Digit Recognizer - CNN LDS.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "_uuid": "a98050d29f40adccc3360218a9b0679b493060fe"
   },
   "source": [
    "## Loading in and checking the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "_cell_guid": "b1076dfc-b9ad-4769-8c92-a6c4dae69d19",
    "_uuid": "8f2839f25d086af736a60e9eeb907d3b93b6e0e5"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using plaidml.keras.backend backend.\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib.image as mpimg\n",
    "%matplotlib inline\n",
    "import os\n",
    "\n",
    "np.random.seed(2)\n",
    "\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import confusion_matrix\n",
    "import itertools\n",
    "\n",
    "os.environ[\"KERAS_BACKEND\"] = \"plaidml.keras.backend\"\n",
    "\n",
    "from keras.utils.np_utils import to_categorical # convert to one-hot-encoding\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout, Flatten, Conv2D, MaxPool2D\n",
    "from keras.optimizers import RMSprop\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.callbacks import ReduceLROnPlateau, ModelCheckpoint, EarlyStopping\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data = np.load('./datasets/aug_data_obj.npy')\n",
    "train_label = np.load('./datasets/aug_labels_obj.npy')\n",
    "test_data = pd.read_csv('./datasets/test.csv')\n",
    "train_data = pd.DataFrame(train_data)\n",
    "train_label = pd.DataFrame(train_label)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_uuid": "31836e63382dad6d5a5f24cec4df24f105bde875"
   },
   "source": [
    "## Splitting the data and its label"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "_cell_guid": "79c7e3d0-c299-4dcb-8224-4455121ee9b0",
    "_uuid": "d629ff2d2480ee46fbb7e2d37f6b5fab8052498a"
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[<matplotlib.axes._subplots.AxesSubplot object at 0x104ade8d0>]],\n",
       "      dtype=object)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "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": [
    "train_label.hist()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "_uuid": "0fc59a6cabf8f5953a4e06314bbd8e30c993590d"
   },
   "outputs": [],
   "source": [
    "train_data[train_data<0] = 0\n",
    "train_data = train_data / 255.0\n",
    "test_data = test_data / 255.0"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_data = train_data.values.reshape(-1,28,28,1)\n",
    "test_data = test_data.values.reshape(-1,28,28,1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "train_label = to_categorical(train_label, num_classes=10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# s_train_data, s_val_data, s_train_label, s_val_label = train_test_split(train_data, train_label, test_size=0.25, stratify=train_label)\n",
    "# train_data=None"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_uuid": "de7c10e92e0ea4fd95b47fce47ac854fa8ba6821"
   },
   "source": [
    "## Checking the image quality of a random value"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "_uuid": "1765ab2eadd35dfdc33a9ceb38de2fd42cc093d8"
   },
   "outputs": [],
   "source": [
    "# from matplotlib import cm\n",
    "\n",
    "# plt.figure(figsize=(15,6))\n",
    "# for i in range(40):  \n",
    "#     plt.subplot(4, 10, i+1)\n",
    "#     plt.imshow(s_train_data[i][:,:,0], cmap=cm.binary)\n",
    "#     plt.title(\"predict=%d\" % np.argmax(s_train_label[i]),y=0.9)\n",
    "#     plt.axis('off')\n",
    "# plt.subplots_adjust(wspace=0.3, hspace=-0.1)\n",
    "# plt.show()\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "_uuid": "44d0b9bd6afb17922e40be8405d150f38d8134a9"
   },
   "source": [
    "# Attempting to solve using a CNN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "_uuid": "02ccf14f347aa6d81e6239ba9864e9534f6886e8"
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:plaidml:Opening device \"metal_amd_radeon_vega_frontier_edition.0\"\n"
     ]
    }
   ],
   "source": [
    "model = Sequential()\n",
    "\n",
    "model.add(Conv2D(filters = 32, kernel_size = (5,5),padding = 'Same', \n",
    "                 activation ='elu', input_shape = (28,28,1)))\n",
    "model.add(Conv2D(filters = 32, kernel_size = (5,5),padding = 'Same', \n",
