3-Small_CNNs.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Small Deep Neural Networks for colonoscopy polyps\n",
    "\n",
    "After extracting polyps and non-polyps images to create a dataset and split the dataset into train and validation subfolders, I shall try different neural networks from small CNNs to VGG/Inception nets using Transfer Learning. I will demonstrate that it is possible to obtain a small CNN classifier with over 90% accuracy in only 2 minutes of training (CPU i7, 16G RAM, GPU Nvidia Titan Xp).\n",
    "\n",
    "Let's import some packages for the next calculations:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "c:\\program files\\python35\\lib\\site-packages\\h5py\\__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import keras\n",
    "from keras import optimizers\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Conv2D, MaxPooling2D\n",
    "from keras.layers import Activation, Dropout, Flatten, Dense\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras import backend as K\n",
    "import numpy as np\n",
    "from matplotlib import pyplot as plt\n",
    "from IPython.display import clear_output\n",
    "\n",
    "from sklearn.metrics import roc_auc_score\n",
    "from numpy.random import seed\n",
    "from tensorflow import set_random_seed\n",
    "\n",
    "import time, os\n",
    "from __future__ import with_statement\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is a function that will help you to visualize the training and validation accuracy during the training inside the browser (source: [live_loss_plot_keras.ipynb](https://gist.github.com/stared/dfb4dfaf6d9a8501cd1cc8b8cb806d2e)).\n",
    "\n",
    "**Note**: if you want to run the training script not inside the notebook, you need to remove this function call from your traning! If not, after each epoch you will need to manually close the new plot!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [],
   "source": [
    "class PlotLosses(keras.callbacks.Callback):\n",
    "    # Show the live training and validation loss during the training\n",
    "    def on_train_begin(self, logs={}):\n",
    "        self.i = 0\n",
    "        self.x = []\n",
    "        self.acc = []\n",
    "        self.val_acc = []\n",
    "        self.fig = plt.figure()\n",
    "        self.logs = []\n",
    "\n",
    "    def on_epoch_end(self, epoch, logs={}):\n",
    "        self.logs.append(logs)\n",
    "        self.x.append(self.i)\n",
    "        self.acc.append(logs.get('acc'))\n",
    "        self.val_acc.append(logs.get('val_acc'))\n",
    "        self.i += 1\n",
    "        clear_output(wait=True)\n",
    "        plt.plot(self.x, self.acc, label=\"ACC_Train\")\n",
    "        plt.plot(self.x, self.val_acc, label=\"ACC_Val\")\n",
    "        plt.legend(bbox_to_anchor=(0., 1.02, 1., .102), loc=3,\n",
    "                   ncol=2, mode=\"expand\", borderaxespad=0.)\n",
    "        plt.grid()\n",
    "        # plt.ylim(0.0, 1.0) # limited Y values to 1.0!\n",
    "        plt.show()\n",
    "\n",
    "plot_losses = PlotLosses() # initialize the class"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Define several constants:\n",
    "* dimensions for our images to be used in the models\n",
    "* folders for train and validation subsets\n",
    "* number of train and validation samples"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Dimensions of our images.\n",
    "img_width, img_height = 150, 150\n",
    "\n",
    "# Train & validation folders\n",
    "train_data_dir      = 'data_polyps/train'\n",
    "validation_data_dir = 'data_polyps/validation'\n",
    "\n",
    "# Folder to save the models\n",
    "modelFolder = 'saved_models'\n",
    "\n",
    "# Train parameters\n",
    "nb_train_samples      = 910\n",
    "nb_validation_samples = 302"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Change the order of the dimensions and channel depending on the input image: (channels, width, height) or (width, height, channels). We are using 3 channels for this study (RGB)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "if K.image_data_format() == 'channels_first':\n",
    "    input_shape = (3, img_width, img_height)\n",
    "else:\n",
    "    input_shape = (img_width, img_height, 3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Small CNNs: 1 Conv + 1 FC\n",
    "\n",
    "We start our calculations using a CNN with a convolution and a fully connected layers. The next section is a function for this type of CNN with 5 parameters: number of filters, dimension of the convolution, epochs, batch size and size of the inputs. You could personalize this function with other parameters such as dropout rate, activation function, etc.\n",
    "\n",
    "`CNN_1Conv_1FC` will have the following characteristics explained into the code:\n",
    "* Uses 2 types of seed for the maximum reproductibility: one for numpy, other for tensorflow; remember that each GPU could still involve some random data.\n",
    "* Conv layer is using `relu` activation function and `MaxPooling2D`.\n",
    "* FC layer is using the same `relu` activation and a `dropout = 0.5`.\n",
    "* The output layer is using `sigmoid` activation for polyps / non-polyps classification.\n",
    "* The loss functions is using the `binary cross entropy` optimization with `rmsprop` algorithm with `accuracy` metrics.\n",
    "* `ImageDataGenerator` form keras is using to create augmented images from the original using different transformations such as flipping, rotation, zoom, etc. I choose these transformations because I consider to be important for the current colonoscopy polyps model. You could add other transformation!\n",
    "* The fitting of the model is using a custom callbacks class to plot the loss and accuracy during the training (`plot_losses`).\n",
    "* The weights and the full model are saved as files.\n",
    "* Show the final trainign and test lose.\n",
    "* Show the live plot during the training similar with Tensorboard."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "def CNN_1Conv_1FC(convs, dims, epochs, batch_size, input_shape):\n",
    "    # CNN with 1 Conv + 1 FC\n",
    "    # convs = no of filters\n",
    "    # dims = dimension of convolution dims x dims\n",
    "    # epochs = training epochs\n",
    "    # batch_size = training batch size\n",
    "    # input_shape = shape of the inputs\n",
    "    \n",
    "    # for reproductibility\n",
    "    #seed(74)            # numpy seed\n",
    "    #set_random_seed(74) # tensorflow seed\n",
    "    \n",
    "    # Start the computational graph for our CNN\n",
    "    model = Sequential()\n",
    "    \n",
    "    # CNN filters\n",
    "    model.add(Conv2D(convs, (dims, dims), input_shape=input_shape))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    # Convert convolution to a fully connected layer\n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(convs))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Dropout(0.5))\n",
    "    model.add(Dense(1))\n",
    "    model.add(Activation('sigmoid'))\n",
    "    \n",
    "    # Compile the model\n",
    "    model.compile(loss='binary_crossentropy',\n",
    "                  optimizer='rmsprop',\n",
    "                  metrics=['accuracy'])\n",
    "\n",
    "    # Aaugmentation configuration for training set\n",
    "    train_datagen = ImageDataGenerator(\n",
    "        rescale=1. / 255,\n",
    "        shear_range=0.2,\n",
    "        zoom_range =0.2,\n",
    "        horizontal_flip=True,\n",
    "        vertical_flip=True,\n",
    "        rotation_range=90)\n",
    "\n",
    "    # Augmentation configuration for testing set: only rescaling!\n",
    "    test_datagen = ImageDataGenerator(rescale=1. / 255)\n",
    "    \n",
    "    train_generator = train_datagen.flow_from_directory(\n",
    "        train_data_dir,\n",
    "        target_size=(img_width, img_height),\n",
    "        batch_size=batch_size,\n",
    "        class_mode='binary')\n",
    "\n",
    "    validation_generator = test_datagen.flow_from_directory(\n",
    "        validation_data_dir,\n",
    "        target_size=(img_width, img_height),\n",
    "        batch_size=batch_size,\n",
    "        class_mode='binary')\n",
    "\n",
    "    model.fit_generator(\n",
    "        train_generator,\n",
    "        steps_per_epoch=nb_train_samples // batch_size,\n",
    "        epochs=epochs,\n",
