deep_learning_project_clean.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "## Useful for importing libraries on Google cloud. libraries are in Notebooks/libs folder\n",
    "import sys\n",
    "sys.path.insert(0, 'libs')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import numpy as np\n",
    "import keras \n",
    "import pandas as pd\n",
    "from sqlalchemy import create_engine\n",
    "\n",
    "from random import randint\n",
    "import matplotlib.pyplot as plt"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "!sudo apt-get install libatlas-base-dev"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "!pip install --upgrade pandas"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "!python3 -m pip install --upgrade pip"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from google.cloud import storage\n",
    "\n",
    "from io import BytesIO\n",
    "\n",
    "client = storage.Client()\n",
    "\n",
    "bucket = \"dataham\""
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "filename_keras = \"/Generated_KERAS/images_augmented.npy\"\n",
    "\n",
    "blob = storage.blob.Blob(filename_keras, bucket)"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "!gsutil -m cp gs://dataham/Generated_KERAS/images_augmented.npy ./data_keras"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "!gsutil -m cp gs://dataham/Generated_KERAS/labels_augmented.npy ./data_keras"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## IMPORTANT FUNCTIONS "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "####### SHUFFLE DATA ####### \n",
    "def shuffle_data(X, Y):\n",
    "    nb_rows = X.shape[0]\n",
    "    index_rand = np.arange(nb_rows)\n",
    "    np.random.shuffle(index_rand)\n",
    "    X_shuffle, Y_shuffle = X[index_rand], Y[index_rand]\n",
    "    return X_shuffle, Y_shuffle"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "def concat_train_label(X1, X2, Y1, Y2):\n",
    "    nb_obs = X1.shape[0] + X2.shape[0]\n",
    "    X_aug = np.empty((nb_obs, 128, 128, 3))\n",
    "    Y_aug = np.empty((nb_obs,))\n",
    "    nb_obs_or = X1.shape[0]\n",
    "    \n",
    "    for k in range(X1.shape[0]):\n",
    "        X_aug[k] = X1[k]\n",
    "        Y_aug[k] = Y1[k]\n",
    "    \n",
    "    iteration = 0\n",
    "    while iteration != Y2.shape[0]:\n",
    "        X_aug[iteration+nb_obs_or] = X2[iteration]\n",
    "        Y_aug[iteration+nb_obs_or] = Y2[iteration]\n",
    "        iteration += 1\n",
    "        \n",
    "    return X_aug, Y_aug"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn.utils.multiclass import unique_labels\n",
    "\n",
    "def plot_confusion_matrix(model, x_val = x_val, y_val = y_val, classes = np.array([0,1]),\n",
    "                          normalize=False,\n",
    "                          title=None,\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",
    "    \n",
    "    \n",
    "    y_pred = model.predict_classes(x_val)\n",
    "    y_true = y_val \n",
    "    if not title:\n",
    "        if normalize:\n",
    "            title = 'Normalized confusion matrix'\n",
    "        else:\n",
    "            title = 'Confusion matrix, without normalization'\n",
    "\n",
    "    # Compute confusion matrix\n",
    "    cm = confusion_matrix(y_true, y_pred)\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",
    "    fig, ax = plt.subplots(figsize = (15,10))\n",
    "    #fig = plt.figure((15,10))\n",
    "    im = ax.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    ax.figure.colorbar(im, ax=ax)\n",
    "    # We want to show all ticks...\n",
    "    ax.set(xticks=np.arange(cm.shape[1]),\n",
    "           yticks=np.arange(cm.shape[0]),\n",
    "           # ... and label them with the respective list entries\n",
    "           xticklabels=classes, yticklabels=classes,\n",
    "           title=title,\n",
    "           ylabel='True label',\n",
    "           xlabel='Predicted label')\n",
    "\n",
    "    # Rotate the tick labels and set their alignment.\n",
    "    plt.setp(ax.get_xticklabels(), rotation=45, ha=\"right\",\n",
    "             rotation_mode=\"anchor\")\n",
    "\n",
    "    # Loop over data dimensions and create text annotations.\n",
    "    fmt = '.2f' if normalize else 'd'\n",
    "    thresh = cm.max() / 2.\n",
    "    for i in range(cm.shape[0]):\n",
    "        for j in range(cm.shape[1]):\n",
    "            ax.text(j, i, format(cm[i, j], fmt),\n",
    "                    ha=\"center\", va=\"center\",\n",
    "                    color=\"white\" if cm[i, j] > thresh else \"black\")\n",
    "    fig.tight_layout()\n",
    "    return ax\n",
    "\n",
    "# Plot non-normalized confusion matrix\n",
    "# plot_confusion_matrix(model_best2)\n",
    "\n",
    "# Plot normalized confusion matrix\n",
    "# plot_confusion_matrix(model_best2, normalize = True)\n",
    "\n",
    "# plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# DATA"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "######### DATASET  ######\n",
    "\n",
    "\n",
    "*** Class MEL: training: 698 , validation: 190\n",
    "\n",
    "*** Class NV: training: 4290 , validation: 1058"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "#### FROM ORIGINAL DATASET WITHOUT KERAS #####\n",
    "\n",
    "#### LOAD DATA FROM LOCAL STORAGE ### \n",
    "\n",
    "###### MEL: class 1 ######\n",
    "x_mel_train, x_mel_test = np.load(\"./dataset/train_mel.npy\"), np.load(\"./dataset/test_mel.npy\")\n",
    "y_mel_train, y_mel_test = np.ones((x_mel_train.shape[0], )), np.ones((x_mel_test.shape[0], ))\n",
    "\n",
    "\n",
    "###### NV: class 0 ######\n",
    "x_nv_train, x_nv_test = np.load(\"./dataset/train_nv.npy\"), np.load(\"./dataset/test_nv.npy\")\n",
    "y_nv_train, y_nv_test = np.zeros((x_nv_train.shape[0], )), np.zeros((x_nv_test.shape[0], ))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "#### FROM ORIGINAL DATASET TRANSFORMED WITH KERAS #####\n",
    "\n",
    "#### LOAD DATA FROM LOCAL STORAGE ### \n",
    "\n",
    "filename_local_images_keras = './data_keras/images_augmented.npy'\n",
    "filename_local_labels_keras = './data_keras/labels_augmented.npy'\n",
    "\n",
    "array_images_ker = np.load(filename_local_images_keras)\n",
    "array_labels_ker = np.load(filename_local_labels_keras)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# del array_images_ker, array_labels_ker"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "We have 5348 images of class nv.\n",
      "=================================\n",
      "We have 888 images of class mel.\n"
     ]
    }
   ],
   "source": [
    "nb_nv = np.sum(array_labels_ker == 0) ## nv with class label 0\n",
    "nb_mel = np.sum(array_labels_ker == 1) ## mel with class label 1\n",
    "\n",
    "print (\"We have {} images of class nv.\".format(nb_nv))\n",
    "\n",
    "print(\"=================================\")\n",
    "\n",
    "print (\"We have {} images of class mel.\".format(nb_mel))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "######### CAREFUL: THE DATA DOWN HERE IS THE DATA AFTER SOME TRANSF WITH KERAS ##############\n",
    "\n",
    "def split_data(factor = 0.8):\n",
    "    \"\"\"\n",
    "    Params \n",
    "    --------------\n",
    "    factor: float \n",
    "    We want 80% data in train set and 20% in validation set\n",
    "    Return\n",
    "    --------------\n",
    "    x_train, y_train, x_val, y_val\n",
    "    \"\"\"\n",
    "    boundary_val = np.int(6236*factor) # 6236 corresponds to total nb of images\n",
    "    x_train, y_train = array_images_ker[0:boundary_val], array_labels_ker[0:boundary_val]\n",
    "    x_val, y_val = array_images_ker[boundary_val:], array_labels_ker[boundary_val:]\n",
    "    return (x_train, y_train), (x_val, y_val)\n",
    "\n",
    "(x_train, y_train), (x_val, y_val) = split_data()    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 77,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "============= ORIGINAL ==================\n",
      "Original data, class nv percentage: 85.76010262989095\n",
      "Original data, class mel percentage: 14.239897370109045\n",
      "============= TRAINING ==================\n",
      "Train data, class nv percentage: 86.00641539695269\n",
      "Train data, class mel percentage: 13.993584603047312\n",
      "============= VALIDATION ==================\n",
      "Val data, class nv percentage: 84.77564102564102\n",
      "Val data, class mel percentage: 15.224358974358973\n"
     ]
    }
   ],
   "source": [
    "######### ORIGINAL ##########\n",
    "percentage_nv_org = (nb_nv/(nb_nv+nb_mel))*100\n",
    "percentage_mel_org = (nb_mel/(nb_nv+nb_mel))*100\n",
    "print(\"============= ORIGINAL ==================\")\n",
    "print(\"Original data, class nv percentage: {}\".format(percentage_nv_org))\n",
    "print(\"Original data, class mel percentage: {}\".format(percentage_mel_org))\n",
    "\n",
    "\n",
    "\n",
    "###### TRAIN ############\n",
    "nb_nv_train = np.sum(y_train == 0)\n",
    "nb_mel_train = np.sum(y_train == 1)\n",
    "percentage_nv_train = (nb_nv_train/(nb_nv_train+nb_mel_train))*100\n",
    "percentage_mel_train= (nb_mel_train/(nb_nv_train+nb_mel_train))*100\n",
    "print(\"============= TRAINING ==================\")\n",
    "print(\"Train data, class nv percentage: {}\".format(percentage_nv_train))\n",
    "print(\"Train data, class mel percentage: {}\".format(percentage_mel_train))\n",
    "\n",
    "\n",
    "########### VAL ##########\n",
    "nb_nv_val = np.sum(y_val == 0)\n",
    "nb_mel_val = np.sum(y_val == 1)\n",
    "percentage_nv_val = (nb_nv_val/(nb_nv_val+nb_mel_val))*100\n",
    "percentage_mel_val = (nb_mel_val/(nb_nv_val+nb_mel_val))*100\n",
    "print(\"============= VALIDATION ==================\")\n",
    "print(\"Val data, class nv percentage: {}\".format(percentage_nv_val))\n",
    "print(\"Val data, class mel percentage: {}\".format(percentage_mel_val))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Simple convolutional Layer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "from keras import backend as K\n",
    "\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Conv2D, MaxPooling2D\n",
    "from keras.layers import Activation, Dropout, Flatten, Dense\n",
    "from collections import Counter \n",
    "from keras import backend as K\n",
    "from keras import optimizers\n",
    "from keras.callbacks import CSVLogger\n",
    "\n",
    "csv_logger = CSVLogger('log.csv', append=True, separator=';')\n",
    "\n",
    "\n",
    "\n",
    "def evaluate_classifier_conv(x_train = x_train, y_train = y_train, x_val = x_val, y_val = y_val, img_rows = 128, img_cols = 128, batch_size = 32, epochs = 20, learning_rate = 0.01, dropout = 0.7, log = True):\n",
    "    \"\"\"\n",
    "    Create simple convolutional model to classify data\n",
    "    Fit data and return fitted model \n",
    "    \"\"\"\n",
    "    \n",
    "    K.clear_session()\n",
    "    np.random.seed(100)\n",
    "\n",
    "    if K.image_data_format() == 'channels_first':\n",
    "        x_train = x_train.reshape(x_train.shape[0], 3, img_rows, img_cols)\n",
    "        x_val = x_val.reshape(x_val.shape[0], 3, img_rows, img_cols)\n",
    "        input_shape = (3, img_rows, img_cols)\n",
    "    else:\n",
    "        x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 3)\n",
    "        x_val = x_val.reshape(x_val.shape[0], img_rows, img_cols, 3)\n",
    "        input_shape = (img_rows, img_cols, 3)\n",
    "\n",
    "    model = Sequential()\n",
    "    model.add(Conv2D(32, (3, 3), input_shape=(128, 128,3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Conv2D(32, (3, 3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Conv2D(64, (3, 3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Flatten())  # this converts our 3D feature maps to 1D feature vectors\n",
    "    model.add(Dense(64))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Dropout(dropout))\n",
    "    model.add(Dense(1))\n",
    "    model.add(Activation('sigmoid'))\n",
    "\n",
    "\n",
    "    # initiate RMSprop optimizer\n",
    "    opt = optimizers.rmsprop(lr=learning_rate, decay = 1e-5)\n",
    "\n",
    "    model.compile(loss='binary_crossentropy',\n",
    "                  optimizer=opt,\n",
    "                  metrics=['accuracy'])\n",
    "    \n",
    "    if log:\n",
    "        # Fit data \n",
    "        model.fit(x_train, y_train,\n",
    "                  batch_size=batch_size,\n",
    "                  epochs=epochs,\n",
    "                  verbose=1,\n",
    "                  validation_data=(x_val, y_val),\n",
    "                    callbacks=[csv_logger])\n",
    "    else:\n",
    "         # Fit data \n",
    "        model.fit(x_train, y_train,\n",
    "                  batch_size=batch_size,\n",
    "                  epochs=epochs,\n",
    "                  verbose=1,\n",
    "                  validation_data=(x_val, y_val))\n",
    "\n",
    "    # Compute final performance on validation data\n",
    "    score = model.evaluate(x_val, y_val, verbose=0)\n",
    "    print('Final validation loss:', score[0])\n",
    "    print('Final validation accuracy:', score[1])\n",
    "    \n",
    "    return model  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf \n",
    "from keras.models import Sequential\n",
    "from keras.layers import Conv2D, MaxPooling2D\n",
    "from keras.layers import Activation, Dropout, Flatten, Dense\n",
    "from collections import Counter \n",
    "from keras import backend as K\n",
    "from keras import optimizers\n",
    "from keras.callbacks import CSVLogger\n",
    "from keras.callbacks import ModelCheckpoint\n",
    "\n",
    "from keras.callbacks import Callback\n",
    "\n",
    "from sklearn.metrics import f1_score\n",
    "\n",
    "\n",
    "csv_logger = CSVLogger('log.csv', append=True, separator=';')\n",
    "\n",
    "\n",
    "\n",
    "    \n",
    "    \n",
    "class EarlyStopByF1(Callback):\n",
    "    def __init__(self, value = 0, verbose = 0):\n",
    "        super(Callback, self).__init__()\n",
    "        self.value = value\n",
    "        self.verbose = verbose\n",
    "\n",
    "    def f1(self, y_true, y_pred):\n",
    "        def recall(y_true, y_pred):\n",
    "            \"\"\"Recall metric.\n",
    "\n",
    "            Only computes a batch-wise average of recall.\n",
    "\n",
    "            Computes the recall, a metric for multi-label classification of\n",
    "            how many relevant items are selected.\n",
    "            \"\"\"\n",
    "            true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "            possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))\n",
    "            recall = true_positives / (possible_positives + K.epsilon())\n",
    "            return recall\n",
    "\n",
    "        def precision(y_true, y_pred):\n",
    "            \"\"\"Precision metric.\n",
    "\n",
    "            Only computes a batch-wise average of precision.\n",
    "\n",
    "            Computes the precision, a metric for multi-label classification of\n",
    "            how many selected items are relevant.\n",
    "            \"\"\"\n",
    "            true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "            predicted_positives = K.sum(K.round(K.clip(y_pred, 0, 1)))\n",
    "            precision = true_positives / (predicted_positives + K.epsilon())\n",
    "            return precision\n",
    "        precision = precision(y_true, y_pred)\n",
    "        recall = recall(y_true, y_pred)\n",
    "        return 2*((precision*recall)/(precision+recall+K.epsilon()))\n",
    "    \n",
    "    \n",
    "    def on_epoch_end(self, epoch, logs={}):\n",
    "        predict = np.asarray(self.model.predict(self.validation_data[0]))\n",
    "        target = self.validation_data[1]\n",
    "        predict = tf.convert_to_tensor(predict, tf.float32)\n",
    "        target = tf.convert_to_tensor(target, tf.float32)\n",
    "        score = self.f1(target, predict) # score is now a tf.tensor (necessary for early stopping)\n",
    "#         print (type(score)) # tensor \n",
    "        tensor_value = tf.constant(self.value, shape = (1,))\n",
    "        with tf.Session() as sess:\n",
    "            tensor_value_score = sess.run(tf.greater(score, tensor_value))\n",
    "#             print(sess.run(tf.greater(score, tensor_value))) ## prints [True] when we go beyond self.value\n",
    " \n",
    "        if np.all(tensor_value_score):# if we go beyond our value, we can stop \n",
    "            if self.verbose > 0: # print\n",
    "                print(\"Epoch %05d: early stopping Threshold\" % epoch)\n",
    "            self.model.stop_training = True\n",
    "\n",
    "\n",
    "########## Customize Early Stopping ###########\n",
    "from keras.callbacks import Callback\n",
    "\n",
    "class EarlyStoppingByAccuracy(Callback):\n",
    "    \"\"\"\n",
    "    When checking the logs we see that sometimes we get a very low val loss/ high val accuracy at some epoch and after we go below such values.\n",
    "    Solution:  We will check when such accuracy is reached and stop at this very moment \n",
    "    \n",
    "    PARAMS\n",
    "    --------------------------\n",
    "    value: int v\n",
    "    Value that is seen to be passed for a given epoch => If we are beyond it we stop training!! \n",
    "    \n",
    "    \"\"\"\n",
    "    def __init__(self, monitor='accuracy', value=0.85, verbose=0):\n",
    "        super(Callback, self).__init__()\n",