    "                 activation ='elu'))\n",
    "model.add(MaxPool2D(pool_size=(2,2)))\n",
    "model.add(Dropout(0.25))\n",
    "\n",
    "\n",
    "model.add(Conv2D(filters = 64, kernel_size = (3,3),padding = 'Same', \n",
    "                 activation ='elu'))\n",
    "model.add(Conv2D(filters = 64, kernel_size = (3,3),padding = 'Same', \n",
    "                 activation ='elu'))\n",
    "model.add(MaxPool2D(pool_size=(2,2), strides=(2,2)))\n",
    "model.add(Dropout(0.25))\n",
    "\n",
    "\n",
    "model.add(Flatten())\n",
    "model.add(Dense(256, activation = \"elu\"))\n",
    "model.add(Dropout(0.5))\n",
    "model.add(Dense(10, activation = \"softmax\"))\n",
    "\n",
    "# Define the optimizer\n",
    "optimizer = RMSprop(lr=0.001, rho=0.9, epsilon=1e-08, decay=0.0)\n",
    "\n",
    "# Compile the model\n",
    "model.compile(optimizer = optimizer , loss = \"categorical_crossentropy\", metrics=[\"accuracy\"])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "_uuid": "387e93667da8c4e60c31fbbd9fa5a2c6b92d703b"
   },
   "outputs": [],
   "source": [
    "# Set a learning rate annealer\n",
    "learning_rate_reduction = ReduceLROnPlateau(monitor='val_acc', \n",
    "                                            patience=3, \n",
    "                                            verbose=1, \n",
    "                                            factor=0.5, \n",
    "                                            min_lr=0.00001)\n",
    "\n",
    "filepath=\"weights-full.best.hdf5\"\n",
    "checkpoint = ModelCheckpoint(filepath, monitor='acc', verbose=1, save_best_only=True, mode='max')\n",
    "earlyStop = EarlyStopping(patience=5, monitor='loss')\n",
    "callbacks_list = [checkpoint, earlyStop]\n",
    "\n",
    "epochs = 150\n",
    "batch_size = 7000 # Warning size choosen based on a GPU with 16GB of memory and utalizes ~95%."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Testing results with given data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/150\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "INFO:plaidml:Analyzing Ops: 230 of 329 operations complete\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      " - 534s - loss: 0.8676 - acc: 0.7197\n",
      "\n",
      "Epoch 00001: acc improved from -inf to 0.71972, saving model to weights-full.best.hdf5\n",
      "Epoch 2/150\n",
      " - 529s - loss: 0.2097 - acc: 0.9356\n",
      "\n",
      "Epoch 00002: acc improved from 0.71972 to 0.93562, saving model to weights-full.best.hdf5\n",
      "Epoch 3/150\n",
      " - 529s - loss: 0.1098 - acc: 0.9668\n",
      "\n",
      "Epoch 00003: acc improved from 0.93562 to 0.96676, saving model to weights-full.best.hdf5\n",
      "Epoch 4/150\n",
      " - 530s - loss: 0.0763 - acc: 0.9767\n",
      "\n",
      "Epoch 00004: acc improved from 0.96676 to 0.97674, saving model to weights-full.best.hdf5\n",
      "Epoch 5/150\n",
      " - 525s - loss: 0.0591 - acc: 0.9820\n",
      "\n",
      "Epoch 00005: acc improved from 0.97674 to 0.98196, saving model to weights-full.best.hdf5\n",
      "Epoch 6/150\n",
      " - 522s - loss: 0.0484 - acc: 0.9853\n",
      "\n",
      "Epoch 00006: acc improved from 0.98196 to 0.98527, saving model to weights-full.best.hdf5\n",
      "Epoch 7/150\n",
      " - 522s - loss: 0.0418 - acc: 0.9871\n",
      "\n",
      "Epoch 00007: acc improved from 0.98527 to 0.98708, saving model to weights-full.best.hdf5\n",
      "Epoch 8/150\n",
      " - 520s - loss: 0.0362 - acc: 0.9888\n",
      "\n",
      "Epoch 00008: acc improved from 0.98708 to 0.98882, saving model to weights-full.best.hdf5\n",
      "Epoch 9/150\n",
      " - 525s - loss: 0.0324 - acc: 0.9900\n",
      "\n",
      "Epoch 00009: acc improved from 0.98882 to 0.98999, saving model to weights-full.best.hdf5\n",
      "Epoch 10/150\n",
      " - 520s - loss: 0.0289 - acc: 0.9908\n",
      "\n",
      "Epoch 00010: acc improved from 0.98999 to 0.99082, saving model to weights-full.best.hdf5\n",
      "Epoch 11/150\n",
      " - 520s - loss: 0.0268 - acc: 0.9915\n",
      "\n",
      "Epoch 00011: acc improved from 0.99082 to 0.99151, saving model to weights-full.best.hdf5\n",
