mnist.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# MNIST best results\n",
    "Result\tMethod\tVenue\tDetails\n",
    "1. 0.21%\tRegularization of Neural Networks using DropConnect\tICML 2013\t\n",
    "2. 0.23%\tMulti-column Deep Neural Networks for Image Classiﬁcation\tCVPR 2012\t\n",
    "3. 0.23%\tAPAC: Augmented PAttern Classification with Neural Networks\tarXiv 2015\t\n",
    "4. 0.24%\tBatch-normalized Maxout Network in Network\tarXiv 2015\tDetails\n",
    "5. 0.29%\tGeneralizing Pooling Functions in Convolutional Neural Networks: Mixed, Gated, and Tree\tAISTATS 2016\tDetails\n",
    "6. 0.31%\tRecurrent Convolutional Neural Network for Object Recognition\tCVPR 2015\t\n",
    "\n",
    "http://rodrigob.github.io/are_we_there_yet/build/classification_datasets_results.html"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Results of this model on MNIST test data: \n",
    "1. 0.22% test error rate with ensemble of 10 CNNs\n",
    "2. 0.21% test error rate with ensemble of 6 selected CNNs\n",
    "3. 0.26% test error rate for best single model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "# uncomment this when using the tensorflow backend\n",
    "#import tensorflow as tf\n",
    "#sess = tf.InteractiveSession()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import keras\n",
    "import numpy as np\n",
    "from keras.datasets import mnist\n",
    "from keras.layers import Dense, Dropout, Conv2D, MaxPooling2D\n",
    "from keras.layers import Flatten, Lambda, BatchNormalization\n",
    "from keras.models import Sequential\n",
    "from keras.optimizers import Adam as Adam\n",
    "\n",
    "# try leaky relu\n",
    "from keras.layers.advanced_activations import LeakyReLU"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Load MNIST dataset"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "(x_train, y_train), (x_test, y_test) = mnist.load_data()\n",
    "\n",
    "# store labels on test set for visualization\n",
    "test_labels = y_test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "rows = 28\n",
    "cols = 28\n",
    "\n",
    "# theano input shape\n",
    "input_shape = (1, rows, cols)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(60000, 28, 28)"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(60000, 1, 28, 28)"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# vgg expects channels - here we have just one\n",
    "x_train = x_train.reshape(x_train.shape[0], 1, 28, 28)\n",
    "x_test = x_test.reshape(x_test.shape[0], 1, 28, 28)\n",
    "x_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# convert to float\n",
    "x_train = x_train.astype(np.float32)\n",
    "x_test = x_test.astype(np.float32)\n",
    "\n",
    "# normalize\n",
    "x_train /= 255\n",
    "x_test /= 255"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(60000,)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(60000, 10)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# labels should be onehot encoded\n",
    "y_train = keras.utils.to_categorical(y_train, 10)\n",
    "y_test = keras.utils.to_categorical(y_test, 10)\n",
    "\n",
    "y_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.,  0.,  0.,  0.,  0.,  1.,  0.,  0.,  0.,  0.],\n",
       "       [ 1.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],\n",
       "       [ 0.,  0.,  0.,  0.,  1.,  0.,  0.,  0.,  0.,  0.],\n",
       "       [ 0.,  1.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.],\n",
       "       [ 0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  0.,  1.]])"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train[:5]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Define model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "# calculate mean and standard deviation\n",
    "mean_px = x_train.mean().astype(np.float32)\n",
    "std_px = x_train.std().astype(np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "# function to normalize input data\n",
    "def norm_input(x): return (x-mean_px)/std_px"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Batchnorm + dropout + data augmentation\n",
    "def create_model():\n",
    "    model = Sequential([\n",
    "        Lambda(norm_input, input_shape=(1,28,28), output_shape=(1,28,28)),\n",
    "        Conv2D(32, (3,3)),\n",
    "        LeakyReLU(),\n",
    "        BatchNormalization(axis=1),\n",
    "        Conv2D(32, (3,3)),\n",
    "        LeakyReLU(),\n",
    "        MaxPooling2D(),\n",
    "        BatchNormalization(axis=1),\n",
    "        Conv2D(64, (3,3)),\n",
    "        LeakyReLU(),\n",
    "        BatchNormalization(axis=1),\n",
    "        Conv2D(64, (3,3)),\n",
    "        LeakyReLU(),\n",
    "        MaxPooling2D(),\n",
    "        Flatten(),\n",
    "        BatchNormalization(),\n",
    "        Dense(512),\n",
    "        LeakyReLU(),\n",
    "        BatchNormalization(),\n",
    "        Dropout(0.3),\n",
    "        Dense(10, activation='softmax')\n",
    "    ])\n",
    "    model.compile(Adam(), loss='categorical_crossentropy', metrics=['accuracy'])\n",
    "    return model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Data augmentation\n",
    "Use keras's data augmentation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "batch_size = 512"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "from keras.preprocessing.image import ImageDataGenerator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "gen = ImageDataGenerator(rotation_range=12, width_shift_range=0.1, shear_range=0.3,\n",
    "                        height_shift_range=0.1, zoom_range=0.1, data_format='channels_first')\n",
    "batches = gen.flow(x_train, y_train, batch_size=batch_size)\n",
    "test_batches = gen.flow(x_test, y_test, batch_size=batch_size)\n",
    "steps_per_epoch = int(np.ceil(batches.n/batch_size))\n",