    "        validation_data=validation_generator,\n",
    "        validation_steps=nb_validation_samples // batch_size,\n",
    "        callbacks=[plot_losses], # using the plotting function for train and validation loss\n",
    "        verbose=0)\n",
    "\n",
    "    # Evaluate final test loss and accuracy scores\n",
    "    score = model.evaluate_generator(validation_generator, nb_validation_samples//batch_size, workers=7)\n",
    "    print('Validation/Test loss    :', score[0])\n",
    "    print('Validation/Test accuracy:', score[1])\n",
    "    \n",
    "    return model # return the model!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Search for a CNN with 32 filter of 3 x 3 pixels:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAEQCAYAAABLMTQcAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAIABJREFUeJzt3Xlc1VX+x/HXYRdkERBFQVFAcd93M01Nc23RMi1bx/acaZtqZmqypuY3TU3WWDNmi1qmZuU+We5b7ooLbiguKIiCoqCs9/z+OCCIF7jAReTyeT4ePuB+7/d+7zkC73u+53vO+SqtNUIIIRyLU1UXQAghhP1JuAshhAOScBdCCAck4S6EEA5Iwl0IIRyQhLsQQjggl6p64+3btwe5uLhMA1ojHzJCCFEaC7A3Jyfn8U6dOiWVtnOVhbuLi8u0+vXrt6hbt+55JycnGWwvhBAlsFgs6uzZsy0TExOnASNK278qW8yt69ate1GCXQghSufk5KTr1q2biuntKH3/Si5Pie8twS6EELbLy0ybclv6uoUQwgHV+HCfMWOGn1Kq086dOz3yt+3evdv91ltvjWjUqFHrpk2bthoyZEjTkydPugCsWrXKs3Pnzs3DwsJaN2nSpNV9993X+NKlS9f9P06ePDkgKiqqZVRUVEtXV9eOzZo1axkVFdXy6aefbmhr2WJjY12HDh3a1D41FUJYU1kZcPDgQbd69eq1zc3NvWZ7VFRUy1WrVnkWV56PP/44YPz48Y0qWq8qu6B6s5g9e7Z/x44d02bOnOnfoUOH05cvX1bDhw+PfO+9906OHTs2FWDRokXeiYmJLgDjxo0LnzFjxtEBAwakWywWpk+fXufChQtO3t7elsLHnThxYvLEiROTARo2bNhmzZo1h4KDg3OKvn92djaurq5WyxYREZG9ZMmSo3avtBDiqsrKgObNm2cFBwdn/fzzz7WHDh2aBrBz506P9PR0p379+l2u7HrV6JZ7amqq07Zt22p/9dVXx3766ac6AFOnTvXv2LFjWv4PFWD48OGXunTpkvHBBx8E3XvvvckDBgxIB3BycuKRRx45Hxoael1ol+T5559vMHbs2MY9e/aMHD16dJN9+/a5d+rUqXmLFi1atmrVqsXKlSu9APbu3eseFRXVEuDDDz8MHDx4cNPevXtHNm7cuPUzzzxj8xmAEMK6ys6AUaNGpcyaNcs///HMmTP977rrrhSAWbNm+bZt2zaqRYsWLXv27Nks/8zAXm6KlvvL86JDDyVeKvY0pTya1fe+/P6odidL2ufbb7/169u3b2rbtm0z/fz8ctevX++5d+/eWh07drT6qRoTE1Nr/PjxyfYo3549ezw3b958wNPTU1+6dMlp3bp1hzw9PfXOnTs9HnroobDdu3cfKPqa/fv3e+7atSvG1dVVR0REtHn55ZeTwsLCsu1RHiGq1PxnQkmKsWsGENTyMndOqdIMGD9+fEqnTp1aZmdnn3B1dWX+/Pl15syZcwRg4MCBaWPGjDng5OTEhx9+GDhp0qT6n3/+eXzZKlm8myLcq8rcuXP9J06cmARwzz33pMycOdO/tNfYy5AhQ857enpqgIyMDPXYY4813r9/v6ezs7M+efKku7XX9O7d+2KdOnUsAE2bNr1y5MgRNwl3IcqvsjOgUaNGOZGRkRkLFy70CQ4OznZxcdFdunTJAIiLi3O78847Q86ePeualZXlFBoammnP974pwr20FnZlSExMdN60aZPPoUOHaj377LPk5uYqpZR+5ZVXEtauXVvb2mtatGhxZdu2bZ4PPPDAhYq+v5eX19X+ubfffrteSEhI1vz58+OysrKUt7d3B2uvcXd3vzp01NnZmZycHFXRcghxUyilhV0ZblQGjB49OuW7777zDwoKyr7nnntS8rc/++yzjSZOnJg4bty41MWLF3tPmjSpgT3qla/G9rnPnDmzzt133518+vTpPadOndqTmJi4OyQkJKtFixYZ27dvrz179mzf/H3nzZvns2XLllovvfRS0ty5cwPy+8QBPv30U/8TJ05U6EMyNTXVOTg4ONvJyYkpU6YEyN2xhKh8NyoDHnzwwfOrVq3yXbBggf9DDz10NdwvXbrk3KhRo2yAr7/+OsDe9aux4f79998H3H333ecLbxs5cuT5mTNn+i9YsCB2ypQpQY0bN24dHh7e6uuvvw4MDg7ODg0NzZkxY8bRl19+OSQsLKx106ZNW61fv947v6ukvF544YWkmTNnBrZr1y7q+PHjbm5ubpLuQlSyG5UBgYGBue3bt08LDAzMjoqKysrf/qc//en0/fffH96pU6fmAQEBZRqUYQtVVa3E6OjoY+3atTtXJW8uhBDVVHR0dGC7du3CStuvxrbchRDCkd0UF1Sru8mTJwd89tln9Qpv69KlS9rMmTNPVFWZhBA3zs2YAdItI4QQ1Uh16JaxWCwWGconhBA2ystMmwZwVGW47z179qyvBLwQQpQu72YdvsBeW/Yvtc9dKfUlMAxI0lpft0i8UkoBk4EhwGXgYa31jtKOm5OT83hiYuK0xMREuc2eEEKU7upt9mzZudQ+d6VUHyANmFFMuA8BnsOEezdgsta6W1lLLYQQwn5KbTFrrdcCKSXsMhIT/FprvQnwU0oF26uAQgghys4e3SENgcLrQsTnbRNCCFFF7DHO3doFUat9PUqpCcAEgFq1anUKDQ0t1xtaLBacnGpeN31NrTfU3LpLvWsWW+p96NChc1rruqUdyx7hHg8UTukQ4LS1HbXWU4GpAJ07d9bbtm0r1xuuXr2avn37luu11VlNrTfU3LpLvWsWW+qtlDpuy7Hs8dG4EBivjO5AqtY6wQ7HFUIIUU62DIX8DugLBCql4oE3AVcArfV/gKWYkTKxmKGQj1RWYYUQQtim1HDXWt9fyvMaeMZuJRJCCFFhNe+KhRBC1AAS7kII4YAk3IUQwgFJuAshhAOScBdCCAck4S6EEA5Iwl0IIRyQhLsQQjggCXchhHBAEu5CCOGAJNyFEMIBSbgLIYQDknAXQggHJOEuhBAOSMJdCCEckIS7EEI4IAl3IYRwQBLuQgjhgCTchRDCAUm4CyGEA5JwF0IIByThLoQQDkjCXQghHJCEuxBCOCAJdyGEcEAS7kII4YAk3IUQ1cPpXXB8I2hd1SWpFlyqugBCCFGq88fg62GQdQnqt4Huz0Dre8DFrapLdtOSlrsQ4uaWmwM/TgClYNB7kJsN85+Ej1rDmvchPbmqS3hTkpa7EOLmtv5fcHIz3D0N2o6G7k/BkZWw6VNY9Q6s+yc0HwJBLSAgHAIiwD8c3GtXdcmrlIS7EOLmFb8dVr8HrUeZYAfTgo/ob/4lHYDNn8HhX2Hfj9e+tnZ9qNcK+rwMjXtUrByb/wvZV6D3723b/+xBWPMP6P8XqBNWsfcuJwl3UXNs+gzO7IWRU6q6JMIWmWnw4+Pg0wCGfmB9n6AoGD7ZfJ91Gc7HQXIsJB8x/46shK8GQ5t7YeAk8AkuezkO/wr/e8V87xsCbUaVXu45D8C5QxC/BR5eCn6hZX/fCpI+d1Ez7JkHP78KO7+BKxequjQ1x5GVMLWfCd6yWvYapMTBXf+BWn6l7+/maVrqLUfCLS/AnVPguW1wy0sQMx8+6WS6eHIybS/DpTPw05MQ1ApCu8OiiXDucPH7aw1LXjAfMIPeM79r04fDxQTb39NOJNyF4zu5BeY/Dd4NzONT26u2PDXJ1i/g9A44ta1MLws8uwl2zDDdIGG9y//+bl6ma+SZzdD0Vlj+V/i0h2mNl8ZigZ8mQFY6jPoSRn8FLu4w96HiP6x2zIDdc6Dva9DjaXjgR0g/CzNGQFpS+etRDhLuwrGdPwbf3W9O7R/9H6AgvmxBI8opMw1il5vvT262/XWXEml+8N8Q3A76vm6fsvg3hfu/g3E/mD77b0fBwudLbsVv/BiOroY7/m66f3wawN1TISkG/vfy9fsn7oGlL0PTfnDLi2ZbaBcY9z2kxsOMkTd0ZI9N4a6UGqyUOqiUilVKvWrl+UZKqVVKqZ1Kqd1KqSH2L6oQZZSRCrPuA0s2jJ1rLmwFtTD9oKLyxS6HnAxwdjdnT7Za8CxOlkwzOsbe49gjB8BTv0HvF2DHdPhysAneouK3w8q3TRdPx4cKtkcMgD4vme69XbMKtmdcNC16T3+4+3Nwci54rnFPuH82pByFmSPhynn71qkYpYa7UsoZmALcAbQE7ldKtSyy25+BuVrrDsAY4FN7F1SIMsnNge8fNn2f986Eus3M9pDOpuVusVTO+yYfgXOxlXPs0lxMMO9/s9i/EDwDoc1oE+62/J9fOAGxv3K88eiCn1k5ZWTnsv7wOXTRGa0ubjDgTbjvW9N//t8+cHRNoRdehB8eBe9gc7FWqWtf3/c1CLsFFr8ASftNP/ui581Z4qgvoXbd6wvT9FbzfmcPwsy7zXtUMlta7l2BWK31Ua11FjAbGFlkHw345H3vC5y2XxGFKCOtzeiGIyth6IfmDytfSFfIuAAplRCC52Lh89vgP73hwBL7H78085+Cz3pC3Nob/95FZWfAoWUQNdQMQ8y4AMklXIjMF7cOgOSAbhUuwluLYnjgi818v91KyxygxTCYsAq86sLMO2H9RwUXRC+chHumQa0617/Oydk85+5tWusbP4Z9P5m+/cY9iy9Q5AC4dwYk7obtX1W4fqWxJdwbAicLPY7P21bYX4EHlFLxwFLgObuUrrrIyYJN/4HMS1Vdkoq5cNKMKqnuNv8Htn0BPZ+HTg9d+1xIF/M1fqt93/NyCswaDU4upn92zgOw5XP7vkdJcrLgxG+mD3nWfXD8txv33tYcXQVZadBiBITmBbUt/e5xa8EzkHSvRhV6+23HUvhuywk8XJ3425L9JF3KsL5jYCQ8vsKUc/mb8Hk/2PO9aZ036l78G3jXNwGffBh+fQMib4eeE0svWPM7zPv1qPyItGWcu7KyrejKPfcDX2utP1BK9QBmKqVaa62vOQ9TSk0AJgDUq1eP1atXl6PIkJaWVu7XVobg07/Q/NAUDsfGcipkWKW9T5nqrTUh8QtJ9W3BJR/bTm+bHplOo5M/su6MJ7kunuUvaCWwte4B57bSeu/fOBfYjX2u/aDoa7SF3s6enNmygMMXGtilbMqSTbvoN/C5eJJd7d8hrXYYLTP/SeDSlzixZyNHmz4IqnxjF2ytt0/qfjrmZHAo8klC4hfhNuMuotu9xSWf5uV634qK2j+VQGcvNpzU6Ph4erl4c27LAg5eLCG0tabHgV9J9W1BWnp6uf/GcyyaNzdeIcBD8VwHN97ZlMEzX6zmmfYexb+o7sOEhNch/Mh0Un1bs8vS8frfHSs8gh+iUfJaYus+gGWtbWdMWmvUIev72jXbtNYl/gN6AMsKPX4NeK3IPvuA0EKPjwJBJR23U6dOurxWrVpV7tfancWi9SddtH7TR+tvRlXqW5Wp3mveN2X68QnbXzP3YfOaUzvKXLbKZlPdE3Zr/U6w1v/po3VmWvH7TR+p9We97FMwi8X8H7/po/Xu7wu252Rrvej3Zvu8x7TOzijX4W3+ma/9wLxX2lmtU09p/VE7rd8N1frUznK9b4XkZGn9XiOtf5hQsO3be7X+pHPJrzt7WOs3fXT6hs/1ipUry/32/155WDf+42K9PCZRa631JysO6cZ/XKyX7U0o/cXnYrXOTC91tz3xF/QjX23Rjf+4WDf+42I95r+/6cTUK6W+LvrkeT3ik3V605FzVp+35ecNbNOl5LbW2qZuma1ApFKqiVLKDXPBdGGRfU4A/QGUUi0AD+BsBT93qofYFXDuIPg2Mv2F2cWc/t1I+34yV/rB+kiA4uTvezNdlLPVpUTTHVHLz4xMcPMqft+QLnBmnxm/XFHrPoDo78yQvcIzF51dTH9//zfNaf4391Tu5KnjG6FuFHgFmiF7Dy0CD1/Tl5y4t/Le15pj60wfe8sRBdtCu5oZm5dTrtk19Uo26w+fY8qqWL6dPROAOxYp/rz+Cgt2nSLXUrblfY+dS+fjFYcZ0qY+/VvUA+CJW8OJqu/NXxbs5WJGdskHCAg3k6GKcSDxIk/M3MawT9az/fh5Xh7UnPfubsOukxe4Y/I6Vh44Y/V1FzOyeXPBXkZO2cDp1AwuZ+WWqV7lUWq3jNY6Ryn1LLAMcAa+1FrvU0pNwnyCLAReBD5XSv0B02XzcN4njOP77RNzVX3wezBnHBzfYNa8qCrx28yMutBuUDsIEnbb/tqr4V5Foz3KKyvdBPuVC/Doz8VOMf96QxwHz6TxXusuoC1wemfJE2Qup5gRN/5NIGKguTDr7l3w/N4fzYdo2/vg1leuf71SZqakT0NY8Izpz+39BzMV3rWELgJg3eGzfLb6CP3r5tK3tPpbcuHEpoK1V8BMd39oIXw1xIyvfmQp1LWti+ZKVi5fboije1N/OjX2NyOP4rdC7K9m3Hfk7dD3uhHRBWIWgqsXhN9WsC2v3z354HrW6I5sPprClmMpxJ0r+ID9ymsLKS5BjOp9C7M3xTJx9i4+XnGY526LZHi7Bjg7WeshLqC15s/z9+Lm7MSbw1td3e7q7MTf72nL3Z9u4B8/H+CdO9sUW+/pvx0j6WImXu7O1HJzxtPVGU83FzzcnPk15gyLd5+mtpsLvx8QyaO9m+Dj4QpAlzB/np21g0e/3sbjvZvwyuAo3Fyc0FqzMPo07yzZz7m0TMZ3b8yLg5pffV1lsmltGa31UsyF0sLb3ij0fQzQy75FqwYS95pf9gF/Nb/Izu6mJV9V4X7hhJmwU7sejJkFv/0bDv5s/vgLj7u1JjcbLuVNkbZHuOd/thcdRmZvFgv89AQkRJtJKsFtre72vz0J/HVRDAATurShCZjheSWF+57vIW6N+cDc/jU4uZqLbJEDwa+RGZ3SqAeM+KTkera7z3zg/Pw6LHwOlr8FXR6DLo+bD+BCziae4Jf53+B/aiWTnQ7xUfIz5J0UFy9xt1nnvHGRP0H/JqYF//UQsxb6A/PMxKASHE9O54mZ20lOPMFx5124BR6mdcYOVOZFUM7g29As5NW4JzTpc/0BLLlwYDE0ux1ca5GZk8uS3QlsO+zKJJz47od5/DPHGR8PF7o28WdUpxDahvjStoEPvlOeg8hBPDegGa2cT5ERGMXHKw7z+zkm5J+9LYIR7Rrg4my9w2HBrtOsjz3H2yNbUc/n2g/P9qF+PNKrCV+sj2NEu4Z0beJ/zfOrDibxxoK9nEy5gpebM5ezc6+7J4inmzNP3RrOhD5N8fO8dvx9RFBt5j/Ti/eW7mfa+jg2x6XwyuDm/HfNUdbHnqNNQ1++eKgzbUNsWEbBTmThsIrY9Cm4ekKnh82pXFgv07rh3RtfloyLpvWakwkPLzan576hZgJP2hlzql6Si6e5ep28ouGuNUzrD/XbwvCPKnas0qx4C/YvMut4NL/D6i77TqfywtxoWgT7cCDxIosOZfJ8QETpM1V3z4F6beB3K+HkJjNlPXa5GR0B6DphqPu+NVPSS9OkDzy5znRZ/PapWTFw/b/MGPBWd2OJ30bKrkXUTd3LOOBSrSDccnPokb6SA4lPEVXfp/hjH99ovlobhhcYYQJ+5t2mFX/vjGIbH6sOJDFx9g7GsIxXPWfiZMkm4YI/C5y6ENxlKF1vuxvl7GaGes5/Bp7eeO2ZDJgziPSz0GIEWmtemBPNkj0J+NZyZYJ7JPfUOUW/0b2Jqu9zbUs8cS9cTr76geGkFEPaBDO4VX1+iUlk8opYXpgbzQe/HGJst0bc1yWUwNoF/+8XLmfx9uIY2of6MbZbY6v1e/H2Zizbl8irP+5m6fO34OHqzJmLGUxaFMOSPQmE1/Vi9oTudG8agNaazBwLl7NyuZyVw+WsXOp5e+DrWXyL28PVmbdGtqZXRCAvz9vNg19swdvdhUkjWzGuW+NSzzzsTZYfKK9LibB7LnR4oGAsbMQA0694/viNLUtuDsx71EyQuHd6wem3b95KdBdOFv/afPldMgGRps+9Ir1qCbvM+i3bvzaTPCrLjhmw4SPo/JhZ49uKc2mZTJixHd9arkx/pAtdGvuzZHeC6XeP31p8PZOPmDq0vddMemnSB25/G57+jeTf7eRNp2d53vUtUlUJoVuUUuY4Y2fDc9vNzMd9P8G398Cav3PifAZzfB4i/r5f8H7tELS+h35OO/l6zYGSj3t8I9RpUvwHeN3m8Phys8+se2Hnt9c8bbFoPlp+iInTV/Op62Re50ucwm+DJzeQMmEnXwW+yH3r6jPum4McTdVw53/gYjws+9P177V/Ibh4QOTtfLE+jiV7Enjp9mbs/MtAwtr3Izgthlb1PK8Puvyx+U1uuWazk5NicOtgljzXm6kPdqJxgCfvLztIj/dW8Nx3O9kSl4LWmveWHuDClWzevatNsSHq6ebCu3e14ehZ0y8/feMx+n+whl/3n+HFgc1YOvEWujcNyPtRKTxcnfH3ciOkjifN6nmXGOyF3d6qPv+beAsvD2rOihdvZXyPsBse7CDhXn5bPgdLzrWhEjHQfM1fT+NGWfaaOWMY+gGE9yvYnr/MaGoZwr3prZB50bS+ymv39+DsZlp1K98p/3FKcvBnWPwH0x12xz+sdotk5Vh46pvtnEvLZOr4TgT5eDC0bTAHz1wiyacNpCeZriyrdZgLKKvLu761NpVZmb34+ZQbd322gePJ5bgwGxAOQ//JgtuW81j2ywx0msbxuxZw7x8mE9KiGyiFe5s78VKZXNjzM0kXi7lQb7GYcC/aJVOUT7Dpdw/rDQueNmcOWpN6OZvHZ2xj5YqfWVX7DXrlboGBb5uL0vVb06qhHz8+1ZO372zNnlOpDP5oHdOO10X3eM5M3y+8AJfFYs6iwvuz+VQm7/3vAINa1eOZfhE4OSlzUTX7sll2uai4teYGG74hVovv5KS4vVV9Zv2uO8tfuJUHujdm9cEk7v3vb/T/cA1ztp3k8d5NaNmg5A/bPs3qcnfHhny6+ghvLtxHh0Z+/PL7PjzXPxJ3l1K6LsuggV8tnukXQZBPyddWKpOEe3lkXTaTZKKGmgWJ8gVGmlEzsStuXFmOrIQtU6HHs9D5kWufy/9DsSnc8/Zpkjebs7xdM5Zc2DvPXHTr8azpf7X3Koxxa2HueHMvzdHTzciUIrTWvLFgL1uPnef90e2u9nXe0bo+SsGvl/LGW1ubzKS16ZJp0ue61vDqg0ksjD7N030jmPlYN5LTsrjr041sO5Zy/XFKsTD6NH9YeIzMJgP54aUR3NUhBFX4Q6pJHzKda3O72syM34o5Gzx3EK6klDwzMp+HDxfvmUVy+N2w6m/snDKekZ+somnsdH7yeAt/T2fUIz9Dr+fBqSAanJ0UD3ZvzIoXb6Vv87q8s2Q/b14aia4bBQueLRgBc3oHXDxFapM7eGbWThr7e/L+6HYFdbo6manIOjO5OWYggrU+fCsigmrz5vBWbH69P/93Txu83FxoXs+biQMibXr9X4a2ZGibYCaPac+MR7sSFljCyKpqTMK9PKJnmcV/ehaZZaaUmWIct8bMGKxsFotZwtS3EfR/4/rn3b3Bw8+24ZCp8eAZYAITyh/ucWtMH3/be82Sp54BsOLt8h3Lmvht5qKxf1OznKqH9Zba9I3HmL31JM/0C2dEu4KADvLxoGuYPzOOeIFLLev97vHbzE0f2t53zeYrWbn8ZcFemtb14ul+4XRvGsBPT/fEx8OFsdM2s2DXKZur8fPeRP4wZxedw/z5fHzn6y7QAeDsyvnArgx23cnsTbFcsTZ87vgG87WYcE9Oy+SL9XE8+MVmur+7grZvr6bTvnv4OOdOOpxbyA+ZT/Jnl5k4N7sd9cQ6s4phMYK8PfjPA5148tZwZmxN5C3n59GXz8H//mh22L8Q7eTC89uDSM/M4T8Pdrp2VIhviBk5VHSmakK0OVss0iVTGk83F+7r0ohFz/Vm2R/64Olm2yXEOl5uTBnXkZHtG177YepgJNzLymIxF8QadipoiRQWMcBMuz65qfLLEvOT+cPo93rxF/X8Qm3vc/cNMaNAnFyvC/fLWTm2lWn3XHD3hchB5sOl9wtmKnremiFltWL/GR79eivHL+aai27f3GMuFo+fb1bgs2L94XO8vWQ/A1rU48WB1w//G9Y2mINnr3C5blvrLffdc0y/cYvh12z+aMUhTqZc4d272lw9hW9atzY/Pd2L9iF+TJy9i09WHL5+oaoiVh1I4rnvdtAuxJcvH+5CLbfiuwPO1u2BpyWdlpnRzNth5UP62AazTn2hW7ll51r4NeYME2Zso9u7K3h7cQxJFzPpGR7AK4ObM218F0b+4VMsQz8iwN0Cg941o6uK+f8szMlJ8eodUfztrtbMPFGHb93vgz1zIWYBxCzkSO1OrDmZw9/vaUOzet7XHyC06/Ut97i8RbvCyhbuomQyWqasDv1sFp0a9aX14W9N+phwPPyrzaeZ5ZKbbfqzg1qaVnJxfENtu8CbGm/6gZ2cTau40ESmL9fH8bel+3lzeEvG9wgr/hhZl02fa6u7CsZxd3kMfptixoM/uszmoZEnUy7z1qJ9LN+fhFKQ4JLAuP1P4OLqCeMXmLU9rNgQe44JM7cRXteLj8a0N329RQxqXZ83F+5jn1NzuiR8Zyae5Zc3Nxv2/mBuuFzorCDm9EWmrYvj3s4hVy+65avj5cbMx7vy6g97+ODXQ2w5lsLYro24rUXQdf246w+f44lvthNV34evHulKbfeS/wTP12mPdqvNOPdd/H19T8Z1bVRQJ61Nf3tYb1CKkymXmb7xGPN3neJcWhaBtd14pFcYozqF0ry+laANeAQ6P1yu4arjujWmgV8tJn5roZPzJpr/+CROOZeZlt2fR3qFMbJ90eWn8oR2MxeRU0+ZYZVgutmCWl43LFRUTM1tuedmwy9/hrXvmz4/W/02xQRmi6ILY+Zx9zZjoSu7333HDLM+dP83Sx7D7htaep+71maf/D76gIir4b4o+jSTFsfg4+HCGwv28fWGuOKPc3CpOWsp3J3hWstM8Dm5GQ7/Umq1MrJzmbz8MAM+XMPGI8m8PiSKVb+L4Evn97h0JZO4Id8We8PhX/Yl8shXWwmt48mLdaFGAAAgAElEQVQ3j3UrNjiDvD3o1iSARckNzVDRxEITvWJXmD7sQnXItWhe+2kPfrVceX1IC6vHdHdx5sN72/H6kCgOJF7iqW930OWd5bz24x62HjMjOrbEpfD4jK00DfRixqNd8a1V+ugLi7Mbqtkg+uqtHD93iRUHCt3NJ+UopCVCWC82HjnH0I/XMf23Y3Ru7M+08Z357bX+/GloS+vBnq8C3RL9mgcx64lbeMv5WbKzs8nViqQG/Yv9PwJMyx0K1tTPyTTDJyuzIVRD1cyWe8ZFc0Hu6CrzOHYl3PN5sVfqr9q/CI6vh9v/ZvUi3lWRA81Y6MKtE3vKSoc1/2cm0DQbVPK+viGmPzMj1UxHtyYj1YTy1XAPh9jlbDx8hhfnRtO1iT9fPNSZF+ZG89dFMVg0PNq7yfXH2T3X9KkWHbnR4QHYMNn0vUcMvOZiXT6tNasPnuWvi/ZxPPkyw9oG86ehLQj2yIGpfclyvszj6i32zUtm9oRLRBY55f9xRzwvz9tN64a+TH+ki/U+7EKGtg1m8vwQJnlgumbyQ2f3HKjlf81Y8G82HSf65AU+uq99icdVSjGhTziP9W7Khthz/LTzFPN3nuK7LScI9a9FSloWDf1q8c3j3ajjVYabULQYjvveHxjiHce0dYEMbGmm1eePb/8lLZxnftpCWIAXXz7chVD/G7foW+uGvnz47P18PPUEfhmneO/B23AtZpIRYOY+uNQyXTOt7jLXN3KuSLhXgpoX7hcTzNKsZ2Jg5BTThbLkBTMxY+SnEGXlJlJnYuDXv5ghjgER0PHBkt8jYoAJ9yMroON4+9dh02fmouW9M0pvefkVGutev5hwz7/gWrjlnpvJW9/8QlhgIz5/sDPeHq5MGduR577bwaTFMWjgscIBn37O1LfHM6Rm5LIg+gSnLxQM32vp/zAjjrzJ/FlT2OVzG8npWaSkZ5KclkVyehbn07PIsWjC63rx7ePd6BURaF64fxEkx3Kw9Z9589axjJm6ifs/38zsCd2JCKoNwIzfjvHGgn30DA9g6vjOpXZ1AAxuXZ83FtQh1T0Y3/x+94yL5uyjw4PgbFrViakZvL/sILdEBjKyvW2rSDo7Kfo0q0ufZnV5584clu1L5Kedp/Cr5ca0hzpfM/nGJhEDwcWDp+rtZWhsOHviU2kT4os+voErLn5M+PkSPcMD+eyBTjadDdhbA79a/P7FN8jMsZT+f+/sCg07FlxUjVtrVswsbSinKLOaFe5JB8y9E6+ch3FzTQiDuTvPvEdg9v3Q7UkYOMlcoLx0Blb9DXbONN0tt/8Nuv6u9BmJQS3NRa7Dv9o/3C+nmFZwsztKXm86X/5EptR4qN/a+j5Xw93sm+QeQhAQ5ZbEq4+Oujp5w83FiX+P7chzs3by9uIYtNY8fkveUNB9P4Elh3+d6cB/31tORrYFN2enqwtGOxFJc+dQ2h3+N39V4fh51bo6QaRdiB8Btd0IC/Dizg4NcXMp1PJLiAblzPk6bWlTtzbf/a4bY6ZuYuznm5jzRA+W7kng/WUHGdCiHv8e2wEPV9vGKgfWdqdneCDbE8PpF7/VFHP/InNbuEJdMm8u3Et2roV37mxdrpEVXu4u3N0xhLs7lnJWWBL32hAxgBbxa/B2v4tp64/y/qh2XIpZzZbMSEZ1CuXdu9pc+/92g7k6O5XcYi8stCts/ASyr5hwD25nFnwTdlVzwv3YBhPeLh7w8BJo0L7guYBweOxX+PVN2PyZOd1tNsiMisnNMoHf52WbRhMApjUd0d+MIMjNvtoKtIv1H5qbglgb+miNrw0TmfKf8w3hfHoWT//vIvOAP3VzJci31jW7ujo78cnYDjz/3U7eWbKf7FxNkLc7rX6ZhrY0YurBWtzVoQHjujWmdcMiZwoHgNlj2TXinO0fegnRUDcKi7P5QI0I8mbW77ozZuomhn28jvSsXO5s34D3R7ezPVzyDG0bzPq4JtyWs96c0e2eY2ZxhnRGa82Hvx5i2b4zvDK4OY0DqngsdIsROB1YzAstLvHObs2VcyeYmp2AV+RY3h/VtnoN6QvtBpZ/mb/J+K1myKywu5pxQXXvj2bp09r1TIgXDvZ8Lu7mLudjvjNht/Z9E9DPbDYrPtoa7PkiB5q+7jLe8WdD7DneW7qfM1ZmJLpnnIXNU6HdGKhX9Da2xfCqa2aLlhju8eDkSpqrP49N38ruVHdyXWsTlGV9fLyrsxMf39+BoW2C+b+fDzB53i9E5RwgrfldbHq9P+/d3fb6YAczAqVBR9Nqs1VC9HWLXTWr5823j3fDp5YrD/cM48N725c52AEGtapPNHkTX/YvMq3Itvdi0eYWbZ+sjGVMl1Ce6BNe5mPbXbNB4OTKKM8dAHgnmguSfQaOrF7BDuZWhwAbJ5sL2tLfXikcv+WekQo/TjD9fPfPLj2ko4ZAw82mT7uYFQZt0uRWs4pe7HLbZg8CO06c57HpW8nItjD9t2M80qsJT/YJv9otEnZsNqDNuHZbOTmZvvRixrofT04n98gBajsF0vudFWTnWvhsXEecN0SUOJHJ1dmJyWPa06lxHQYkbUHvVnQZNgFK6vNVyqzxvfyvpnuptJ/FpcS8n0M7KPJZ1yLYh42v3lahYPP3csOnSUey4l1wXf0eCk1u69G8+sNuvt8ez+O9m/CnoS1ujvCs5QdN++J9dCmfjp1Ipz2L4Jgv1Cumq+1m5hVgruvErTW3JWzUo6pL5JAcP9wT95jWQZ9XbG99e9cz/yqilp85/Tz8q01dKEfOpvHY11up5+PBv+5rz4yNx/jPmiN8u+k4T/WN4JHIK9RPXAndnzQTjcrA4hNCTspxDp9OJSU9i5T0LKJPprL6YBJHz6XzvdtRUl0DeKBbY4a2DaZT4zpwILzUVRNdnJ14tFcY/HuxGWtd2mgjKHQP021mWdiSJESbr8HtIC7zuqftEbqD2zVm74kwOl6JxdKgE8//coklexL4w4BmPN8/4uYI9nwtR8DC5xgUkATJ28w1l9KWcr5ZhXYzjYeQLiXfWEWUm+OH+9WAqEArvLwiB8CKSebCbAkfFkkXM3joyy04KcX0R8xaFx0b1WFCn3D++ctB3v85ht5rJhHu5ElSyydpZNFWJ+cA5ORa2HnyAqsOJLHm0FlOpFzmLzmKPs5HGfrx+qv7ubk40aNpAON7NKbDb2m4hN9Kh+GFunoCIkx3Vk5myReQT+8wf6S9bLg5MECDDmZ0RPxWG8NdmQvBcXZenybP7S3rM39hJB2JZdaV7iw5msCfh7YouFB8M2k+FNRE2PqFWX20/biqLlH5hXaFXd9Kl0wlqhnh7t2gama/ReSFe+xy6GD9D/FSRjYPf7WVlPQsZk/ofs0iRi0b+PDlw104sfgfNNp2iImZT7Pg0714exygTUNf2ob40TbEl8ig2uyOT2XVwSTWHjrLxYwcXJwUnRrX4Z6OITQ+14x6J9by37GtqePtjb+XGw39aplp77k5sDzx+lZ3QASgISUOgqKKr+Pu781NSlqMKH6fwty8oF4r265FJESbchRdM9yO6ni5kRDcnyOJu/hXYhv+fncbxnQt25nRDeMVYIYM7jS3o6vWwwfD+5vJaLb+3ogyqwHhvrtqWu1gJmz4NYZtX0L7sdeNSc/KsfDkN9s5dOYS04q7S8u5WBrt+gDdbDCdaw+gZ3Ak0fGp7IlP5Yv1R8nOLVjHpK63O4Nb16df8yB6RQYWLNq0sz2c0AwKsYB/ka6pSwnmlnPXhXveRcTk2OLDPTfHrADZbFDZhrKFdIE988w6PVYmNF2VEG19/R4763DLMO6YE8IHY9oxvJ1tY9mrTMuR5oYfrp6l3lXppuYXChOjq7oUDs2xwz3rslkStWUVtQ6Ugt6/N+uOH119zVrrFovm5XnRbIhN5p+j29G3uZUzC4sFFj4LLu6oYR8RuuMgfbs04r68buvMnFwOJFzi0JlLtAj2oWWwj/XumsLDIf2LzCwtOoEpn39euKeUcLPs4xvMuu9W1jwvUUhX84F37iAEFTNVPT3ZlLfrhLIduxzuaBNM/xb1qnScuM2ihsHSl8wHpEsZZrmKGqca/DZXwJl9plValS2c9uPMDbTXvn91k9aad5bsZ8Gu07w8qDmjOhVzIXLLVDjxGwz+u9WbPru7ONMu1I/RnUNp3dC32H74gnXdrQxtLDKB6apafmYYZUlL/+5faFqQ+TcpsdXVi6oldM0k7DJfb9DPrloEO5jfg9v+YtZcF6IE1eQ3upzyA6J+FXXLgLkY2WuiaeUe20CuRfOn+Xv5ckMcj/QK4+m+xYyhTjlqhgxG3g7t7q9YGfLD3dpwyPzx7z5W1sAptIDYdfLvuhM50Nw/tiwCws068yWGexVeCL/Z9XmpYHa1EMVw7HBP3G0WgbJliF5l6vgQeNXFsvafvDB3F7M2n+CpvuG8Mayl9aF2FgsseM7MbB32UYVW7gPMB0ztetYnMqXGm3vAute+/rmA8OJb7ifz5gKU54KYUnn3MC1hqGVCtLlekX9/WiFEmTh2uOfPbqzqscpunmR3ewanoys5Fr2WVwY354+Do4ofQ73tC7P65KB37beqZHFL/+bfpMOagAgT4BkXr39u/0IzSqa0VSmLE9LF3Dzb2rHB/OyszSQWQtjEccM9J8us5ngTjChIz8zhif3tOK9r8++QFTzdN6L4nc8fM2vchPc3S+Xai29I8X3uRfvb8wXklbPoRVWtIWahuTl1eYcphnQGtBknX9SVC+Y2dzfBz06I6spxw/3sATMz9Qb02aZn5jBpUQzvLzvAd1tOsO7wWeLOpZOZk0vq5Wwe+GIza45nkBD1MKFn15jhmdYkH4EZd5pJPsMn2/eMwy/UBHnRW8CV1nLPL1dhp3bAxXgzLK+8GnYyX631uyfuMV8l3IUoN8cdCnn1glzln9p/ujqWLzfE4eykyLVcG561XJ3JtWimjO1Iy/Ce8NEMWPdPsxZ7YfHbYNa9Jnwf/LFgHXZ78Q01y9mmn4Padc22jFTITC0+3Os0AdT1/e4x882aIM0Hl788tfwgsDmctBLu+T+7+hLuQpSXY4e7m3deQFWekymX+XxdHHd1aMj7o9qSeDGDU+evEJ/3L/FiBiPbNyi472bXCbDuA7O2fP7koANLYd6j5r6gD/xQMIHInq