    "        self.monitor = monitor\n",
    "        self.value = value\n",
    "        self.verbose = verbose\n",
    "\n",
    "    def on_epoch_end(self, epoch, logs={}):\n",
    "        current = logs.get(self.monitor)\n",
    "        if current is None:\n",
    "            warnings.warn(\"Early stopping requires %s available!\" % self.monitor, RuntimeWarning)\n",
    "\n",
    "        if current >= self.value:\n",
    "            if self.verbose > 0:\n",
    "                print(\"Epoch %05d: early stopping THR\" % epoch)\n",
    "            self.model.stop_training = True\n",
    "            \n",
    "def evaluate_classifier_conv_custom_fscore(x_train, y_train, x_val, y_val, img_rows = 128, img_cols = 128, \n",
    "                                           batch_size = 32, epochs = 20, learning_rate = 0.01, \n",
    "                                           dropout = 0.7, log = True, threshold_acc = 0.87, name_model = 'stack', threshold_fscore = 0.5):\n",
    "    \"\"\"\n",
    "    Create simple convolutional model to classify data\n",
    "    Fit data and return fitted model \n",
    "    \n",
    "    Custom callback\n",
    "    \"\"\"\n",
    "    \n",
    "    K.clear_session()\n",
    "    np.random.seed(100)\n",
    "\n",
    "    if K.image_data_format() == 'channels_first':\n",
    "        x_train = x_train.reshape(x_train.shape[0], 3, img_rows, img_cols)\n",
    "        x_val = x_val.reshape(x_val.shape[0], 3, img_rows, img_cols)\n",
    "        input_shape = (3, img_rows, img_cols)\n",
    "    else:\n",
    "        x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 3)\n",
    "        x_val = x_val.reshape(x_val.shape[0], img_rows, img_cols, 3)\n",
    "        input_shape = (img_rows, img_cols, 3)\n",
    "\n",
    "    model = Sequential()\n",
    "    model.add(Conv2D(32, (3, 3), input_shape=(128, 128,3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Conv2D(32, (3, 3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Conv2D(64, (3, 3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Flatten())  # this converts our 3D feature maps to 1D feature vectors\n",
    "    model.add(Dense(64))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Dropout(dropout))\n",
    "    model.add(Dense(1))\n",
    "    model.add(Activation('sigmoid'))\n",
    "    \n",
    "    def recall(y_true, y_pred):\n",
    "            \"\"\"Recall metric.\n",
    "\n",
    "            Only computes a batch-wise average of recall.\n",
    "\n",
    "            Computes the recall, a metric for multi-label classification of\n",
    "            how many relevant items are selected.\n",
    "            \"\"\"\n",
    "            true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "            possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))\n",
    "            recall = true_positives / (possible_positives + K.epsilon())\n",
    "            return recall\n",
    "    \n",
    "    def f1(y_true, y_pred):\n",
    "        def recall(y_true, y_pred):\n",
    "            \"\"\"Recall metric.\n",
    "\n",
    "            Only computes a batch-wise average of recall.\n",
    "\n",
    "            Computes the recall, a metric for multi-label classification of\n",
    "            how many relevant items are selected.\n",
    "            \"\"\"\n",
    "            true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "            possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))\n",
    "            recall = true_positives / (possible_positives + K.epsilon())\n",
    "            return recall\n",
    "\n",
    "        def precision(y_true, y_pred):\n",
    "            \"\"\"Precision metric.\n",
    "\n",
    "            Only computes a batch-wise average of precision.\n",
    "\n",
    "            Computes the precision, a metric for multi-label classification of\n",
    "            how many selected items are relevant.\n",
    "            \"\"\"\n",
    "            true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "            predicted_positives = K.sum(K.round(K.clip(y_pred, 0, 1)))\n",
    "            precision = true_positives / (predicted_positives + K.epsilon())\n",
    "            return precision\n",
    "        precision = precision(y_true, y_pred)\n",
    "        recall = recall(y_true, y_pred)\n",
    "        return 2*((precision*recall)/(precision+recall+K.epsilon()))\n",
    "\n",
    "\n",
    "\n",
    "    # initiate RMSprop optimizer\n",
    "    opt = optimizers.rmsprop(lr=learning_rate, decay = 1e-5)\n",
    "\n",
    "    model.compile(loss='binary_crossentropy',\n",
    "                  optimizer=opt,\n",
    "                  metrics=['accuracy', f1, recall])\n",
    "    \n",
    "    \n",
    "    kfold_weights_path = \"best_model{}.hdf5\".format(name_model)\n",
    "    callbacks = [\n",
    "                EarlyStoppingByAccuracy(monitor='val_acc', value=threshold_acc, verbose=1), # early stopping based on validation accuracy \n",
    "                ModelCheckpoint(kfold_weights_path, monitor='f1', save_best_only=True, verbose=0),\n",
    "                EarlyStopByF1(value = threshold_fscore, verbose = 1) # early stopping based on f1 score\n",
    "                ]\n",
    "\n",
    "    if log:\n",
    "        callbacks.append(csv_logger)\n",
    "        # Fit data \n",
    "        model.fit(x_train, y_train,\n",
    "                  batch_size=batch_size,\n",
    "                  epochs=epochs,\n",
    "                  verbose=1,\n",
    "                  validation_data=(x_val, y_val),\n",
    "                    callbacks = callbacks)\n",
    "        \n",
    "    else:\n",
    "        \n",
    "        # Fit data \n",
    "        model.fit(x_train, y_train,\n",
    "                  batch_size=batch_size,\n",
    "                  epochs=epochs,\n",
    "                  verbose=1,\n",
    "                  validation_data=(x_val, y_val),\n",
    "                  callbacks = callbacks)\n",
    "\n",
    "        \n",
    "    # Compute final performance on validation data\n",
    "    score = model.evaluate(x_val, y_val, verbose=0)\n",
    "    print('Final validation loss:', score[0])\n",
    "    print('Final validation accuracy:', score[1])\n",
    "    \n",
    "    return model  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# MOBILENET VERSION2 "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def get_one_hot(y_in):\n",
    "    one_hot = np.zeros((y_in.shape[0], 2))\n",
    "    for i in np.arange(y_in.shape[0]):\n",
    "        one_hot[i][int(y_in[i])] = 1\n",
    "    return one_hot\n",
    "\n",
    "y_train_one_hot = get_one_hot(y_train)\n",
    "y_val_one_hot = get_one_hot(y_val)\n",
    "\n",
    "y_train_one_hot_aug_stacked = get_one_hot(y_train_aug_stacked)\n",
    "y_train_one_hot_aug_simple = get_one_hot(y_train_aug_simple)\n",
    "y_train_one_hot_aug_strong = get_one_hot(y_train_aug_strong)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "def build_model():\n",
    "    input_tensor = Input(shape=(128, 128, 3))\n",
    "    base_model = MobileNetV2(\n",
    "        include_top=False,\n",
    "        weights='imagenet',\n",
    "        input_tensor=input_tensor,\n",
    "        input_shape=(128, 128, 3),\n",
    "        pooling='avg')\n",
    "\n",
    "    for layer in base_model.layers:\n",
    "        layer.trainable = True\n",
    "    for layer in base_model.layers[:FREEZE_NO]:\n",
    "        layer.trainable = False\n",
    "\n",
    "    op = Dense(128, activation='relu')(base_model.output)\n",
    "    op = Dropout(DROPOUT_RATIO)(op)\n",
    "\n",
    "    output_tensor = Dense(2, activation='softmax')(op)\n",
    "\n",
    "    model = Model(inputs=input_tensor, outputs=output_tensor)\n",
    "    model.compile(optimizer=Adam(lr=0.000005),\n",
    "              loss='categorical_crossentropy',\n",
    "              metrics=['categorical_accuracy'])\n",
    "    \n",
    "    return model\n",
    "\n",
    "model_noaug = build_model()\n",
    "model_stacked = build_model()\n",
    "model_simple = build_model()\n",
    "model_strong = build_model()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def train(model, train_data, val_data):\n",
    "    history = model.fit(train_data[0], train_data[1],\n",
    "            batch_size=BATCH_SIZE,\n",
    "            epochs=EPOCH_NO,\n",
    "            validation_data=(val_data))\n",
    "    return history\n",
    "history_strong = train(model_strong, (x_train_aug_strong, y_train_one_hot_aug_strong), (x_val, y_val_one_hot))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### WITHOUT DATA AUGMENTATION`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "##### PARAMS FOR ORIGINAL DATASET #####\n",
    "model_params_org = { \n",
    "                    \"x_train\": x_train, \n",
    "                    \"y_train\": y_train, \n",
    "                    \"x_val\": x_val, \n",
    "                    \"y_val\": y_val\n",
    "                    \"img_rows\": 128,\n",
    "                    \"img_cols\": 128,\n",
    "                    \"batch_size\": 32, \n",
    "                    \"epochs\": 20, \n",
    "                    \"learning_rate\": 0.01\n",
    "                    \"dropout\": 0.7\n",
    "                    }"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model_original_conv = evaluate_classifier_conv(**model_params_org)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 85,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>epoch</th>\n",
       "      <th>acc</th>\n",
       "      <th>loss</th>\n",
       "      <th>val_acc</th>\n",
       "      <th>val_loss</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>0.859663</td>\n",
       "      <td>2.254287</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>5</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>6</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>7</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>8</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>9</th>\n",
       "      <td>9</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>10</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>11</th>\n",
       "      <td>11</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <td>12</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>13</th>\n",
       "      <td>13</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>14</th>\n",
       "      <td>14</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <td>15</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>16</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>17</th>\n",
       "      <td>17</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>18</th>\n",
       "      <td>18</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>19</th>\n",
       "      <td>19</td>\n",
       "      <td>0.860064</td>\n",
       "      <td>2.255499</td>\n",
       "      <td>0.847756</td>\n",
       "      <td>2.453877</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    epoch       acc      loss   val_acc  val_loss\n",
       "0       0  0.859663  2.254287  0.847756  2.453877\n",
       "1       1  0.860064  2.255499  0.847756  2.453877\n",
       "2       2  0.860064  2.255499  0.847756  2.453877\n",
       "3       3  0.860064  2.255499  0.847756  2.453877\n",
       "4       4  0.860064  2.255499  0.847756  2.453877\n",
       "5       5  0.860064  2.255499  0.847756  2.453877\n",
       "6       6  0.860064  2.255499  0.847756  2.453877\n",
       "7       7  0.860064  2.255499  0.847756  2.453877\n",
       "8       8  0.860064  2.255499  0.847756  2.453877\n",
       "9       9  0.860064  2.255499  0.847756  2.453877\n",
       "10     10  0.860064  2.255499  0.847756  2.453877\n",
       "11     11  0.860064  2.255499  0.847756  2.453877\n",
       "12     12  0.860064  2.255499  0.847756  2.453877\n",
       "13     13  0.860064  2.255499  0.847756  2.453877\n",
       "14     14  0.860064  2.255499  0.847756  2.453877\n",
       "15     15  0.860064  2.255499  0.847756  2.453877\n",
       "16     16  0.860064  2.255499  0.847756  2.453877\n",
       "17     17  0.860064  2.255499  0.847756  2.453877\n",
       "18     18  0.860064  2.255499  0.847756  2.453877\n",
       "19     19  0.860064  2.255499  0.847756  2.453877"
      ]
     },
     "execution_count": 85,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "log_training = pd.read_csv(\"./log.csv\", sep = \";\")\n",
    "log_training"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 92,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "============ Class NV ========= \n",
      "Accuracy on images on nv class fromn validation data: 1.0\n",
      "============ Class MEL ========= \n",
      "Accuracy on images on mel class fromn validation data: 0.0\n"
     ]
    }
   ],
   "source": [
    "index_nv_val = np.where( y_val == 0 )\n",
    "index_mel_val = np.where( y_val == 1 )\n",
    "\n",
    "score_nv = model.evaluate(x_val[index_nv_val], y_val[index_nv_val], verbose=0)\n",
    "score_mel = model.evaluate(x_val[index_mel_val], y_val[index_mel_val], verbose=0)\n",
    "\n",
    "print(\"============ Class NV ========= \")\n",
    "print(\"Accuracy on images on nv class fromn validation data: {}\".format(score_nv[1]))\n",
    "\n",
    "print(\"============ Class MEL ========= \")\n",
    "print(\"Accuracy on images on mel class fromn validation data: {}\".format(score_mel[1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### WITH DATA AUGMENTATION`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "############ TRANSFORMATIONS WITH NUMPY ##########\n",
    "import numpy as np\n",
    "from PIL import Image\n",
    "from scipy.misc import imresize\n",
    "from scipy.ndimage.interpolation import rotate\n",
    "\n",
    "\n",
    "def resize(image, size):\n",
    "    size = check_size(size)\n",
    "    image = imresize(image, size)\n",
    "    return image\n",
    "\n",
    "def check_size(size):\n",
    "    if type(size) == int:\n",
    "        size = (size, size)\n",
    "    if type(size) != tuple:\n",
    "        raise TypeError('size is int or tuple')\n",
    "    return size\n",
    "\n",
    "def random_crop(image, crop_size):\n",
    "    crop_size = check_size(crop_size)\n",
    "    h, w, _ = image.shape\n",
    "    top = np.random.randint(0, h - crop_size[0])\n",
    "    left = np.random.randint(0, w - crop_size[1])\n",
    "    bottom = top + crop_size[0]\n",
    "    right = left + crop_size[1]\n",
    "    image = image[top:bottom, left:right, :]\n",
    "    return image\n",
    "\n",
    "\n",
    "def horizontal_flip(image, rate=0.5):\n",
    "    if np.random.rand() < rate:\n",
    "        image = image[:, ::-1, :]\n",
    "    return image\n",
    "\n",
    "\n",
    "def vertical_flip(image, rate=0.5):\n",
    "    if np.random.rand() < rate:\n",
    "        image = image[::-1, :, :]\n",
    "    return image\n",
    "\n",
    "\n",
    "def scale_augmentation(image, scale_range, crop_size):\n",
    "    scale_size = np.random.randint(*scale_range)\n",
    "    image = imresize(image, (scale_size, scale_size))\n",
    "    image = random_crop(image, crop_size)\n",
    "    return image\n",
    "\n",
    "\n",
    "def random_rotation(image, angle_range=(0, 180)):\n",
    "    h, w, _ = image.shape\n",
    "    angle = np.random.randint(*angle_range)\n",
    "    image = rotate(image, angle)\n",
    "    image = resize(image, (h, w))\n",
    "    return image"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "transf_arr = [\"random_rotation\", \"vertical_flip\", \"horizontal_flip\" ]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## DO NOT RUN AGAIN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python3.5/dist-packages/ipykernel_launcher.py:10: DeprecationWarning: `imresize` is deprecated!\n",
      "`imresize` is deprecated in SciPy 1.0.0, and will be removed in 1.3.0.\n",
      "Use Pillow instead: ``numpy.array(Image.fromarray(arr).resize())``.\n",
      "  # Remove the CWD from sys.path while we load stuff.\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "==========================\n",
      "Nice! 1000 steps done already :) \n"
     ]
    }
   ],
   "source": [
    "#### We are going to augment only the training data so that both classes in the training data have the same proportion ####\n",
    "x_train_copy, y_train_copy = np.copy(x_train), np.copy(y_train)\n",
    "\n",
    "iteration = 0 \n",
    "for i in range(x_train_copy.shape[0]):\n",
    "    iteration += 1\n",
    "    if iteration % 1000 == 0:\n",
    "        print(\"==========================\")\n",
    "        print(\"Nice! {} steps done already :) \".format(iteration))\n",
    "    label = y_train_copy[i]\n",
    "    if label == 1: # if class mel, then duplicate 6 times \n",
    "        image = x_train_copy[i]\n",
    "        for k in range(6):\n",
    "            random_int = randint(0,2) # random index in transformation array \n",
    "            if random_int == 0: # transf is random_rotation\n",
    "                image_new = random_rotation(image)\n",
    "            elif random_int == 1:\n",
    "                image_new = vertical_flip(image)\n",
    "            elif random_int == 2: \n",
    "                image_new = horizontal_flip(image)\n",
    "    #         print (image_new.shape)\n",
    "            image_new = image_new.reshape(1,128,128,3)\n",
    "            x_train_copy = np.append(x_train_copy, image_new, axis = 0)\n",