    "validation_steps = int(np.ceil(test_batches.n/batch_size))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.image.AxesImage at 0x7f815356acf8>"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f815aa7dbe0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# load ONE image from training set to display on screen\n",
    "img = x_train[1]\n",
    "\n",
    "# visualize original image\n",
    "plt.imshow(img[0], cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# trick our generator into believing img has enough dimensions\n",
    "# and get some augmented images for our single test image\n",
    "img = np.expand_dims(img, axis=0)\n",
    "aug_iter = gen.flow(img)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1, 28, 28)"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "aug_img = next(aug_iter)[0].astype(np.float32)\n",
    "aug_img.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f815357a0f0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# show augmented images\n",
    "f = plt.figure(figsize=(12,6))\n",
    "for i in range(8):\n",
    "    sp = f.add_subplot(2, 26//3, i+1)\n",
    "    sp.axis('Off')\n",
    "    aug_img = next(aug_iter)[0].astype(np.float32)\n",
    "    plt.imshow(aug_img[0], cmap='gray')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Create models with saved weights"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "models = []\n",
    "\n",
    "for i in range(10):\n",
    "    m = create_model()\n",
    "    m.load_weights(\"weights/weights_leaky_relu_model_0.22_\"+str(i)+'.pkl')\n",
    "    models.append(m)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Ensembling\n",
    "Here we create ten models and fit them to our augmented data. We use the mean prediction from all ten classifiers as our final prediction. This should average out errors in our classifiers and enable us to predict with higher accuracy."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "# fit the model on batches with real-time data augmentation\n",
    "def fit_model(m):\n",
    "    m.fit_generator(batches, steps_per_epoch=steps_per_epoch, epochs=1, verbose=0,\n",
    "                   validation_data=test_batches, validation_steps=validation_steps)\n",
    "    m.optimizer.lr = 0.1\n",
    "    m.fit_generator(batches, steps_per_epoch=steps_per_epoch, epochs=4, verbose=0,\n",
    "                   validation_data=test_batches, validation_steps=validation_steps)\n",
    "    m.optimizer.lr = 0.01\n",
    "    m.fit_generator(batches, steps_per_epoch=steps_per_epoch, epochs=12, verbose=0,\n",
    "                   validation_data=test_batches, validation_steps=validation_steps)\n",
    "    m.optimizer.lr = 0.001\n",
    "    m.fit_generator(batches, steps_per_epoch=steps_per_epoch, epochs=18, verbose=0,\n",
    "                   validation_data=test_batches, validation_steps=validation_steps)\n",
    "    return m"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Create and train ten models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "models = []\n",
    "\n",
    "for i in range(10):\n",
    "    print(\"Training model {}\".format(i))\n",
    "    m = fit_model(create_model())\n",
    "    models.append(m)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# train models some more if accuracy is not satisfying enough\n",
    "for i, m in enumerate(models):\n",
    "    m.optimizer.lr = 0.0001\n",
    "    print(\"Model {}\".format(i))\n",
    "    m.fit_generator(batches, steps_per_epoch=steps_per_epoch, epochs=2,\n",
    "                   validation_data=test_batches, validation_steps=validation_steps)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Evaluate Model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "eval_batch_size = 512"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10000/10000 [==============================] - 3s 306us/step\n",
      "10000/10000 [==============================] - 2s 235us/step\n",
      "10000/10000 [==============================] - 2s 244us/step\n",
      "10000/10000 [==============================] - 2s 237us/step\n",
      "10000/10000 [==============================] - 2s 238us/step\n",
      "10000/10000 [==============================] - 2s 235us/step\n",
      "10000/10000 [==============================] - 2s 237us/step\n",
      "10000/10000 [==============================] - 2s 234us/step\n",
      "10000/10000 [==============================] - 2s 236us/step\n",
      "10000/10000 [==============================] - 2s 234us/step\n"
     ]
    }
   ],
   "source": [
    "# evaluate every model\n",
    "evals = np.array([m.evaluate(x_test,y_test, batch_size=eval_batch_size) for m in models])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 0.01117255,  0.99643   ])"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# calculate mean from evaluations\n",
    "evals.mean(axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_preds = np.stack([m.predict(x_test, batch_size=eval_batch_size) for m in models])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10, 10000, 10)"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_preds.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10000, 10)"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "avg_preds = all_preds.mean(axis=0)\n",
    "avg_preds.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.2200007438659668"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# display error rate in percent\n",
    "(1 - keras.metrics.categorical_accuracy(y_test, avg_preds).eval().mean()) * 100"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Show test error rate of every model separately"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Results for .22 average error rate model weights (Model # / error rate)\n",
    "1. 0.370001792908\n",
    "2. 0.330001115799\n",
    "3. 0.429999828339\n",
    "4. 0.349998474121\n",
    "5. 0.370001792908\n",
    "6. 0.340002775192\n",
    "7. 0.370001792908\n",
    "8. 0.360000133514\n",
    "9. 0.29000043869\n",
    "10. 0.360000133514"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "1 0.370001792908\n",
      "2 0.330001115799\n",
      "3 0.429999828339\n",
      "4 0.349998474121\n",
      "5 0.370001792908\n",
      "6 0.340002775192\n",
      "7 0.370001792908\n",
      "8 0.360000133514\n",
      "9 0.29000043869\n",
      "10 0.360000133514\n"
     ]
    }
   ],
   "source": [
    "for i, m in enumerate(models):\n",
    "    pred = np.array(m.predict(x_test, batch_size=eval_batch_size))\n",
    "    print(i+1, (1 - keras.metrics.categorical_accuracy(y_test, pred).eval().mean()) * 100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "9 0.360000133514\n"
     ]
    }
   ],
   "source": [
    "# calculate accuracy of a single model\n",
    "pred = np.array(models[i].predict(x_test, batch_size=eval_batch_size))\n",
    "print(i, (1 - keras.metrics.categorical_accuracy(y_test, pred).eval().mean()) * 100)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Plot Training History"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Here we create and train a single model from scratch and create a plot of the training history."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# create and train a single model\n",
    "m = create_model()\n",
    "history = m.fit_generator(batches, steps_per_epoch=steps_per_epoch, epochs=10, verbose=0,\n",
    "                         validation_data=test_batches, validation_steps=validation_steps)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Jason Brownlee's script to plot model training history:\n",
    "https://machinelearningmastery.com/display-deep-learning-model-training-history-in-keras/"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "dict_keys(['val_loss', 'val_acc', 'loss', 'acc'])\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYUAAAEWCAYAAACJ0YulAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fcc4c4b05f8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXwAAAEWCAYAAABliCz2AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nUiT+P3bghKP0E4TjAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fcc4c4ddf28>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "\n",
    "axes = plt.gca()\n",
    "#axes.set_xlim([1,5])\n",
    "#axes.set_ylim([0.2,1.0])\n",
    "\n",
    "# list all data in history\n",
    "print(history.history.keys())\n",
    "# summarize history for accuracy\n",
    "plt.plot(history.history['acc'])\n",
    "plt.plot(history.history['val_acc'])\n",
    "plt.title('model accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')\n",
    "plt.show()\n",
    "\n",
    "# summarize history for loss\n",
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('model loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'test'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train single model some more"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Select a single model, train for a few epochs, and check accuracy."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# which model to train\n",
    "i = 8"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# reload weights in case of overfitting\n",
    "models[i].load_weights(\"weights/model_0.26.pkl\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/1\n",
      "118/118 [==============================] - 11s - loss: 0.0176 - acc: 0.9944 - val_loss: 0.0215 - val_acc: 0.9932\n"
     ]
    }
   ],
   "source": [
    "models[i].optimizer.lr = 0.0001\n",
    "history = models[i].fit_generator(batches, steps_per_epoch=steps_per_epoch, epochs=1,\n",
    "                   validation_data=test_batches, validation_steps=validation_steps)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 120,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "8 0.260001420975\n"
     ]
    }
   ],
   "source": [
    "# calculate accuracy of a single model\n",
    "pred = np.array(models[i].predict(x_test, batch_size=eval_batch_size))\n",
    "print(i, (1 - keras.metrics.categorical_accuracy(y_test, pred).eval().mean()) * 100)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "models[i].save_weights(\"weights/weights_leaky_relu_model_0.2x\"+str(i)+'.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Try evaluation only with best performing models\n",
    "Ensemble with models 0,4,5,7,8,9 gives a test error rate of 0.21%. State of the art."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "best_model_numbers = [0,4,5,7,8,9]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "best_models = []\n",
    "\n",
    "for i in best_model_numbers:\n",
    "    best_models.append(models[i])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "10000/10000 [==============================] - 3s 268us/step\n",
      "10000/10000 [==============================] - 2s 209us/step\n",
      "10000/10000 [==============================] - 2s 215us/step\n",
      "10000/10000 [==============================] - 2s 212us/step\n",
      "10000/10000 [==============================] - 2s 210us/step\n",
      "10000/10000 [==============================] - 2s 213us/step\n"
     ]
    }
   ],
   "source": [
    "# evaluate every model\n",
    "evals = np.array([m.evaluate(x_test,y_test, batch_size=eval_batch_size) for m in best_models])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 0.010873  ,  0.99651667])"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# calculate mean from evaluations\n",
    "evals.mean(axis=0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "all_preds = np.stack([m.predict(x_test, batch_size=eval_batch_size) for m in best_models])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(6, 10000, 10)"