6OdT9REO6pp/KeKybcXT3Mh0zhcNfa9Lc3ubXiFztDu5i6a33tWUpCtBm9k19OIUSZOXa4129T8gxIO3h36X6cleKPg6NwcXYipI4nIXU8KXZeZfenzZ2U1n1gbu23dRosfdmcYYydW3mBVng4ZH6XSHFj3AvzLzJiJnGPuS7Q+w8VL1NIF9j5jRn2WfgDzcoyv0KIsnHMPndLLpzZW+kB8duRZP63N5Gn+4ZT39fDthd5BUCXR820/YXPwZIXIXIQPLy4cluqfoXuyJSv0E06ipU/1j2/rz5mgbkmEDWs4mUqvEJkvqx0syiWhLsQFeKY4Z4cC9mXKzUgci2aSYtjaOhXi9/1KeMKgj2eM/du3TEDOj8K931T+cueeviBW+1rh0Omxpu+89olLG8cEGFuOpJ+1jzev9AsWOUVWPEy1Y0yZSp8UTVxL6Al3IWoIMfslim8DnglmbP1JPsTLjJlbEeb79t5lXc9Mxom+7IJ9xsxDl+pvLHuhVvu8eDToOQ1wQuPmLmcYlrV9lrvxcnZ3ESlcLjfgJ+dEDWB44a7iwcENquUw6deyeafvxyka5g/Q9rUL99B2ldwSYHy8A2BCycKHpc0xj1f4dUhL54GlH26ZPKFdDE3/M66bJYxSIg2a9p4X7+WjhDCdo7ZLZMQbdYMdy7fZ9fk5Ye5c8oGlu5JwFJkaCPAJysOc/5yFm8ML2b5gJuVn5WWe2lLM/g1Ml1IybGmSya0m9VFzMotpAtYcgrWAbpZZhULUc05XrhrnbeGe/lO61cdTOJfyw9xJCmNp7/dwR2T17F49+mrIX/kbBpfbzzGfXkrMVYrvqFwJcVctLTkwsVTpYe7kzP4NzW3Czyz1/7LJzfsbL7Gb4XsDDi7X7pkhLADm8JdKTVYKXVQKRWrlHq1mH3uVUrFKKX2KaVm2beYZXD+mJmYU46VIJMuZvDS3Gii6nuz+U/9mTymPTkWC8/O2smgj9ayMPo07yyOwcPVmRdvb27/slc230IjZi4lgs61bVG1gAhI2me+bzHcvmWqXdfMRYjfCkkxphUv4S5EhZXab6GUcgamAAOBeGCrUmqh1jqm0D6RwGtAL631eaVUFdzTLk85L8jlWjS/n7OLy1m5/HtsBzzdXBjZviHD2jZgyZ4EPllxmOe/2wnAa3dEUde7hPuK3qzyh0NeOFmwJkxpfe5Q0O/eoGOZb85tk5AuELf2hq/hLoQjs6VTuisQq7U+CqCUmg2MBGIK7fM7YIrW+jyA1jrJ3gW1WeJuM7wvqGXp+xbynzVH2HgkmX/c05aIoILFsJydFCPaNWBYm2CW7k1g14kLPNwrzM6FvkGu3rSjcLjb2HKHyrujVUgX2DPXzFb18DVL/QohKsSWcG8IFF4rNh6um4DZDEAptQFwBv6qtf7ZLiUsq4RoM37atWBS0fHkdAJru+Plbr2624+n8OGvhxjergGjO1sPOycnxbC2DRjW9ia/x2ZJvINBOeeFu4/ZZu0mHUU1ucXMam1zb+WUKySv3z12uXkvuZgqRIXZEu7W/tKKDiFxASKBvkAIsE4p1VprfeGaAyk1AZgAUK9ePVavXl3W8gKQlpZm/bVa0/P4VpIDOnEw7/mzly28uu4K7s5wWyNXBjRywc+j4FJDerbmjQ1X8HeHOwIvsGbNmnKV6UYott5l0N0tgAuHtpPr7EmQixcbNllZcteayDdg52HgcIXe3xplyaa3kxvOlixO5Phz1Eod7VH36kjqXbPYs962hHs8ULhjNgQ4bWWfTVrrbCBOKXUQE/bXLPmntZ4KTAXo3Lmz7tu3b7kKvXr1aqy+9mICrEkluMNggruZ5yctikGpY/RuFsTS/Wf45Xgud3ZowO9uaUpEUG2e/nYHqVlXmPdUT9qH+pWrPDdKsfUui7hI6uss8KgFuU0qfjx7iesIJzfRqOtwGrXpe93Tdql7NST1rlnsWW9bwn0rEKmUagKcAsYAY4vsMx+4H/haKRWI6aY5apcSlkWRi6kXM7KZs/UEw9oG89GYDhxPTmfauji+336SudviaRviy+74VF4fEnXTB7vd+IbA8Y2Qeanqbz9YWEhnOLlJLqYKYSelhrvWOkcp9SywDNOf/qXWep9SahKwTWu9MO+525VSMUAu8LLWOrkyC25VQjSgoF5rAOZsOUl6Vi6P9TZrvzQO8OLtO1vzh4HNmPnbcWb8doz+UUE83ruMa8NUZ76hZqZpRio0Knbtyhuvy2Nm/Zv8i7dCiAqxaQqn1nopsLTItjcKfa+BF/L+VZ2EaBMO7rXJybXw1YY4ujXxp03ItZON/L3cmDggkuduM0Hi5FSDLuD5hpjx7ZkXb66Wu39TuPXlqi6FEA7DsWaoJhbMTP3f3kROp2bw+C3Ft8qdnFTNCna49g5PtoxxF0JUS44T7unJZohfcFu01kxbd5SwAE/6R1XdfKqbUuFAv5la7kIIu3KccD++wXxt0IHtx88THZ/KY72b1LyWeWkKB7qEuxAOy3HCfdcsqF0fGvVk2ro4fGu5ck8nCa/ruHlBLX8zmal2OZcrFkLc9Bwj3C+dgcO/QLsxHL+QybKYRMZ1a4Snm2MuV19hfqHmJh3lXBJZCHHzc4y/7t2zzQiQDg/w1YZjuDgpHuoZVtWlunk1G2zuqiSEcFjVP9y1hp3fQGg3Ur3CmLttBcPbNqCej403rK6J+r1e1SUQQlSy6t8tE7/N3Nez/ThmbznB5axcHu3dpKpLJYQQVar6h/vOmeDqSWbUCL7eeIweTQOq3x2ShBDCzqp3uGddRu/9kRP1BzBgyk4SUjOYcGsNWkpACCGKUW3DXWvN3uUzUVmXeDm2HT4erkx/tCv9msukJSGEqJYXVPcn5/LxZxt5KeFrTrnU58Ex9zOkTQOZsCSEEHmqXct9yqpY/m9rBpw/Rk/nGOrf+hjD2jWUYBdCiEKqXbgPalWf+5q7MbdbHKBw7lB0aXkhhBDVLtwjgmpzR5gzLntmQ3g/WR9FCCGsqHbhDlDn/G6zAmT7cVVdFCGEuClVy3Cvn7gCPHwhalhVF0UIIW5K1S/cr1wg8NwmaDMaXGWJASGEsKb6hfveH3C2ZEGHB6q6JEIIcdOqfuFevw0nQ+6E4PZVXRIhhLhpVb9JTKFdORJxmVAl49qFEKI41a/lLoQQolQS7kII4YAk3IUQwgFJuAshhAOScBdCCAck4S6EEA5Iwl0IIRyQhLsQQjggCXchhHBAEu5CCOGAJNyFEMIBSbgLIYQDknAXQggHJOEuhBAOyKZwV0oNVkodVErFKqVeLWG/UUoprZTqbL8iCiGEKKtSw10p5QxMAe4AWgL3K6VaWtnPG3ge2GzvQgohhCgbW1ruXYFYrfVRrXUWMBsYaWW/t4F/ABl2LJ8QQohysCXcGwInCz2Oz9t2lVKqAxCqtV5sx7IJIYQoJ1tus2ftfnb66pNKOQH/Ah4u9UBKTQAmANSrV4/Vq1fbVMii0tLSyv3a6qym1htqbt2l3jWLXeuttS7xH9ADWFbo8WvAa4Ue+wLngGN5/zKA00Dnko7bqVMnXV6rVq0q92urs5pab61rbt2l3jWLLfUGtulScltrbVO3zFYgUinVRCnlBowBFhb6cEjVWgdqrcO01mHAJmCE1nqbPT58hBBClF2p4a61zgGeBZYB+4G5Wut9SqlJSqkRlV1AIYQQZWdLnzta66XA0iLb3ihm374VL5YQQoiKkBmqQgjhgCTchRDCAUm4CyGEA5JwF0IIByThLoQQDkjCXQghHJCEuxBCOCAJdyGEcEAS7kII4YAk3IUQwgFJuAshhAOScBdCCAck4S6EEA5Iwl0IIRyQhLsQQjggCXchhHBAEu5CCOGAJNyFEMIBSbgLIYQDknAXQggHJOEuhBAOSMJdCCEckIS7EEI4IAl3IYRwQBLuQgjhgCTchRDCAUm4CyGEA5JwF0IIByThLoQQDkjCXQghHJCEuxBCOCAJdyGEcEAS7kII4YAk3IUQwgHZFO5KqcFKqYNKqVil1KtWnn9BKRWjlNqtlFqhlGps/6IKIYSwVanhrpRyBqYAdwAtgfuVUi2L7LYT6Ky1bgvMA/5h74IKIYSwnS0t965ArNb6qNY6C5gNjCy8g9Z6ldb6ct7DTUCIfYsphBCiLGwJ94bAyUKP4/O2Fecx4H8VKZQQQoiKcbFhH2Vlm7a6o1IPAJ2BW4t5fgIwAaBevXqsXr3atlIWkZaWVu7XVmc1td5Qc+su9a5Z7FlvW8I9Hggt9DgEOF10J6XUAOBPwK1a60xrB9JaTwWmAnTu3Fn37du3rOUFYPXq1ZT3tdVZTa031Ny6S71rFnvW25Zuma1ApFKqiVLKDRgDLCy8g1KqA/BfYITWOskuJRNCCFFupYa71joHeBZYBuwH5mqt9ymlJimlRuTt9j5QG/heKbVLKbWwmMMJIYS4AWzplkFrvRRYWmTbG4W+H2DncgkhhKgAmaEqhBAOSMJdCCEckIS7EEI4IAl3IYRwQBLuQgjhgCTchRDCAUm4CyGEA5JwF0IIByThLoQQDkjCXQghHJCEuxBCOCAJdyGEcEAS7kII4YAk3IUQwgFJuAshhAOScBdCCAck4S6EEA5Iwl0IIRyQhLsQQjggCXchhHBAEu5CCOGAJNyFEMIBSbgLIYQDknAXQggHJOEuhBAOSMJdCCEckIS7EEI4IAl3IYRwQBLuQgjhgCTchRDCAUm4CyGEA5JwF0IIByThLoQQDkjCXQghHJCEuxBCOCCbwl0pNVgpdVApFauUetXK8+5KqTl5z29WSoXZu6BCCCFsV2q4K6WcgSnAHUBL4H6lVMsiuz0GnNdaRwD/Av7P3gUVQghhO1ta7l2BWK31Ua11FjAbGFlkn5HA9Lzv5wH9lVLKfsUUQghRFraEe0PgZKHH8XnbrO6jtc4BUoEAexRQCCFE2bnYsI+1Frguxz4opSYAE/IepimlDtrw/tYEAufK+drqrKbWG2pu3aXeNYst9W5sy4FsCfd4ILTQ4xDgdDH7xCulXABfIKXogbTWU4GpthSsJEqpbVrrzhU9TnVTU+sNNbfuUu+axZ71tqVbZisQqZRqopRyA8YAC4vssxB4KO/7UcBKrfV1LXchhBA3Rqktd611jlLqWWAZ4Ax8qbXep5SaBGzTWi8EvgBmKqViMS32MZVZaCGEECWzpVsGrfVSYGmRbW8U+j4DGG3fopWowl071VRNrTfU3LpLvWsWu9VbSe+JEEI4Hll+QAghHFC1C/fSlkJwFEqpL5VSSUqpvYW2+SulflVKHc77Wqcqy1gZlFKhSqlVSqn9Sql9SqmJedsduu5KKQ+l1BalVHRevd/K294kb0mPw3lLfLhVdVkrg1LKWSm1Uym1OO+xw9dbKXVMKbVHKbVLKbUtb5vdfs+rVbjbuBSCo/gaGFxk26vACq11JLAi77GjyQFe1Fq3ALoDz+T9jB297pnAbVrrdkB7YLBSqjtmKY9/5dX7PGapD0c0Edhf6HFNqXc/rXX7QsMf7fZ7Xq3CHduWQnAIWuu1XD9XoPAyD9OBO29ooW4ArXWC1npH3veXMH/wDXHwumsjLe+ha94/DdyGWdIDHLDeAEqpEGAoMC3vsaIG1LsYdvs9r27hbstSCI6sntY6AUwIAkFVXJ5Klbe6aAdgMzWg7nldE7uAJOBX4AhwIW9JD3Dc3/ePgFcAS97jAGpGvTXwi1Jqe97sfbDj77lNQyFvIjYtcyCqP6VUbeAH4Pda64s1YR06rXUu0F4p5Qf8BLSwttuNLVXlUkoNA5K01uvECv4AAAGFSURBVNuVUn3zN1vZ1aHqnaeX1vq0UioI+FUpdcCeB69uLXdblkJwZGeUUsEAeV+Tqrg8lUIp5YoJ9m+11j/mba4RdQfQWl8AVmOuOfjlLekBjvn73gsYoZQ6hulmvQ3Tknf0eqO1Pp33NQnzYd4VO/6eV7dwt2UpBEdWeJmHh4AFVViWSpHX3/oFsF9r/WGhpxy67kqpunktdpRStYABmOsNqzBLeoAD1ltr/ZrWOkRrHYb5e16ptR6Hg9dbKeWllPLO/x64HdiLHX/Pq90kJqXUEMwne/5SCH+r4iJVCqXUd0BfzCpxZ4A3gfnAXKARcAIYrbW+boG26kwp1RtYB+yhoA/2dUy/u8PWXSnVFnMBzRnT6JqrtZ6klGqKadH6AzuBB7TWmVVX0sqT1y3zktZ6mKPXO69+P+U9dAFmaa3/ppQKwE6/59Uu3IUQQpSuunXLCCGEsIGEuxBCOCAJdyGEcEAS7kII4YAk3IUQwgFJuAshhAOScBdCCAck4S6EEA7o/wFaw1cwDpz6BwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation/Test loss    : 0.4046796038746834\n",
      "Validation/Test accuracy: 0.82421875\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_3 (Conv2D)            (None, 148, 148, 32)      896       \n",
      "_________________________________________________________________\n",
      "activation_7 (Activation)    (None, 148, 148, 32)      0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_3 (MaxPooling2 (None, 74, 74, 32)        0         \n",
      "_________________________________________________________________\n",
      "flatten_3 (Flatten)          (None, 175232)            0         \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 32)                5607456   \n",
      "_________________________________________________________________\n",
      "activation_8 (Activation)    (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dropout_3 (Dropout)          (None, 32)                0         \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 1)                 33        \n",
      "_________________________________________________________________\n",
      "activation_9 (Activation)    (None, 1)                 0         \n",
      "=================================================================\n",
      "Total params: 5,608,385\n",
      "Trainable params: 5,608,385\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "---> Save model ...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "print('--> Training Conv-FC ver 1 ...')\n",
    "model = CNN_1Conv_1FC(convs=32, dims=3, epochs=50, batch_size=64, input_shape= input_shape)\n",
    "print(model.summary())\n",
    "\n",
    "# Save the weights and the full model\n",
    "print('---> Save model ...')\n",
    "model.save_weights(os.path.join(modelFolder,'model1_1_Conv-FC_weigths.h5'))\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Conclusion**: with a `Conv + FC topology with 32 filters of 3 x 3 pixels`, we are able to obtain in `50 epochs` a classifier with over `80% test accuracy` for the polyps vs non-polyps prediction! Remember, we are not using any transfer learning, we are training a simple CNN. We are training `5,608,385 parameters`!\n",
