.4622 - acc: 0.7743 - val_loss: 0.5465 - val_acc: 0.7362\n",
      "Epoch 136/150\n",
      "514/514 [==============================] - 0s 198us/step - loss: 0.4644 - acc: 0.7743 - val_loss: 0.5401 - val_acc: 0.7402\n",
      "Epoch 137/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.4714 - acc: 0.7763 - val_loss: 0.5386 - val_acc: 0.7402\n",
      "Epoch 138/150\n",
      "514/514 [==============================] - 0s 182us/step - loss: 0.4634 - acc: 0.7840 - val_loss: 0.5374 - val_acc: 0.7402\n",
      "Epoch 139/150\n",
      "514/514 [==============================] - 0s 175us/step - loss: 0.4667 - acc: 0.7918 - val_loss: 0.5512 - val_acc: 0.7677\n",
      "Epoch 140/150\n",
      "514/514 [==============================] - 0s 169us/step - loss: 0.4684 - acc: 0.7840 - val_loss: 0.5337 - val_acc: 0.7480\n",
      "Epoch 141/150\n",
      "514/514 [==============================] - 0s 173us/step - loss: 0.4629 - acc: 0.7724 - val_loss: 0.5304 - val_acc: 0.7402\n",
      "Epoch 142/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.4667 - acc: 0.7588 - val_loss: 0.5440 - val_acc: 0.7480\n",
      "Epoch 143/150\n",
      "514/514 [==============================] - 0s 194us/step - loss: 0.4660 - acc: 0.7704 - val_loss: 0.5357 - val_acc: 0.7362\n",
      "Epoch 144/150\n",
      "514/514 [==============================] - 0s 170us/step - loss: 0.4571 - acc: 0.7860 - val_loss: 0.5368 - val_acc: 0.7441\n",
      "Epoch 145/150\n",
      "514/514 [==============================] - 0s 173us/step - loss: 0.4576 - acc: 0.7685 - val_loss: 0.5388 - val_acc: 0.7441\n",
      "Epoch 146/150\n",
      "514/514 [==============================] - 0s 187us/step - loss: 0.4607 - acc: 0.7665 - val_loss: 0.5345 - val_acc: 0.7402\n",
      "Epoch 147/150\n",
      "514/514 [==============================] - 0s 181us/step - loss: 0.4615 - acc: 0.7782 - val_loss: 0.5299 - val_acc: 0.7480\n",
      "Epoch 148/150\n",
      "514/514 [==============================] - 0s 203us/step - loss: 0.4543 - acc: 0.7821 - val_loss: 0.5287 - val_acc: 0.7559\n",
      "Epoch 149/150\n",
      "514/514 [==============================] - 0s 186us/step - loss: 0.4622 - acc: 0.7743 - val_loss: 0.5314 - val_acc: 0.7559\n",
      "Epoch 150/150\n",
      "514/514 [==============================] - 0s 177us/step - loss: 0.4572 - acc: 0.7802 - val_loss: 0.5360 - val_acc: 0.7677\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x7f5edaf3f710>"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# MLP with manual validation set\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense\n",
    "from sklearn.cross_validation import train_test_split\n",
    "import numpy\n",
    "# fix random seed for reproducibility\n",
    "seed = 7\n",
    "numpy.random.seed(seed)\n",
    "# load pima indians dataset\n",
    "dataset = numpy.loadtxt(\"pima-indians-diabetes.csv\", delimiter=\",\")\n",
    "# split into input (X) and output (Y) variables\n",
    "X = dataset[:,0:8]\n",
    "Y = dataset[:,8]\n",
    "# split into 67% for train and 33% for test\n",
    "X_train, X_test, y_train, y_test = train_test_split(X, Y, test_size=0.33, random_state=seed) # create model\n",
    "model = Sequential()\n",
    "model.add(Dense(12, input_dim=8, init='uniform', activation='relu'))\n",
    "model.add(Dense(8, init='uniform', activation='relu'))\n",
    "model.add(Dense(1, init='uniform', activation='sigmoid'))\n",
    "# Compile model\n",
    "model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy'])\n",
    "# Fit the model\n",
    "model.fit(X_train, y_train, validation_data=(X_test,y_test), nb_epoch=150, batch_size=10)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### 8.3 Manual k-Fold Cross Validation\n",