      ]
     },
     "execution_count": 34,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_preds.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10000, 10)"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "avg_preds = all_preds.mean(axis=0)\n",
    "avg_preds.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.20999908447265625"
      ]
     },
     "execution_count": 36,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# display error rate in percent\n",
    "(1 - keras.metrics.categorical_accuracy(y_test, avg_preds).eval().mean()) * 100"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Visualize model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "lambda_1 (Lambda)            (None, 1, 28, 28)         0         \n",
      "_________________________________________________________________\n",
      "conv2d_1 (Conv2D)            (None, 32, 26, 26)        320       \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_1 (LeakyReLU)    (None, 32, 26, 26)        0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_1 (Batch (None, 32, 26, 26)        128       \n",
      "_________________________________________________________________\n",
      "conv2d_2 (Conv2D)            (None, 32, 24, 24)        9248      \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_2 (LeakyReLU)    (None, 32, 24, 24)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_1 (MaxPooling2 (None, 32, 12, 12)        0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_2 (Batch (None, 32, 12, 12)        128       \n",
      "_________________________________________________________________\n",
      "conv2d_3 (Conv2D)            (None, 64, 10, 10)        18496     \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_3 (LeakyReLU)    (None, 64, 10, 10)        0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_3 (Batch (None, 64, 10, 10)        256       \n",
      "_________________________________________________________________\n",
      "conv2d_4 (Conv2D)            (None, 64, 8, 8)          36928     \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_4 (LeakyReLU)    (None, 64, 8, 8)          0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_2 (MaxPooling2 (None, 64, 4, 4)          0         \n",
      "_________________________________________________________________\n",
      "flatten_1 (Flatten)          (None, 1024)              0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_4 (Batch (None, 1024)              4096      \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 512)               524800    \n",
      "_________________________________________________________________\n",
      "leaky_re_lu_5 (LeakyReLU)    (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "batch_normalization_5 (Batch (None, 512)               2048      \n",
      "_________________________________________________________________\n",
      "dropout_1 (Dropout)          (None, 512)               0         \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 10)                5130      \n",
      "=================================================================\n",
      "Total params: 601,578\n",
      "Trainable params: 598,250\n",
      "Non-trainable params: 3,328\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "models[0].summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/svg+xml": [
       "<svg height=\"1578pt\" viewBox=\"0.00 0.00 268.00 1578.00\" width=\"268pt\" xmlns=\"http://www.w3.org/2000/svg\" xmlns:xlink=\"http://www.w3.org/1999/xlink\">\n",
       "<g class=\"graph\" id=\"graph0\" transform=\"scale(1 1) rotate(0) translate(4 1574)\">\n",
       "<title>G</title>\n",
       "<polygon fill=\"white\" points=\"-4,4 -4,-1574 264,-1574 264,4 -4,4\" stroke=\"none\"/>\n",
       "<!-- 140041003808696 -->\n",
       "<g class=\"node\" id=\"node1\"><title>140041003808696</title>\n",
       "<polygon fill=\"none\" points=\"43.5,-1533.5 43.5,-1569.5 216.5,-1569.5 216.5,-1533.5 43.5,-1533.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-1547.8\">lambda_1_input: InputLayer</text>\n",
       "</g>\n",
       "<!-- 140041005806088 -->\n",
       "<g class=\"node\" id=\"node2\"><title>140041005806088</title>\n",
       "<polygon fill=\"none\" points=\"68.5,-1460.5 68.5,-1496.5 191.5,-1496.5 191.5,-1460.5 68.5,-1460.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-1474.8\">lambda_1: Lambda</text>\n",
       "</g>\n",
       "<!-- 140041003808696&#45;&gt;140041005806088 -->\n",
       "<g class=\"edge\" id=\"edge1\"><title>140041003808696-&gt;140041005806088</title>\n",
       "<path d=\"M130,-1533.31C130,-1525.29 130,-1515.55 130,-1506.57\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-1506.53 130,-1496.53 126.5,-1506.53 133.5,-1506.53\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041002078728 -->\n",
       "<g class=\"node\" id=\"node3\"><title>140041002078728</title>\n",
       "<polygon fill=\"none\" points=\"68.5,-1387.5 68.5,-1423.5 191.5,-1423.5 191.5,-1387.5 68.5,-1387.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-1401.8\">conv2d_1: Conv2D</text>\n",
       "</g>\n",
       "<!-- 140041005806088&#45;&gt;140041002078728 -->\n",
       "<g class=\"edge\" id=\"edge2\"><title>140041005806088-&gt;140041002078728</title>\n",
       "<path d=\"M130,-1460.31C130,-1452.29 130,-1442.55 130,-1433.57\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-1433.53 130,-1423.53 126.5,-1433.53 133.5,-1433.53\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041005805808 -->\n",
       "<g class=\"node\" id=\"node4\"><title>140041005805808</title>\n",
       "<polygon fill=\"none\" points=\"46,-1314.5 46,-1350.5 214,-1350.5 214,-1314.5 46,-1314.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-1328.8\">leaky_re_lu_1: LeakyReLU</text>\n",