    "            y_train_copy = np.append(y_train_copy, np.array([label]), axis = 0)\n",
    "            \n",
    "filename_X_train_local = './data_keras/Xtrain_aug.npy'\n",
    "filename_Y_train_local = './data_keras/Ytrain_aug.npy'\n",
    "# np.save(filename_X_train_local, x_train_copy)\n",
    "# np.save(filename_Y_train_local, y_train_copy)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## ==============================="
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "============= TRAINING: AUGMENTED DATASET ==================\n",
      "Train data, class nv percentage: 46.752397558849175\n",
      "Train data, class mel percentage: 53.24760244115083\n"
     ]
    }
   ],
   "source": [
    "###### LOAD AUGMENTED DATA ######\n",
    "filename_X_train_local = './data_keras/Xtrain_aug.npy'\n",
    "filename_Y_train_local = './data_keras/Ytrain_aug.npy'\n",
    "x_train_aug, y_train_aug = np.load(filename_X_train_local), np.load(filename_Y_train_local)\n",
    "\n",
    "##### Shuffle data #####\n",
    "x_train_aug, y_train_aug = shuffle_data(x_train_aug, y_train_aug)\n",
    "\n",
    "###### TRAIN ############\n",
    "nb_nv_train_aug = np.sum(y_train_aug == 0)\n",
    "nb_mel_train_aug = np.sum(y_train_aug == 1)\n",
    "percentage_nv_train = (nb_nv_train_aug/(nb_nv_train_aug+nb_mel_train_aug))*100\n",
    "percentage_mel_train= (nb_mel_train_aug/(nb_nv_train_aug+nb_mel_train_aug))*100\n",
    "print(\"============= TRAINING: AUGMENTED DATASET ==================\")\n",
    "print(\"Train data, class nv percentage: {}\".format(percentage_nv_train))\n",
    "print(\"Train data, class mel percentage: {}\".format(percentage_mel_train))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "##### PARAMS FOR AUGMENTED DATASET #####\n",
    "model_params_aug = { \n",
    "                    \"x_train\": x_train_aug, \n",
    "                    \"y_train\": y_train_aug, \n",
    "                    \"x_val\": x_val, \n",
    "                    \"y_val\": y_val,\n",
    "                    \"img_rows\": 128,\n",
    "                    \"img_cols\": 128,\n",
    "                    \"batch_size\": 100, \n",
    "                    \"epochs\": 2, \n",
    "                    \"learning_rate\": 0.01,\n",
    "                    \"dropout\": 0.7\n",
    "                    }\n",
    "\n",
    "csv_logger = CSVLogger('log_conv_aug.csv', append=True, separator=';')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_aug_conv = evaluate_classifier_conv(**model_params_aug)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "============ Class NV: aug ========= \n",
      "Accuracy on images on nv class fromn validation data: 1.0\n",
      "============ Class MEL: aug ========= \n",
      "Accuracy on images on mel class fromn validation data: 0.0\n"
     ]
    }
   ],
   "source": [
    "index_nv_val = np.where( y_val == 0 )\n",
    "index_mel_val = np.where( y_val == 1 )\n",
    "\n",
    "score_nv = model_aug_conv.evaluate(x_val[index_nv_val], y_val[index_nv_val], verbose=0)\n",
    "score_mel = model_aug_conv.evaluate(x_val[index_mel_val], y_val[index_mel_val], verbose=0)\n",
    "\n",
    "print(\"============ Class NV: aug ========= \")\n",
    "print(\"Accuracy on images on nv class fromn validation data: {}\".format(score_nv[1]))\n",
    "\n",
    "print(\"============ Class MEL: aug ========= \")\n",
    "print(\"Accuracy on images on mel class fromn validation data: {}\".format(score_mel[1]))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "### More advanced transformations ### \n",
    "source: https://github.com/albu/albumentations\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "## Useful for importing libraries on Google cloud. libraries are in Notebooks/libs folder\n",
    "import sys\n",
    "sys.path.insert(0, 'libs')\n",
    "import cv2\n",
    "import albumentations as A"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## NOTES: Here we work with image of type uint8!!! \n",
    "So we need to scale them maybe before feeding them to our network. Our network works with images in the range [0,1] and here we produce image in the range [0, 255]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "BOX_COLOR = (255, 0, 0)\n",
    "TEXT_COLOR = (255, 255, 255)\n",
    "\n",
    "def visualize_bbox(img, bbox, color=BOX_COLOR, thickness=2, **kwargs):\n",
    "    #height, width = img.shape[:2]\n",
    "\n",
    "    x_min, y_min, w, h = bbox\n",
    "    x_min, x_max, y_min, y_max = int(x_min), int(x_min + w), int(y_min), int(y_min + h)\n",
    "    \n",
    "    cv2.rectangle(img, (x_min, y_min), (x_max, y_max), color=color, thickness=thickness)\n",
    "    return img\n",
    "\n",
    "def visualize_titles(img, bbox, title, color=BOX_COLOR, thickness=2, font_thickness = 2, font_scale=0.35, **kwargs):\n",
    "    #height, width = img.shape[:2]\n",
    "    x_min, y_min, w, h = bbox\n",
    "    x_min, x_max, y_min, y_max = int(x_min), int(x_min + w), int(y_min), int(y_min + h)\n",
    "    \n",
    "    ((text_width, text_height), _) = cv2.getTextSize(title, cv2.FONT_HERSHEY_SIMPLEX, font_scale, font_thickness)\n",
    "    cv2.rectangle(img, (x_min, y_min - int(1.3 * text_height)), (x_min + text_width, y_min), BOX_COLOR, -1)\n",
    "    cv2.putText(img, title, (x_min, y_min - int(0.3 * text_height)), cv2.FONT_HERSHEY_SIMPLEX, font_scale, TEXT_COLOR,\n",
    "                font_thickness, lineType=cv2.LINE_AA)\n",
    "    return img\n",
    "\n",
    "\n",
    "def augment_and_show(aug, image, mask=None, bboxes=[], categories=[], category_id_to_name=[], filename=None, \n",
    "                     font_scale_orig=0.35, \n",
    "                     font_scale_aug=0.35, show_title=True, **kwargs):\n",
    "\n",
    "    augmented = aug(image=image, mask=mask, bboxes=bboxes, category_id=categories)\n",
    "\n",
    "    image = cv2.cvtColor(image, cv2.COLOR_BGR2RGB)\n",
    "    image_aug = cv2.cvtColor(augmented['image'], cv2.COLOR_BGR2RGB)\n",
    "\n",
    "    for bbox in bboxes:\n",
    "        visualize_bbox(image, bbox, **kwargs)\n",
    "\n",
    "    for bbox in augmented['bboxes']:\n",
    "        visualize_bbox(image_aug, bbox, **kwargs)\n",
    "\n",
    "    if show_title:\n",
    "        for bbox,cat_id in zip(bboxes, categories):\n",
    "            visualize_titles(image, bbox, category_id_to_name[cat_id], font_scale=font_scale_orig, **kwargs)\n",
    "        for bbox,cat_id in zip(augmented['bboxes'], augmented['category_id']):\n",
    "            visualize_titles(image_aug, bbox, category_id_to_name[cat_id], font_scale=font_scale_aug, **kwargs)\n",
    "\n",
    "    \n",
    "    if mask is None:\n",
    "        f, ax = plt.subplots(1, 2, figsize=(16, 8))\n",
    "        \n",
    "        ax[0].imshow(image)\n",
    "        ax[0].set_title('Original image')\n",
    "        \n",
    "        ax[1].imshow(image_aug)\n",
    "        ax[1].set_title('Augmented image')\n",
    "    else:\n",
    "        f, ax = plt.subplots(2, 2, figsize=(16, 16))\n",
    "        \n",
    "        if len(mask.shape) != 3:\n",
    "            mask = label2rgb(mask, bg_label=0)            \n",
    "            mask_aug = label2rgb(augmented['mask'], bg_label=0)\n",
    "        else:\n",
    "            mask = cv2.cvtColor(mask, cv2.COLOR_BGR2RGB)\n",
    "            mask_aug = cv2.cvtColor(augmented['mask'], cv2.COLOR_BGR2RGB)\n",
    "            \n",
    "        ax[0, 0].imshow(image)\n",
    "        ax[0, 0].set_title('Original image')\n",
    "        \n",
    "        ax[0, 1].imshow(image_aug)\n",
    "        ax[0, 1].set_title('Augmented image')\n",
    "        \n",
    "        ax[1, 0].imshow(mask, interpolation='nearest')\n",
    "        ax[1, 0].set_title('Original mask')\n",
    "\n",
    "        ax[1, 1].imshow(mask_aug, interpolation='nearest')\n",
    "        ax[1, 1].set_title('Augmented mask')\n",
    "\n",
    "    f.tight_layout()\n",
    "    if filename is not None:\n",
    "        f.savefig(filename)\n",
    "        \n",
    "    return augmented['image'], augmented['mask'], augmented['bboxes']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "from skimage import img_as_ubyte\n",
    "\n",
    "np.random.seed(42)\n",
    "image = x_train[0]\n",
    "image = img_as_ubyte(image)\n",
    "light = A.Compose([\n",
    "    A.RandomBrightnessContrast(p=1),    \n",
    "    A.RandomGamma(p=1),    \n",
    "    A.CLAHE(p=1),    \n",
    "], p=1)\n",
    "\n",
    "medium = A.Compose([\n",
    "    A.CLAHE(p=1),\n",
    "    A.HueSaturationValue(hue_shift_limit=20, sat_shift_limit=50, val_shift_limit=50, p=1),\n",
    "], p=1)\n",
    "\n",
    "\n",
    "strong = A.Compose([\n",
    "    A.ChannelShuffle(p=1),\n",
    "], p=1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x576 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "r = augment_and_show(strong, image, filename = './transforms/augmented_image_strong')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "source: https://github.com/albu/albumentations/blob/master/notebooks/example.ipynb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "from albumentations import (\n",
    "    HorizontalFlip, IAAPerspective, ShiftScaleRotate, CLAHE, RandomRotate90,\n",
    "    Transpose, ShiftScaleRotate, Blur, OpticalDistortion, GridDistortion, HueSaturationValue,\n",
    "    IAAAdditiveGaussianNoise, GaussNoise, MotionBlur, MedianBlur, RandomBrightnessContrast, IAAPiecewiseAffine,\n",
    "    IAASharpen, IAAEmboss, Flip, OneOf, Compose\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def strong_aug(p=.5):\n",
    "    return A.Compose([\n",
    "        RandomRotate90(),\n",
    "        Flip(),\n",
    "        Transpose(),\n",
    "        OneOf([\n",
    "            IAAAdditiveGaussianNoise(),\n",
    "            GaussNoise(),\n",
    "        ], p=0.2),\n",
    "        OneOf([\n",
    "            MotionBlur(p=.2),\n",
    "            MedianBlur(blur_limit=3, p=0.1),\n",
    "            Blur(blur_limit=3, p=0.1),\n",
    "        ], p=0.2),\n",
    "        ShiftScaleRotate(shift_limit=0.0625, scale_limit=0.2, rotate_limit=45, p=0.2),\n",
    "        OneOf([\n",
    "            OpticalDistortion(p=0.3),\n",
    "            GridDistortion(p=.1),\n",
    "            IAAPiecewiseAffine(p=0.3),\n",
    "        ], p=0.2),\n",
    "        OneOf([\n",
    "            CLAHE(clip_limit=2),\n",
    "            IAASharpen(),\n",
    "            IAAEmboss(),\n",
    "            RandomBrightnessContrast(),            \n",
    "        ], p=0.3),\n",
    "        HueSaturationValue(p=0.3),\n",
    "    ], p=p)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def augment_and_show(aug, image, verbose = True ):\n",
    "    image = aug(image=image)['image']\n",
    "    if verbose: \n",
    "        plt.figure(figsize=(10, 10))\n",
    "        plt.imshow(image)\n",
    "    return image "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 72,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "aug = strong_aug(p=1)\n",
    "image_aug = augment_and_show(aug, image)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## DO NOT RUN AGAIN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 83,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from skimage import img_as_ubyte\n",
    "from skimage import img_as_float\n",
    "\n",
    "\"\"\" PROBLEM MEMORY \"\"\"\n",
    "\n",
    "def augment_data(X, Y, class_label = 1, filename = None ):\n",
    "    \"\"\"\n",
    "    PARAMS\n",
    "    --------------\n",
    "    class_label: int \n",
    "    label of the observations that must be replicated: class mel of label 1 here \n",
    "    \n",
    "    filename: list \n",
    "    list[0] = file_X, list[1] = file_Y\n",
    "    \n",
    "    X, Y: numpy array\n",
    "    data that must be augmented symmetrically\n",
    "    \n",
    "    Returns\n",
    "    -------------------\n",
    "    Augmented data X_aug, Y_aug\n",
    "    \"\"\"\n",
    "    \n",
    "    X_copy = np.copy(X)\n",
    "    Y_copy = np.copy(Y)\n",
    "    iteration = 0\n",
    "    for i in range(X.shape[0]):\n",
    "        iteration += 1\n",
    "        if iteration % 500 == 0:\n",
    "            print(\"==========================\")\n",
    "            print(\"Nice! {} steps done already :) \".format(iteration))\n",
    "       \n",
    "        label = Y[i]\n",
    "        if label == class_label: ## class mel \n",
    "            ## we replicate 6 times with different ways \n",
    "            image = X[i]\n",
    "            # transform to uint8 \n",
    "            image = img_as_ubyte(image)\n",
    "            for k in range(6):\n",
    "                aug = strong_aug(p=1) # random transform of the image\n",
    "                # transform it \n",
    "                image_aug = augment_and_show(aug, image, verbose = False)\n",
    "                # convert image in float before adding in to data set \n",
    "                image_aug = img_as_float(image_aug)\n",
    "                # add image to data set: each row is a new image so axis = 0\n",
    "                image_aug = image_aug.reshape(1,128,128,3)\n",
    "                X_copy = np.append(X_copy, image_aug, axis = 0)\n",
    "                Y_copy = np.append(Y_copy, np.array([label]), axis = 0)\n",
    "                \n",
    "    if filename is not None:\n",
    "        filename_data, filename_label = filename[0], filename[1]\n",
    "        np.save(filename_data, X_copy)\n",
    "        np.save(filename_label, Y_copy)\n",
    "    \n",
    "    return X_copy, Y_copy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## DO NOT RUN AGAIN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 102,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from skimage import img_as_ubyte\n",
    "from skimage import img_as_float\n",
    "\n",
    "\"\"\" SOLUTION \"\"\"\n",
    "\n",
    "def augment_data_mel(X, Y, class_label = 1, filename = None ):\n",
    "    \"\"\"\n",
    "    PARAMS\n",
    "    --------------\n",
    "    class_label: int \n",
    "    label of the observations that must be replicated: class mel of label 1 here \n",
    "    \n",
    "    filename: list \n",
    "    list[0] = file_X, list[1] = file_Y\n",
    "    \n",
    "    X, Y: numpy array\n",
    "    data that must be augmented symmetrically\n",
    "    \n",
    "    Returns\n",
    "    -------------------\n",
    "    Augmented data X_aug, Y_aug\n",
    "    \"\"\"\n",
    "    n_mel = 698 # nb of instances of mel in training data\n",
    "    n_new = 6*n_mel\n",
    "    X_new = np.ones((n_new, 128, 128, 3))\n",
    "    Y_new = np.ones((n_new,))\n",
    "    \n",
    "    index_mel = np.where( Y == class_label ) # by default it is mel \n",
    "    X_mel = X[index_mel]\n",
    "    \n",
    "    iter_mel = 0\n",
    "    for i in range(X_mel.shape[0]):\n",
    "        if i%500 == 0 and i!=0:\n",
    "            print(\"==========================\")\n",
    "            print(\"Nice! {} images augmented already :) \".format(i))\n",
    "        \n",
    "        ## we replicate 6 times with different ways \n",
    "        image = X_mel[i]\n",
    "        # transform to uint8 \n",
    "        image = img_as_ubyte(image)\n",
    "        for k in range(6):\n",
    "            aug = strong_aug(p=1) # random transform of the image\n",
    "            # transform it \n",
    "            image_aug = augment_and_show(aug, image, verbose = False)\n",
    "            # convert image in float before adding in to data set \n",
    "            image_aug = img_as_float(image_aug)\n",
    "            # add image to data set: each row is a new image so axis = 0\n",
    "            image_aug = image_aug.reshape(1,128,128,3)\n",
    "\n",
    "            X_new[iter_mel] = image_aug\n",
    "            Y_new[iter_mel] = class_label\n",
    "            iter_mel += 1\n",
    "            \n",
    "            \n",
    "    if filename is not None:\n",
    "        filename_data, filename_label = filename[0], filename[1]\n",
    "        np.save(filename_data, X_new)\n",
    "        np.save(filename_label, Y_new)\n",
    "        \n",
    "    return X_new, Y_new"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "filename_aug_strong_X, filename_aug_strong_Y = \"./data_keras/X_train_aug_strong.npy\" , \"./data_keras/Y_train_aug_strong.npy\"\n",
    "filename_arr = [filename_aug_strong_X, filename_aug_strong_Y]\n",
    "\n",
    "X_train_aug_strong, Y_train_aug_strong = augment_data_mel(x_train, y_train, class_label = 1, filename = filename_arr )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "import gc\n",
    "def clean_variables():\n",
    "    gc.collect()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "### LOAD STRONGLY AUGMENTED DATA FOR MEL CLASS #### \n",
    "filename_mel_strong_X, filename_mel_strong_Y = \"./data_keras/X_train_mel_strong.npy\" , \"./data_keras/Y_train_mel_strong.npy\"\n",
    "# np.save(filename_mel_strong_X, X_train_aug_strong)\n",
    "# np.save(filename_mel_strong_Y, Y_train_aug_strong)\n",
    "\n",
    "X_train_aug_mel = np.load(filename_mel_strong_X)\n",
    "Y_train_aug_mel = np.load(filename_mel_strong_Y)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 75,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(4988, 4188)"
      ]
     },
     "execution_count": 75,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_train.shape[0], Y_train_aug_mel.shape[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "########## Customize Early Stopping ###########\n",
    "from keras.callbacks import Callback\n",