      "Epoch 12/150\n",
      " - 520s - loss: 0.0247 - acc: 0.9922\n",
      "\n",
      "Epoch 00012: acc improved from 0.99151 to 0.99216, saving model to weights-full.best.hdf5\n",
      "Epoch 13/150\n",
      " - 523s - loss: 0.0230 - acc: 0.9926\n",
      "\n",
      "Epoch 00013: acc improved from 0.99216 to 0.99265, saving model to weights-full.best.hdf5\n",
      "Epoch 14/150\n",
      " - 525s - loss: 0.0218 - acc: 0.9931\n",
      "\n",
      "Epoch 00014: acc improved from 0.99265 to 0.99313, saving model to weights-full.best.hdf5\n",
      "Epoch 15/150\n",
      " - 521s - loss: 0.0203 - acc: 0.9936\n",
      "\n",
      "Epoch 00015: acc improved from 0.99313 to 0.99358, saving model to weights-full.best.hdf5\n",
      "Epoch 16/150\n",
      " - 522s - loss: 0.0191 - acc: 0.9939\n",
      "\n",
      "Epoch 00016: acc improved from 0.99358 to 0.99395, saving model to weights-full.best.hdf5\n",
      "Epoch 17/150\n",
      " - 528s - loss: 0.0184 - acc: 0.9941\n",
      "\n",
      "Epoch 00017: acc improved from 0.99395 to 0.99410, saving model to weights-full.best.hdf5\n",
      "Epoch 18/150\n",
      " - 524s - loss: 0.0175 - acc: 0.9945\n",
      "\n",
      "Epoch 00018: acc improved from 0.99410 to 0.99445, saving model to weights-full.best.hdf5\n",
      "Epoch 19/150\n",
      " - 521s - loss: 0.0171 - acc: 0.9945\n",
      "\n",
      "Epoch 00019: acc improved from 0.99445 to 0.99455, saving model to weights-full.best.hdf5\n",
      "Epoch 20/150\n",
      " - 525s - loss: 0.0166 - acc: 0.9948\n",
      "\n",
      "Epoch 00020: acc improved from 0.99455 to 0.99479, saving model to weights-full.best.hdf5\n",
      "Epoch 21/150\n",
      " - 524s - loss: 0.0159 - acc: 0.9949\n",
      "\n",
      "Epoch 00021: acc improved from 0.99479 to 0.99486, saving model to weights-full.best.hdf5\n",
      "Epoch 22/150\n",
      " - 528s - loss: 0.0156 - acc: 0.9951\n",
      "\n",
      "Epoch 00022: acc improved from 0.99486 to 0.99507, saving model to weights-full.best.hdf5\n",
      "Epoch 23/150\n",
      " - 519s - loss: 0.0150 - acc: 0.9952\n",
      "\n",
      "Epoch 00023: acc improved from 0.99507 to 0.99520, saving model to weights-full.best.hdf5\n",
      "Epoch 24/150\n",
      " - 526s - loss: 0.0150 - acc: 0.9953\n",
      "\n",
      "Epoch 00024: acc improved from 0.99520 to 0.99526, saving model to weights-full.best.hdf5\n",
      "Epoch 25/150\n",
      " - 530s - loss: 0.0143 - acc: 0.9955\n",
      "\n",
      "Epoch 00025: acc improved from 0.99526 to 0.99549, saving model to weights-full.best.hdf5\n",
      "Epoch 26/150\n",
      " - 531s - loss: 0.0141 - acc: 0.9956\n",
      "\n",
      "Epoch 00026: acc improved from 0.99549 to 0.99564, saving model to weights-full.best.hdf5\n",
      "Epoch 27/150\n",
      " - 530s - loss: 0.0135 - acc: 0.9957\n",
      "\n",
      "Epoch 00027: acc improved from 0.99564 to 0.99571, saving model to weights-full.best.hdf5\n",
      "Epoch 28/150\n",
      " - 526s - loss: 0.0133 - acc: 0.9958\n",
      "\n",
      "Epoch 00028: acc improved from 0.99571 to 0.99584, saving model to weights-full.best.hdf5\n",
      "Epoch 29/150\n",
      " - 530s - loss: 0.0133 - acc: 0.9958\n",
      "\n",
      "Epoch 00029: acc improved from 0.99584 to 0.99585, saving model to weights-full.best.hdf5\n",
      "Epoch 30/150\n",
      " - 530s - loss: 0.0132 - acc: 0.9959\n",
      "\n",
      "Epoch 00030: acc improved from 0.99585 to 0.99586, saving model to weights-full.best.hdf5\n",
      "Epoch 31/150\n",
      " - 530s - loss: 0.0123 - acc: 0.9960\n",
      "\n",
      "Epoch 00031: acc improved from 0.99586 to 0.99605, saving model to weights-full.best.hdf5\n",
      "Epoch 32/150\n",
      " - 527s - loss: nan - acc: 0.7839\n",
      "\n",
      "Epoch 00032: acc did not improve from 0.99605\n",
      "Epoch 33/150\n",
      " - 526s - loss: nan - acc: 0.0000e+00\n",
      "\n",
      "Epoch 00033: acc did not improve from 0.99605\n",
      "Epoch 34/150\n",
      " - 525s - loss: nan - acc: 0.0000e+00\n",
      "\n",
      "Epoch 00034: acc did not improve from 0.99605\n",