       "</g>\n",
       "<!-- 140041002078728&#45;&gt;140041005805808 -->\n",
       "<g class=\"edge\" id=\"edge3\"><title>140041002078728-&gt;140041005805808</title>\n",
       "<path d=\"M130,-1387.31C130,-1379.29 130,-1369.55 130,-1360.57\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-1360.53 130,-1350.53 126.5,-1360.53 133.5,-1360.53\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041002525640 -->\n",
       "<g class=\"node\" id=\"node5\"><title>140041002525640</title>\n",
       "<polygon fill=\"none\" points=\"0,-1241.5 0,-1277.5 260,-1277.5 260,-1241.5 0,-1241.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-1255.8\">batch_normalization_1: BatchNormalization</text>\n",
       "</g>\n",
       "<!-- 140041005805808&#45;&gt;140041002525640 -->\n",
       "<g class=\"edge\" id=\"edge4\"><title>140041005805808-&gt;140041002525640</title>\n",
       "<path d=\"M130,-1314.31C130,-1306.29 130,-1296.55 130,-1287.57\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-1287.53 130,-1277.53 126.5,-1287.53 133.5,-1287.53\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041002525584 -->\n",
       "<g class=\"node\" id=\"node6\"><title>140041002525584</title>\n",
       "<polygon fill=\"none\" points=\"68.5,-1168.5 68.5,-1204.5 191.5,-1204.5 191.5,-1168.5 68.5,-1168.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-1182.8\">conv2d_2: Conv2D</text>\n",
       "</g>\n",
       "<!-- 140041002525640&#45;&gt;140041002525584 -->\n",
       "<g class=\"edge\" id=\"edge5\"><title>140041002525640-&gt;140041002525584</title>\n",
       "<path d=\"M130,-1241.31C130,-1233.29 130,-1223.55 130,-1214.57\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-1214.53 130,-1204.53 126.5,-1214.53 133.5,-1214.53\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041005468584 -->\n",
       "<g class=\"node\" id=\"node7\"><title>140041005468584</title>\n",
       "<polygon fill=\"none\" points=\"46,-1095.5 46,-1131.5 214,-1131.5 214,-1095.5 46,-1095.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-1109.8\">leaky_re_lu_2: LeakyReLU</text>\n",
       "</g>\n",
       "<!-- 140041002525584&#45;&gt;140041005468584 -->\n",
       "<g class=\"edge\" id=\"edge6\"><title>140041002525584-&gt;140041005468584</title>\n",
       "<path d=\"M130,-1168.31C130,-1160.29 130,-1150.55 130,-1141.57\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-1141.53 130,-1131.53 126.5,-1141.53 133.5,-1141.53\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041005469032 -->\n",
       "<g class=\"node\" id=\"node8\"><title>140041005469032</title>\n",
       "<polygon fill=\"none\" points=\"26.5,-1022.5 26.5,-1058.5 233.5,-1058.5 233.5,-1022.5 26.5,-1022.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-1036.8\">max_pooling2d_1: MaxPooling2D</text>\n",
       "</g>\n",
       "<!-- 140041005468584&#45;&gt;140041005469032 -->\n",
       "<g class=\"edge\" id=\"edge7\"><title>140041005468584-&gt;140041005469032</title>\n",
       "<path d=\"M130,-1095.31C130,-1087.29 130,-1077.55 130,-1068.57\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-1068.53 130,-1058.53 126.5,-1068.53 133.5,-1068.53\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041005466064 -->\n",
       "<g class=\"node\" id=\"node9\"><title>140041005466064</title>\n",
       "<polygon fill=\"none\" points=\"0,-949.5 0,-985.5 260,-985.5 260,-949.5 0,-949.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-963.8\">batch_normalization_2: BatchNormalization</text>\n",
       "</g>\n",
       "<!-- 140041005469032&#45;&gt;140041005466064 -->\n",
       "<g class=\"edge\" id=\"edge8\"><title>140041005469032-&gt;140041005466064</title>\n",
       "<path d=\"M130,-1022.31C130,-1014.29 130,-1004.55 130,-995.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-995.529 130,-985.529 126.5,-995.529 133.5,-995.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041005469368 -->\n",
       "<g class=\"node\" id=\"node10\"><title>140041005469368</title>\n",
       "<polygon fill=\"none\" points=\"68.5,-876.5 68.5,-912.5 191.5,-912.5 191.5,-876.5 68.5,-876.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-890.8\">conv2d_3: Conv2D</text>\n",
       "</g>\n",
       "<!-- 140041005466064&#45;&gt;140041005469368 -->\n",
       "<g class=\"edge\" id=\"edge9\"><title>140041005466064-&gt;140041005469368</title>\n",
       "<path d=\"M130,-949.313C130,-941.289 130,-931.547 130,-922.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-922.529 130,-912.529 126.5,-922.529 133.5,-922.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041005466232 -->\n",
       "<g class=\"node\" id=\"node11\"><title>140041005466232</title>\n",
       "<polygon fill=\"none\" points=\"46,-803.5 46,-839.5 214,-839.5 214,-803.5 46,-803.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-817.8\">leaky_re_lu_3: LeakyReLU</text>\n",
       "</g>\n",
       "<!-- 140041005469368&#45;&gt;140041005466232 -->\n",
       "<g class=\"edge\" id=\"edge10\"><title>140041005469368-&gt;140041005466232</title>\n",
       "<path d=\"M130,-876.313C130,-868.289 130,-858.547 130,-849.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-849.529 130,-839.529 126.5,-849.529 133.5,-849.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041005266368 -->\n",
       "<g class=\"node\" id=\"node12\"><title>140041005266368</title>\n",
       "<polygon fill=\"none\" points=\"0,-730.5 0,-766.5 260,-766.5 260,-730.5 0,-730.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-744.8\">batch_normalization_3: BatchNormalization</text>\n",
       "</g>\n",
       "<!-- 140041005466232&#45;&gt;140041005266368 -->\n",
       "<g class=\"edge\" id=\"edge11\"><title>140041005466232-&gt;140041005266368</title>\n",