    "The gold standard for machine learning model evaluation is k-fold cross validation. It provides a robust estimate of the performance of a model on unseen data. \n",
    "\n",
    "It does this by splitting the training dataset into k subsets and takes turns training models on all subsets except one which is held out, and evaluating model performance on the held out validation dataset. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Cross validation is often not used for evaluating deep learning models because of the greater computational expense. \n",
    "\n",
    "In the example below we use the handy StratifiedKFold class1 from the scikit-learn Python machine learning library to split up the training dataset into 10 folds. The folds are stratified, meaning that the algorithm attempts to balance the number of instances of each class in each fold.\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:22: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(12, input_dim=8, activation=\"relu\", kernel_initializer=\"uniform\")`\n",
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:23: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(8, activation=\"relu\", kernel_initializer=\"uniform\")`\n",
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:24: UserWarning: Update your `Dense` call to the Keras 2 API: `Dense(1, activation=\"sigmoid\", kernel_initializer=\"uniform\")`\n",
      "/home/nbuser/anaconda3_501/lib/python3.6/site-packages/ipykernel/__main__.py:27: UserWarning: The `nb_epoch` argument in `fit` has been renamed `epochs`.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "acc: 74.03%\n",
      "acc: 75.32%\n",
      "acc: 77.92%\n",
      "acc: 77.92%\n",
      "acc: 77.92%\n",
      "acc: 72.73%\n",
      "acc: 76.62%\n",
      "acc: 74.03%\n",
      "acc: 72.37%\n",
      "acc: 72.37%\n",
      "75.12% (+/- 2.22%)\n"
     ]
    }
   ],
   "source": [
    "# MLP for Pima Indians Dataset with 10-fold cross validation\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense\n",
    "from sklearn.cross_validation import StratifiedKFold\n",
    "import numpy\n",
    "# fix random seed for reproducibility\n",
    "seed = 7\n",
    "numpy.random.seed(seed)\n",
    "# load pima indians dataset\n",
    "dataset = numpy.loadtxt(\"pima-indians-diabetes.csv\", delimiter=\",\")\n",
    "\n",
    "# split into input (X) and output (Y) variables\n",
    "X = dataset[:,0:8]\n",
    "Y = dataset[:,8]\n",
    "# define 10-fold cross validation test harness\n",
    "kfold = StratifiedKFold(y=Y, n_folds=10, shuffle=True, random_state=seed)\n",
    "cvscores = []\n",
    "\n",
    "for i, (train, test) in enumerate(kfold):\n",
    "  # create model\n",
    "    model = Sequential()\n",
    "    model.add(Dense(12, input_dim=8, init='uniform', activation='relu')) \n",
    "    model.add(Dense(8, init='uniform', activation='relu'))\n",
    "    model.add(Dense(1, init='uniform', activation='sigmoid'))\n",
    "    # Compile model\n",
    "    model.compile(loss='binary_crossentropy', optimizer='adam', metrics=['accuracy']) # Fit the model\n",
    "    model.fit(X[train], Y[train], nb_epoch=150, batch_size=10, verbose=0)\n",
    "    # evaluate the model\n",
    "    scores = model.evaluate(X[test], Y[test], verbose=0)\n",
    "    print(\"%s: %.2f%%\" % (model.metrics_names[1], scores[1]*100)) \n",
    "    cvscores.append(scores[1] * 100)\n",
    "print(\"%.2f%% (+/- %.2f%%)\" % (numpy.mean(cvscores), numpy.std(cvscores)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Notice that we had to re-create the model each loop to then fit and evaluate it with the data for the fold. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.3"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}