      "Epoch 35/150\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-11-ae0fe2cc90e2>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m\u001b[0m\n\u001b[1;32m      3\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      4\u001b[0m history = model.fit(train_data, train_label, batch_size = batch_size, epochs = epochs, \n\u001b[0;32m----> 5\u001b[0;31m                     verbose = 2, callbacks=callbacks_list)\n\u001b[0m",
      "\u001b[0;32m/anaconda3/envs/ml_env/lib/python3.6/site-packages/keras/engine/training.py\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, x, y, batch_size, epochs, verbose, callbacks, validation_split, validation_data, shuffle, class_weight, sample_weight, initial_epoch, steps_per_epoch, validation_steps, **kwargs)\u001b[0m\n\u001b[1;32m   1035\u001b[0m                                         \u001b[0minitial_epoch\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0minitial_epoch\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1036\u001b[0m                                         \u001b[0msteps_per_epoch\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0msteps_per_epoch\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1037\u001b[0;31m                                         validation_steps=validation_steps)\n\u001b[0m\u001b[1;32m   1038\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1039\u001b[0m     def evaluate(self, x=None, y=None,\n",
      "\u001b[0;32m/anaconda3/envs/ml_env/lib/python3.6/site-packages/keras/engine/training_arrays.py\u001b[0m in \u001b[0;36mfit_loop\u001b[0;34m(model, f, ins, out_labels, batch_size, epochs, verbose, callbacks, val_f, val_ins, shuffle, callback_metrics, initial_epoch, steps_per_epoch, validation_steps)\u001b[0m\n\u001b[1;32m    197\u001b[0m                     \u001b[0mins_batch\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mi\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mins_batch\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mi\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mtoarray\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    198\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 199\u001b[0;31m                 \u001b[0mouts\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mf\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mins_batch\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    200\u001b[0m                 \u001b[0mouts\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mto_list\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mouts\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    201\u001b[0m                 \u001b[0;32mfor\u001b[0m \u001b[0ml\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mo\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mzip\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mout_labels\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mouts\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/anaconda3/envs/ml_env/lib/python3.6/site-packages/plaidml/keras/backend.py\u001b[0m in \u001b[0;36m__call__\u001b[0;34m(self, inputs)\u001b[0m\n\u001b[1;32m    163\u001b[0m             \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_invoker\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mset_output\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mname\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mt\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    164\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 165\u001b[0;31m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_invoker\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0minvoke\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    166\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    167\u001b[0m         \u001b[0;32mreturn\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mas_ndarray\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0m_ctx\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mt\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mtensors\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/anaconda3/envs/ml_env/lib/python3.6/site-packages/plaidml/__init__.py\u001b[0m in \u001b[0;36minvoke\u001b[0;34m(self)\u001b[0m\n\u001b[1;32m   