    "\n",
    "Let's check a second CNN but with filter of `5 x 5` pixels:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation/Test loss    : 0.5873111039400101\n",
      "Validation/Test accuracy: 0.7421875\n",
      "---> Save model ...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "print('--> Training Conv-FC ver 2 ...')\n",
    "model = CNN_1Conv_1FC(convs=32, dims=5, epochs=50, batch_size=64, input_shape= input_shape)\n",
    "# Save the weights and the full model\n",
    "print('---> Save model ...')\n",
    "model.save_weights(os.path.join(modelFolder,'model1_2_Conv-FC_weigths.h5'))\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Conclusion**: in the case of the `Conv + FC topology, 32 filters 5 x 5` we can obtain less than 80% accuracy in 50 epochs (`5,459,393 parameters`). This means bigger filters are not able to find the best patterns for this classification.\n",
    "\n",
    "Maybe it could improve with the increasing of the number of filters 3 x 3. Let's test with 64 filters (double than the previous ones):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation/Test loss    : 0.4850814938545227\n",
      "Validation/Test accuracy: 0.82421875\n",
      "---> Save model ...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "print('--> Training Conv-FC ver 3 ...')\n",
    "model = CNN_1Conv_1FC(convs=64, dims=3, epochs=100, batch_size=64, input_shape= input_shape)\n",
    "# Save the weights and the full model\n",
    "print('---> Save model ...')\n",
    "model.save_weights(os.path.join(modelFolder,'model1_3_Conv-FC_weigths.h5'))\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's try `128` filters with `3 x 3` pixels (double than previous calculation and 4 times greater than the first model):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation/Test loss    : 0.3926382213830948\n",
      "Validation/Test accuracy: 0.859375\n",
      "---> Save model ...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "print('--> Training Conv-FC ver 4 ...')\n",
    "model = CNN_1Conv_1FC(convs=128, dims=3, epochs=100, batch_size=64, input_shape= input_shape)\n",
    "# Save the weights and the full model\n",
    "print('---> Save model ...')\n",
    "model.save_weights(os.path.join(modelFolder,'model1_4_Conv-FC_weigths.h5'))\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Conclusions**: Only few % of ACC improvement was obtained by using a double/triple quantity of filters 3 x 3 but using `22,431,617` / `89,722,625` parameters compared with `5,608,385` params for 32 filters.\n",
    "\n",
    "In the next step, we add extra convolutional layers to the CNN.\n",
    "\n",
    "### Small CNNs: 2 Conv + 1 FC\n",
    "\n",
    "We continue our study by adding an extra Conv layer. The main idea is to try to have an extra layer that could find more general patters from the basic ones in the first Conv layer. Thus, we define a second CNN topology with **Conv-Conv-FC** (`CNN_2Conv_1FC`). I will use `adam` optimizer for faster calculations."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [],
   "source": [
    "def CNN_2Conv_1FC(convs, dims, epochs, batch_size, input_shape):\n",
    "    # CNN with 2 Conv + 1 FC\n",
    "    # convs = no of filters\n",
    "    # dims = dimension of convolution dims x dims\n",
    "    # epochs = training epochs\n",
    "    # batch_size = training batch size\n",
    "    # input_shape = shape of the inputs\n",
    "    \n",
    "    # for reproductibility\n",
    "    #seed(1)            # numpy seed\n",
    "    #set_random_seed(2) # tensorflow seed\n",
    "    \n",
    "    # Start the computational graph for our CNN\n",
    "    model = Sequential()\n",
    "    \n",
    "    # Conv 1 filters\n",
    "    model.add(Conv2D(convs, (dims, dims), input_shape=input_shape))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    \n",
    "    # Conv 2 filters\n",
    "    model.add(Conv2D(convs*2, (dims, dims))) # second Conv has 2x filters!\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    # Convert convolution to a fully connected layer\n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(convs*2)) # FC has 2x neurons!\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Dropout(0.5))\n",
    "    model.add(Dense(1))\n",
    "    model.add(Activation('sigmoid'))\n",
    "    \n",
    "    # Compile the model\n",
    "    model.compile(loss='binary_crossentropy',\n",
    "                  optimizer= 'adam', # 'rmsprop',\n",
    "                  metrics=['accuracy'])\n",
    "\n",
    "    # Aaugmentation configuration for training set\n",
    "    train_datagen = ImageDataGenerator(\n",
    "        rescale=1. / 255,\n",
    "        shear_range=0.2,\n",
    "        zoom_range =0.2,\n",
    "        horizontal_flip=True,\n",
    "        vertical_flip=True,\n",
    "        rotation_range=90)\n",
    "\n",
    "    # Augmentation configuration for testing set: only rescaling!\n",
    "    test_datagen = ImageDataGenerator(rescale=1. / 255)\n",
    "    \n",
    "    train_generator = train_datagen.flow_from_directory(\n",
    "        train_data_dir,\n",
    "        target_size=(img_width, img_height),\n",
    "        batch_size=batch_size,\n",
    "        class_mode='binary')\n",
    "\n",
    "    validation_generator = test_datagen.flow_from_directory(\n",
    "        validation_data_dir,\n",
    "        target_size=(img_width, img_height),\n",
    "        batch_size=batch_size,\n",
    "        class_mode='binary')\n",
    "\n",
    "    model.fit_generator(\n",
    "        train_generator,\n",
    "        steps_per_epoch=nb_train_samples // batch_size,\n",
    "        epochs=epochs,\n",
    "        validation_data=validation_generator,\n",
    "        validation_steps=nb_validation_samples // batch_size,\n",
    "        callbacks=[plot_losses], # using the plotting function for train and validation loss\n",
    "        verbose=0)\n",
    "\n",
    "    # Evaluate final test loss and accuracy scores\n",
    "    score = model.evaluate_generator(validation_generator, nb_validation_samples//batch_size, workers=7)\n",
    "    print('Validation/Test loss    :', score[0])\n",
    "    print('Validation/Test accuracy:', score[1])\n",
    "    \n",
    "    return model # return the model!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's see the classifier for this topology using `32 filters - 64 filters - 64 FC`. All filters are 3 x 3 pixels."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation/Test loss    : 0.35756915807724\n",
      "Validation/Test accuracy: 0.8671875\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_19 (Conv2D)           (None, 148, 148, 32)      896       \n",
      "_________________________________________________________________\n",
      "activation_53 (Activation)   (None, 148, 148, 32)      0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_19 (MaxPooling (None, 74, 74, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_20 (Conv2D)           (None, 72, 72, 64)        18496     \n",
      "_________________________________________________________________\n",
      "activation_54 (Activation)   (None, 72, 72, 64)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_20 (MaxPooling (None, 36, 36, 64)        0         \n",
      "_________________________________________________________________\n",
      "flatten_18 (Flatten)         (None, 82944)             0         \n",
      "_________________________________________________________________\n",
      "dense_35 (Dense)             (None, 64)                5308480   \n",
      "_________________________________________________________________\n",
      "activation_55 (Activation)   (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dropout_18 (Dropout)         (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_36 (Dense)             (None, 1)                 65        \n",
      "_________________________________________________________________\n",
      "activation_56 (Activation)   (None, 1)                 0         \n",
      "=================================================================\n",
      "Total params: 5,327,937\n",
      "Trainable params: 5,327,937\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "---> Save model ...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "print('--> Training Conv-Conv-FC ver 1 ...')\n",