       "<path d=\"M130,-803.313C130,-795.289 130,-785.547 130,-776.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-776.529 130,-766.529 126.5,-776.529 133.5,-776.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041005268776 -->\n",
       "<g class=\"node\" id=\"node13\"><title>140041005268776</title>\n",
       "<polygon fill=\"none\" points=\"68.5,-657.5 68.5,-693.5 191.5,-693.5 191.5,-657.5 68.5,-657.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-671.8\">conv2d_4: Conv2D</text>\n",
       "</g>\n",
       "<!-- 140041005266368&#45;&gt;140041005268776 -->\n",
       "<g class=\"edge\" id=\"edge12\"><title>140041005266368-&gt;140041005268776</title>\n",
       "<path d=\"M130,-730.313C130,-722.289 130,-712.547 130,-703.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-703.529 130,-693.529 126.5,-703.529 133.5,-703.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041003973824 -->\n",
       "<g class=\"node\" id=\"node14\"><title>140041003973824</title>\n",
       "<polygon fill=\"none\" points=\"46,-584.5 46,-620.5 214,-620.5 214,-584.5 46,-584.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-598.8\">leaky_re_lu_4: LeakyReLU</text>\n",
       "</g>\n",
       "<!-- 140041005268776&#45;&gt;140041003973824 -->\n",
       "<g class=\"edge\" id=\"edge13\"><title>140041005268776-&gt;140041003973824</title>\n",
       "<path d=\"M130,-657.313C130,-649.289 130,-639.547 130,-630.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-630.529 130,-620.529 126.5,-630.529 133.5,-630.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041002645488 -->\n",
       "<g class=\"node\" id=\"node15\"><title>140041002645488</title>\n",
       "<polygon fill=\"none\" points=\"26.5,-511.5 26.5,-547.5 233.5,-547.5 233.5,-511.5 26.5,-511.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-525.8\">max_pooling2d_2: MaxPooling2D</text>\n",
       "</g>\n",
       "<!-- 140041003973824&#45;&gt;140041002645488 -->\n",
       "<g class=\"edge\" id=\"edge14\"><title>140041003973824-&gt;140041002645488</title>\n",
       "<path d=\"M130,-584.313C130,-576.289 130,-566.547 130,-557.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-557.529 130,-547.529 126.5,-557.529 133.5,-557.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041002159688 -->\n",
       "<g class=\"node\" id=\"node16\"><title>140041002159688</title>\n",
       "<polygon fill=\"none\" points=\"75,-438.5 75,-474.5 185,-474.5 185,-438.5 75,-438.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-452.8\">flatten_1: Flatten</text>\n",
       "</g>\n",
       "<!-- 140041002645488&#45;&gt;140041002159688 -->\n",
       "<g class=\"edge\" id=\"edge15\"><title>140041002645488-&gt;140041002159688</title>\n",
       "<path d=\"M130,-511.313C130,-503.289 130,-493.547 130,-484.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-484.529 130,-474.529 126.5,-484.529 133.5,-484.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041002160080 -->\n",
       "<g class=\"node\" id=\"node17\"><title>140041002160080</title>\n",
       "<polygon fill=\"none\" points=\"0,-365.5 0,-401.5 260,-401.5 260,-365.5 0,-365.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-379.8\">batch_normalization_4: BatchNormalization</text>\n",
       "</g>\n",
       "<!-- 140041002159688&#45;&gt;140041002160080 -->\n",
       "<g class=\"edge\" id=\"edge16\"><title>140041002159688-&gt;140041002160080</title>\n",
       "<path d=\"M130,-438.313C130,-430.289 130,-420.547 130,-411.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-411.529 130,-401.529 126.5,-411.529 133.5,-411.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041004048736 -->\n",
       "<g class=\"node\" id=\"node18\"><title>140041004048736</title>\n",
       "<polygon fill=\"none\" points=\"79,-292.5 79,-328.5 181,-328.5 181,-292.5 79,-292.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-306.8\">dense_1: Dense</text>\n",
       "</g>\n",
       "<!-- 140041002160080&#45;&gt;140041004048736 -->\n",
       "<g class=\"edge\" id=\"edge17\"><title>140041002160080-&gt;140041004048736</title>\n",
       "<path d=\"M130,-365.313C130,-357.289 130,-347.547 130,-338.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-338.529 130,-328.529 126.5,-338.529 133.5,-338.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041343449016 -->\n",
       "<g class=\"node\" id=\"node19\"><title>140041343449016</title>\n",
       "<polygon fill=\"none\" points=\"46,-219.5 46,-255.5 214,-255.5 214,-219.5 46,-219.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-233.8\">leaky_re_lu_5: LeakyReLU</text>\n",
       "</g>\n",
       "<!-- 140041004048736&#45;&gt;140041343449016 -->\n",
       "<g class=\"edge\" id=\"edge18\"><title>140041004048736-&gt;140041343449016</title>\n",
       "<path d=\"M130,-292.313C130,-284.289 130,-274.547 130,-265.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-265.529 130,-255.529 126.5,-265.529 133.5,-265.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041003319368 -->\n",
       "<g class=\"node\" id=\"node20\"><title>140041003319368</title>\n",
       "<polygon fill=\"none\" points=\"0,-146.5 0,-182.5 260,-182.5 260,-146.5 0,-146.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-160.8\">batch_normalization_5: BatchNormalization</text>\n",
       "</g>\n",
       "<!-- 140041343449016&#45;&gt;140041003319368 -->\n",
       "<g class=\"edge\" id=\"edge19\"><title>140041343449016-&gt;140041003319368</title>\n",
       "<path d=\"M130,-219.313C130,-211.289 130,-201.547 130,-192.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-192.529 130,-182.529 126.5,-192.529 133.5,-192.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041003320376 -->\n",
       "<g class=\"node\" id=\"node21\"><title>140041003320376</title>\n",
       "<polygon fill=\"none\" points=\"67.5,-73.5 67.5,-109.5 192.5,-109.5 192.5,-73.5 67.5,-73.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-87.8\">dropout_1: Dropout</text>\n",