    "\n",
    "class EarlyStoppingByAccuracy(Callback):\n",
    "    \"\"\"\n",
    "    When checking the logs we see that sometimes we get a very low val loss/ high val accuracy at some epoch and after we go below such values.\n",
    "    Solution:  We will check when such accuracy is reached and stop at this very moment \n",
    "    \n",
    "    PARAMS\n",
    "    --------------------------\n",
    "    value: int v\n",
    "    Value that is seen to be passed for a given epoch => If we are beyond it we stop training!! \n",
    "    \n",
    "    \"\"\"\n",
    "    def __init__(self, monitor='accuracy', value=0.85, verbose=0):\n",
    "        super(Callback, self).__init__()\n",
    "        self.monitor = monitor\n",
    "        self.value = value\n",
    "        self.verbose = verbose\n",
    "\n",
    "    def on_epoch_end(self, epoch, logs={}):\n",
    "        current = logs.get(self.monitor)\n",
    "        if current is None:\n",
    "            warnings.warn(\"Early stopping requires %s available!\" % self.monitor, RuntimeWarning)\n",
    "\n",
    "        if current >= self.value:\n",
    "            if self.verbose > 0:\n",
    "                print(\"Epoch %05d: early stopping THR\" % epoch)\n",
    "            self.model.stop_training = True"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Conv2D, MaxPooling2D\n",
    "from keras.layers import Activation, Dropout, Flatten, Dense\n",
    "from collections import Counter \n",
    "from keras import backend as K\n",
    "from keras import optimizers\n",
    "from keras.callbacks import CSVLogger\n",
    "from keras.callbacks import ModelCheckpoint\n",
    "\n",
    "\n",
    "csv_logger = CSVLogger('log.csv', append=True, separator=';')\n",
    "\n",
    "\n",
    "\n",
    "def evaluate_classifier_conv_custom(x_train = x_train, y_train = y_train, x_val = x_val, y_val = y_val, img_rows = 128, img_cols = 128, batch_size = 32, epochs = 20, learning_rate = 0.01, dropout = 0.7, log = True, threshold_acc = 0.87):\n",
    "    \"\"\"\n",
    "    Create simple convolutional model to classify data\n",
    "    Fit data and return fitted model \n",
    "    \n",
    "    Custom callback\n",
    "    \"\"\"\n",
    "    \n",
    "    K.clear_session()\n",
    "    np.random.seed(100)\n",
    "\n",
    "    if K.image_data_format() == 'channels_first':\n",
    "        x_train = x_train.reshape(x_train.shape[0], 3, img_rows, img_cols)\n",
    "        x_val = x_val.reshape(x_val.shape[0], 3, img_rows, img_cols)\n",
    "        input_shape = (3, img_rows, img_cols)\n",
    "    else:\n",
    "        x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 3)\n",
    "        x_val = x_val.reshape(x_val.shape[0], img_rows, img_cols, 3)\n",
    "        input_shape = (img_rows, img_cols, 3)\n",
    "\n",
    "    model = Sequential()\n",
    "    model.add(Conv2D(32, (3, 3), input_shape=(128, 128,3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Conv2D(32, (3, 3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Conv2D(64, (3, 3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Flatten())  # this converts our 3D feature maps to 1D feature vectors\n",
    "    model.add(Dense(64))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Dropout(dropout))\n",
    "    model.add(Dense(1))\n",
    "    model.add(Activation('sigmoid'))\n",
    "\n",
    "\n",
    "    # initiate RMSprop optimizer\n",
    "    opt = optimizers.rmsprop(lr=learning_rate, decay = 1e-5)\n",
    "\n",
    "    model.compile(loss='binary_crossentropy',\n",
    "                  optimizer=opt,\n",
    "                  metrics=['accuracy'])\n",
    "    \n",
    "    kfold_weights_path = \"best_model{}.hdf5\".format(3)\n",
    "    callbacks = [\n",
    "                EarlyStoppingByAccuracy(monitor='val_acc', value=threshold_acc, verbose=1), # early stopping based on validation accuracy \n",
    "                ModelCheckpoint(kfold_weights_path, monitor='val_loss', save_best_only=True, verbose=0)\n",
    "                \n",
    "                ]\n",
    "\n",
    "    if log:\n",
    "        callbacks.append(csv_logger)\n",
    "        # Fit data \n",
    "        model.fit(x_train, y_train,\n",
    "                  batch_size=batch_size,\n",
    "                  epochs=epochs,\n",
    "                  verbose=1,\n",
    "                  validation_data=(x_val, y_val),\n",
    "                    callbacks = callbacks)\n",
    "        \n",
    "    else:\n",
    "        \n",
    "        # Fit data \n",
    "        model.fit(x_train, y_train,\n",
    "                  batch_size=batch_size,\n",
    "                  epochs=epochs,\n",
    "                  verbose=1,\n",
    "                  validation_data=(x_val, y_val),\n",
    "                  callbacks = callbacks)\n",
    "\n",
    "        \n",
    "    # Compute final performance on validation data\n",
    "    score = model.evaluate(x_val, y_val, verbose=0)\n",
    "    print('Final validation loss:', score[0])\n",
    "    print('Final validation accuracy:', score[1])\n",
    "    \n",
    "    return model  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "##### PARAMS FOR AUGMENTED DATASET #####\n",
    "model_params_aug = { \n",
    "                    \"x_train\": X_aug, \n",
    "                    \"y_train\": Y_aug, \n",
    "                    \"x_val\": x_val, \n",
    "                    \"y_val\": y_val,\n",
    "                    \"img_rows\": 128,\n",
    "                    \"img_cols\": 128, \n",
    "                    \"batch_size\": 16, \n",
    "                    \"epochs\": 40, \n",
    "                    \"learning_rate\": 0.0001, \n",
    "                    \"dropout\": 0.50,\n",
    "                    \"log\":False,\n",
    "                    \"threshold_acc\": 0.87\n",
    "                    }\n",
    "\n",
    "csv_logger = CSVLogger('log_conv_strong_augOK3.csv', append=True, separator=';')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model_strong_conv = evaluate_classifier_conv_custom(**model_params_aug)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "### LOAD BEST MODEL ###\n",
    "from keras.models import load_model\n",
    "\n",
    "filename_best_model = \"best_model{}.hdf5\".format(2)\n",
    "model_best2 = load_model(filename_best_model)\n",
    "model_best2.evaluate(x_val, y_val)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_graphs(model = model_best2, nb_epochs = 6) -> None:\n",
    "    \n",
    "    dict_history = model.history.history\n",
    "    train_costs, val_costs = dict_history['loss'], dict_history['val_loss']\n",
    "    acc_arr_train, acc_arr_val = dict_history['acc'], dict_history['val_acc']\n",
    "    ########  Cross-entropy  ###########\n",
    "    plt.figure(figsize = (15,10))\n",
    "    ax = plt.gca()\n",
    "\n",
    "    epochs = range(nb_epochs)\n",
    "    ax.plot(epochs, train_costs, color = \"green\")\n",
    "    ax.plot(epochs, val_costs, color = \"purple\")\n",
    "    ax.grid(True)\n",
    "    ax.set_ylabel(\"Cross-entropy\",fontsize=14, color='blue')\n",
    "   \n",
    "\n",
    "    plt.title(\"Training and Validation cross-entropy vs epochs \", fontsize = 15)\n",
    "    plt.show()\n",
    "    #########  Accuracy #############\n",
    "    plt.figure(figsize = (15,10))\n",
    "    ax = plt.gca()\n",
    "    ax2 = ax.twinx()\n",
    "\n",
    "    epochs = range(nb_epochs)\n",
    "    ax.plot(epochs, acc_arr_train, color = \"green\") \n",
    "    ax.plot(epochs, acc_arr_val, color = \"purple\")\n",
    "    ax.grid(True)\n",
    "    ax.set_ylabel(\"accuracy\",fontsize=14,color='green')\n",
    "\n",
    "\n",
    "    plt.title(\"Training and Validation accuracy vs epochs \", fontsize = 15)\n",
    "    plt.show()\n",
    "    return ;"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def acc_val_mel_nv(model) -> None:\n",
    "    index_nv_val = np.where( y_val == 0 )\n",
    "    index_mel_val = np.where( y_val == 1 )\n",
    "\n",
    "    score_nv = model.evaluate(x_val[index_nv_val], y_val[index_nv_val], verbose=0)\n",
    "    score_mel = model.evaluate(x_val[index_mel_val], y_val[index_mel_val], verbose=0)\n",
    "\n",
    "    print(\"============ Class NV: aug ========= \")\n",
    "    print(\"Accuracy on images on nv class fromn validation data: {}\".format(score_nv[1]))\n",
    "\n",
    "    print(\"============ Class MEL: aug ========= \")\n",
    "    print(\"Accuracy on images on mel class fromn validation data: {}\".format(score_mel[1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "============ Class NV: aug ========= \n",
      "Accuracy on images on nv class fromn validation data: 0.9187145557655955\n",
      "============ Class MEL: aug ========= \n",
      "Accuracy on images on mel class fromn validation data: 0.6526315758102819\n"
     ]
    }
   ],
   "source": [
    "acc_val_mel_nv(model_best2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from sklearn.metrics import recall_score\n",
    "from sklearn.metrics import f1_score\n",
    "\n",
    "\n",
    "def compute_recall(model):\n",
    "    Y_pred = model.predict_classes(x_val)\n",
    "    \n",
    "    recall_strong = recall_score(y_val, Y_pred.reshape(-1,))\n",
    "    print (\"Recall on class MEL is {}\".format(recall_strong))\n",
    "    return recall_strong\n",
    "    \n",
    "def compute_fscore(model):\n",
    "    Y_pred = model.predict_classes(x_val)\n",
    "    f1_val = f1_score(y_true = y_val, y_pred = Y_pred)\n",
    "    print (\"F1 score on class MEL with validation data is {}\".format(f1_val))\n",
    "    return f1_val"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Recall on class MEL is 0.6526315789473685\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.6526315789473685"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "compute_recall(model = model_best2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score on class MEL with validation data is 0.6200000000000001\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "0.6200000000000001"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "compute_fscore(model = model_best2 )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn.utils.multiclass import unique_labels\n",
    "\n",
    "def plot_confusion_matrix(model, x_val = x_val, y_val = y_val, classes = np.array([0,1]),\n",
    "                          normalize=False,\n",
    "                          title=None,\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",
    "    \n",
    "    \n",
    "    y_pred = model.predict_classes(x_val)\n",
    "    y_true = y_val \n",
    "    if not title:\n",
    "        if normalize:\n",
    "            title = 'Normalized confusion matrix'\n",
    "        else:\n",
    "            title = 'Confusion matrix, without normalization'\n",
    "\n",
    "    # Compute confusion matrix\n",
    "    cm = confusion_matrix(y_true, y_pred)\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",
    "    fig, ax = plt.subplots(figsize = (15,10))\n",
    "    #fig = plt.figure((15,10))\n",
    "    im = ax.imshow(cm, interpolation='nearest', cmap=cmap)\n",
    "    ax.figure.colorbar(im, ax=ax)\n",
    "    # We want to show all ticks...\n",
    "    ax.set(xticks=np.arange(cm.shape[1]),\n",
    "           yticks=np.arange(cm.shape[0]),\n",
    "           # ... and label them with the respective list entries\n",
    "           xticklabels=classes, yticklabels=classes,\n",
    "           title=title,\n",
    "           ylabel='True label',\n",
    "           xlabel='Predicted label')\n",
    "\n",
    "    # Rotate the tick labels and set their alignment.\n",
    "    plt.setp(ax.get_xticklabels(), rotation=45, ha=\"right\",\n",
    "             rotation_mode=\"anchor\")\n",
    "\n",
    "    # Loop over data dimensions and create text annotations.\n",
    "    fmt = '.2f' if normalize else 'd'\n",
    "    thresh = cm.max() / 2.\n",
    "    for i in range(cm.shape[0]):\n",
    "        for j in range(cm.shape[1]):\n",
    "            ax.text(j, i, format(cm[i, j], fmt),\n",
    "                    ha=\"center\", va=\"center\",\n",
    "                    color=\"white\" if cm[i, j] > thresh else \"black\")\n",
    "    fig.tight_layout()\n",
    "    return ax\n",
    "\n",
    "# Plot non-normalized confusion matrix\n",
    "# plot_confusion_matrix(model_best2)\n",
    "\n",
    "# Plot normalized confusion matrix\n",
    "# plot_confusion_matrix(model_best2, normalize = True)\n",
    "\n",
    "# plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "index_nv_val = np.where( y_val == 0 )\n",
    "index_mel_val = np.where( y_val == 1 )\n",
    "\n",
    "score_nv = model_strong_conv.evaluate(x_val[index_nv_val], y_val[index_nv_val], verbose=0)\n",
    "score_mel = model_strong_conv.evaluate(x_val[index_mel_val], y_val[index_mel_val], verbose=0)\n",
    "\n",
    "print(\"============ Class NV: aug ========= \")\n",
    "print(\"Accuracy on images on nv class fromn validation data: {}\".format(score_nv[1]))\n",
    "\n",
    "print(\"============ Class MEL: aug ========= \")\n",
    "print(\"Accuracy on images on mel class fromn validation data: {}\".format(score_mel[1]))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "@COMPILE: define a specific metric for unbalanced data \n",
    "\"\"\"\n",
    "\n",
    "def recall_np(y_true, y_pred):\n",
    "    \"\"\"Recall metric\n",
    "    Only computes a batch-wise average of recall. Computes the recall, a metric\n",
    "    for multi-label classification of how many relevant items are selected.\n",
    "    \"\"\"\n",
    "    epsilon = np.finfo(np.float32).eps\n",
    "    y_pred = y_pred.reshape(-1,)\n",
    "    true_positives = np.sum(y_true * y_pred)\n",
    "    print(true_positives)\n",
    "    possible_positives = np.sum(y_true)\n",
    "    recall = true_positives / (possible_positives + epsilon)\n",
    "    return recall"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Recall on class MEL is 0.5842105263157895\n"
     ]
    }
   ],
   "source": [
    "Y_pred = model_strong_conv.predict_classes(x_val)\n",
    "from sklearn.metrics import recall_score\n",
    "recall_strong = recall_score(y_val, Y_pred.reshape(-1,))\n",
    "print (\"Recall on class MEL is {}\".format(recall_strong))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "F1 score on class MEL with validation data is 0.5606060606060606\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import f1_score\n",
    "\n",
    "f1_val = f1_score(y_true = y_val, y_pred = Y_pred)\n",
    "print (\"F1 score on class MEL with validation data is {}\".format(f1_val))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## AUGMENT CLASS NV "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "from skimage import img_as_ubyte\n",
    "from skimage import img_as_float\n",
    "\n",
    "\"\"\" SOLUTION \"\"\"\n",
    "\n",
    "def augment_data_nv(X, Y, class_label = 0, factor = 2, filename = None ):\n",
    "    \"\"\"\n",
    "    PARAMS\n",
    "    --------------\n",
    "    class_label: int \n",
    "    label of the observations that must be replicated: class nv of label 0 here \n",
    "    \n",
    "    filename: list \n",
    "    list[0] = file_X, list[1] = file_Y\n",
    "    \n",
    "    X, Y: numpy array\n",
    "    data that must be augmented symmetrically\n",
    "    \n",
    "    Returns\n",
    "    -------------------\n",
    "    Augmented data X_aug, Y_aug\n",
    "    \"\"\"\n",
    "    n_nv = 4290 # nb of instances of mel in training data\n",
    "    n_new = factor*n_nv\n",
    "    X_new = np.ones((n_new, 128, 128, 3))\n",
    "    Y_new = np.ones((n_new,))\n",
    "    \n",
    "    index_nv = np.where( Y == class_label ) # by default it is mel \n",
    "    X_nv = X[index_nv]\n",
    "    \n",
    "    iter_nv = 0\n",
    "    for i in range(X_nv.shape[0]):\n",
    "        if i%500 == 0 and i!=0:\n",
    "            print(\"==========================\")\n",
    "            print(\"Nice! {} images augmented already :) \".format(i))\n",
    "        \n",
    "        ## we replicate 2 times with different ways \n",
    "        image = X_nv[i]\n",
    "        # transform to uint8 \n",
    "        image = img_as_ubyte(image)\n",
    "        for k in range(factor):\n",
    "            aug = strong_aug(p=1) # random transform of the image\n",
    "            # transform it \n",
    "            image_aug = augment_and_show(aug, image, verbose = False)\n",
    "            # convert image in float before adding in to data set \n",
    "            image_aug = img_as_float(image_aug)\n",
    "            # add image to data set: each row is a new image so axis = 0\n",
    "            image_aug = image_aug.reshape(1,128,128,3)\n",
    "\n",
    "            X_new[iter_nv] = image_aug\n",
    "            Y_new[iter_nv] = class_label\n",
    "            iter_nv += 1\n",
    "            \n",
    "            \n",
    "    if filename is not None:\n",
    "        filename_data, filename_label = filename[0], filename[1]\n",
    "        np.save(filename_data, X_new)\n",
    "        np.save(filename_label, Y_new)\n",
    "        \n",
    "    return X_new, Y_new"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "==========================\n",
      "Nice! 500 images augmented already :) \n",
      "==========================\n",
      "Nice! 1000 images augmented already :) \n",
      "==========================\n",
      "Nice! 1500 images augmented already :) \n",
      "==========================\n",