1430\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1431\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0minvoke\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1432\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mInvocation\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_ctx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1433\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1434\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/anaconda3/envs/ml_env/lib/python3.6/site-packages/plaidml/__init__.py\u001b[0m in \u001b[0;36m__init__\u001b[0;34m(self, ctx, invoker)\u001b[0m\n\u001b[1;32m   1436\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1437\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m__init__\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mctx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minvoker\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m-> 1438\u001b[0;31m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_as_parameter_\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_lib\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mplaidml_schedule_invocation\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mctx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0minvoker\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m   1439\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_free\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_lib\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mplaidml_free_invocation\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m   1440\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/anaconda3/envs/ml_env/lib/python3.6/site-packages/plaidml/__init__.py\u001b[0m in \u001b[0;36m_check_err\u001b[0;34m(self, result, func, args)\u001b[0m\n\u001b[1;32m    707\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mplaidml_compute_grad_wrt\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0merrcheck\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_check_err\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    708\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 709\u001b[0;31m     \u001b[0;32mdef\u001b[0m \u001b[0m_check_err\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mresult\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    710\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0mresult\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    711\u001b[0m             \u001b[0;32mreturn\u001b[0m \u001b[0mresult\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "# history = model.fit(s_train_data, s_train_label, batch_size = batch_size, epochs = epochs, \n",
    "# validation_data = (s_val_data, s_val_label), verbose = 2, callbacks=callbacks_list)\n",
    "\n",
    "history = model.fit(train_data, train_label, batch_size = batch_size, epochs = epochs, \n",
    "                    verbose = 2, callbacks=callbacks_list)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Building our results"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "model.load_weights(filepath)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "results = model.predict(test_data)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "results = np.argmax(results,axis = 1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x432 with 40 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Preview predictions\n",
    "plt.figure(figsize=(15,6))\n",
    "for i in range(40):  \n",
    "    plt.subplot(4, 10, i+1)\n",
    "    plt.imshow(test_data[i][:,:,0], cmap=\"gray\")\n",
    "    plt.title(\"predict=%d\" % results[i],y=0.9)\n",
    "    plt.axis('off')\n",
    "plt.subplots_adjust(wspace=0.3, hspace=-0.1)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Exporting our results for Kaggle\n",
    "df = pd.DataFrame(results)\n",
    "df.index.name='ImageId'\n",
    "df.index+=1\n",
    "df.columns=['Label']\n",
    "df.to_csv('./datasets/results-cnn-lds.csv', header=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "scored 99.585% on Kaggle"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.7"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}