       "</g>\n",
       "<!-- 140041003319368&#45;&gt;140041003320376 -->\n",
       "<g class=\"edge\" id=\"edge20\"><title>140041003319368-&gt;140041003320376</title>\n",
       "<path d=\"M130,-146.313C130,-138.289 130,-128.547 130,-119.569\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-119.529 130,-109.529 126.5,-119.529 133.5,-119.529\" stroke=\"black\"/>\n",
       "</g>\n",
       "<!-- 140041003270096 -->\n",
       "<g class=\"node\" id=\"node22\"><title>140041003270096</title>\n",
       "<polygon fill=\"none\" points=\"79,-0.5 79,-36.5 181,-36.5 181,-0.5 79,-0.5\" stroke=\"black\"/>\n",
       "<text font-family=\"Times,serif\" font-size=\"14.00\" text-anchor=\"middle\" x=\"130\" y=\"-14.8\">dense_2: Dense</text>\n",
       "</g>\n",
       "<!-- 140041003320376&#45;&gt;140041003270096 -->\n",
       "<g class=\"edge\" id=\"edge21\"><title>140041003320376-&gt;140041003270096</title>\n",
       "<path d=\"M130,-73.3129C130,-65.2895 130,-55.5475 130,-46.5691\" fill=\"none\" stroke=\"black\"/>\n",
       "<polygon fill=\"black\" points=\"133.5,-46.5288 130,-36.5288 126.5,-46.5289 133.5,-46.5288\" stroke=\"black\"/>\n",
       "</g>\n",
       "</g>\n",
       "</svg>"
      ],
      "text/plain": [
       "<IPython.core.display.SVG object>"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# fix for python3\n",
    "import pydotplus as pydot\n",
    "\n",
    "from keras.utils import plot_model\n",
    "plot_model(models[0], to_file='model.png')\n",
    "\n",
    "# visualize our model\n",
    "from IPython.display import SVG\n",
    "from keras.utils.vis_utils import model_to_dot\n",
    "SVG(model_to_dot(models[0]).create(prog='dot', format='svg'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Visualize predictions for ensemble\n",
    "As well as looking at the overall metrics, it's also a good idea to look at examples of:\n",
    "1. A few correct labels at random\n",
    "2. A few incorrect labels at random"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "# decode a one hot encoded string\n",
    "def one_hot_decode(encoded_seq):\n",
    "    return [np.argmax(vector) for vector in encoded_seq]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10000, 10)"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "avg_preds.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(10000,)"
      ]
     },
     "execution_count": 39,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "preds = np.array(one_hot_decode(avg_preds))\n",
    "preds.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "correct = np.where(preds==test_labels)[0]\n",
    "incorrect = np.where(preds!=test_labels)[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(9979,)"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "correct.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(21,)"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "incorrect.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "# number of images to view\n",
    "n_view = incorrect.shape[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [],
   "source": [
    "from numpy.random import permutation\n",
    "\n",
    "# get a random selection of correctly predicted images\n",
    "idx = permutation(correct)[:30]\n",
    "idx_incorrect = incorrect[:n_view]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [],
   "source": [
    "preds_correct = preds[idx]\n",
    "preds_incorrect = preds[idx_incorrect]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "imgs = x_test[idx]\n",
    "imgs_incorrect = x_test[idx_incorrect]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(30, 1, 28, 28)"
      ]
     },
     "execution_count": 47,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "imgs.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### A few correct labels at random"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f81499588d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "f = plt.figure(figsize=(12,4))\n",
    "\n",
    "for i in range(30):\n",
    "    sp = f.add_subplot(3, 10, i+1)\n",
    "    sp.axis('Off')\n",
    "    sp.set_title(preds_correct[i])\n",
    "    img = imgs[i].astype('float32')\n",
    "    plt.imshow(img[0], cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([3, 0, 6, 5, 1, 4, 8, 6, 2, 9, 2, 5, 8, 4, 7, 4, 7, 1, 8, 2, 3, 8, 0,\n",
       "       7, 3, 9, 1, 6, 5, 3], dtype=uint8)"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# ground truth labels\n",
    "test_labels[idx]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### All incorrect labels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f81480989e8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "f = plt.figure(figsize=(12,4))\n",
    "\n",
    "plt.subplots_adjust(left=None, bottom=None, right=None, top=1.15,\n",
    "                wspace=None, hspace=None)\n",
    "\n",
    "for i in range(n_view): \n",
    "    sp = f.add_subplot(3, 10, i+1)\n",
    "    sp.axis('Off')\n",
    "    sp.text(0.8, 1.2, \"Pred.: {}\\nLabel: {}\"\n",
    "                 .format(preds_incorrect[i],test_labels[idx_incorrect][i]),\n",
    "            horizontalalignment='right', verticalalignment='center',\n",
    "            transform=sp.transAxes)\n",
    "    plt.text(0.5, -0.12, \"#{}\".format(idx_incorrect[i]), \n",
    "             horizontalalignment='center', verticalalignment='center',\n",
    "             transform=sp.transAxes)\n",
    "    img = imgs_incorrect[i].astype('float32')\n",
    "    plt.imshow(img[0], cmap='gray')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([4, 2, 8, 9, 7, 9, 5, 4, 2, 5, 6, 6, 5, 6, 2, 6, 7, 5, 7, 8, 5], dtype=uint8)"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# ground truth labels\n",
    "test_labels[idx_incorrect]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 447,  659,  947, 1232, 1260, 1901, 2040, 2130, 2462, 2597, 2654,\n",