      "Nice! 2000 images augmented already :) \n",
      "==========================\n",
      "Nice! 2500 images augmented already :) \n",
      "==========================\n",
      "Nice! 3000 images augmented already :) \n",
      "==========================\n",
      "Nice! 3500 images augmented already :) \n",
      "==========================\n",
      "Nice! 4000 images augmented already :) \n"
     ]
    }
   ],
   "source": [
    "filename_aug_strong_X, filename_aug_strong_Y = \"./data_keras/X_nv_strong.npy\" , \"./data_keras/Y_nv_strong.npy\"\n",
    "filename_arr = [filename_aug_strong_X, filename_aug_strong_Y]\n",
    "\n",
    "X_nv_strong, Y_nv_strong = augment_data_nv( x_train, y_train, class_label = 0, factor = 2, filename = filename_arr )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## AUGMENTED DATA WITH STACKED AUTOENCODER"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "### LOAD DATA ####\n",
    "filename_stack_aug = './data_keras/stacked_vae_normdist_3499-8-8-32-30.npy'\n",
    "X_mel_stack_aug = np.load(filename_stack_aug)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "### ADD AUGMENTED DATA FOR CLASS MEL TO TRAINING DATA ###\n",
    "\n",
    "## GOAL: just stack the augmented mel data with training data and shuffle \n",
    "# PROBLEM: memory error \n",
    "# Solution: Pre allocate memory by stating shape of matrix \n",
    "\n",
    "### nb_obs = nb observations total so obs in X1 + obs in X2\n",
    "def concat_train_label(X1, X2 , Y1, Y2):\n",
    "    nb_obs = X1.shape[0] + X2.shape[0]\n",
    "    X_aug = np.empty((nb_obs, 128, 128, 3))\n",
    "    Y_aug = np.empty((nb_obs,))\n",
    "    nb_obs_or = X1.shape[0]\n",
    "    \n",
    "    for k in range(X1.shape[0]):\n",
    "        X_aug[k] = X1[k]\n",
    "        Y_aug[k] = Y1[k]\n",
    "    \n",
    "    iteration = 0\n",
    "    while iteration != Y2.shape[0]:\n",
    "        X_aug[iteration+nb_obs_or] = X2[iteration]\n",
    "        Y_aug[iteration+nb_obs_or] = Y2[iteration]\n",
    "        iteration += 1\n",
    "        \n",
    "    return X_aug, Y_aug\n",
    "\n",
    "X_aug, Y_aug = concat_train_label(x_train, X_mel_stack_aug, y_train, np.ones(X_mel_stack_aug.shape[0], ))\n",
    "\n",
    "# # ## Shuffle data ##\n",
    "X_aug, Y_aug = shuffle_data(X_aug, Y_aug)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "##### PARAMS FOR AUGMENTED DATASET #####\n",
    "model_params_aug = { \n",
    "                    \"x_train\": X_aug, \n",
    "                    \"y_train\": Y_aug, \n",
    "                    \"x_val\": x_val, \n",
    "                    \"y_val\": y_val,\n",
    "                    \"img_rows\": 128,\n",
    "                    \"img_cols\": 128, \n",
    "                    \"batch_size\": 16, \n",
    "                    \"epochs\": 40, \n",
    "                    \"learning_rate\": 0.0001, \n",
    "                    \"dropout\": 0.50,\n",
    "                    \"log\":False,\n",
    "                    \"threshold_acc\": 0.86\n",
    "                    }\n",
    "\n",
    "csv_logger = CSVLogger('log_conv_strong_augOK3.csv', append=True, separator=';')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model__conv = evaluate_classifier_conv_custom(**model_params_aug)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7ff369a13630>"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAx8AAALICAYAAAAe6E2GAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4wLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvqOYd8AAAIABJREFUeJzs3XuYXWV5N/7vnYQQJIBoAIWEgwIKWFs14qmv9VxUlNZ6AKvW1yqtV1GrtRbbSn2p9vfWttpW6UEv0NZW8dBaUWnRvvV8gmCRCooGBJOASJDzISHh+f0xO3EySWYGzForzP58vPbFXnut/axnz+Aw93zvZ61qrQUAAKBr84aeAAAAMB4UHwAAQC8UHwAAQC8UHwAAQC8UHwAAQC8UHwAAQC8UH8CcV1Vvrqp/Gj0/sKpurqr5O/gcl1fVU3bkmLM45yur6urR57nvTzHOzVX1gB05t6FU1UVV9YSh5wHAtik+gJ/a6Bfvq6tq90mvvbyqPjfgtLaptfaD1tri1trGoefy06iqXZK8PcnTRp/n2rs71uj9l+242e14VfW+qnrLTMe11o5qrX2uhykBcDcoPoAdZUGS1/y0g9QEP5tmtl+SRUkuGnoiO4OqWjD0HACYmf/AAzvKnyV5fVXde1s7q+qxVXVeVd0w+udjJ+37XFW9taq+nOTWJA8YvfaWqvrKqC3oE1V136r656q6cTTGwZPG+KuqWjXad35V/a/tzOPgqmpVtaCqHjMae9Pj9qq6fHTcvKo6uaouraprq+rDVXWfSeO8uKquGO37g+m+MFW1W1X9xej4G6rqS1W122jfs0etQtePPvMRk953eVW9vqouHL3vQ1W1qKoOT3LJ6LDrq+q/Jn+uKV/Xl4+eH1pVnx+Ns7aqPjTpuFZVh46e71VV/1hV14zm+4ebisGqeulo7n9eVddV1fer6unTfO7Lq+p3R/O/papOr6r9qurfq+qmqvrPqtp70vEfqaofjub4hao6avT6iUl+NckbNv27MGn836uqC5PcMvqebm5/q6qzq+ovJo3/oao6Y7rvFQDdUnwAO8qKJJ9L8vqpO0a/tH8qyV8nuW8m2oU+VVuuU3hxkhOT7JHkitFrx49ePyDJA5N8Ncl7k9wnybeT/NGk95+X5OdG+z6Q5CNVtWi6CbfWvjpqOVqcZO8kX0vywdHuVyf5pSS/kGT/JNclOW30eY5M8rejue0/+kxLpznVnyd5RJLHjub3hiR3joqIDyb57ST7JDk7ySeqauGk9z4/yTFJDkny0CQvba19N8lRo/33bq09abrPOfLHST49+pxLk7xzO8e9M8leSR4w+uwvSfK/J+1/VCYKnyVJ3pbk9Kqqac77K0memuTwJM9K8u9Jfn/0/nmZ+Dpv8u9JDkuyb5JvJPnnJGmtvXv0/G2j79ezJr3nhCTPzMTXYcOUc78syYur6klV9atJHpkdkM4BcPcpPoAd6ZQkr6qqfaa8/swk32utvb+1tqG19sEk38nEL6ObvK+1dtFo/x2j197bWru0tXZDJn4xvbS19p+jXzI/kuRhm97cWvun1tq1o/f/RZJdkzzoLsz9r5PckmRTivEbSf6gtba6tbYuyZuTPHeULDw3ySdba18Y7XtTkju3NegoNXhZkte01ta01ja21r4yet8LknyqtfaZ0Wf+8yS7ZaJI2Tyv1tqVrbUfJ/lEJgqsu+OOJAcl2b+1dntr7UvbmOv80Zze2Fq7qbV2eZK/yESRtckVrbX3jNbM/EOS+2eiBWx73tlau7q1tibJF5N8vbX236PP/7Fs+T08Y3TeTV/vn62qvWb4XH/dWlvVWrtt6o7W2g+T/OZonn+V5CWttZtmGA+ADik+gB2mtfatJJ9McvKUXfvnJ2nGJldkItHYZNU2hrx60vPbtrG9eNNGVf1OVX171LJzfSb+er9kNvOuqt9I8oQkL2ytbSoiDkrysVE71PWZSFo2ZuIX7f0nz7e1dkuS7S34XpKJtRmXbmPfFl+X0blXZcuvyw8nPb81kz7zXfSGJJXk3FGb18u2M9eF2fJ7NfX7tHk+rbVbR0+nm9OsvodVNb+q/u+oze3GJJdPmtN0tvXvzWSfTDI/ySXbKrgA6JfiA9jR/ijJK7LlL6xXZuKX+ckOTLJm0na7uyccre/4vUy0KO3dWrt3khsy8cv2bN77x0mOGyUsm6xK8vTW2r0nPRaN/oJ/VZJlk8a4VyZar7ZlbZLbM9E2NtUWX5dR+9KybPl1ma1bRv+816TX7rfpSWvth621V7TW9s9EqvM3m9Z5TJnrpoRkk6nfp668MMlxSZ6SicLx4NHrm76H2/v3Y6Z/b96aicLx/lV1wk85RwB+SooPYIdqra1M8qFs2ct/dpLDq+qFo0XBL0hyZCb+Kr0j7JFkQ5JrkiyoqlOS7DnTm6pq2WiuLxmto5js75K8taoOGh27T1UdN9r30STHVtXPj9ZnnJrt/DwdpRlnJHl7Ve0/+gv/Y6pq1yQfTvLMqnpyTVw693eSrEvylbv06SfOc00mioQXjc7xskwqeKrqeVW1aV3KdZn4pX3jlDE2jub01qraY/TZX5fkn+7qfO6GPTLx2a/NRAH1J1P2X52JdSizVlWPz8R6lZeMHu+sqgOmfxcAXVJ8AF04Ncnme36M7kFxbCZ+ub42Ey1Ax7bW1u6g852TiTUh381Em9DtmbkdJ0menIl04KP1kytebbp07V8lOSvJp6vqpkwsRn/U6PNclOS3MrGw/apM/DK/eprzvD7J/2RiUfyPk/xpknmttUuSvCgTi7zXZmINzLNaa+tn+bmnekWS383E1/iobFnEPDLJ16vq5tHnek1r7fvbGONVmUhRLkvypdFn7OMKUf+Yie/dmiQXZ+LrPdnpSY4ctcH920yDVdWeozFPGq21+dJojPfOsEAegA5Va3e70wEAAGDWJB8AAEAvFB8AAEAvFB8AAEAvFB8AAEAvFgw9gclqwW6tFu4x9DQAdgoPO+LAoacAsFO44orLs3bt2jl1pbr5ex7U2obbBp1Du+2ac1prx/R5zp2r+Fi4R3Z90POHngbATuHLX3/X0FMA2Ck87lHLh57CDtc23Db47723X3Dakr7Pqe0KAADoxU6VfAAAwHiopMYvBxi/TwwAAAxC8QEAAPRC2xUAAPStktScuoDXrEg+AACAXig+AACAXmi7AgCAIbjaFQAAQDckHwAAMAQLzgEAALqh+AAAAHqh7QoAAHpXFpwDAAB0RfIBAABDsOAcAAAgqapjquqSqlpZVSdvY/+BVfXZqvrvqrqwqp4x05iKDwAAYAtVNT/JaUmenuTIJCdU1ZFTDvvDJB9urT0syfFJ/mamcbVdAQBA3yo7+4Lzo5OsbK1dliRVdWaS45JcPOmYlmTP0fO9klw506CKDwAAYKoDkqyatL06yaOmHPPmJJ+uqlcl2T3JU2YadKcutwAAYG6qiQXnQz6SJVW1YtLjxC0nuJU2ZfuEJO9rrS1N8owk76+aPs6RfAAAwHha21pbvp19q5Msm7S9NFu3Vf16kmOSpLX21apalGRJkh9t74SSDwAAYKrzkhxWVYdU1cJMLCg/a8oxP0jy5CSpqiOSLEpyzXSDSj4AAGAIO/GC89bahqo6Kck5SeYnOaO1dlFVnZpkRWvtrCS/k+Q9VfXaTLRkvbS1NrU1awuKDwAAYCuttbOTnD3ltVMmPb84yePuypiKDwAAGII7nAMAAHRD8QEAAPRC2xUAAPSuduoF510Zv08MAAAMQvIBAAB9q1hwDgAA0BXFBwAA0AttVwAAMAQLzgEAALoh+QAAgN651C4AAEBnFB8AAEAvtF0BAMAQ5rnPBwAAQCcUHwAAQC+0XQEAQN8qrnYFAADQFckHAAAMoSw4BwAA6ITiAwAA6IW2KwAA6F1ZcA4AANAVyQcAAAzBgnMAAIBuKD4AAIBeaLsCAIAhWHAOAADQDckHAAD0rcqCcwAAgK4oPgAAgF5ouwIAgCFYcA4AANANyQcAAAzBgnMAAIBuKD4AAIBeaLsCAIDelQXnAAAAXZF8AADAECw4BwAA6IbiAwAA6IW2KwAA6FvFgnMAAICuSD4AAKB3LrULAADQGcUHAADQC21XAAAwBPf5AAAA6IbiAwAA6IW2KwAAGIKrXQEAAHRD8gEAAEOw4BwAAKAbig8AAKAX2q4AAKBvVRacAwAAdEXyAQAAQ7DgHAAAoBuKDwAAoBfargAAYACl7QoAAKAbkg8AAOhZRfIBAADQGcUHAADQC21XAADQtxo9xozkAwAA6IXkAwAAelcWnAMAAHRF8QEAAPRC2xUAAAxA2xUAAEBHJB8AADAAyQcAAEBHFB8AAEAvtF0BAMAAtF0BAAB0RPIBAAB9q9FjzEg+AACAXig+AACAXmi7AgCAnlXKgnMAAICuKD4AAIBeaLsCAIABaLsCAADoiOQDAAAGIPkAAADoiOIDAADohbYrAAAYgLYrAACAjkg+AACgbzV6jBnJBwAA0AvFBwAA0AttVwAAMAALzgEAAJJU1TFVdUlVrayqk7ex/x1VdcHo8d2qun6mMSUfAADQs0rt1MlHVc1PclqSpyZZneS8qjqrtXbxpmNaa6+ddPyrkjxspnElHwAAwFRHJ1nZWrustbY+yZlJjpvm+BOSfHCmQRUfAAAwnpZU1YpJjxMn7TsgyapJ26tHr22lqg5KckiS/5rphNquAABgADtB29Xa1try7ezb1uTado49PslHW2sbZzqh5AMAAJhqdZJlk7aXJrlyO8cen1m0XCWSDwAAGMbgwce0zktyWFUdkmRNJgqMF049qKoelGTvJF+dzaCSDwAAYAuttQ1JTkpyTpJvJ/lwa+2iqjq1qp496dATkpzZWtteS9YWJB8AAMBWWmtnJzl7ymunTNl+810ZU/EBAAB9q51iwXnvtF0BAAC9kHwAAMAAJB8AAAAdUXwAAAC90HYFAAAD0HYFAADQEckHAAD0rFKSDwAAgK4oPgAAgF5ouwIAgCGMX9eV5AMAAOiH4gMAAOiFtisAAOhbuc8HAABAZyQfAAAwAMkHAABARxQfAABAL7RdAQDAALRdAQAAdETyAQAAQxi/4EPyAQAA9EPxAQAA9ELbFQAADMCCcwAAgI5IPgAAoGdVJfkAAADoiuIDAADohbYrAAAYgLYrAACAjkg+AABgAJIPmOOe+tgj8s2PvSnf+vgf5fX/+6lb7V92v73zH+9+db76wd/LuR96Y37x549MkjzpUQ/Ol//5DTnvw7+fL//zG/ILjzy876kD7HCfPuc/8tCjHpSjHnxo/uxt/3er/V/64hfymEc+PIsXLci//stHt9j3+ye/IQ//2aPycz9zRF73269Oa62vaQP3YJ0WH1V1TFVdUlUrq+rkLs8FM5k3r/KXJz8/x530N3nYr7wlzzvmEXnwA+63xTG/9/Jj8i+f+UYec8Kf5iVvfG/+6o0vSJJce/3Nee5v/30e+fw/yStOeX/OeMtLhvgIADvMxo0b89uv/q18/BP/nv++8OJ85MwP5tsXX7zFMcuWHZh3n/6+vOD4F27x+le/8pV89StfznnfuDDnX/CtnL/ivHzxC5/vc/rAPVRnbVdVNT/JaUmemmR1kvOq6qzW2sXTvxO68ciHHJxLV63N5WuuTZJ85Jxv5NgnPDTfueyHm49prWXP3RclSfZavFuuuuaGJMk3L1m9+ZiLL70quy7cJQt3WZD1d2zo8RMA7DjnnXtuHvjAQ3PIAx6QJHneC47PJz/x8Rxx5JGbjzno4IOTJPPmbfm3yqrKuttvz/r169Nay4Y77si+++7X29xhzhi/rqtOk4+jk6xsrV3WWluf5Mwkx3V4PpjW/vvuldVXX7d5e83V1+WAffba4pi3/v3ZOf4ZR2flf/xxPvbOV+Z1f/qRrcb55af8XL55ySqFB3CPduWVa7J06bLN2wccsDRr1qyZ1Xsf/ZjH5PFPeGIOWXb/HLLs/nnK034xDz7iiK6mCswhXRYfByRZNWl79ei1LVTViVW1oqpWtA23dTgdxl1t488LUzuUn3/M8vzTJ76WQ495U375VX+b09/yki0Wgx3xgPvlLa8+Lie95cyOZwvQrW2t0Zjt4tdLV67MJd/5dlZevjqXXrEmn/vsf+VLX/zCjp4izHmb7nI+1GMIXRYf2/pEW/2ka629u7W2vLW2vBbs1uF0GHdrfnR9lu639+btA/bbO1eO2qo2+bVfekz+5dPfSJJ8/cLvZ9HCXbLk3rtPHL/vvfOht5+Yl7/p/fn+6rX9TRygAwccsDSrV//kb4Rr1qzO/vvvP6v3fvzjH8vRj3p0Fi9enMWLF+cXj3l6vv71r3U1VWAO6bL4WJ1k2aTtpUmu7PB8MK0VF12RQw/cJwftf9/ssmB+nveLD8+nPnfhFses+uGP84SjH5QkedAh+2XRrrvkmutuzl6Ld8u/vvM3c8o7z8pXv3nZENMH2KGWP/KRWbnye7n8+9/P+vXr85EPnZlnHvvsWb132bID88UvfD4bNmzIHXfckS9+4fN58IO1XQEz6/I+H+clOayqDkmyJsnxSV44/VugOxs33pnX/umH84m/+a3Mn1f5h49/Ld++7Id50yufmW9c/IN86vP/k5Pf/rH8zZtOyKte9MS0lrzilPcnSX7z+Mfngcv2ycmvOCYnv+KYJMmzXvmuXHPdzUN+JIC7bcGCBXnHX70rz3rmL2bjxo35tZe+LEcedVROffMpefgjlufYZz07K847Ly943i/n+uuuy9mf+kTecuof5RvfvCjP+ZXn5vOf/a8sf9jPpKry1Kcdk2ce+6yhPxLcs9R43uejurwud1U9I8lfJpmf5IzW2lunO37evfZtuz7o+Z3NB+Ce5Lrz3jX0FAB2Co971PKcf/6KOfWb+q73O6wt/dW/HnQOl739Gee31pb3ec5O73DeWjs7ydldngMAAO5pKskYBh/ucA4AAPRD8QEAAPSi07YrAABgW4a718aQJB8AAEAvFB8AAEAvtF0BAMAAxrDrSvIBAAD0Q/IBAAADsOAcAACgI4oPAACgF9quAACgb2XBOQAAQGckHwAA0LNKMm/e+EUfkg8AAKAXig8AAKAX2q4AAGAAFpwDAAB0RPIBAAADcIdzAACAjig+AACAXmi7AgCAvrnDOQAAQHckHwAA0LOKBecAAACdUXwAAAC90HYFAAC9K21XAAAAXZF8AADAAMYw+JB8AAAA/VB8AAAAvdB2BQAAA7DgHAAAoCOSDwAA6FtZcA4AANAZxQcAANALbVcAANCzigXnAAAAnVF8AAAAvdB2BQAAAxjDrivJBwAA0A/JBwAADMCCcwAAgI4oPgAAgF5ouwIAgAGMYdeV5AMAAOiH5AMAAPpWFpwDAAB0RvEBAAD0QtsVAAD0rGLBOQAAQGckHwAA0Luy4BwAACBJquqYqrqkqlZW1cnbOeb5VXVxVV1UVR+YaUzJBwAAsIWqmp/ktCRPTbI6yXlVdVZr7eJJxxyW5I1JHtdau66q9p1pXMUHAAAMYCfvujo6ycrW2mVJUlVnJjkuycWTjnlFktNaa9clSWvtRzMNqu0KAADG05KqWjHpceKkfQckWTVpe/XotckOT3J4VX25qr5WVcfMdELJBwAADGAnWHC+trW2fDv7tjW5NmV7QZLDkjwhydIkX6yqh7TWrt/eCSUfAADAVKuTLJu0vTTJlds45uOttTtaa99PckkmipHtUnwAAABTnZfksKo6pKoWJjk+yVlTjvm3JE9Mkqpakok2rMumG1TbFQAA9K127gXnrbUNVXVSknOSzE9yRmvtoqo6NcmK1tpZo31Pq6qLk2xM8ruttWunG1fxAQAAbKW1dnaSs6e8dsqk5y3J60aPWVF8AABAzyo7xYLz3lnzAQAA9ELxAQAA9ELbFQAADEDbFQAAQEckHwAAMIAxDD4kHwAAQD8UHwAAQC+0XQEAwAAsOAcAAOiI4gMAAOiFtisAAOhbudoVAABAZyQfAADQs0pZcA4AANAVxQcAANALbVcAADCAMey6knwAAAD9kHwAAMAA5o1h9CH5AAAAeqH4AAAAeqHtCgAABjCGXVeSDwAAoB+SDwAA6FlV3OEcAACgK4oPAACgF9quAABgAPPGr+tK8gEAAPRD8gEAAAOw4BwAAKAjig8AAKAX2q4AAGAAY9h1JfkAAAD6IfkAAICeVZLK+EUfkg8AAKAXig8AAKAX2q4AAGAA7nAOAADQEckHAAD0rcodzgEAALqi+AAAAHqh7QoAAAYwhl1Xkg8AAKAfig8AAKAX2q4AAKBnlWTeGPZdST4AAIBeSD4AAGAAYxh8SD4AAIB+KD4AAIBeaLsCAIAB1Bj2XUk+AACAXkg+AACgZ1UWnAMAAHRG8QEAAPRC2xUAAAzAHc4BAAA6IvkAAIABjF/uIfkAAAB6ovgAAAB6oe0KAAAG4A7nAAAAHZF8AABAzyrJvPELPiQfAABAPxQfAABAL7RdAQBA36osOAcAAOiK5AMAAAYwhsGH5AMAAOiH4gMAAOiFtisAABiABecAAAAdkXwAAEDP3OEcAACgQ9tNPqpqz+ne2Fq7ccdPBwAAmKuma7u6KEnLRCq0yabtluTADucFAABz2jguON9u8dFaW9bnRAAAgLltVms+qur4qvr90fOlVfWIbqcFAADMNTMWH1X1riRPTPLi0Uu3Jvm7LicFAABzXQ38GMJsLrX72Nbaw6vqv5OktfbjqlrY8bwAAIA5ZjbFxx1VNS8Ti8xTVfdNcmenswIAgDmsKpk3hgvOZ7Pm47Qk/5Jkn6r6P0m+lORPO50VAAAw58yYfLTW/rGqzk/ylNFLz2utfavbaQEAAHPNbNqukmR+kjsy0XrlrugAAPBTGsOuq1ld7eoPknwwyf5Jlib5QFW9seuJAQAAc8tsko8XJXlEa+3WJKmqtyY5P8n/1+XEAABgLhvHO5zPpoXqimxZpCxIclk30wEAAOaq7SYfVfWOTKzxuDXJRVV1zmj7aZm44hUAAMCsTdd2temKVhcl+dSk17/W3XQAAGA8jGHX1faLj9ba6X1OBAAAmNtmXHBeVQ9M8tYkRyZZtOn11trhHc4LAADmrEq5w/l2vC/Je5NUkqcn+XCSMzucEwAAMAfNpvi4V2vtnCRprV3aWvvDJE/sdloAAMBcM5v7fKyriYsQX1pVv5lkTZJ9u50WAADMYTWeC85nk3y8NsniJK9O8rgkr0jysi4nBQAADKuqjqmqS6pqZVWdvI39L62qa6rqgtHj5TONOWPy0Vr7+ujpTUlefNenDQAATLUz3+G8quYnOS3JU5OsTnJeVZ3VWrt4yqEfaq2dNNtxp7vJ4McycVPBbWqtPWe2J5mt/Zftl9e847U7eliAe6Tnnn7u0FMA2ClcuvaWoacwjo5OsrK1dlmSVNWZSY5LMrX4uEumSz7e9dMMDAAA7NSWVNWKSdvvbq29e/T8gCSrJu1bneRR2xjjV6rq8Um+m+S1rbVV2zhms+luMvj/ZjdnAADgrprN4uuOrW2tLd/Ovm31hE3tivpEkg+21taNLkz1D0meNN0Jd4LPDAAA7GRWJ1k2aXtpkisnH9Bau7a1tm60+Z4kj5hp0NlcahcAANiBKjv3gvMk5yU5rKoOycStNo5P8sLJB1TV/VtrV402n53k2zMNOuvio6p2nVTZAAAAc1RrbUNVnZTknCTzk5zRWruoqk5NsqK1dlaSV1fVs5NsSPLjJC+dadwZi4+qOjrJ6Un2SnJgVf1skpe31l51tz8NAACwU2utnZ3k7CmvnTLp+RuTvPGujDmb5OOvkxyb5N9GJ/lmVT3xrpwEAADY0ryduuuqG7NZcD6vtXbFlNc2djEZAABg7ppN8rFq1HrVRnc6fFUmruMLAADcTZKPbXtlktclOTDJ1UkePXoNAABg1mZMPlprP8rEpbUAAADuttlc7eo92fpuhmmtndjJjAAAYI6r2unv89GJ2az5+M9Jzxcl+eUkq7qZDgAAMFfNpu3qQ5O3q+r9ST7T2YwAAIA5adZ3OJ/kkCQH7eiJAADAOBnHq13NZs3HdfnJmo95mbh1+sldTgoAAJh7pi0+amIVzM8mWTN66c7W2laLzwEAgLtmDNebT3+fj1Gh8bHW2sbRQ+EBAADcLbO5yeC5VfXwzmcCAADMadttu6qqBa21DUl+PskrqurSJLckqUyEIgoSAAC4GyrJvDHsu5puzce5SR6e5Jd6mgsAADCHTVd8VJK01i7taS4AADA2ZrP+Ya6ZrvjYp6pet72drbW3dzAfAABgjpqu+JifZHFGCQgAAMBPY7ri46rW2qm9zQQAAMbIGK43n7bVbAy/HAAAQFemSz6e3NssAABgjFTVWF5qd7vJR2vtx31OBAAAmNvG8QpfAADAAKZruwIAADoyhl1Xkg8AAKAfkg8AABjAPMkHAABANxQfAABAL7RdAQBAzypxnw8AAICuSD4AAGAAYxh8SD4AAIB+KD4AAIBeaLsCAIC+lft8AAAAdEbyAQAAA6iMX/Qh+QAAAHqh+AAAAHqh7QoAAHo2cYfzoWfRP8kHAADQC8kHAAAMQPIBAADQEcUHAADQC21XAAAwgKrx67uSfAAAAL1QfAAAAL3QdgUAAD1znw8AAIAOST4AAKBvlYzhenPJBwAA0A/FBwAA0AttVwAAMIB5Y9h3JfkAAAB6IfkAAICeudQuAABAhxQfAABAL7RdAQDAAMZwvbnkAwAA6IfkAwAAeleZl/GLPiQfAABALxQfAABAL7RdAQBAzyoWnAMAAHRG8gEAAH0rdzgHAADojOIDAADohbYrAAAYwLwxXHEu+QAAAHoh+QAAgJ651C4AAECHFB8AAEAvtF0BAMAALDgHAADoiOQDAAAGMIbBh+QDAADoh+IDAADohbYrAADoWWU8U4Bx/MwAAMAAFB8AAEAvtF0BAEDfKqkxvNyV5AMAAOiF5AMAAAYwfrmH5AMAAOiJ4gMAAOiFtisAAOhZJZlnwTkAAEA3JB8AADCA8cs9JB8AAEBPFB8AAEAvtF0BAMAAxnC9ueQDAADoh+QDAAB6V6kxjD4kHwAAwFaq6piquqSqVlbVydMc99yqalW1fKYxFR8AAMCYqFtbAAATX0lEQVQWqmp+ktOSPD3JkUlOqKojt3HcHkleneTrsxlX8QEAAD2rTPwiPuRjBkcnWdlau6y1tj7JmUmO28Zxf5zkbUlun83nVnwAAMB4WlJVKyY9Tpy074AkqyZtrx69tllVPSzJstbaJ2d7QgvOAQBgADvBgvO1rbXtrdPY1uTa5p1V85K8I8lL78oJJR8AAMBUq5Msm7S9NMmVk7b3SPKQJJ+rqsuTPDrJWTMtOld8AAAAU52X5LCqOqSqFiY5PslZm3a21m5orS1prR3cWjs4ydeSPLu1tmK6QbVdAQDAAAZvuppGa21DVZ2U5Jwk85Oc0Vq7qKpOTbKitXbW9CNsm+IDAADYSmvt7CRnT3ntlO0c+4TZjKn4AACAvtVOseC8d9Z8AAAAvVB8AAAAvdB2BQAAPdt0h/NxM46fGQAAGIDkAwAABmDBOQAAQEcUHwAAQC+0XQEAwADGr+lK8gEAAPRE8QEAAPRC2xUAAAxgDC92JfkAAAD6IfkAAICeTdzhfPyiD8kHAADQC8UHAADQC21XAAAwAAvOAQAAOiL5AACA3lXKgnMAAIBuKD4AAIBeaLsCAIABWHAOAADQEckHAAD0zB3OAQAAOqT4AAAAeqHtCgAA+lYWnAMAAHRG8gEAAAOQfAAAAHRE8QEAAPRC2xUAAAyg3OcDAACgG5IPAADoWSWZN37Bh+QDAADoh+IDAADohbYrAAAYgAXnAAAAHZF8AADAANzhHAAAoCOKDwAAoBfargAAYAAWnAMAAHRE8QEAAPRC2xUAAPSskswbv64ryQcAANAPyQcAAPSuLDgHAADoiuIDAADohbYrAADoWyU1fl1Xkg8AAKAfkg8AABjAGAYfig/Gy8H32S1PPnRJqioXXnVjzv3B9ds87vB9ds9xR90v/3j+6lx907rsuWhBXvbIZbnutjuSJFfeeHs+8921fU4dYId6+LK9cuJjD8y8qnz6O9fkoxdctcX+Jx++JC979LJce8v6JMknL/pRPv2da5Ik+yxemFc9/pDss3hhWpI3n31JfnTz+r4/AnAP1FnxUVVnJDk2yY9aaw/p6jwwW5XkqYftkw9/88rctG5DXvyIpbl07S259tY7tjhul/mVhx+wV6688fYtXr/+9g35hxWre5wxQDfmVfLKxx2UP/zUJbn2lvV5x3OOytcvvy6rrt/y594XL/1x/u7LV2z1/tc98QH50DeuzAVrbsyiBfPS+po4cI/X5ZqP9yU5psPx4S65/5675rrb7sgNt2/InS35zo9uzqFLdt/quJ8/5D45d9X12XCn/5wCc9Ph+y7OVTeuy9U3rcuGO1u+sPLaPPrgvWf13mX3XpR5VblgzY1Jkts33Jl1G+7scrowJ03c4bwGfQyhs+SjtfaFqjq4q/Hhrlq864LctG7D5u2b1m3I/fdctMUx+y5emD13XZDLrr01j1x27y327bVoQV7yiKVZv/HOfPH7P86aG7b8CyHAPcV977VLrrl53ebttbesz4P2XbzVcY89ZO8cdf89cuUNt+c9X/lB1t6yPgfce1FuWb8xv/+0Q7PfHrvmgjU35h++vir+XgPMxuBXu6qqE6tqRVWtuOWGHw89HcbNlP9YPvHQJfnspdduddgt6zbk7796Rf7x/NX57Mq1OfaI/bJw/jguEwPmhG38+JpaO5x7xfV52Qe+mVd99Fu5YPUNee0TH5AkmV+Vo+63OKd/dVVe+68X5X577JonH76k+znDHFQDP4YwePHRWnt3a215a2357nvdZ+jpMIfdvG5D9tj1J2HfHrsuyM3rf5KELJw/L0t2X5jjf27/nPjoA7P/nrvmOQ+5X/bbY9dsbBOtBUly9c3rc/3td2Tvey3s/TMA7AjX3nJH9lm86+btJbsvzI9v2XLB+E3rNmxuPz3nO9fk0CX3SjKRklx27a25+qZ1ubMlX7v8ujxwn61bWAG2ZfDiA/py1U3rsvduu2SvRQsyr5IH77s4K9fesnn/+o135rQvX553f+0HeffXfpArb1yXf/3WD3P1Teuy2y7zNv+FYK9FC7L3brvkhtvu2PaJAHZy3/3Rzdl/r12z3x4Ls2Be5fGH3jdfv2LLq//tfa9dNj9/1EF7b16M/r1rbsniXRdkz0UTf8x56AF7ZtV1t/U3eeAezaV2GRutJf/5vbV57kPvn3lV+Z+rbsy1t96Rxx28d35407pceu2t233vsr12y+MOuU/ubC2tJZ/57jWbkxCAe5o7W/J3X7oipz7jwZlXyWcuuSY/uO62/OryA/K9a27JuVdcn2c/ZL8cfdC9c2dLbrp9Q/7yc5dtfu/pX/1B3nrsg1NJVq69Jed8+5phPxDcU41hB3e11s0Ksar6YJInJFmS5Ookf9RaO3269yx90M+01/z9v3UyH4B7ms9fsvX6I4Bx9KU/+bVcf8W359Sv6kf8zMPae//ts4PO4TGH7n1+a215n+fs8mpXJ3Q1NgAA3NPVGEYf1nwAAAC9UHwAAAC9sOAcAAAGMNBNxgcl+QAAAHoh+QAAgAGMYfAh+QAAAPqh+AAAAHqh7QoAAIYwhn1Xkg8AAKAXkg8AAOhZxR3OAQAAOqP4AAAAeqHtCgAA+lbucA4AANAZxQcAANALbVcAADCAMey6knwAAAD9kHwAAMAQxjD6kHwAAAC9UHwAAAC90HYFAAC9q9QY9l1JPgAAgF5IPgAAYADucA4AANARxQcAANALbVcAANCzylje5kPyAQAA9EPyAQAAQxjD6EPyAQAA9ELxAQAA9ELxAQAAA6iB/zfj/KqOqapLqmplVZ28jf2/WVX/U1UXVNWXqurImcZUfAAAAFuoqvlJTkvy9CRHJjlhG8XFB1prP9Na+7kkb0vy9pnGteAcAAAGsJPf4fzoJCtba5clSVWdmeS4JBdvOqC1duOk43dP0mYaVPEBAABMdUCSVZO2Vyd51NSDquq3krwuycIkT5ppUG1XAAAwnpZU1YpJjxMn7dtWLrNVstFaO6219sAkv5fkD2c6oeQDAAAGsBN0Xa1trS3fzr7VSZZN2l6a5Mppxjozyd/OdELJBwAAMNV5SQ6rqkOqamGS45OcNfmAqjps0uYzk3xvpkElHwAA0LfKThF9bE9rbUNVnZTknCTzk5zRWruoqk5NsqK1dlaSk6rqKUnuSHJdkl+baVzFBwAAsJXW2tlJzp7y2imTnr/mro6p7QoAAOiF5AMAAAYwm7uMzzWSDwAAoBeSDwAA6Fllp7/DeSckHwAAQC8UHwAAQC+0XQEAwADGsOtK8gEAAPRD8QEAAPRC2xUAAAxhDPuuJB8AAEAvJB8AADAAdzgHAADoiOIDAADohbYrAAAYQI1f15XkAwAA6IfkAwAABjCGwYfkAwAA6IfiAwAA6IW2KwAAGMIY9l1JPgAAgF5IPgAAoGcVdzgHAADojOIDAADohbYrAADoW7nDOQAAQGckHwAAMIAxDD4kHwAAQD8UHwAAQC+0XQEAwBDGsO9K8gEAAPRC8gEAAL0rdzgHAADoiuIDAADohbYrAAAYgDucAwAAdETyAQAAPauM5ZV2JR8AAEA/FB8AAEAvtF0BAMAQxrDvSvIBAAD0QvEBAAD0QtsVAAAMoMaw70ryAQAA9ELyAQAAA3CHcwAAgI4oPgAAgF5ouwIAgAGMYdeV5AMAAOiH5AMAAPpWFpwDAAB0RvEBAAD0QtsVAAAMYvz6riQfAABALyQfAADQs4oF5wAAAJ1RfAAAAL3QdgUAAAMYw64ryQcAANAPyQcAAAzAgnMAAICOKD4AAIBeaLsCAIAB1BguOZd8AAAAvZB8AADAEMYv+JB8AAAA/VB8AAAAvdB2BQAAAxjDrivJBwAA0A/JBwAA9KzKHc4BAAA6o/gAAAB6oe0KAAAG4A7nAAAAHVF8AAAAvdB2BQAAQxi/rivJBwAA0A/JBwAADGAMgw/JBwAA0A/FBwAA0AttVwAAMIAaw74ryQcAANALyQcAAPSu3OEcAACgK4oPAACgF9quAACgZxULzgEAADqj+AAAAHqh+AAAAHqh+AAAAHphwTkAAAzAgnMAAICOSD4AAGAA7nAOAACQpKqOqapLqmplVZ28jf2vq6qLq+rCqvp/VXXQTGMqPgAAgC1U1fwkpyV5epIjk5xQVUdOOey/kyxvrT00yUeTvG2mcRUfAADQt5pYcD7kYwZHJ1nZWrustbY+yZlJjpt8QGvts621W0ebX0uydKZBFR8AADCellTVikmPEyftOyDJqknbq0evbc+vJ/n3mU5owTkAAPSsRo+BrW2tLd/Ovm1Nr23zwKoXJVme5BdmOqHiAwAAmGp1kmWTtpcmuXLqQVX1lCR/kOQXWmvrZhpU2xUAADDVeUkOq6pDqmphkuOTnDX5gKp6WJK/T/Ls1tqPZjOo5AMAAIawE/RdbU9rbUNVnZTknCTzk5zRWruoqk5NsqK1dlaSP0uyOMlHamIF+w9aa8+eblzFBwAAsJXW2tlJzp7y2imTnj/lro6p+AAAgAG4wzkAAEBHFB8AAEAvtF0BAMAAZnGX8TlH8gEAAPRC8QEAAPRC2xUAAAxgDLuuJB8AAEA/JB8AADCEMYw+JB8AAEAvFB8AAEAvtF0BAMAAagz7riQfAABALyQfAADQs4o7nAMAAHSmWmtDz2GzqromyRVDz4OxtyTJ2qEnAbCT8DORncFBrbV9hp7EjlRV/5GJ/38NaW1r7Zg+T7hTFR+wM6iqFa215UPPA2Bn4GcisCNpuwIAAHqh+AAAAHqh+ICtvXvoCQDsRPxMBHYYaz4AAIBeSD4AAIBeKD4AAIBeKD4AAIBeKD5gpKoeVFWPqapdqmr+0PMBGJqfhcCOZsE5JKmq5yT5kyRrRo8VSd7XWrtx0IkBDKCqDm+tfXf0fH5rbePQcwLmBskHY6+qdknygiS/3lp7cpKPJ1mW5A1VteegkwPoWVUdm+SCqvpAkrTWNkpAgB1F8QET9kxy2Oj5x5J8MsnCJC+sqhpsVgA9qqrdk5yU5LeTrK+qf0oUIMCOo/hg7LXW7kjy9iTPqar/1Vq7M8mXklyQ5OcHnRxAj1prtyR5WZIPJHl9kkWTC5Ah5wbMDYoPmPDFJJ9O8uKqenxrbWNr7QNJ9k/ys8NODaA/rbUrW2s3t9bWJvmNJLttKkCq6uFV9eBhZwjcky0YegKwM2it3V5V/5ykJXnj6D+u65Lsl+SqQScHMJDW2rVV9RtJ/qyqvpNkfpInDjwt4B5M8QEjrbXrquo9SS7OxF/7bk/yotba1cPODGA4rbW1VXVhkqcneWprbfXQcwLuuVxqF7ZhtLCyjdZ/AIytqto7yYeT/E5r7cKh5wPcsyk+AIBpVdWi1trtQ88DuOdTfAAAAL1wtSsAAKAXig8AAKAXig8AAKAXig8AAKAXig+AJFW1saouqKpvVdVH/v927i3UqiqKw/j3pV00LXvJKAJN0wqpo2JEUUnIoeiCRUFShCSaBlFEgpBBQZDgW0R0BYkgCkqQIqR6MJVjFCfNwhsVPvVgL1JmBDZ62FPYndSzj8qp4P+DBWvPOdYca66HDWPPubY6/hTGmq9+2M7vUledIHaS+uhJ5HhWfarX9iEx69R7R5BrivrtSO8xIiJiqBQfEREdh6uqr6pmAX8Ay7s77Rjxd2ZVbaiqNScImQSMuPiIiIj4P0rxERHxT5uB6e0X/13qy8AgcKnarw6og22FZAKAequ6W90C3HN0IHWx+lI7n6yuV3e043pgDTCtrbqsbXEr1S/Vb9TnusZ6Wt2jfgrMHG4S6tI2zg71/SGrOQvUzepe9Y4WP0Zd25X7kVN9kBEREd1SfEREdFHHArcBO1vTTOCtqpoNHAJWAwuqag7wFfCkeg7wOnAncCNw0XGGfxHYVFXXAHOA74BVwPdt1WWl2g9cDlwL9AFz1ZvUucD9wGw6xc28HqbzQVXNa/l2AUu6+qYANwO3A6+0OSwBDlbVvDb+UnVqD3kiIiJ6MvbfvoGIiP+Icer2dr4ZeBO4GNhfVdta+3XAVcBWFeAsYAC4AvixqvYBqG8Dy46R4xbgIYCqOgIcVC8YEtPfjq/b5wl0ipGJwPqq+q3l2NDDnGapz9PZ2jUB2NjV915V/QnsU39oc+gHru56H+T8lntvD7kiIiKGleIjIqLjcFX1dTe0AuNQdxPwSVUtGhLXB9Rpug+BF6rq1SE5njiJHOuAhVW1Q10MzO/qGzpWtdyPVVV3kYI6ZYR5IyIijinbriIiercNuEGdDqCOV2cAu4Gp6rQWt+g4138GrGjXjlHPA36hs6px1Ebg4a53SS5RLwQ+B+5Wx6kT6WzxGs5E4Cf1TOCBIX33qWe0e74M2NNyr2jxqDPUc3vIExER0ZOsfERE9KiqDrQVhHfUs1vz6qraqy4DPlJ/BrYAs44xxOPAa+oS4AiwoqoG1K3tr2w/bu99XAkMtJWXX4EHq2pQfRfYDuynszVsOM8AX7T4nfy9yNkDbAImA8ur6nf1DTrvggzaSX4AWNjb04mIiBieVadrp0BERERERMTxZdtVRERERESMihQfERERERExKlJ8RERERETEqEjxERERERERoyLFR0REREREjIoUHxERERERMSpSfERERERExKj4C3hsgEIAT5OyAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 1080x720 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot normalized confusion matrix\n",
    "plot_confusion_matrix(model__conv, normalize = True)\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### CUSTOM EARLY STOPPING BASED ON FSCORE "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "\"\"\"\n",
    "@COMPILE: define a specific metric for unbalanced data \n",
    "\"\"\"\n",
    "\n",
    "def recall_np(y_true, y_pred):\n",
    "    \"\"\"Recall metric\n",
    "    Only computes a batch-wise average of recall. Computes the recall, a metric\n",
    "    for multi-label classification of how many relevant items are selected.\n",
    "    \"\"\"\n",
    "    epsilon = np.finfo(np.float32).eps\n",
    "    y_pred = y_pred.reshape(-1,)\n",
    "    true_positives = np.sum(y_true * y_pred)\n",
    "    print(true_positives)\n",
    "    possible_positives = np.sum(y_true)\n",
    "    recall = true_positives / (possible_positives + epsilon)\n",
    "    return recall\n",
    "\n",
    "\n",
    "def precision(y_true, y_pred):\n",
    "    prod = y_true * y_pred\n",
    "    nb_common_ones = np.sum(prod)\n",
    "    nb_pred_ones = np.sum(y_pred)\n",
    "    precision = nb_common_ones/nb_pred_ones\n",
    "    return precision \n",
    "\n",
    "def f_score_main(y_true, y_pred): \n",
    "    \"\"\"\n",
    "    fscore = 2(recall*precision)/(recall + precision)\n",
    "    \"\"\"\n",
    "    recall = recall_np(y_true, y_pred)\n",
    "    precision = precision(y_ture, y_pred)\n",
    "    return 2*(precision*recall)/(precision+recall)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "import tensorflow as tf \n",
    "from keras.models import Sequential\n",
    "from keras.layers import Conv2D, MaxPooling2D\n",
    "from keras.layers import Activation, Dropout, Flatten, Dense\n",
    "from collections import Counter \n",
    "from keras import backend as K\n",
    "from keras import optimizers\n",
    "from keras.callbacks import CSVLogger\n",
    "from keras.callbacks import ModelCheckpoint\n",
    "\n",
    "from keras.callbacks import Callback\n",
    "\n",
    "from sklearn.metrics import f1_score\n",
    "\n",
    "\n",
    "csv_logger = CSVLogger('log.csv', append=True, separator=';')\n",
    "\n",
    "\n",
    "\n",
    "    \n",
    "    \n",
    "class EarlyStopByF1(Callback):\n",
    "    def __init__(self, value = 0, verbose = 0):\n",
    "        super(Callback, self).__init__()\n",
    "        self.value = value\n",
    "        self.verbose = verbose\n",
    "\n",
    "    def f1(self, y_true, y_pred):\n",
    "        def recall(y_true, y_pred):\n",
    "            \"\"\"Recall metric.\n",
    "\n",
    "            Only computes a batch-wise average of recall.\n",
    "\n",
    "            Computes the recall, a metric for multi-label classification of\n",
    "            how many relevant items are selected.\n",
    "            \"\"\"\n",
    "            true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "            possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))\n",
    "            recall = true_positives / (possible_positives + K.epsilon())\n",
    "            return recall\n",
    "\n",
    "        def precision(y_true, y_pred):\n",
    "            \"\"\"Precision metric.\n",
    "\n",
    "            Only computes a batch-wise average of precision.\n",
    "\n",
    "            Computes the precision, a metric for multi-label classification of\n",
    "            how many selected items are relevant.\n",
    "            \"\"\"\n",
    "            true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "            predicted_positives = K.sum(K.round(K.clip(y_pred, 0, 1)))\n",
    "            precision = true_positives / (predicted_positives + K.epsilon())\n",
    "            return precision\n",
    "        precision = precision(y_true, y_pred)\n",
    "        recall = recall(y_true, y_pred)\n",
    "        return 2*((precision*recall)/(precision+recall+K.epsilon()))\n",
    "    \n",
    "    \n",
    "    def on_epoch_end(self, epoch, logs={}):\n",
    "        predict = np.asarray(self.model.predict(self.validation_data[0]))\n",
    "        target = self.validation_data[1]\n",
    "        predict = tf.convert_to_tensor(predict, tf.float32)\n",
    "        target = tf.convert_to_tensor(target, tf.float32)\n",
    "        score = self.f1(target, predict) # score is now a tf.tensor (necessary for early stopping)\n",
    "#         print (type(score)) # tensor \n",
    "        tensor_value = tf.constant(self.value, shape = (1,))\n",
    "        with tf.Session() as sess:\n",
    "            tensor_value_score = sess.run(tf.greater(score, tensor_value))\n",
    "#             print(sess.run(tf.greater(score, tensor_value))) ## prints [True] when we go beyond self.value\n",
    " \n",
    "        if np.all(tensor_value_score):# if we go beyond our value, we can stop \n",
    "            if self.verbose > 0: # print\n",
    "                print(\"Epoch %05d: early stopping Threshold\" % epoch)\n",
    "            self.model.stop_training = True\n",
    "\n",
    "\n",
    "########## Customize Early Stopping ###########\n",
    "from keras.callbacks import Callback\n",
    "\n",
    "class EarlyStoppingByAccuracy(Callback):\n",
    "    \"\"\"\n",
    "    When checking the logs we see that sometimes we get a very low val loss/ high val accuracy at some epoch and after we go below such values.\n",
    "    Solution:  We will check when such accuracy is reached and stop at this very moment \n",
    "    \n",
    "    PARAMS\n",
    "    --------------------------\n",
    "    value: int v\n",
    "    Value that is seen to be passed for a given epoch => If we are beyond it we stop training!! \n",
    "    \n",
    "    \"\"\"\n",
    "    def __init__(self, monitor='accuracy', value=0.85, verbose=0):\n",
    "        super(Callback, self).__init__()\n",
    "        self.monitor = monitor\n",
    "        self.value = value\n",
    "        self.verbose = verbose\n",
    "\n",
    "    def on_epoch_end(self, epoch, logs={}):\n",
    "        current = logs.get(self.monitor)\n",
    "        if current is None:\n",
    "            warnings.warn(\"Early stopping requires %s available!\" % self.monitor, RuntimeWarning)\n",
    "\n",
    "        if current >= self.value:\n",
    "            if self.verbose > 0:\n",
    "                print(\"Epoch %05d: early stopping THR\" % epoch)\n",
    "            self.model.stop_training = True\n",
    "            \n",
    "def evaluate_classifier_conv_custom_fscore(x_train, y_train, x_val, y_val, img_rows = 128, img_cols = 128, \n",
    "                                           batch_size = 32, epochs = 20, learning_rate = 0.01, \n",
    "                                           dropout = 0.7, log = True, threshold_acc = 0.87, name_model = 'stack', threshold_fscore = 0.5):\n",
    "    \"\"\"\n",
    "    Create simple convolutional model to classify data\n",
    "    Fit data and return fitted model \n",
    "    \n",
    "    Custom callback\n",
    "    \"\"\"\n",
    "    \n",
    "    K.clear_session()\n",
    "    np.random.seed(100)\n",
    "\n",
    "    if K.image_data_format() == 'channels_first':\n",
    "        x_train = x_train.reshape(x_train.shape[0], 3, img_rows, img_cols)\n",
    "        x_val = x_val.reshape(x_val.shape[0], 3, img_rows, img_cols)\n",
    "        input_shape = (3, img_rows, img_cols)\n",
    "    else:\n",
    "        x_train = x_train.reshape(x_train.shape[0], img_rows, img_cols, 3)\n",
    "        x_val = x_val.reshape(x_val.shape[0], img_rows, img_cols, 3)\n",
    "        input_shape = (img_rows, img_cols, 3)\n",
    "\n",
    "    model = Sequential()\n",
    "    model.add(Conv2D(32, (3, 3), input_shape=(128, 128,3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Conv2D(32, (3, 3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Conv2D(64, (3, 3)))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    model.add(Flatten())  # this converts our 3D feature maps to 1D feature vectors\n",
    "    model.add(Dense(64))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Dropout(dropout))\n",
    "    model.add(Dense(1))\n",
    "    model.add(Activation('sigmoid'))\n",
    "    \n",
    "    def recall(y_true, y_pred):\n",
    "            \"\"\"Recall metric.\n",
    "\n",
    "            Only computes a batch-wise average of recall.\n",
    "\n",
    "            Computes the recall, a metric for multi-label classification of\n",
    "            how many relevant items are selected.\n",
    "            \"\"\"\n",
    "            true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "            possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))\n",
    "            recall = true_positives / (possible_positives + K.epsilon())\n",
    "            return recall\n",
    "    \n",
    "    def f1(y_true, y_pred):\n",
    "        def recall(y_true, y_pred):\n",
    "            \"\"\"Recall metric.\n",
    "\n",
    "            Only computes a batch-wise average of recall.\n",
    "\n",
    "            Computes the recall, a metric for multi-label classification of\n",
    "            how many relevant items are selected.\n",
    "            \"\"\"\n",
    "            true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "            possible_positives = K.sum(K.round(K.clip(y_true, 0, 1)))\n",
    "            recall = true_positives / (possible_positives + K.epsilon())\n",
    "            return recall\n",
    "\n",
    "        def precision(y_true, y_pred):\n",
    "            \"\"\"Precision metric.\n",
    "\n",
    "            Only computes a batch-wise average of precision.\n",
    "\n",
    "            Computes the precision, a metric for multi-label classification of\n",
    "            how many selected items are relevant.\n",
    "            \"\"\"\n",
    "            true_positives = K.sum(K.round(K.clip(y_true * y_pred, 0, 1)))\n",
    "            predicted_positives = K.sum(K.round(K.clip(y_pred, 0, 1)))\n",
    "            precision = true_positives / (predicted_positives + K.epsilon())\n",
    "            return precision\n",
    "        precision = precision(y_true, y_pred)\n",
    "        recall = recall(y_true, y_pred)\n",
    "        return 2*((precision*recall)/(precision+recall+K.epsilon()))\n",
    "\n",
    "\n",
    "\n",
    "    # initiate RMSprop optimizer\n",
    "    opt = optimizers.rmsprop(lr=learning_rate, decay = 1e-5)\n",
    "\n",
    "    model.compile(loss='binary_crossentropy',\n",
    "                  optimizer=opt,\n",
    "                  metrics=['accuracy', f1, recall])\n",
    "    \n",
    "    \n",
    "    kfold_weights_path = \"best_model{}.hdf5\".format(name_model)\n",
    "    callbacks = [\n",
    "                EarlyStoppingByAccuracy(monitor='val_acc', value=threshold_acc, verbose=1), # early stopping based on validation accuracy \n",
    "                ModelCheckpoint(kfold_weights_path, monitor='f1', save_best_only=True, verbose=0),\n",
    "                EarlyStopByF1(value = threshold_fscore, verbose = 1) # early stopping based on f1 score\n",
    "                ]\n",
    "\n",
    "    if log:\n",
    "        callbacks.append(csv_logger)\n",
    "        # Fit data \n",
    "        model.fit(x_train, y_train,\n",
    "                  batch_size=batch_size,\n",
    "                  epochs=epochs,\n",
    "                  verbose=1,\n",
    "                  validation_data=(x_val, y_val),\n",
    "                    callbacks = callbacks)\n",
    "        \n",
    "    else:\n",
    "        \n",
    "        # Fit data \n",
    "        model.fit(x_train, y_train,\n",
    "                  batch_size=batch_size,\n",
    "                  epochs=epochs,\n",
    "                  verbose=1,\n",
    "                  validation_data=(x_val, y_val),\n",
    "                  callbacks = callbacks)\n",
    "\n",
    "        \n",
    "    # Compute final performance on validation data\n",
    "    score = model.evaluate(x_val, y_val, verbose=0)\n",
    "    print('Final validation loss:', score[0])\n",
    "    print('Final validation accuracy:', score[1])\n",
    "    \n",
    "    return model  "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "##### PARAMS FOR AUGMENTED DATASET #####\n",
    "## We add early stopping based on f score \n",
    "model_params_aug_custom = { \n",
    "                    \"x_train\": X_aug, \n",
    "                    \"y_train\": Y_aug, \n",
    "                    \"x_val\": x_val, \n",
    "                    \"y_val\": y_val,\n",
    "                    \"img_rows\": 128,\n",
    "                    \"img_cols\": 128, \n",
    "                    \"batch_size\": 16, \n",
    "                    \"epochs\": 2, \n",
    "                    \"learning_rate\": 0.0001, \n",
    "                    \"dropout\": 0.7,\n",
    "                    \"log\":False,\n",
    "                    \"threshold_acc\": 0.86, \n",
    "                    \"name_model\" : 'stack', \n",
    "                    \"threshold_fscore\" : 0.4\n",
    "                    }\n",
    "\n",
    "csv_logger = CSVLogger('log_conv_strong_augOK3.csv', append=True, separator=';')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "model_stack_conv = evaluate_classifier_conv_custom_fscore(**model_params_aug_custom)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## AUGMENTED DATA WITH CONVOLUTIONAL VAE "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-------- LOAD VAE DATA : class MEL ------------\n",
      "Nb of observations generated for this class: 9261\n",
      "-------- LOAD VAE DATA : class NV ------------\n",
      "Nb of observations generated for this class: 4913\n"
     ]
    }
   ],
   "source": [
    "###### LOAD DATA GENERATED FROM VAE ##########\n",
    "\n",
    "### --------- CLASS MEL ---------- ####\n",
    "\n",
    "filename_mel_VAE = \"./augmented_data/images_VAE_mel_10000.npy\"\n",