    "model = CNN_2Conv_1FC(convs=32, dims=3, epochs=50, batch_size=64, input_shape= input_shape)\n",
    "print(model.summary())\n",
    "\n",
    "# Save the weights and the full model\n",
    "print('---> Save model ...')\n",
    "model.save_weights(os.path.join(modelFolder,'model2_1_Conv-Conv-FC_weigths.h5'))\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Conclusions**: By adding the second Conv layer we can see accuracy improvment to over 85% using `5,327,937 parameters`.\n",
    "\n",
    "In the next step, we shall try the same topology Conv 64 - Conv 128  - FC (with a double number of filters):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation/Test loss    : 0.413276731967926\n",
      "Validation/Test accuracy: 0.8359375\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_21 (Conv2D)           (None, 148, 148, 64)      1792      \n",
      "_________________________________________________________________\n",
      "activation_57 (Activation)   (None, 148, 148, 64)      0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_21 (MaxPooling (None, 74, 74, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_22 (Conv2D)           (None, 72, 72, 128)       73856     \n",
      "_________________________________________________________________\n",
      "activation_58 (Activation)   (None, 72, 72, 128)       0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_22 (MaxPooling (None, 36, 36, 128)       0         \n",
      "_________________________________________________________________\n",
      "flatten_19 (Flatten)         (None, 165888)            0         \n",
      "_________________________________________________________________\n",
      "dense_37 (Dense)             (None, 128)               21233792  \n",
      "_________________________________________________________________\n",
      "activation_59 (Activation)   (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dropout_19 (Dropout)         (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_38 (Dense)             (None, 1)                 129       \n",
      "_________________________________________________________________\n",
      "activation_60 (Activation)   (None, 1)                 0         \n",
      "=================================================================\n",
      "Total params: 21,309,569\n",
      "Trainable params: 21,309,569\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "---> Save model ...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "print('--> Training Conv-Conv-FC ver 2 ...')\n",
    "model = CNN_2Conv_1FC(convs=64, dims=3, epochs=50, batch_size=64, input_shape= input_shape)\n",
    "print(model.summary())\n",
    "\n",
    "# Save the weights and the full model\n",
    "print('---> Save model ...')\n",
    "model.save_weights(os.path.join(modelFolder,'model2_2_Conv-Conv-FC_weigths.h5'))\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Conclusions**: By doubling the number of filters for `Conv-Conv-FC` with `21,309,569 parameters` no better results are obtained.\n",
    "\n",
    "### Small CNNs: 3 Conv - 1FC\n",
    "\n",
    "In the next step, we shall add the 3rd Conv layer to obtain a similar topology with the LeCun net: `Conv-Conv-Conv-FC`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "def CNN_3Conv_1FC(convs, dims, epochs, batch_size, input_shape):\n",
    "    # CNN with 3 Conv + 1 FC\n",
    "    # convs = no of filters\n",
    "    # dims = dimension of convolution dims x dims\n",
    "    # epochs = training epochs\n",
    "    # batch_size = training batch size\n",
    "    # input_shape = shape of the inputs\n",
    "    \n",
    "    # for reproductibility\n",
    "    seed(1)            # numpy seed\n",
    "    set_random_seed(2) # tensorflow seed\n",
    "    \n",
    "    # Start the computational graph for our CNN\n",
    "    model = Sequential()\n",
    "    \n",
    "    # Conv 1 filters\n",
    "    model.add(Conv2D(convs, (dims, dims), input_shape=input_shape))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    \n",
    "    # Conv 2 filters\n",
    "    model.add(Conv2D(convs, (dims, dims), input_shape=input_shape))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    \n",
    "    # Conv 3 filters\n",
    "    model.add(Conv2D(convs*2, (dims, dims))) # second Conv has 2x filters!\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    # Convert convolution to a fully connected layer\n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(convs*2)) # FC has 2x neurons!\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Dropout(0.5))\n",
    "    model.add(Dense(1))\n",
    "    model.add(Activation('sigmoid'))\n",
    "    \n",
    "    # Compile the model\n",
    "    model.compile(loss='binary_crossentropy',\n",
    "                  optimizer='adam',\n",
    "                  metrics=['accuracy'])\n",
    "\n",
    "    # Aaugmentation configuration for training set\n",
    "    train_datagen = ImageDataGenerator(\n",
    "        rescale=1. / 255,\n",
    "        shear_range=0.2,\n",
    "        zoom_range =0.2,\n",
    "        horizontal_flip=True,\n",
    "        vertical_flip=True,\n",
    "        rotation_range=90)\n",
    "\n",
    "    # Augmentation configuration for testing set: only rescaling!\n",
    "    test_datagen = ImageDataGenerator(rescale=1. / 255)\n",
    "    \n",
    "    train_generator = train_datagen.flow_from_directory(\n",
    "        train_data_dir,\n",
    "        target_size=(img_width, img_height),\n",
    "        batch_size=batch_size,\n",
    "        class_mode='binary')\n",
    "\n",
    "    validation_generator = test_datagen.flow_from_directory(\n",
    "        validation_data_dir,\n",
    "        target_size=(img_width, img_height),\n",
    "        batch_size=batch_size,\n",
    "        class_mode='binary')\n",
    "\n",
    "    model.fit_generator(\n",
    "        train_generator,\n",
    "        steps_per_epoch=nb_train_samples // batch_size,\n",
    "        epochs=epochs,\n",
    "        validation_data=validation_generator,\n",
    "        validation_steps=nb_validation_samples // batch_size,\n",
    "        callbacks=[plot_losses], # using the plotting function for train and validation loss\n",
    "        verbose=0)\n",
    "\n",
    "    # Evaluate final test loss and accuracy scores\n",
    "    score = model.evaluate_generator(validation_generator, nb_validation_samples//batch_size, workers=7)\n",
    "    print('Validation/Test loss    :', score[0])\n",
    "    print('Validation/Test accuracy:', score[1])\n",
    "    \n",
    "    return model # return the model!"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's try `Conv-Conv-Conv-FC` with 32, 32 and 64 filters (3 x 3):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation/Test loss    : 0.33883558958768845\n",
      "Validation/Test accuracy: 0.87890625\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_25 (Conv2D)           (None, 148, 148, 32)      896       \n",
      "_________________________________________________________________\n",
      "activation_65 (Activation)   (None, 148, 148, 32)      0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_25 (MaxPooling (None, 74, 74, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_26 (Conv2D)           (None, 72, 72, 32)        9248      \n",
      "_________________________________________________________________\n",
      "activation_66 (Activation)   (None, 72, 72, 32)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_26 (MaxPooling (None, 36, 36, 32)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_27 (Conv2D)           (None, 34, 34, 64)        18496     \n",
      "_________________________________________________________________\n",
      "activation_67 (Activation)   (None, 34, 34, 64)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_27 (MaxPooling (None, 17, 17, 64)        0         \n",
      "_________________________________________________________________\n",
      "flatten_21 (Flatten)         (None, 18496)             0         \n",
      "_________________________________________________________________\n",
      "dense_41 (Dense)             (None, 64)                1183808   \n",
      "_________________________________________________________________\n",
      "activation_68 (Activation)   (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dropout_21 (Dropout)         (None, 64)                0         \n",