       "       3422, 3558, 3762, 4176, 4699, 5654, 5937, 6576, 6625, 9729])"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "idx_incorrect"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Print confusion matrix"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Plot unnormalized and normalized confusion matrices. Code adapted from \n",
    "http://scikit-learn.org/stable/auto_examples/model_selection/plot_confusion_matrix.html\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def plot_confusion_matrix(cm, classes,\n",
    "                          normalize=False,\n",
    "                          title='Confusion matrix',\n",
    "                          cmap=plt.cm.Blues):\n",
    "    \"\"\"\n",
    "    This function prints and plots the confusion matrix.\n",
    "    Normalization can be applied by setting `normalize=True`.\n",
    "    \"\"\"\n",
    "    if normalize:\n",
    "        cm = cm.astype('float') / cm.sum(axis=1)[:, np.newaxis]\n",
    "        print(\"Normalized confusion matrix\")\n",
    "    else:\n",
    "        print('Confusion matrix, without normalization')\n",
    "\n",
    "    print(cm)\n",
    "\n",
    "    plt.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    plt.title(title)\n",
    "    plt.colorbar()\n",
    "    tick_marks = np.arange(len(classes))\n",
    "    plt.xticks(tick_marks, classes, rotation=45)\n",
    "    plt.yticks(tick_marks, classes)\n",
    "\n",
    "    fmt = '.2f' if normalize else 'd'\n",
    "    thresh = cm.max() / 2.\n",
    "    for i, j in itertools.product(range(cm.shape[0]), range(cm.shape[1])):\n",
    "        plt.text(j, i, format(cm[i, j], fmt),\n",
    "                 horizontalalignment=\"center\",\n",
    "                 color=\"white\" if cm[i, j] > thresh else \"black\")\n",
    "\n",
    "    plt.tight_layout()\n",
    "    plt.ylabel('True label')\n",
    "    plt.xlabel('Predicted label')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "# calculate confusion matrix\n",
    "from sklearn.metrics import confusion_matrix\n",
    "cnf_matrix = confusion_matrix(test_labels, preds)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Confusion matrix, without normalization\n",
      "[[ 980    0    0    0    0    0    0    0    0    0]\n",
      " [   0 1135    0    0    0    0    0    0    0    0]\n",
      " [   1    0 1029    0    0    0    0    2    0    0]\n",
      " [   0    0    0 1010    0    0    0    0    0    0]\n",
      " [   0    0    0    0  980    0    0    0    0    2]\n",
      " [   1    0    0    2    0  887    2    0    0    0]\n",
      " [   1    2    0    0    0    0  954    0    1    0]\n",
      " [   0    2    1    0    0    0    0 1025    0    0]\n",
      " [   0    0    1    0    0    0    0    0  972    1]\n",
      " [   0    0    0    0    2    0    0    0    0 1007]]\n",
      "Normalized confusion matrix\n",
      "[[  1.00e+00   0.00e+00   0.00e+00   0.00e+00   0.00e+00   0.00e+00\n",
      "    0.00e+00   0.00e+00   0.00e+00   0.00e+00]\n",
      " [  0.00e+00   1.00e+00   0.00e+00   0.00e+00   0.00e+00   0.00e+00\n",
      "    0.00e+00   0.00e+00   0.00e+00   0.00e+00]\n",
      " [  9.69e-04   0.00e+00   9.97e-01   0.00e+00   0.00e+00   0.00e+00\n",
      "    0.00e+00   1.94e-03   0.00e+00   0.00e+00]\n",
      " [  0.00e+00   0.00e+00   0.00e+00   1.00e+00   0.00e+00   0.00e+00\n",
      "    0.00e+00   0.00e+00   0.00e+00   0.00e+00]\n",
      " [  0.00e+00   0.00e+00   0.00e+00   0.00e+00   9.98e-01   0.00e+00\n",
      "    0.00e+00   0.00e+00   0.00e+00   2.04e-03]\n",
      " [  1.12e-03   0.00e+00   0.00e+00   2.24e-03   0.00e+00   9.94e-01\n",
      "    2.24e-03   0.00e+00   0.00e+00   0.00e+00]\n",
      " [  1.04e-03   2.09e-03   0.00e+00   0.00e+00   0.00e+00   0.00e+00\n",
      "    9.96e-01   0.00e+00   1.04e-03   0.00e+00]\n",
      " [  0.00e+00   1.95e-03   9.73e-04   0.00e+00   0.00e+00   0.00e+00\n",
      "    0.00e+00   9.97e-01   0.00e+00   0.00e+00]\n",
      " [  0.00e+00   0.00e+00   1.03e-03   0.00e+00   0.00e+00   0.00e+00\n",
      "    0.00e+00   0.00e+00   9.98e-01   1.03e-03]\n",
      " [  0.00e+00   0.00e+00   0.00e+00   0.00e+00   1.98e-03   0.00e+00\n",
      "    0.00e+00   0.00e+00   0.00e+00   9.98e-01]]\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAdAAAAFuCAYAAAAxnUkFAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nB3M9TqhJuU0AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f1758cce4e0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAcYAAAFuCAYAAAAbLbiqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f1758ccecf8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import itertools\n",
    "\n",
    "class_names = [i for i in range(10)]\n",
    "np.set_printoptions(precision=2)\n",
    "\n",
    "# Plot non-normalized confusion matrix\n",
    "plt.figure(figsize=(20,5))\n",
    "plot_confusion_matrix(cnf_matrix, classes=class_names,\n",
    "                      title='Confusion matrix, without normalization')\n",
    "\n",
    "# Plot normalized confusion matrix\n",
    "plt.figure(figsize=(20,5))\n",
    "plot_confusion_matrix(cnf_matrix, classes=class_names, normalize=True,\n",
    "                      title='Normalized confusion matrix')\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Save weights if model is accurate enough"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 100,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# save the weights for every classifier\n",
    "for i, m in enumerate(models):\n",
    "    m.save_weights(\"weights/weights_leaky_relu_model_0.22_additional_epochs\"+str(i)+'.pkl')"
   ]
  }
 ],
 "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": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}