    "x_mel_aug_VAE = np.load(filename_mel_VAE)\n",
    "print (\"-------- LOAD VAE DATA : class MEL ------------\")\n",
    "print('Nb of observations generated for this class: {}'.format(x_mel_aug_VAE.shape[0]))\n",
    "#######################################\n",
    "\n",
    "\n",
    "### --------- CLASS NV ---------- ####\n",
    "filename_nv_VAE = \"./augmented_data/images_VAE_nv.npy\"\n",
    "x_nv_aug_VAE = np.load(filename_nv_VAE)\n",
    "print (\"-------- LOAD VAE DATA : class NV ------------\")\n",
    "print('Nb of observations generated for this class: {}'.format(x_nv_aug_VAE.shape[0]))\n",
    "#######################################"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "### CLASSIFICATION WITH ONLY AUGMENTED DATA"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### DATA "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "####### RECALL: DATA ORIGINAL FOR BOTH CLASSES  #######\n",
    "\n",
    "###### MEL: class 1 ######\n",
    "x_mel_train, x_mel_test = np.load(\"./dataset/train_mel.npy\"), np.load(\"./dataset/test_mel.npy\")\n",
    "y_mel_train, y_mel_test = np.ones((x_mel_train.shape[0], )), np.ones((x_mel_test.shape[0], ))\n",
    "\n",
    "\n",
    "###### NV: class 0 ######\n",
    "x_nv_train, x_nv_test = np.load(\"./dataset/train_nv.npy\"), np.load(\"./dataset/test_nv.npy\")\n",
    "y_nv_train, y_nv_test = np.zeros((x_nv_train.shape[0], )), np.zeros((x_nv_test.shape[0], ))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "### VALIDATION ORIGINAL DATASET MIX CLASSES ###### \n",
    "\n",
    "##### We need to concatenate also x_mel_test and y_mel_test ###### \n",
    "x_val_org, y_val_org = concat_train_label(x_mel_test, x_nv_test, y_mel_test, y_nv_test)\n",
    "x_val_org, y_val_org = shuffle_data(x_val_org, y_val_org)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train_org, y_train_org = concat_train_label(x_mel_train, x_nv_train, y_mel_train, y_nv_train)\n",
    "x_train_org, y_train_org = shuffle_data(x_train_org, y_train_org)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "#### We are going to classify our images but using only augmented images from both classes ### \n",
    "### First, we shuffle data\n",
    "### Second, we take the same nb of observations for both classes \n",
    "### Third, we concatenate those two classes and shuffle again the data\n",
    "\n",
    "#### Shuffle data ####\n",
    "y_mel_aug_VAE = np.ones((x_mel_aug_VAE.shape[0],))\n",
    "x_mel_aug_VAE, _ = shuffle_data(x_mel_aug_VAE, y_mel_aug_VAE)\n",
    "\n",
    "y_nv_aug_VAE = np.zeros((x_nv_aug_VAE.shape[0],))\n",
    "x_nv_aug_VAE, _ = shuffle_data(x_nv_aug_VAE, y_nv_aug_VAE)\n",
    "\n",
    "#### Take same nb of observations for both classes --- expected: 4,913 obs  #### \n",
    "x_mel_VAE, y_mel_VAE = x_mel_aug_VAE[np.arange(x_nv_aug_VAE.shape[0])], y_mel_aug_VAE[np.arange(x_nv_aug_VAE.shape[0])]\n",
    "\n",
    "### Concatenate both arrays and shuffle the concatenate ### \n",
    "X_aug_only, Y_aug_only = concat_train_label(x_nv_aug_VAE, x_mel_VAE, y_nv_aug_VAE, y_mel_VAE)\n",
    "\n",
    "## Shuffle data ##\n",
    "X_aug_only, Y_aug_only = shuffle_data(X_aug_only, Y_aug_only)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "####### AUGMENTED DATA MEL + ORG MEL ##########\n",
    "\n",
    "### sample data from mel augmented until we have same nb of observations from class MEL and class NV ### \n",
    "\n",
    "### IF AUGMENTED MEL NOT SHUFFLED BEFORE ###\n",
    "y_mel_aug_VAE = np.ones((x_mel_aug_VAE.shape[0],))\n",
    "x_mel_aug_VAE, _ = shuffle_data(x_mel_aug_VAE, y_mel_aug_VAE)\n",
    "#######################################\n",
    "nb_obs_add = x_nv_train.shape[0]-x_mel_train.shape[0]\n",
    "index_aug_mel = np.arange(nb_obs_add)\n",
    "x_mel_aug_VAE_add = x_mel_aug_VAE[index_aug_mel] ## only take a certain nb of observations from MEL augmented dataset\n",
    "y_mel_aug_VAE_add = np.ones((x_mel_aug_VAE_add.shape[0],)) ## mel of class 1\n",
    "\n",
    "## concatenate MEL org with MEL aug : MIX org and augmented ==> Now, we have obs with aug and org of class MEL \n",
    "x_mel_train_mix, y_mel_train_mix =  concat_train_label(x_mel_train, x_mel_aug_VAE_add, y_mel_train, y_mel_aug_VAE_add)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(5365, 128, 128, 3)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x_mel_train_mix.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### ONLY COMPUTE ONCE "
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "##### SAVE DATA TO LOCAL AND STORE IT IN CLOUD ######\n",
    "filename_aug_only_X = './augmented_data/X_aug_only_nv_mel'\n",
    "filename_aug_only_Y = './augmented_data/Y_aug_only_nv_mel'\n",
    "\n",
    "\n",
    "np.save(filename_aug_only_X, X_aug_only)\n",
    "np.save(filename_aug_only_Y, Y_aug_only)"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "######## STORE IT IN CLOUD ###########\n",
    "!gsutil -m cp  ./augmented_data/X_aug_only_nv_mel.npy  gs://dataham/VAE_DATA/"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### PARAMS "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "##### PARAMS FOR AUGMENTED DATASET #####\n",
    "## We add early stopping based on f score \n",
    "model_params_VAE_aug_custom = { \n",
    "                    \"x_train\": X_aug_only, \n",
    "                    \"y_train\": Y_aug_only, \n",
    "                    \"x_val\": x_val_org, \n",
    "                    \"y_val\": y_val_org,\n",
    "                    \"img_rows\": 128,\n",
    "                    \"img_cols\": 128, \n",
    "                    \"batch_size\": 32, \n",
    "                    \"epochs\": 20, \n",
    "                    \"learning_rate\": 0.0001, \n",
    "                    \"dropout\": 0.7,\n",
    "                    \"log\":False,\n",
    "                    \"threshold_acc\": 0.86, \n",
    "                    \"name_model\" : 'VAE_only', \n",
    "                    \"threshold_fscore\" : 0.5\n",
    "                    }\n",
    "\n",
    "csv_logger = CSVLogger('log_conv_strong_augOK3.csv', append=True, separator=';')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#### CLASSIFY AUGMENTED DATA #######\n",
    "model_VAE_conv = evaluate_classifier_conv_custom_fscore(**model_params_VAE_aug_custom)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "####### LOAD BEST MODEL THANKS TO MODELCHECKPOINT ##########\n",
    "from keras.models import load_model\n",
    "\n",
    "model_best_VAE_conv_AUG_only = load_model(\"./best_modelVAE_only.hdf5\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Confusion matrix, without normalization\n",
      "Normalized confusion matrix\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot non-normalized confusion matrix\n",
    "plot_confusion_matrix(model_VAE_conv, x_val = x_val_org, y_val=y_val_org)\n",
    "\n",
    "# Plot normalized confusion matrix\n",
    "plot_confusion_matrix(model_VAE_conv, x_val = x_val_org, y_val=y_val_org, normalize = True)\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "### CLASSIFICATION WITH  AUGMENTED DATA FROM MEL ADDED TO ORIGINAL DATA"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### DATA "
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "x_mel_train_mix, y_mel_train_mix"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [],
   "source": [
    "### Concatenate both arrays and shuffle the concatenate ### \n",
    "X_aug_mix, Y_aug_mix = concat_train_label(x_nv_train, x_mel_train_mix, y_nv_train, y_mel_train_mix)\n",
    "\n",
    "## Shuffle data ##\n",
    "X_aug_mix, Y_aug_mix = shuffle_data(X_aug_mix, Y_aug_mix)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### PARAMS "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "##### PARAMS FOR AUGMENTED DATASET #####\n",
    "## We add early stopping based on f score \n",
    "model_params_VAE_aug_custom_mix = { \n",
    "                    \"x_train\": X_aug_mix, \n",
    "                    \"y_train\": Y_aug_mix, \n",
    "                    \"x_val\": x_val_org,  # x_val \n",
    "                    \"y_val\": y_val_org, # y_val\n",
    "                    \"img_rows\": 128,\n",
    "                    \"img_cols\": 128, \n",
    "                    \"batch_size\": 32, \n",
    "                    \"epochs\": 5, \n",
    "                    \"learning_rate\": 1e-3, \n",
    "                    \"dropout\": 0.3,\n",
    "                    \"log\":False,\n",
    "                    \"threshold_acc\": 0.86, \n",
    "                    \"name_model\" : 'VAE_only', \n",
    "                    \"threshold_fscore\" : 0.8\n",
    "                    }\n",
    "\n",
    "csv_logger = CSVLogger('log_conv_strong_augOK3.csv', append=True, separator=';')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### CLASSIFICATION"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#### CLASSIFY AUGMENTED DATA where MEL is from ORG and AUGMENTED_VAE  #######\n",
    "model_VAE_conv_mix = evaluate_classifier_conv_custom_fscore(**model_params_VAE_aug_custom_mix)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7ff36976b710>"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1080x720 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "##### PLOT CONFUSION MATRIX ######\n",
    "\n",
    "# Plot normalized confusion matrix\n",
    "plot_confusion_matrix(model_VAE_conv_mix, x_val = x_val_org, y_val=y_val_org, normalize = True)\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "### SMART AUGMENTATION"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "-------- LOAD VAE DATA : class MEL ------------\n",
      "Nb of observations generated for this class: 4913\n"
     ]
    }
   ],
   "source": [
    "###### LOAD DATA GENERATED FROM SMART VAE ##########\n",
    "\n",
    "### --------- CLASS MEL ---------- ####\n",
    "\n",
    "filename_mel_smart = \"./augmented_data/images_smart_mel.npy\"\n",
    "x_mel_smart = np.load(filename_mel_smart)\n",
    "print (\"-------- LOAD VAE DATA : class MEL ------------\")\n",
    "print('Nb of observations generated for this class: {}'.format(x_mel_smart.shape[0]))\n",
    "#######################################\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "####### AUGMENTED DATA MEL + ORG MEL ##########\n",
    "\n",
    "### sample data from mel augmented until we have same nb of observations from class MEL and class NV ### \n",
    "\n",
    "## concatenate MEL org with MEL aug : MIX org and augmented ==> Now, we have obs with aug and org of class MEL \n",
    "x_mel_smart_mix, y_mel_smart_mix =  concat_train_label(x_mel_train, x_mel_smart, y_mel_train, np.ones(x_mel_smart.shape[0],))\n",
    "\n",
    "\n",
    "X_aug_smart, Y_aug_smart  = concat_train_label(x_mel_smart_mix, x_nv_train, y_mel_smart_mix, y_nv_train)\n",
    "\n",
    "X_aug_smart, Y_aug_smart  = shuffle_data(X_aug_smart, Y_aug_smart)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [],
   "source": [
    "##### PARAMS FOR AUGMENTED DATASET #####\n",
    "## We add early stopping based on f score \n",
    "model_params_smart_custom_mix = { \n",
    "                    \"x_train\": X_aug_smart, \n",
    "                    \"y_train\": Y_aug_smart, \n",
    "                    \"x_val\": x_val_org,  # x_val \n",
    "                    \"y_val\": y_val_org, # y_val\n",
    "                    \"img_rows\": 128,\n",
    "                    \"img_cols\": 128, \n",
    "                    \"batch_size\": 32, \n",
    "                    \"epochs\": 20, \n",
    "                    \"learning_rate\": 1e-5, \n",
    "                    \"dropout\": 0.3,\n",
    "                    \"log\":False,\n",
    "                    \"threshold_acc\": 0.86, \n",
    "                    \"name_model\" : 'VAE_only', \n",
    "                    \"threshold_fscore\" : 0.8\n",
    "                    }\n",
    "\n",
    "csv_logger = CSVLogger('log_conv_strong_augOK3.csv', append=True, separator=';')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### CLASSIFICATION WITH SMART_AUG_MEL + ORG"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "#### CLASSIFY AUGMENTED DATA where MEL is from ORG and AUGMENTED_VAE  #######\n",
    "model_VAE_smart_mix = evaluate_classifier_conv_custom_fscore(**model_params_smart_custom_mix)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 121,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7fd6d73f2f98>"
      ]
     },
     "execution_count": 121,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAx8AAALICAYAAAAe6E2GAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4wLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvqOYd8AAAIABJREFUeJzs3XmcXmV5P/7PNYkRVMQlyJIECBVk0SqKoFYpbi0UBGutglXrz9alLda61GLrVtQubq1taRULpVoV0NaKGEW7qKVVS9wFi4ZFEjYNixtftuT+/TFP4mSYzAyQc07I837zel48Z3nuc52ZMMyV67rPXa21AAAAdG1i6AAAAIDxIPkAAAB6IfkAAAB6IfkAAAB6IfkAAAB6IfkAAAB6IfkAtnlV9Yaq+qfR+92r6sdVtWALX+PSqnrSlhxzHtf8raq6enQ/978T4/y4qvbakrENparOr6rDho4DgJlJPoA7bfSL99VVdc8p+36zqj4zYFgzaq1d1lq7V2tt3dCx3BlVdbck70jyC6P7ueaOjjX6/MVbLrotr6pOq6o3zXVea+2A1tpneggJgDtA8gFsKQuTvPTODlKT/Gya285Jtkty/tCBbA2qauHQMQAwN/+DB7aUtyZ5ZVXdZ6aDVfWYqjqvqn4w+vdjphz7TFW9uar+O8kNSfYa7XtTVf3PqC3oY1V1/6p6f1X9cDTGnlPGeGdVrR4d+1JVPW4zcexZVa2qFlbVo0djb3jdWFWXjs6bqKoTquqiqrqmqs6sqvtNGec5VfXd0bE/mu0LU1XbV9XbR+f/oKrOrartR8eOHrUKXT+65/2mfO7SqnplVX199Lkzqmq7qtonyYWj066vqv+Yel/Tvq6/OXr/wKr67GictVV1xpTzWlU9cPR+x6p6b1V9fxTvazYkg1X1vFHsb6uq66rqkqo6Ypb7vrSqfn8U/0+q6pSq2rmqPlFVP6qqf6uq+045/0NVddUoxs9V1QGj/S9M8mtJXrXhz8KU8f+gqr6e5Cej7+nG9reqWlFVb58y/hlVdeps3ysAuiX5ALaUlUk+k+SV0w+Mfmn/eJK/SnL/TLYLfbw2nafwnCQvTLJDku+O9h072r8kyc8k+XySf0hyvyTfSvL6KZ8/L8nDRsc+kORDVbXdbAG31j4/ajm6V5L7JvlCkg+ODv9ukqcm+fkkuyW5LslJo/vZP8nfjWLbbXRPS2e51NuSPCLJY0bxvSrJ+lES8cEkv5dkpyQrknysqhZN+ewzkhyeZHmSn03yvNbat5McMDp+n9baE2a7z5E3JvnU6D6XJvnrzZz310l2TLLX6N6fm+T/m3L8kEwmPouTvCXJKVVVs1z3V5I8Ock+SZ6S5BNJ/nD0+YlMfp03+ESSvZM8IMmXk7w/SVprJ4/ev2X0/XrKlM8cl+TITH4dbp127ecneU5VPaGqfi3JI7MFqnMA3HGSD2BLel2Sl1TVTtP2H5nkO62197XWbm2tfTDJ/2Xyl9ENTmutnT86fsto3z+01i5qrf0gk7+YXtRa+7fRL5kfSnLghg+31v6ptXbN6PNvT3L3JA+6HbH/VZKfJNlQxXhRkj9qra1prd2U5A1Jnj6qLDw9ydmttc+Njr02yfqZBh1VDZ6f5KWttctba+taa/8z+twzk3y8tfbp0T2/Lcn2mUxSNsbVWruitXZtko9lMsG6I25JskeS3VprN7bWzp0h1gWjmF7dWvtRa+3SJG/PZJK1wXdba+8ZzZn5xyS7ZrIFbHP+urV2dWvt8iT/leSLrbWvjO7/I9n0e3jq6Lobvt4Praod57ivv2qtrW6t/b/pB1prVyV58SjOdyZ5bmvtR3OMB0CHJB/AFtNa+2aSs5OcMO3QbvlpNWOD72ayorHB6hmGvHrK+/83w/a9NmxU1Suq6lujlp3rM/m394vnE3dVvSjJYUme1VrbkETskeQjo3ao6zNZaVmXyV+0d5sab2vtJ0k2N+F7cSbnZlw0w7FNvi6ja6/Opl+Xq6a8vyFT7vl2elWSSvK/ozav528m1kXZ9Hs1/fu0MZ7W2g2jt7PFNK/vYVUtqKo/G7W5/TDJpVNims1Mf26mOjvJgiQXzpRwAdAvyQewpb0+yQuy6S+sV2Tyl/mpdk9y+ZTtdkcvOJrf8QeZbFG6b2vtPkl+kMlftufz2TcmOWZUYdlgdZIjWmv3mfLabvQ3+FcmWTZljHtksvVqJmuT3JjJtrHpNvm6jNqXlmXTr8t8/WT073tM2bfLhjettataay9ore2WyarO326Y5zEt1g0Vkg2mf5+68qwkxyR5UiYTxz1H+zd8Dzf352OuPzdvzmTiuGtVHXcnYwTgTpJ8AFtUa21VkjOyaS//iiT7VNWzRpOCn5lk/0z+rfSWsEOSW5N8P8nCqnpdknvP9aGqWjaK9bmjeRRTvSvJm6tqj9G5O1XVMaNjH05yVFU9djQ/48Rs5ufpqJpxapJ3VNVuo7/hf3RV3T3JmUmOrKon1uSjc1+R5KYk/3O77n7yOt/PZJLw7NE1np8pCU9V/WpVbZiXcl0mf2lfN22MdaOY3lxVO4zu/eVJ/un2xnMH7JDJe78mkwnUn0w7fnUm56HMW1Udmsn5Ks8dvf66qpbM/ikAuiT5ALpwYpKNa36M1qA4KpO/XF+TyRago1pra7fQ9c7J5JyQb2eyTejGzN2OkyRPzGR14MP10ydebXh07TuTnJXkU1X1o0xORj9kdD/nJ/mdTE5svzKTv8yvmeU6r0zyjUxOir82yZ8nmWitXZjk2Zmc5L02k3NgntJau3me9z3dC5L8fia/xgdk0yTmkUm+WFU/Ht3XS1trl8wwxksyWUW5OMm5o3vs4wlR783k9+7yJBdk8us91SlJ9h+1wf3rXINV1b1HYx4/mmtz7miMf5hjgjwAHarW7nCnAwAAwLypfAAAAL2QfAAAAL2QfAAAAL2QfAAAAL1YOHQAU9X2C1vuvWjoMAC2Cg9ZMtPSIADjZ81ll+fatdduU0+qq8Xbtdy8fu4Tu/SjW85prR3e5yW3quQj916Uhc/aZ+goALYKZ7/xw0OHALBVOOrQXx46hC3v5vXJIQ8YNoZ/u3xx35fUdgUAAPRi66p8AADAuBjDNU9VPgAAgF5IPgAAgF5ouwIAgL5VxrIMMIa3DAAADEHyAQAA9ELbFQAADMHTrgAAALqh8gEAAEMYv8KHygcAANAPyQcAANALbVcAANC7MuEcAACgKyofAADQNyucAwAAdEfyAQAA9ELbFQAADMGEcwAAgG6ofAAAwBDGr/Ch8gEAAPRD8gEAAPRC2xUAAPStkkyMX9+VygcAANALlQ8AABjC+BU+VD4AAIB+SD4AAIBeaLsCAIAhWOEcAACgGyofAAAwhPErfKh8AAAA/ZB8AAAAvdB2BQAAfbPCOQAAQHdUPgAAYAjjV/hQ+QAAAPoh+QAAAHqh7QoAAHpXVjgHAADoiuQDAADohbYrAADom3U+AAAAuiP5AACAIdTAr7nCqzq8qi6sqlVVdcIMx3evqv+sqq9U1der6pfmGlPyAQAAbKKqFiQ5KckRSfZPclxV7T/ttNckObO1dmCSY5P87VzjSj4AAIDpDk6yqrV2cWvt5iSnJzlm2jktyb1H73dMcsVcg5pwDgAAQxh+nY/FVbVyyvbJrbWTR++XJFk95diaJIdM+/wbknyqql6S5J5JnjTXBSUfAAAwnta21g7azLGZMqM2bfu4JKe11t5eVY9O8r6qenBrbf3mLij5AACAIQxe+JjVmiTLpmwvzW3bqn4jyeFJ0lr7fFVtl2Rxku9tblBzPgAAgOnOS7J3VS2vqkWZnFB+1rRzLkvyxCSpqv2SbJfk+7MNKvkAAAA20Vq7NcnxSc5J8q1MPtXq/Ko6saqOHp32iiQvqKqvJflgkue11qa3Zm1C2xUAAPTtLrDCeWttRZIV0/a9bsr7C5L83O0ZU+UDAADohcoHAAAMYesufHRC5QMAAOiF5AMAAOiFtisAABjC8Cuc907lAwAA6IXKBwAADGEMywBjeMsAAMAQJB8AAEAvtF0BAEDfqkw4BwAA6IrKBwAADGH8Ch8qHwAAQD8kHwAAQC+0XQEAwBBMOAcAAOiGygcAAAxhDMsAY3jLAADAECQfAABAL7RdAQBA3yomnAMAAHRF8gEAAPRC2xUAAAxh/LquVD4AAIB+qHwAAMAQJsav9KHyAQAA9ELyAQAA9ELbFQAADME6HwAAAN1Q+QAAgL5VPGoXAACgK5IPAACgF9quAACgd5UaeMJ5G+CaKh8AAEAvVD4AAGAAKh8AAAAdkXwAAAC90HYFAAADGMMFzlU+AACAfqh8AABAzyrJxMClj3UDXFPlAwAA6IXkAwAA6IW2KwAA6FsNv87HEFQ+AACAXqh8AADAAFQ+AAAAOiL5AAAAeqHtCgAAelfargAAALqi8gEAAAMYw8KHygcAANAPyQcAANALbVcAANCzinU+AAAAOiP5AAAAeqHtCgAA+lbargAAADqj8gEAAAOoqHwAAAB0QvIBAAD0QtsVAAAMwIRzAACAjqh8AADAAMaw8KHyAQAA9EPyAQAA9ELbFQAA9KxSmRjDviuVDwAAoBcqHwAAMACP2gUAAOiI5AMAAOiFtisAAOhbabsCAADojMoHAAAMYAwLHyofAABAPyQfAABAL7RdAQBAzyomnAMAAHRG5QMAAAag8gEAANARyQcAANALbVcAANC70nYFAADQFZUPAADoW5lwDgAAkCSpqsOr6sKqWlVVJ8xw/C+q6quj17er6vq5xlT5AAAANlFVC5KclOTJSdYkOa+qzmqtXbDhnNbay6ac/5IkB841rsoHAAAMoGrY1xwOTrKqtXZxa+3mJKcnOWaW849L8sG5BpV8AAAA0y1JsnrK9prRvtuoqj2SLE/yH3MNqu0KAADG0+KqWjll++TW2smj9zPVRtpmxjk2yYdba+vmuqDkAwAAelbZKp52tba1dtBmjq1JsmzK9tIkV2zm3GOT/M58LqjtCgAAmO68JHtX1fKqWpTJBOOs6SdV1YOS3DfJ5+czqMoHAAAMYCuofGxWa+3Wqjo+yTlJFiQ5tbV2flWdmGRla21DInJcktNba5trydqE5AMAALiN1tqKJCum7XvdtO033J4xtV0BAAC9