      "_________________________________________________________________\n",
      "dense_42 (Dense)             (None, 1)                 65        \n",
      "_________________________________________________________________\n",
      "activation_69 (Activation)   (None, 1)                 0         \n",
      "=================================================================\n",
      "Total params: 1,212,513\n",
      "Trainable params: 1,212,513\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "---> Save model ...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "print('--> Training Conv-Conv-Conv-FC ver 1 ...')\n",
    "model = CNN_3Conv_1FC(convs=32, dims=3, epochs=50, batch_size=64, input_shape= input_shape)\n",
    "print(model.summary())\n",
    "\n",
    "# Save the weights and the full model\n",
    "print('---> Save model ...')\n",
    "model.save_weights(os.path.join(modelFolder,'model3_1_Conv-Conv-Conv-FC_weigths.h5'))\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's try `Conv-Conv-Conv-FC` with double number of filters: 64, 64 and 128 filters (3 x 3):"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAEQCAYAAABLMTQcAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAIABJREFUeJzt3XlclVUex/HP4bIjgiwiCoKKCu4KLqmllpZZLpXmmi2WbabTtExNTVNW075Yo2mWueY65VKWWoJLbuC+4AKuqIiCosgOZ/54QFFZLngBufzerxcvuM997vOcw/K9h/Occx6ltUYIIYR1sansAgghhLA8CXchhLBCEu5CCGGFJNyFEMIKSbgLIYQVknAXQggrZFtZJ966dWttW1vb74AWyJuMEEKUJBfYk52d/WRoaGhCSTtXWrjb2tp+V6dOnRBvb+/zNjY2MtheCCGKkZubq86ePdssPj7+O6BfSftXZou5hbe390UJdiGEKJmNjY329vZOxujtKHn/ci5PseeWYBdCCPPlZaZZuS193UIIYYWqfbjPnDnTXSkVun37dsf8bbt27XLo1q1bUP369Vs0bNiweZ8+fRqeOHHCFiA8PNw5LCysaWBgYIsGDRo0Hzx4cMClS5du+D5OmDDBMzg4uFlwcHAzOzu7dk2aNGkWHBzc7LnnnqtnbtliYmLs7rvvvoaWqakQojDllQEHDhyw9/HxaZWTk3PN9uDg4Gbh4eHORZXnq6++8hw5cmT9m61XpV1QvVXMmzfPo127dimzZs3yaNu27anU1FTVt2/fxh988MGJYcOGJQMsW7bMNT4+3hZg+PDhjWbOnHm4Z8+el3Nzc5kxY0atCxcu2Li6uuYWPO64ceMSx40blwhQr169lmvWrDno6+ubff35s7KysLOzK7RsQUFBWb/++uthi1daCHFFeWVA06ZNM319fTN///33Gvfdd18KwPbt2x0vX75s06NHj9Tyrle1brknJyfbREVF1fjhhx+O/vzzz7UAvv32W4927dql5P9QAfr27Xupffv26Z999lnthx9+OLFnz56XAWxsbHj88cfP+/v73xDaxRk7dmzdYcOGBXTu3LnxoEGDGuzdu9chNDS0aUhISLPmzZuHrF692gVgz549DsHBwc0APv/8c6/evXs37Nq1a+OAgIAWzz//vNn/AQghClfeGTBw4MCkH3/80SP/8axZszweeOCBJIAff/zRrVWrVsEhISHNOnfu3CT/PwNLuSVa7q8s2ul/MP5Skf+mlEWTOq6pnwxsfaK4febMmePevXv35FatWmW4u7vnrF+/3nnPnj1O7dq1K/Rddd++fU4jR45MtET5du/e7bx58+b9zs7O+tKlSzbr1q076OzsrLdv3+746KOPBu7atWv/9a+Jjo523rFjxz47OzsdFBTU8pVXXkkIDAzMskR5hKhUi5/3J2GfRTOA2s1SGTCxUjNg5MiRSaGhoc2ysrKO29nZsXjx4lrz58+PBejVq1fKkCFD9tvY2PD55597jR8/vs7UqVPjSlfJot0S4V5ZFixY4DFu3LgEgIceeihp1qxZHiW9xlL69Olz3tnZWQOkp6erUaNGBURHRzubTCZ94sQJh8Je07Vr14u1atXKBWjYsGFabGysvYS7EGVX3hlQv3797MaNG6cvXbq0pq+vb5atra1u3759OsCRI0fsBwwY4Hf27Fm7zMxMG39//wxLnvuWCPeSWtjlIT4+3rRp06aaBw8edBozZgw5OTlKKaVfffXV02vXrq1R2GtCQkLSoqKinEeMGHHhZs/v4uJypX/u3Xff9fHz88tcvHjxkczMTOXq6tq2sNc4ODhcGTpqMpnIzs5WN1sOIW4JJbSwy0NFZcCgQYOS5s6d61G7du2shx56KCl/+5gxY+qPGzcufvjw4cm//PKL6/jx4+taol75qm2f+6xZs2o9+OCDiadOndp98uTJ3fHx8bv8/PwyQ0JC0rdu3Vpj3rx5bvn7Llq0qOaWLVucXn755YQFCxZ45veJA0yaNMnj+PHjN/UmmZycbPL19c2ysbFh4sSJnnJ3LCHKX0VlwCOPPHI+PDzcbcmSJR6PPvrolXC/dOmSqX79+lkA06dP97R0/aptuC9cuNDzwQcfPF9wW//+/c/PmjXLY8mSJTETJ06sHRAQ0KJRo0bNp0+f7uXr65vl7++fPXPmzMOvvPKKX2BgYIuGDRs2X79+vWt+V0lZ/f3vf0+YNWuWV+vWrYOPHTtmb29vL+kuRDmrqAzw8vLKadOmTYqXl1dWcHBwZv72N95449TQoUMbhYaGNvX09CzVoAxzqMpqJe7cufNo69atz1XKyYUQoorauXOnV+vWrQNL2q/attyFEMKa3RIXVKu6CRMmeH7zzTc+Bbe1b98+ZdasWccrq0xCiIpzK2aAdMsIIUQVUhW6ZXJzc3NlKJ8QQpgpLzPNGsBRmeG+5+zZs24S8EIIUbK8m3W4AXvM2b/EPnel1DTgfiBBa33DIvFKKQVMAPoAqcBjWuttJR03Ozv7yfj4+O/i4+PlNntCCFGyK7fZM2fnEvvclVJ3ACnAzCLCvQ/wAka4dwQmaK07lrbUQgghLKfEFrPWei2QVMwu/TGCX2utNwHuSilfSxVQCCFE6VmiO6QeUHBdiLi8bUIIISqJJca5F3ZBtNC+HqXUaGA0gJOTU6i/v3+ZTpibm4uNTfXrpq+u9YbqW3epd/ViTr0PHjx4TmvtXdKxLBHucUDBlPYDThW2o9b6W+BbgLCwMB0VFVWmE0ZERNC9e/cyvbYqq671hupbd6l39WJOvZVSx8w5liXeGpcCI5WhE5CstT5tgeMKIYQoI3OGQs4FugNeSqk44N+AHYDWejKwHGOkTAzGUMjHy6uwQgghzFNiuGuth5bwvAaet1iJhBBC3LTqd8VCCCGqAQl3IYSwQhLuQghhhSTchRDCCkm4CyGEFZJwF0IIKyThLoQQVkjCXQghrJCEuxBCWCEJdyGEsEIS7kIIYYUk3IUQwgpJuAshqoacLMjNqexSVBkS7kKIW19GCnzbHWb2l4A3k4S7EKLiaQ1n9poX1FrDsrFwZg8cXQcbvi7/8pWnvT9D2oVyP42EuxCi4m2bAd90hsXPlRzwW76FPf+Du96CkL4Q/j6c2Vcx5bS0hGhY+Bhsn13up5JwF0JUrPNHYcUb4FoXds27IeBjElI4n55rPDixxdi3yb3Q5UW47wtwqAk/P230wVc1O+aAjS20Glzup5JwF0JUnNxcWPw8oGDUCujxhhHwS55H52Tz9Z+H6Pn5Gl6MSOOJicu5PHsE2TV84YFvwMYGanhD3y8hfhes/bSya1M6OVmwcz40vseoRzkr8TZ7Qghxg9xcuHgSkmIhMQYSY8HNDzo+a4RwUbZMgWProf9EcK8P3V41toe/T+SRJL5IeIQBbfwwpcQzPP5DbLOS6H/xHZxnRNOnpS/3tfKldkhfaDUE1n4CTXtD3bYVU+fCnN4Jsauhy99AqeL3jfkTLidAm2EVUjQJdyFEybSGk1th63Q4uc0I9ez0q8+bHCAnA+IiYcBksHO88RjnDsEfb0OT3tBm+NXNoeP4Y8sxhlycza8B9gQPms6xmZMIzN7BmR6fck9Od5bvPs07y/bx/q/R9G5Rh0fb/oOwI2tRPz8Do9cUfr6KsPp9OLQC6oVCgzuK33fHbHD2gib3VEjRJNyFKE9pFyD2Tzi4EtBG8BXXsi0oN7dci2aWjEuwawFE/QBndoOdCwR2hUY9wLMReDQCzyBw9YVNE2Hlm3DpDAyZA84eV4+Tk230k9s5Qd8JV1q5B+IvMWpGJOdS7qddC3dCov8Lsx8g8NhaaDMCn+5PMRYYe1djYhIuMW/LCRZEneCXXacZ6jGaD86+TeYf72J/7/sV/725fA5i/jC+XvNx8eF+OREO/A4dRoPJrkKKJ+EuhCVpDecOwsHfjUA/vhF0jhGKWZch+H5o1q/k4yQdgWm9qVNvINDdvHOvfq/oURi1GtwYuMWJ3wORU2HXQqPcPi3hvs+h5SBwrHnD7lprcjo+j62rLyx+Fqb1hhGLjK4XgL++NFr+A6eBax0Awg8k8MKP23G2N7Hg6dto4ncvhNeANR+S4tKAGvdd26ceVNuVN+9vxkt3N2XZrlPM2ujGnIt3MXTTRBalt2HgA4PMq5ul7PnJ+Nm2GWG0yo9vgvqdCt9390LIzaqwLhmQcBfCsn55Ebb+YHzt0xK6/s24gFa3LUzqaPQTh/QtuX92zceQEk/jQ9/B+aegVmDx+x9cCWs/4UKdzmTU8AOuHl+Ri/fhxajFz8KQuSX/53A4AmY9CCZ7aPEQhD1udDsUUebk1Cyenh1FTEIKr9/biQdH/ISaPxy+6wnDF4KygYgPofkD0OIhktOy+GLVQWZuPEqIb02+ezQMXzcn42A9Xod67dh1JJXOdk6Fns/J3sTDYf48HObP7thAzv14F622/5sL9/TH3dm++LpZ0u4FULs59PkYDv5mXOAdsQgw3uxiz16mkbcLSikj/H1bQ50WFVY8CXchLCX9ojHUrfkDcPd7xgXGgm5/CZY8D4dWQZO7iz7O2YPGCJIWA9HRv8KSMTByadGhnJpE1uLnOUZ9+hx9mkxu/Lf/MZMTbx+cScLKT6nd+9Wiz50YCwseBe+m8NivJbb0TySl8vj0SI4nphJUuwYvLdzJvMBafNjvJxqteAx+6AM1aoNTLXSfT/lpaxwf/BZN0uVMRnQK4LV7g3G2vy6GmtxD5qmIYs+br2UjP461G02TyPGE742mR/vWZr3upiXGGtcXer4D9i5w2/Pw53g4tR3qtuWLPw7x1Z+HaOPvztvtc2gTvxvu/aRiypZHwl0ISzm0EnIyoeMzNwY7GGObIz6CtR9D415FtoRzIz5Amxz5zmU0Lep60+XoNxD1PXR4qtD9T88bi9flRN5z+JhvBt92JSw1GrSxz7ZjjVm57gB3bvyACafqMKD/QwR4ulx7oPRkmDvEaGkPnVtisO+OS+bx6ZFkZucwa1QH2gd6sHDrCT78bT93z87m+dCJjDvzT0wJe4jrPY0XZx0k8uh52vi7M/3xDrSo51b899NMdVvcAZFwZt9aqKhw370IUNByoPG4/VPw1wRY+ykrW37GV38eomuQF7FnU9i29Fua29qx36MXLSumdICEuxCWs28J1KgDfh0Kf95kZ3TT/Pp3OLIGGna/8lROrmbLkSS2blnHmAM/8XX2AD5bcw5FV1Z4baXxqrdQQT3Bo8E1hwz/6Vt6HF/GHJcRfPzsCGq7Fj5qpHOQFxfa/cilyd0YfOzf9PvMnp7tmzP2zsbUcXM0JhEtGgVJh2HkkhK7gVbvP8Pzc7bj4WLPvNEdCartCsDg9vW5p3kdPl5xgK+3HGexy2s82CCFr5c6UdMxhY8easmgUH9sbErolioFu3qtycQOm5NbLXbMYmkNu+YbF5bz38QdaxrDQNd8yDfR3WlZL4TvHg2DnEz0Z6OIyArlqe/3cXezRF66uylN67iWezFlEpMQlpCZaoycCLm/+D7tNsONkSV5E3BOJ6fxr8V76PifPxk6dRPB+78m1caFxv1fY/M/7+LuADueSBpJSpYmftaT5OQYMzlzcjWf/G8drXaO56hDUx544dMigz2fey0vaj06Bx/bFOZ4TmNR1DHu+DichydvZMvUFyBmFem9PjRCqxg/bj7OkzOiaFTbhZ+f63wl2K+cx9me/zzQksXPdcHdvRYT9tdkcHt/wl/uzuD29S0a7ADY2pNUM4TA9H2cuZhe8v43K38oaKuHr9mc0vZJUnHiaZufmfxIKI52JhyP/IFT1nm6Dvobf+/VhI2xifSesJbZm46VezGl5S6EJcT8AVmpEFLCSBg7R+g8Fla8zl+rl/HsWnsysnPpGeLD0Hpn6RoRBd3epHf7EACGhjhQ98EBzJ0by+jzXzDp09dp+eArzPjrKA/H/hs32wzcn5iByanwi483qNsG1ftDGv/6d7bcsZtJOQNw3jefDilzmJHdi3eW1qVZ1DrCAjxwdbQlK0eTlZOb96FJTMlg5b4z9GjqzX+HtcPFoegIae3vzs/PdSHxckaJbzw3y7Z+B1olz2DFoXj6hwaW67nYvcAY11/gZ6215uVlx2mT04unTctQ2XFAY9g+B2rUwSm4F2Ob2/JIpwAmr4mla5BX+ZYRCXchLCN6KTh5QECXEne92GI46o+PyA7/mIZ1PuTLwW0I9HKBWQ8Yx+j0zDX7N/FxpfHYt0iYEsXjZ6Zzz7Rg2tsc4G67rdDrPfAJKV1Zw56AY3/hvulj/tnTBdImkhNwO0Gdv2bM8UtsOZLIvMjjZGTnYmeywd5kg51JYWuywc5G8XiXQN7oE4KtqeR//E02qtyDHcCjaRds9nzHkb1boDzDPSfbWMSsyT3g5H5l86SIWH7fG0/XXuNQG1fBus+h59vGdZjOY8BkRG0tF3te71PKn1cZSbgLcbOyM+DgCmjW/8ofcVEijybxt3k76JfRm3/YzqNLPwdsvVzg6F/GNPZe74LDjf2xysaG2sOmoCd1ZGHN7/FMOwa+t0Gn50pfXqWMiUSnd8Kqf0GtBpgGz6SLswddmvoCRktUlTRc8xZi498egJzjW9B6UPmV/XAEXD57zcJfaw6e5dOVB+jbui7D72wDWY/D5inGKJr8cfCVQPrchbhZh9dAxkUj3IuQmZ3LJyv2M3jKRkw2irsfexMc3bFd/7lxgW71e1DDB9o/WfR53Oqh7vkAn4t7sCUHBkwCG1PZyuzgCg/PgsZ3w9B5N4yMqUrBDoCbH6kO3jTM2MfxpNTyO8+u+eDobox2Ao6eu8zYudtp6uPKRw+1NL5vnV8wfi6RU8GvPXg3Kb/yFENa7qLaCD+QwJYjSQwM9aORd41i99VaszH2HBH747mnZT1CA4oZFhi9xFiGtkG3Qp8+kZTKmLnb2XniAoPD/PlX32bUcLA1Wt0R/4GN/4XjG4xx0PbOxVei7Qg4fwTqtgOPhiVVuXg+zYxJRtZAKXLrtadtzFY2xCbeOMzTEjJSYP8vxoVUWwfWHzrHC3O3ATDlkdCr4/Vr1oW2jxjDVwusoVPRJNxFtfDTtjheWbSLnFzNNxGxdAny5JFOAfQM8cH2aAT8/AxkpKB1Drk52SidQ2c0bbU94zY8j1Or/rx2b/DVmZT5crJh/3JjMSzbG2dH/r7nNK8s2gXApOHt6NPS9+qTHUcbdxVa+Sa4+UPooyVXRCnjphXiBi4NO1Hj8HIm749haIf6xe6788QFQnxrYm9bis6LA8shKxXd8mG+iYjh0xUHaORdgymPhN74ZtL9NWMdnetG1FQkCXdh9WZtPMq/luylS5An/3mgJct2nuLHzcd5ZvY2/