UPkAAIABTGzFbVddUfkAAAB6ofIBAAB9m98q49sclQ8AAKAXkg8AAKAX2q4AAKBnldqq1/noisoHAADQC5UPAAAYQEXlAwAAoBOSDwAAoBfargAAYAAmnAMAAHRE5QMAAAYwjpUPyQdj5Rf2fUze8cuvyoKayKlf/Eje+u//sMnxtz31lTnsgY9Mkmx/t+3ygB3ul53+8HFJkhvf/qV888pVSZLLrrsyTzvl9/oNHmAL+synP5c/ftWbsm79uhz73Gfkt1/xok2Ov+evT83p/3hmFi5cmPstvl/e+rd/mqW7L0mSLN/xQdn3gH2SJLst3S2nnPnu3uMH7po6TT6q6vAk70yyIMnft9b+rMvrwWwmaiJ/9SuvzhHvenHWXH91vvCy9+fsb34237r64o3nvPJf37bx/e887tg8bMm+G7f/3y035aC3PbPXmAG6sG7durz2FW/I+z96WnZZskuO/vlfyZOOfEL22Xfvjecc8ND9c/bnPpLt77F93vf378+fvvYtOekf35kk2W777fKJ//nYUOEDd2GdzfmoqgVJTkpyRJL9kxxXVft3dT2Yy8G7PzgXrV2dS665PLesuzVnfOWcPOXBh232/GceeERO//In+wsQoCdfXfn17LnXHtl9+e5ZtGhRnvIrR+bTZ//7Juc85tBHZft7bJ8kOfCRD8uVl181RKiwTasa9jWELiecH5xkVWvt4tbazUlOT3JMh9eDWe12nwdkzfU//Z/n5T+4Okt2fMCM5+5+312z5/13y39+53837ttu4aJ84eXvz7kvfW+OfvDjO48XoCtXXXlVdl2y68btXZfskquuvHqz55/x3g/nsF84dOP2TTfelKMO/eU89fFPzzkf+3SnsQLbli7brpYkWT1le02SQ6afVFUvTPLCJMkOd+swHMbdTKuItrQZz33Ggb+Yf/nav2V9W79x314nHpErf/j9LL//knzqt9+Tb175nVx8zZrO4gXozAw/+jY38fVfTv9ovvHlb+SMT75/477Pf+uz2XnXnXPZJZfluKOem30P2Cd77LVHV9HCNmmy+jB+E867rHzM9NW8zY+71trJrbWDWmsHZXvz3+nO5ddfnaX32WXj9pIdd84VP/j+jOc+88DDc8a0lqsrfzh57iXXXJ7PrVqZhy3dd6aPAmz1dtltl1x5+ZUbt6+8/KrsvMttK8Hn/ud/52/e+rf5+zPfnbvf/e4b9++8685Jkt2X755HPfbgfPPrF3QfNLBN6DL5WJNk2ZTtpUmu6PB6MKvzVp+fB+60e/a8326524KFeeaBv5izz//sbc7bZ6c9cp973Dufv/RrG/fdZ/sdsmjBZGXu/ve8Tx69/GH51lUX3+azAHcFD33EQ3LJRZfmsktX5+abb87H/vnjefKRT9zknG9+7fy8+qWvzSlnvDuLd7r/xv0/uO4Huemmm5Ik1669Niu/+OXsve8De40fuOvqstRwXpK9q2p5ksuTHJvkWR1eD2a1bv26vPSf/ywff9HfZcHERE774kdzwVUX5fWH/1a+tPqCjYnIMx9+RM78yqZVj/123it/+6uvyfq2PhM1kbf++6mbPCUL4K5k4cKFOfFtr89zn/r8rFu/Ls94ztOzz3575+1v+sv87IEPyZOPfGL+5DVvyQ0/viG//dyXJPnpI3W/c+FF+cOXvjYTExNZv359futlL9rkKVnAfNVYtl1VazP3vG+Rwat+KclfZvJRu6e21t486/k736MtfNY+ncUDcFdy0Rs/PHQIAFuFow795Xz9y9/Ypn5T327Zjm3Z7z1q0BhWvfJTX2qtHdTnNTudZNFaW5FkRZfXAACAu6JxrHx0OecDAABgI8kHAADQC8+2BQCAAYxh15XKBwAA0A/JBwAA0AttVwAAMABPuwIAAOiIygcAAPSsSuUDAACgM5IPAACgF9quAABgANquAAAAOqLyAQAAAxjDwofKBwAA0A/JBwAA0AttVwAA0Lsy4RwAAKArKh8AADAAlQ8AAICOSD4AAIBeaLsCAICeVWm7AgAA6IzKBwAADGAMCx8qHwAAQD8kHwAAQC+0XQEAwABMOAcAAOiIygcAAAxB5QMAAKAbkg8AAKAX2q4AAKB3ZcI5AABAV1Q+AACgbzWW881VPgAAgH5IPgAAgF5ouwIAgJ5VrHAOAADQGckHAADQC21XAAAwAG1XAAAAHVH5AACAAah8AAAAdETyAQAA9ELbFQAADGAMu65UPgAAgH6ofAAAQN+qTDgHAADoiuQDAADohbYrAADoWcU6HwAAAJ1R+QAAgAGofAAAAHRE8gEAAPRC2xUAAAxA2xUAAEBHVD4AAKBvlYxh4UPlAwAA6IfkAwAA6IW2KwAAGIAJ5wAAAB2RfAAAQM8qlaphX3PGWHV4VV1YVauq6oTNnPOMqrqgqs6vqg/MNaa2KwAAYBNVtSDJSUmenGRNkvOq6qzW2gVTztk7yauT/Fxr7bqqesBc46p8AAAA0x2cZFVr7eLW2s1JTk9yzLRzXpDkpNbadUnSWvveXIOqfAAAwAC2ggnni6tq5ZTtk1trJ4/eL0myesqxNUkOmfb5fZKkqv47yYIkb2itfXK2C0o+AABgPK1trR20mWMzZUZt2vbCJHsnOSzJ0iT/VVUPbq1dv7kLSj4AAGAAwxc+ZrUmybIp20uTXDHDOV9ord2S5JKqujCTych5mxvUnA8AAGC685LsXVXLq2pRkmOTnDXtnH9N8vgkqarFmWzDuni2QSUfAADAJlprtyY5Psk5Sb6V5MzW2vlVdWJVHT067Zwk11TVBUn+M8nvt9aumW1cbVcAANC32iomnM+qtbYiyYpp+1435X1L8vLRa15UPgAAgF5IPgAAgF5ouwIAgCFs5W1XXVD5AAAAeqHyAQAAA9jaJ5x3QeUDAADoheQDAADohbYrAADoWSWZGL+uK5UPAACgHyofAADQuzLhHAAAoCuSDwAAoBfargAAoG+VTGi7AgAA6IbKBwAA9KxihXMAAIDOSD4AAIBeaLsCAIABjGMVYBzvGQAAGIDKBwAADMCjdgEAADoi+QAAAHqh7QoAAHpmnQ8AAIAOqXwAAEDvyoRzAACArkg+AACAXmi7AgCAvpUJ5wAAAJ1R+QAAgJ5VxrMKMI73DAAADEDyAQAA9ELbFQAADMA6HwAAAB2RfAAAAL3QdgUAAAOwzgcAAEBHVD4AAKBnFRPOAQAAOiP5AAAAeqHtCgAABjB+TVcqHwAAQE9UPgAAoHdlwjkAAEBXJB8AAEAvtF0BAEDPqqzzAQAA0BmVDwAAGECpfAAAAHRD8gEAAPRC2xUAAAzAhHMAAICOqHwAAEDPavQaNyofAABALyQfAABAL7RdAQDAAEw4BwAA6IjKBwAA9K5UPgAAALoi+QAAAHqh7QoAAHpWlZS2KwAAgG6ofAAAwABMOAcAAOjIZisfVXXv2T7YWvvhlg8HAADYVs3WdnV+kpZkaj1ow3ZLsnuHcQEAwDZt/JquZkk+WmvL+gwEAADYts1rzkdVHVtVfzh6v7SqHtFtWAAAwLZmzqddVdXfJLlbkkOT/EmSG5K8K8kjuw0NAAC2TZXxfNrVfB61+5jW2sOr6itJ0lq7tqoWdRwXAACwjZlP8nFLVU1kcpJ5qur+SdZ3GhUAAGzjxrHyMZ85Hycl+eckO1XVHyc5N8mfdxoVAACwzZmz8tFae29VfSnJk0a7frW19s1uwwIAALY182m7SpIFSW7JZOuVVdEBAOBOqZS2q9uqqj9K8sEkuyVZmuQDVfXqrgMDAAC2LfOpfDw7ySNaazckSVW9OcmXkvxpl4EBAMC2qjKe7UTzuefvZtMkZWGSi7sJBwAA2FZttvJRVX+RyTkeNyQ5v6rOGW3/QiafeAUAADBvs7VdbXii1flJPj5l/xe6CwcAAMZAZSwnnG82+WitndJnIAAAwLZtzgnnVfUzSd6cZP8k223Y31rbp8O4AABgm2aF85mdluQfMjkp/4gkZyY5vcOYAACAbdB8ko97tNbOSZLW2kWttdckeXy3YQEAANua+azzcVNNzoa5qKpenOTyJA/oNiwAANh2VbRdbc7Lktwrye8m+bkkL0jy/C6DAgAAhlVVh1fVhVW1qqpOmOH486rq+1X11dHrN+cac87KR2vti6O3P0rynNsfNgAAMN3W/KjdqlqQ5KQkT06yJsl5VXVWa+2Caaee0Vo7fr7jzrbI4EcyuajgjFprT5vvRebr4cv2zX+/zfqFAEmy/eEeKgiQJFn1vaEjGEcHJ1nVWrs4Sarq9CTHJJmefNwus1U+/ubODAwAAGzVFlfVyinbJ7fWTh69X5Jk9ZRja5IcMsMYv1JVhyb5dpKXtdZWz3DORrMtMvjv84sZAAC4fSoTGbztam1r7aDNHJspuOldUR9L8sHW2k2jB1P9Y5InzHbB+Uw4BwAAxsuaJMumbC9NcsXUE1pr17TWbhptvifJI+YadD6P2gUAALawrXnCeZLzkuxdVcszudTGsUmeNfWEqtq1tXblaPPoJN+aa9B5Jx9VdfcpmQ0AALCNaq3dWlXHJzknyYIkp7bWzq+qE5OsbK2dleR3q+roJLcmuTbJ8+Yad87ko6oOTnJKkh2T7F5VD03ym621l9zhuwEAALZqrbUVSVZM2/e6Ke9fneTVt2fM+VQ+/irJUUn+dXSRr1XV42/PRQAAgJ+qssL5Zs9prX132r51XQQDAABsu+ZT+Vg9ar1qo5UOX5LJ5/gCAAB3UA3/qN3ezafy8VtJXp5k9yRXJ3nUaB8AAMC8zVn5aK19L5OP1gIAALjD5vO0q/fktqsZprX2wk4iAgCAMbCVr/PRifnM+fi3Ke+3S/LLSVZ3Ew4AALCtmk/b1RlTt6vqfUk+3VlEAADANmneK5xPsTzJHls6EAAAGBeVGst1PuYz5+O6/HTOx0Qml04/ocugAACAbc+syUdNzoJ5aJLLR7vWt9ZuM/kcAAC4fWpeq15sW2a941Gi8ZHW2rrRS+IBAADcIfNJt/63qh7eeSQAAMA2bbNtV1W1sLV2a5LHJnlBVV2U5CdJKpNFEQkJAADcQSacb+p/kzw8yVN7igUAANiGzZZ8VJK01i7qKRYAABgbVjjf1E5V9fLNHWytvaODeAAAgG3UbMnHgiT3yqgCAgAAcGfMlnxc2Vo7sbdIAABgTNTon3Ez26N2x++rAQAAdGa2yscTe4sCAADGSY3no3Y3W/lorV3bZyAAAMC2bT4rnAMAANxps7VdAQAAHRnHdT5UPgAAgF6ofAAAQM8qycQY1gHG744BAIBBSD4AAIBeaLsCAIDelQnnAAAAXVH5AACAAah8AAAAdETyAQAA9ELbFQAADGAi2q4AAAA6ofIBAAA9q5hwDgAA0BnJBwAA0AttVwAA0LdKJrRdAQAAdEPlAwAAelcpj9oFAADohuQDAADohbYrAADoWSWZqPGrA4zfHQMAAIOQfAAAAL3QdgUAAAMo63wAAAB0Q+UDAAAGYJ0PAACAjkg+AACAXmi7AgCA3lUmTDgHAADohsoHAAD0rGLCOQAAQGckHwAAQC+0XQEAwABMOAcAAOiIygcAAPStkqrxqwOM3x0DAACDkHwAAAC90HYFAAC9K+t8AAAAdEXlAwAAelbxqF0AAIDOSD4AAIBeaLsCAIABlLYrAACAbqh8AADAACY8ahcAAKAbkg8AAKAX2q4AAKBnFRPOAQAAOqPyAQAAvatUjV8dYPzuGAAAGITkAwAA6IW2KwAAGIB1PgAAADoi+QAAAHqh7QoAAHpWZZ0PAACAzqh8AADAAMqEcwAAgG5IPgAAgF5IPgAAoHeVqmFfc0ZYdXhVXVhVq6rqhFnOe3pVtao6aK4xJR8AAMAmqmpBkpOSHJFk/yTHVdX+M5y3Q5LfTfLF+Ywr+QAAgAFMpAZ9zeHgJKtaaxe31m5OcnqSY2Y4741J3pLkxvndMwAAMI4WV9XKKa+pKissAAATCUlEQVQXTjm2JMnqKdtrRvs2qqoDkyxrrZ093wt61C4AAIynta21zc3TmKk00jYerJpI8hdJnnd7Lij5AACAnlWSyd/ft1prkiybsr00yRVTtndI8uAknxlNXt8lyVlVdXRrbeXmBt2q7xgAABjEeUn2rqrlVbUoybFJztpwsLX2g9ba4tbanq21PZN8IcmsiUei8gEAAAOorXqF89barVV1fJJzkixIcmpr7fyqOjHJytbaWbOPMDPJBwAAcButtRVJVkzb97rNnHvYfMbUdgUAAPRC5QMAAAYwn1XGtzUqHwAAQC9UPgAAYABb84Tzrqh8AAAAvZB8AAAAvdB2BQAAAzDhHAAAoCMqHwAA0LNKMmHCOQAAQDckHwAAQC+0XQEAQN+qTDgHAADoisoHAAAMoMawDjB+dwwAAAxC8gEAAPRC2xUAAAzAhHMAAICOSD4AAIBeaLsCAICeVZKKtisAAIBOqHwAAEDvKhMmnAMAAHRD8gEAAPRC2xUAAAzAhHMAAICOqHwAAMAArHAOAADQEckHAADQC21XAADQs8kVzsevDjB+dwwAAAxC5QMAAHpXJpwDAAB0RfIBAAD0QtsVAAAMYMIK5wAAAN1Q+QAAgL6VFc4BAAA6I/kAAAB6oe0KAAB6NrnCubYrAACATqh8AADAAEw4BwAA6IjkAwAA6IW2KwAA6F2lxrAOMH53DAAADELlAwAABjBhwjkAAEA3JB8AAEAvtF0BAEDPrHAOAADQIckHAADQC21XAAAwgPK0KwAAgG6ofAAAQO/KhHMAAICuSD4AAIBeaLsCAIABmHAOAADQEZUPAADoWSWZGMM6wPjdMWPtU5/8VH52/4flgAc9JG/987fd5vi5nzs3j37kY3Kvu987//LPH9nk2D0X7ZBDHvGoHPKIR+XpT/3VvkIG6MwvHnRY/u/Uz+Y7p52bP3jm79zm+LKddst/vPXMfPnvPpmvvfvTOeLgJyRJ9th5aW44e1W+8q5z8pV3nZO/e+mf9h06cBfVWeWjqk5NclSS77XWHtzVdWC+1q1bl9/73Zfn45/8WJYsXZLHPupxOeopR2a//ffbeM6y3Zfl5FPenb98xztv8/ntt98+X/zSF/oMGaAzExMTOeklb8qT/+BZWbP2ypz3Nx/PWZ//VL512Xc2nvOaX3tpzvzsx/Kus9+X/XbfOyve/N4sf86jkyQXXXFpDnzxLw4VPnAX1WXl47Qkh3c4Ptwu5/3vyvzMz+yV5Xstz6JFi/Krz3h6zj7r7E3O2WPPPfKQn31IJiYUBYFt28EPelhWXXFpLrnqstxy6y05/TMfzTGP+YVNzmmt5d733CFJsuM9d8gV11w9RKiwbarJCedDvobQ2W9YrbXPJbm2q/Hh9rriiiuydNnSjdtLli7J5VdcOe/P33jjjfm5Qx6bQx9zWM766Me6CBGgN0sW75rV3//pz8A1a6/KksW7bnLOG973jjz7iU/L6g+clxVvfm9ectJrNx5bvsvu+fLffTKfefuH89gHH9xb3MBd2+ATzqvqhUlemEy2vEBXWmu32Xd7sv5vX3Jhdttt11xy8SU5/Mm/lAc/+IDs9TN7bckQAXoz04+/6T8nj3v8MTntU2fmHR8+OY/a7+F53x+8Mw9+wRNz5bXfy+6/dnCu/dH1efjeD8m/vuGUHPCCJ+RHN/y4p+hhW2CF80G01k5urR3UWjtop50WDx0O27AlS5Zkzeo1G7cvX3N5dtt1l3l/frfdJv9GcPley3Pozz8uX/3q17Z4jAB9WfP9K7Nsp59WOpYu3iVXXHPVJuf8xuHH5szPTlZ6v/CtL2e7RXfP4h3vl5tvuTnX/uj6JMmXv/ONXHTld7PPUn8ZA8xt8OQD+nLQIx+RVasuyqWXXJqbb745HzrzwznyKUfO67PXXXddbrrppiTJ2rVr8/n/+UL222/fLsMF6NR5F34tey9Znj13WZa7Lbxbjj3smJz1+U9vcs5l37siTzzwsUmSfXd/YLZbdPd8//prsnjH+22cG7d8l92z95LlufjKy3q/B+CuZ/C2K+jLwoUL8xfvfHue8kvHZN26dfn15z03+x+wf058/Rvz8IMenqOecmRWnvelPPPpx+b6667PirM/kTf98Zvz5a+vzP9968K85LdfkomJiaxfvz6vfNUrNnlKFsBdzbr163L837w25/zp+7NgYiKnnnNGLvjut/PHv/7KrPz21/Kxz386r3j3iXnPy9+Slz3tBWlped5bX54kOfQhj8qJv/6K3LpuXdatX5cXv/OEXDeqhADzN44rnNdMffBbZOCqDyY5LMniJFcneX1r7ZTZPvOIgx7e/vuL53YSD8BdzfaH7zN0CABbhy9+L+2HN29Tv6nv97B92z9++j2DxnDIAw79UmvtoD6v2Vnlo7V2XFdjAwDAXZ0J5wAAAB2RfAAAAL0w4RwAAHpW0XYFAADQGZUPAAAYwhg+alflAwAA6IXkAwAA6IW2KwAA6F2ZcA4AANAVlQ8AABhAmXAOAADQDckHAADQC21XAAAwABPOAQAAOiL5AAAAbqOqDq+qC6tqVVWdMMPxF1fVN6rqq1V1blXtP9eYkg8AABhADfzPrLFVLUhyUpIjkuyf5LgZkosPtNYe0lp7WJK3JHnHXPcs+QAAAKY7OMmq1trFrbWbk5ye5JipJ7TWfjhl855J2lyDmnAOAAA9q2wV63wsrqqVU7ZPbq2dPHq/JMnqKcfWJDlk+gBV9TtJXp5kUZInzHVByQcAAIynta21gzZzbKbM6DaVjdbaSUlOqqpnJXlNkl+f7YLargAAgOnWJFk2ZXtpkitmOf/0JE+da1DJBwAA9G7o6eZztnydl2TvqlpeVYuSHJvkrE3uoGrvKZtHJvnOXINquwIAADbRWru1qo5Pck6SBUlOba2dX1UnJlnZWjsryfFV9aQktyS5LnO0XCWSDwAAGMQ8qg+Daq2tSLJi2r7XTXn/0ts7prYrAACgF5IPAACgF9quAACgb7VVrPPRO5UPAACgFyofAAAwgK19wnkXVD4AAIBeSD4AAIBeaLsCAICeVUw4BwAA6IzKBwAA9K5MOAcAAOiK5AMAAOiFtisAABiAtisAAICOqHwAAMAAPGoXAACgI5IPAACgF9quAABgACacAwAAdETlAwAAelZR+QAAAOiM5AMAAOiFtisAAOhdWecDAACgK5IPAACgF9quAABgENquAAAAOqHyAQAAfauYcA4AANAVyQcAANALbVcAADCAMuEcAACgGyofAAAwAJUPAACAjkg+AACAXmi7AgCAnlXKOh8AAABdUfkAAIABmHAOAADQEckHAADQC21XAAAwAG1XAAAAHVH5AACAAXjULgAAQEckHwAAQC+0XQEAwABMOAcAAOiIygcAAPSsUiacAwAAdEXyAQAA9ELbFQAADMCEcwAAgI6ofAAAwCBUPgAAADoh+QAAAHqh7QoAAAYwfk1XKh8AAEBPJB8AAEAvtF0BAMAAqsav8UrlAwAA6IXKBwAADELlAwAAoBOSDwAAoBfargAAYADj13Sl8gEAAPRE5QMAAHpXGcfah8oHAADQC8kHAADQC21XAADQsyornAMAAHRG8gEAAPRC8gEAAPRC8gEAAPTChHMAABhAWecDAACgGyofAAAwAJUPAACAjkg+AACAXkg+AACAXkg+AACAXphwDgAAA6gy4RwAAKATkg8AAKAXkg8AAKAXkg8AAOA2qurwqrqwqlZV1QkzHH95VV1QVV+vqn+vqj3mGlPyAQAAvavB/5k1uqoFSU5KckSS/ZMcV1X7TzvtK0kOaq39bJIPJ3nLXHct+QAAAKY7OMmq1trFrbWbk5ye5JipJ7TW/rO1dsNo8wtJls41qOQDAADG0+KqWjnl9cIpx5YkWT1le81o3+b8RpJPzHVB63wAAMAgBl/nY21r7aDNHJspuDbjiVXPTnJQkp+f64KSDwAAYLo1SZZN2V6a5IrpJ1XVk5L8UZKfb63dNNeg2q4AAIDpzkuyd1Utr6pFSY5NctbUE6rqwCTvTnJ0a+178xlU5QMAAHpW2QqarmbRWru1qo5Pck6SBUlOba2dX1UnJlnZWjsryVuT3CvJh6oqSS5rrR0927iSDwAA4DZaayuSrJi273VT3j/p9o4p+QAAgAGMqgVjxZwPAACgF5IPAACgF9quAABgENquAAAAOqHyAQAAAxi/uofKBwAA0BPJBwAA0AttVwAAMIjxa7xS+QAAAHqh8gEAAL0rK5wDAAB0RfIBAAD0QvIBAAD0QvIBAAD0woRzAADoWSUpj9oFAADohuQDAADohbYrAAAYhLYrAACATqh8AADAAMav7qHyAQAA9ETyAQAA9ELbFQAADKBq/BqvVD4AAIBeqHwAAEDvKuM45VzlAwAA6IXkAwAA6IW2KwAAGMD4NV2pfAAAAD2RfAAAAL3QdgUAAIMYv8YrlQ8AAKAXKh8AANC3ssI5AABAZyQfAABALyQfAABALyQfAABAL0w4BwCAnlWS8qhdAACAblRrbegYNqqq7yf57tBxMPYWJ1k7dBAAWwk/E9ka7NFa22noILakqvpkJv/7GtLa1trhfV5wq0o+YGtQVStbawcNHQfA1sDPRGBL0nYFAAD0QvIBAAD0QvIBt3Xy0AEAbEX8TAS2GHM+AACAXqh8AAAAvZB8AAAAvZB8AAAAvZB8wEhVPaiqHl1Vd6uqBUPHAzA0PwuBLc2Ec0hSVU9L8idJLh+9ViY5rbX2w0EDAxhAVe3TWvv26P2C1tq6oWMCtg0qH4y9qrpbkmcm+Y3W2hOTfDTJsiSvqqp7DxocQM+q6qgkX62qDyRJa22dCgiwpUg+YNK9k+w9ev+RJGcnWZTkWVVVg0UF0KOqumeS45P8XpKbq+qfEgkIsOVIPhh7rbVbkrwjydOq6nGttfVJzk3y1SSPHTQ4gB611n6S5PlJPpDklUm2m5qADBkbsG2QfMCk/0ryqSTPqapDW2vrWmsfSLJbkocOGxpAf1prV7TWftxaW5vkRUm235CAVNXDq2rfYSME7soWDh0AbA1aazdW1fuTtCSvHv3P9aYkOye5ctDgAAbSWrumql6U5K1V9X9JFiR5/MBhAXdhkg8Yaa1dV1XvSXJBJv+278Ykz26tXT1sZADDaa2traqvJzkiyZNba2uGjgm46/KoXZjBaGJlG83/ABhbVXXfJGcmeUVr7etDxwPctUk+AIBZVdV2rbUbh44DuOuTfAAAAL3wtCsAAKAXkg8AAKAXkg8AAKAXkg8AAKAXkg+AJFW1rqq+WlXfrKoPVdU97sRYh1XV2aP3R1fVCbOce5+q+u07cI03VNUr57t/2jmnVdXTb8e19qz/v527CbWyiOM4/v2lvVjeXjYZRaBpWiF1VYwoeiHkUlRgQZAUIYnmXUQRCEG2CIIEdxHRK0QEUVCCECHVwlSuUdy8WfhGhasWtpEyI7B/izPC6aTdc0tOBd8PPPCcmXnmP/MsDvzPzJzkq6mOUZKkXiYfktRxtKqGq2oh8CuwtrsyHVP+zqyqzVW14S+anA9MOfmQJOn/yORDkv5sGzCv/eK/J8kLwDhwaZKRJGNJxtsKyUyAJLcl2ZtkO3DP8Y6SrEzyfLuflWRTkol2XQ9sAOa2VZeNrd26JJ8l+TLJ0119PZlkX5KPgAWTTSLJ6tbPRJJ3e1ZzliXZlmR/kjtb+2lJNnbFfvifvkhJkrqZfEhSlyTTgduB3a1oAfBGVS0CjgDrgWVVtRj4HHg8yVnAK8BdwI3ARSfp/jlga1VdAywGvgaeAL5pqy7rkowAlwPXAsPAkiQ3JVkC3AcsopPcLO1jOu9V1dIWbw+wqqtuNnAzcAfwYpvDKuBwVS1t/a9OMqePOJIk9WX6vz0ASfqPmJFkV7vfBrwGXAwcrKqdrfw64CpgRxKAM4Ax4Argu6o6AJDkTWDNCWLcCjwIUFXHgMNJLuhpM9KuL9rnmXSSkSFgU1X93GJs7mNOC5M8Q2dr10xgS1fdO1X1G3AgybdtDiPA1V3nQc5rsff3EUuSpEmZfEhSx9GqGu4uaAnGke4i4MOqWtHTbhioUzSOAM9W1Us9MR77GzFeB5ZX1USSlcAtXXW9fVWL/UhVdScpJJk9xbiSJJ2Q264kqX87gRuSzANIcnaS+cBeYE6Sua3dipM8/zEw2p6dluRc4Ec6qxrHbQEe6jpLckmSC4FPgLuTzEgyRGeL12SGgO+TnA7c31N3b5LT2pgvA/a12KOtPUnmJzmnjziSJPXFlQ9J6lNVHWorCG8lObMVr6+q/UnWAO8n+QHYDiw8QRePAi8nWQUcA0araizJjvZXth+0cx9XAmNt5eUn4IGqGk/yNrALOEhna9hkngI+be1388ckZx+wFZgFrK2qX5K8SucsyHg6wQ8By/t7O5IkTS5Vp2qngCRJkiSdnNuuJEmSJA2EyYckSZKkgTD5kCRJkjQQJh+SJEmSBsLkQ5IkSdJAmHxIkiRJGgiTD0mSJEkD8TsWKzp4lWuS7gAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 1080x720 with 2 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "##### PLOT CONFUSION MATRIX IN GREEN ######\n",
    "\n",
    "# Plot normalized confusion matrix\n",
    "plot_confusion_matrix(model_VAE_smart_mix, x_val = x_val_org, y_val=y_val_org, normalize = True, cmap=plt.cm.Greens)\n",
    "\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### CLASSIFICATION WITH SMART_AUG_MEL + VAE_AUG_NV +  ORG"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": [
    "############# ADD NV agumented to data ##############\n",
    "\n",
    "X_aug_smart, Y_aug_smart  = concat_train_label(X_aug_smart, x_nv_aug_VAE, Y_aug_smart, np.zeros(x_nv_aug_VAE.shape[0]))\n",
    "X_aug_smart, Y_aug_smart = shuffle_data(X_aug_smart, Y_aug_smart)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### CLASSIFICATION WITH SIMPLE AUGMENTATION -- CHECK FSCORE ---"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [],
   "source": [
    "#### MEL augmented data from keras\n",
    "\n",
    "x_train_mel = x_train[np.where(y_train == 1 )]\n",
    "y_train_mel = np.ones((x_train_mel.shape[0],))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "x_train_aug, y_train_aug = concat_train_label(x_train_org, x_train_mel, y_train_org, y_train_mel)\n",
    "x_train_aug, y_train_aug = shuffle_data(x_train_aug, y_train_aug)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "##### PARAMS FOR AUGMENTED DATASET #####\n",
    "## We add early stopping based on f score \n",
    "model_params_simple_aug = { \n",
    "                    \"x_train\": x_train_aug, \n",
    "                    \"y_train\": y_train_aug, \n",
    "                    \"x_val\": x_val_org,  # x_val \n",
    "                    \"y_val\": y_val_org, # y_val\n",
    "                    \"img_rows\": 128,\n",
    "                    \"img_cols\": 128, \n",
    "                    \"batch_size\": 32, \n",
    "                    \"epochs\": 10, \n",
    "                    \"learning_rate\": 1e-3, \n",
    "                    \"dropout\": 0.3,\n",
    "                    \"log\":False,\n",
    "                    \"threshold_acc\": 0.86, \n",
    "                    \"name_model\" : 'simple_aug', \n",
    "                    \"threshold_fscore\" : 0.8\n",
    "                    }\n",
    "\n",
    "csv_logger = CSVLogger('log_conv_strong_augOK3.csv', append=True, separator=';')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 6950 samples, validate on 1566 samples\n",
      "Epoch 1/10\n",
      "6950/6950 [==============================] - 47s 7ms/step - loss: 0.5232 - acc: 0.7679 - f1: 0.0092 - recall: 0.0143 - val_loss: 0.3917 - val_acc: 0.8557 - val_f1: 0.0000e+00 - val_recall: 0.0000e+00\n",
      "Epoch 2/10\n",
      "6950/6950 [==============================] - 45s 7ms/step - loss: 0.4330 - acc: 0.7721 - f1: 0.0060 - recall: 0.0042 - val_loss: 0.4125 - val_acc: 0.8557 - val_f1: 0.0000e+00 - val_recall: 0.0000e+00\n",
      "Epoch 3/10\n",
      "6950/6950 [==============================] - 45s 7ms/step - loss: 0.4123 - acc: 0.7755 - f1: 0.0449 - recall: 0.0348 - val_loss: 0.3491 - val_acc: 0.8557 - val_f1: 0.0000e+00 - val_recall: 0.0000e+00\n",
      "Epoch 4/10\n",
      "6950/6950 [==============================] - 46s 7ms/step - loss: 0.4000 - acc: 0.7928 - f1: 0.3171 - recall: 0.2737 - val_loss: 0.3941 - val_acc: 0.8550 - val_f1: 0.1823 - val_recall: 0.1357\n",
      "Epoch 5/10\n",
      "6950/6950 [==============================] - 45s 7ms/step - loss: 0.3900 - acc: 0.7993 - f1: 0.4354 - recall: 0.3978 - val_loss: 0.3555 - val_acc: 0.8442 - val_f1: 0.1917 - val_recall: 0.1504\n",
      "Epoch 6/10\n",
      "6950/6950 [==============================] - 46s 7ms/step - loss: 0.3802 - acc: 0.8118 - f1: 0.4599 - recall: 0.4110 - val_loss: 0.3796 - val_acc: 0.7708 - val_f1: 0.3811 - val_recall: 0.5197\n",
      "Epoch 7/10\n",
      "6950/6950 [==============================] - 45s 7ms/step - loss: 0.3707 - acc: 0.8160 - f1: 0.5159 - recall: 0.4792 - val_loss: 0.3868 - val_acc: 0.8123 - val_f1: 0.3764 - val_recall: 0.4116\n",
      "Epoch 8/10\n",
      "6950/6950 [==============================] - 45s 7ms/step - loss: 0.3591 - acc: 0.8227 - f1: 0.5135 - recall: 0.4706 - val_loss: 0.3614 - val_acc: 0.8429 - val_f1: 0.3589 - val_recall: 0.3366\n",
      "Epoch 9/10\n",
      "6950/6950 [==============================] - 45s 7ms/step - loss: 0.3593 - acc: 0.8268 - f1: 0.5399 - recall: 0.5095 - val_loss: 0.3308 - val_acc: 0.8563 - val_f1: 0.2760 - val_recall: 0.2248\n",
      "Epoch 10/10\n",
      "6950/6950 [==============================] - 45s 7ms/step - loss: 0.3470 - acc: 0.8371 - f1: 0.5646 - recall: 0.5344 - val_loss: 0.4855 - val_acc: 0.7778 - val_f1: 0.4009 - val_recall: 0.5410\n",
      "Final validation loss: 0.4854503223341848\n",
      "Final validation accuracy: 0.777777777473284\n"
     ]
    }
   ],
   "source": [
    "model_simple = evaluate_classifier_conv_custom_fscore(**model_params_simple_aug)"
   ]
  }
 ],
 "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": "3.5.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}