Fxt+MX0Mo62DkS6dedgQirpuQrvms409/OkaVI4X174gTv3BHPn3jM8170RT93REEe7vO6QY+shLemG9WIysnP4z6/RzNh4jNZ+bnw9tB31Pa9rlTvVMgJ+3WfG0re2DhXzDbFSKq/fPf3oZrTuUWTXUvj+BB6fHsmD7erx2aDW5ndB7ZpPbk1/nl9rz2/7DnBfK18+fqhV4SOGatSGeyphMbMCJNxFhUrPyiE1MwcPl4pZA2RSRAwf/36AniE+/HdYWxztTIy5szHPdg9i9f4Ezq38FPcLcYzM/AdbTG0Z0KYeIzoFXJ09mRCN7ZQ7WN10CS/xEp+tOsj8qBO80SeE3i3qoPYtBTtnaHTXlXMePXeZMXO3sefkRZ7o0oDX7g0uuoXY9e9G90qrIRXw3bByvm3IVSYaZe7nwJlLBNcpfIGzL/88hJ1J8dO2k7SrX4sRnQJKPnZKAjo2nHl2D7By/1nevC+EUV0b3NLXJswKd6VUb2ACYAK+01p/eN3z9YEZgHvePq9prZdbuKyiikvJyGbgNxs4l5LByhe7lWvAa6356PcDTF4TS/82dfl0UGvsCgzdM9koegWYIG0+aQF3MrDdY3zdxBs3p+vWZakdAj3+ifMfb/PNQw+xoXN3xi/bx7NztmFno9lo/z+i7doyeeYuPFwcqOloy5IdpzDZKKaODKNXM5/iC+pQw+hHFzfP3pls72a0PX2IDTGJhYb72kPn2HniAu8NaMEf0Wd4Z9lemtetSdv6tYo9dPSqHwjROfyc04XZozpyWyPP8qqFxZTY4aSUMgETgXuBZsBQpVSz63Z7E1igtW4LDAEmWbqgomrLydWMm7udQwkpXEjN4p1le8vtXLm5mn8t2cPkNbEM61ifLx5uc02wXxH+H8hMwen+D+nXuu6NwZ6v81hj1MOvL9G5dja/vNCVLwe34e02KXhxgeha3UnPymV33AWW7jxFi3o1+XVs15KDXVicfUBH2poOszHmzA3Paa2Z8MdB6ro58nCYP18OboNPTUeem7ONxJSMQo+ntWbyb1G47viWWNsgvho7tEoEO5jXcu8AxGitDwMopeYB/YGCtx/XQP7bpBtwypKFFFXfR7/v58/9CbzbvzmJlzP58o9D3N+qrsUD8ED8JT5ZsZ8/ohN4ultDXusdXPi/zgnRxtK8YaOMFRCLY2OCAd/A5K6w7G/YDp3LgLb14MwuMDkwetSzjC5kbLqoBH4dcIn8jsQju8nO6XDNRKsNsYlsO36Bdwe0wN7WBntbeyaPCOXBbzYwdt52Zj7REVOBpRHSMnN4ZeF2Htj/MnVsk8kdOQ/76y+o38LMuVRcDzhR4HFc3raC3gZGKKXigOXACxYpnbhKaxzSz1Z2KcpkfuRxvl17mEdvC+CR2wJ5rnsQwXVceePn3SSnWeYO9rvjknl6VhT3fLmWDbGJvNEnpOhgB1jxBti7QvfXzTuBV2NjlMrB32DnXGNsevQyaHRnoZOORCXxCwOgSfZ+dp9MvrLZaLUfok5NRx4Ou7piZ4t6brw3oAV/xSTy2coDV7afTk5j0JQN+EdP5S7Tdky9/4N9/fYVVw8LMKflXthfh77u8VBgutb6M6XUbcAspVQLrfU19wlTSo0GRgP4+PgQERFRhiJDSkpKmV9bVfnEr6bT/q/YpHNJd6o6/+7vT8rhk8h0WniauMP17JWf25AGOYzflMGY71YzqmXJo0Ru+JlrjV/cUnbYtmLmSV92ncvByRb6NbLj7gA7auQeZ82a44UeyyNxK61i/ySm0RPERe42vzI6mDZuzaix7GWiY07QMvk40b4PcKYcfxer4+863ES9teY2W1faZscwZ1UkyY2M6zrRiTlsOZrO8BB7Nq5fd81LagPd/WyZFBGLbXIcbg6Kr7Zn0CpnH6/YLuBM7duJTm0MFfBzsOjPW2td7AdwG7CiwOPXgdev22cv4F/g8WGgdnHHDQ0N1WUVHh5e5tdWWVN7av3vmlpH/1rZJTHb0XMpuvU7K/Sdn4brC6mZNzz/0W/ROuAfv+iIAwklHqvgz/xCaqb+a8l3Wv+7pj78r8a609tL9H9XH9LJaTee4wbZmVp/3V7rCW20zsooTXUM52K0fq+O1uO9tX7HQ+vLiaU/RilUy991fZP1nj1IH32nuR42deOVTUOmbNTt31ul0zKzC31JWma27vv1Ot3sX7/pxm8s1/0+XKSzPgrS+qtQrdMvlr0spWROvYEoXUJua63N6paJBBorpRoopewxLpguvW6f48BdAEqpEMARqJp9CLeiswcgbovxdWJM5ZbFTMlpWTwxPRKA7x9tX+jFyrF3NaaRtwuv/28Xl9KL757JzNH8tvs0T8+KovN7v1M36iMSlCcNbM6wtt0anu8RRE1HM248vHU6nDtgrOFSyKSjEnk2Mu6+k5MBgbebf09SUXH82hOQe4IDR+NIz8phy5EkNh5O5Oluja7OT7iOo52JScPb4WRvS6ifK4u8vsc28xI8PLPKdruV2C2jtc5WSo0BVmAMc5ymtd6rlBqP8Q6yFHgJmKqUehGjy+axvHcYYQnbZ4GNLdnKFtuk2HI7TXJaFieSUm/6DjmXM7J5dvZWjiWmMvvJjsaKh4VwtDPx8cDWDJy8gY9+3897A669T82l9Cz+ijnHqn0JLN+VSlr2NrxdHfi0YRSBJ86ghy2CmD+x2/wNNO97zc0vCpV23hghE3g7BN9X9gq2fxIunjTWZRG3nrx+9+DcGLYdP8+k8Fi8ajgwrIRZq361nFn/jx44rHkftf4v4yK6z/UDA6sOs8a5a2PM+vLrtr1V4Ot9QMlrnYrSy8mCnfOgSW8unzqEW2L5hPvuuGSemb2VkxfS+HJwG2M0SBkkpmTwxPRIdp9M5tNBrenUsPhhY6EBtXiiSwO+X3+E+1rWxauGPeEHEgjff5bIo0lk52pcHW0J87HlmXtD6VTXhOnrp6BBN+PORAFdIGaVcZ/RZ/8CxyLemLIzYdnfjIC/5z8l36C6ODY20Oudsr9elK967dAoQm0OMSk8lvUx53ijTwhO9iUvsuZ45E9Y/5mxNkybYRVQ2PIjM1RvdQd/N5YYbfsIaecn45Z4oOTXlNL8yOP8a8levFzsCQ2oxUsLd1LDwZaepRymGHc+lZHfb+HkhTSmPGLGBJ48L9/dlD+izzDi+83k5Br/8AXXceWpOxrSo2lt2tV3Z/26tXQJ8oJV/zYC+u53jYC2d4YBk2Ha3bDin9B/4o0nSE+G+SPgyFroNR58W5WqXqKKcXRDeQdzR/JRJsScw8PFnuGdim+1A8Y67UvHgU9L6FOxN7MuDxLut7rts437cgb1JG3jUjgTbtzSraTVA82QnpXD20v3Mi/yBF2DvPhqaFvsbW0YNnUTz/+4jRlPdCix5Z1vf/xFHp22hbTMHGY/2ZH2geb3RTvZm/hqSFum/XWEjg086d7Um7ruhYwnvnACNn1jrKXtW+BGyP7tocvfYP3nENwXmva++lzySZgzyOhnf2AKtJZp/tWCf3uaJS0GNE/d3vDqio2FyUqD31835j34dYBB041Fv6o4Wc/9VnbxtHEnlzZDwWRLqnNdY3vS4Zs+9ImkVAZN3si8yBOM6RHEjCc64OFiTw0HW6Y/3gG/Wk48OSOKPQXGChdly5EkBk3eCMDCZzqXKtjztfZ3Z8KQtgzrWL/wYAcIz1uI6c43b3yu+2tQuzksGwupSca2M3vhu55w4TjbvC/gAAAb7ElEQVQMXyTBXp34tccp5yJPN9eMvK2YtWPOxcB3vYxg7zIOHl8ObmXrkrzVSLjfynbOBZ1r9P8BaU754X5z/e4bYxPp+9/1HE28zNSRYbx8T9NrZuZ5uNgz+8mOuDnZMXLaFmISUgo9jtbGCJZHvt+Mt6sD/3u2c9nv6p5+0Vg6Nzuz0KdrXDpsXHvo9Ay4+9+4g60DPDDZCPZfXzJuoDGtN6Dhid+gUY+ylUtUTX7GhKPXW6YUfZ/XXQvh227GxfFhC40uO5MZI66qCOmWuYXk5GpW7o3nxPlUHmpbD8/ts6F+Z2P4HZDmVMfY8SaGQ2bl5PLywp14ONsz7bH2RY5k8XVzYvaTHRk0eQMjv9/Mwmc741vTkYMJl9gUm8jmI0lsOZJE4uVMWvu788Nj7W9uIbAVrxtdUDX94PYXjTe0/CVwtabh4enGErld/170MXxbQfd/GHc12rcEvJrAiEXg5lf0a4R18mpq3EAlLtL4zzdfbi7E74Qt38GO2eDfCQZOs5rWekES7uUp4xJsm2kMmfNqXORuWTm5LN1xikkRMcSevQzA2lVLmW2K5ULYOPJvw5tj62z0vyeWvVvml12nOHkhje8fDSsy2PM18HJhxhMdGPLtJp7+7xIyc+BwmhPZ2FLP3YluTb3p1NCTvq3q4mRnAylnjTeepFij6yioJwR0LrlQ5w7Bjh8h+H5ISTBa3us+h655IX90PR7nd0LvD6+5KXGhurxoXDi1sTP+aEvaX1gnGxuo184I94xLxr1PD64wujlTzoCyMRoKPd4o8b63VZV11qqyaQ27FsCqtyAlHo5tgCFzbtgtPSuHRVvjmLwmlrjzaQTXceXroW1pWseV5B9ncumCE3f86sp98bt4ppvResezUZm7ZbTWTFlzmCY+NejRtLZZr2le141fO+yh/pbxxgZHyHVww8bFGy55wcFasC3eeMPJuK5/fttMGBNptLiLE/6+sSZ63wng7Gn8Ia75CJa/bNzIwmRHmmMdnMJGlVxgky2MXHpzQx2FdfBrD2s/hY8bQk6m0ZJvdKdx16ygnlDDu7JLWK4k3C3t1A747VU4sdm4z6VfmNFiSE26Zjbjoq1xfLJiP2cuZtDa3523+zbnrpDaxkJX6RchdS0pLR6kn11jFkTFMT/yBG28TbiY3Am5tIEPft6NUmCjFE72JkZ1bUBtV8diixZx8Cz74y/x6aDW2NiYGX5H1lI/8n0I6mWMQrmciE3qOWN45uVzkHwCXLyN24l5NgLPIOPmE+kXjIuZf46H+78o+vind8Len+GOV8HFy9jWqIcxIenoOoj4CI6tJ7b5a7Qwd0apBLsAaP4AHFlnjKZqfA/U72RVfeolkXC3lMuJsPpdY3q7syf0+69xc9z4XcZNdfctgbDH0VozMTyGT1ceJDSgFp8NakOXIM9rVy/c+xNkpVKj0+O859+SsXc25rv1R/jfliNstPGgfe551u0+TIpyRmvNxfRs9p26yMwnOhR7Z5jJEbH4ujnSr3Vd8+qUHAcLHzdCe+A0cLzx5gfF6viMMXSx9TDjD6wwq98z7ibfecy125WCBncYH5fPcS5yT+nOLYRPcxi1orJLUWkk3G9WchwX1kzCtG06LiodOj6DTffXrvb1+rY2Lu7sWoAOfYz/LI9m6rojPNC2Hh8PbFX4TSS2zQLv4CvTqGvXdOSffULo7HyG7j73wPxZrH2yPtRtC8DMjUd5a8leFm6N4+GwAiNJ5g4zWitdxrL9+Hk2H0nizftCzLspcFa6MfEnOwMGzyl9sAP0+CfsXQy/vAijI27s2zy20egD7fl20TNL4WqLXghhNhkKWRZaG8G0YCT6y1bU3PYNf+U25570D3jwcF9iLhUIMaWg1SA4voEP565i6rojjLwtgM+uu+3bFQnRcDLKuPVaYa1wj7y+9wLLEIzoGECHQA/e+2UfZy6mGxtTEuDAr/DH2xC3lSlrDlPT0ZYhJayvcaV+y1+CU9vhwSng3cT8701BDq5w74dwZjdsmXLjOVa/CzV8oMPosh1fCFEkCffSyM2BHXONsbE/9IbDa4gNeozb078g4d6pjBl8P0cTL9Pnq3VMWRN7ZSp9ZshAAEx7F/HCnUG806950X3ekd+DjW3RN0z2aGB8LhDuNjaKjwa2IiM7lzd+3mMsuxwXZTxpsifzf6OJ2HeckbcFUqOoMb8FRU0zhiXe8crNLbAFENLPGC20+n2jmydf7Go49pdxDvviR+0IIUpPwr00/vg3LH7G6Kq4/wsuPbeTYUf74FEviOEdA+jfph4rX7yD7k28+eC3/QyavIE9J5N5ctlZInOb8JR7JC/1alJ0v/iF40affZthRV/Jt3MCN/8bRsw08HLhpbub8Ef0GZbtOm0MAbOxhUE/YH8+hlftFvJo58CS63hiC/z2D+MCqrl3KSqOUsY6HTrXOC4YrfY/x4N7fWj36M2fQwhxAwl3cx3bCBv+C6GPwXObIOwJvlx7irMpGbw7oMWVGZ61XR2Z8kgoE4a0IfbsZe7/ej3rD51FtRpMrcuHIb6YO/9EfGSMv+32j+LL4tGw0IlMo7o2NEbeLN1L1rEtUKclCb49mJNzN4/ZLMc7Mar44148DfMfMSb9PDTVuHeoJdQKhG6vGheWD/xm3J7u9A7jzaMsa6oLIUok4W6OzMuw+FmjpXn3+6AU0acvMn3DUYZ2qE8b/2snyiil6N+mHqtevIOhHerzzYhQwvo8YbSkd80v/BxnD8LOH421wkuaUenZ6JpumXwmG8UnA1uRmp5OblwU+LVn2l9H+SB7CDluAUYdMgpfSoAja+Hb7pBx0RiTX9LY9NK6bYxxkXj5K0Zfu1dTYwEwIUS5kHA3xx9vw/kjMGASONQgN1fz5uI9uDnZ8eo9TYt8We2ajnzwYEvuaV7HGOPe+G7Yvcjou79exH+MiTy3FzO9Pp9nkDGOPH+BrAKa+LjyVgcbHHQ6kdmNmLPpGN1aNMDuwclGt8+qf137gpxsoz98Rj/jAuiolcYQMkuztYf7PjfGxZ87CHe+Ybn/DIQQN5BwL8nhNbDlW+j4LAR2BWDRtji2HjvPa/cG4+5cim6FVg8bM1aPXnuD3isTeTo9Z96wvysjZgpfY2awbzwAL2+051JGNk93awgBtxljyaOmQcwfxo7JcTCjL6z92BiT//QaqNOy0GNaRGAXY+W9Jr2NC61CiHIj4V6c9IvGHX48GsFdxo2nLqRm8uFv+wkLqMXAdqVckKpJb7B3NZYmKKioiTxF8bxxOGRBplNbyXb0IE750LmRJ6388rqNerxpdIcseQF2zofJXY1JVg9OhQETK2bUSq/xMGy+zCIVopxJuBdn5ZtwMc5YSjbv5hgfrzhAcloW7w5oYf4U/nx2TtCsP+xbatwgAK5O5On6YvETeQpyDwBlKnqNmbhIbOt3ZOEznflqaNsC53c06pJyBn4ebVxDeHqt8R+FEMKqSLgX5dAfsG0GdH4B/DuQk6tZEHmCuVuO81jnQEJ8yzBjE4wJTZmXjFEj+UMCSzuRx9beCObCumXSzht92n5htKtfC68aDtc+X6+dsUBXt3/AqFVX/wsQQlgVWX6gMGkXYOkL4B2M7v46q/bG8+nKAxw8k0Jrf3f+1rPo5XtLFHg7uPrC7oXGlP7jG6DPp6W/bV4RI2Y4udX47FfEWi4A7R4p3bmEEFWOhHth/hwPKWfYdfs3/HvqNrYfv0BDLxcmDmvHvS3qlL47piAbE7QcaCyolXSk7BN5PIPg+Caj9V+w/zouClBGC10IUW1V63B/fs42oo4l4eHigKeLPR4u9gTaXWDc3hlEuPRh1E+XqVPTkQ8fbMnAUD9sC1sLpixaPgwbvoaz0TDgm7JN5PFoBJkpRv+5a52r209sgdrNjGGNQohqq9qG+9Zj5/l192m6BHniZGci8XImJ86n0iFlCrloPr18L//sE8zI2wJxtLPweOw6LcGnBeRml30ij2dD43Ni7NVwz801Fh1rNsAy5RRCVFnVNtwnr4nF3dmOqSPDcLbP+zZcOgMTwslpPoSl/YYWvmqjJSgFI/5nLDVQ1ok8nkHG56RYY/w4GBdY05PBv4NlyimEqLKqZbjHJFxi1b4zjL2r8dVgB9j4X8jJxHTHS5jKK9jzFexKKQs3fzDZXztiJi7S+FzcxVQhRLVQLYdCTllzGEc7Gx4ruEpiapKx3G6LgVVjeKCNyViQq+CImbhIcHADz5sYzSOEsArVLtxPJ6exeMdJBof54+FS4ELmpm8g6zLc/lLlFa60PIMg6fDVx3FR4Bdq3PldCFGtVbsUmLb+CLkanry94dWN6cmweYqx3knt4MorXGl5NDTCPTfXWO0xYa90yQghgGrW556cmsWPm49zfytf/D0KTBra8i1kJMMdL1de4crCsxFkp8PFk8aqlTpXwl0IAVSzcJ+9+RiXM3N4+o4CfeoZKbBxkrGol2/ryitcWRQcMZM/M7VeaOWVRwhxy6g23TLpWTn88NcRujXxplndAuvCRE2DtCS4vYq12uHapX/joowLqc4elVsmIcQtodqE+6KtcZxLyeSZbgVa7VlpxkzRht3Bvwp2Z7j6gq0TJB42RspIl4wQIk+16JbJydVMXXeY1n5udGpYoGW7bRZcToA7plda2W6KjY1xUTV2NVw+C35hlV0iIcQtwqyWu1Kqt1LqgFIqRin1WhH7PKyU2qeU2quU+tGyxbw5v+05zbHEVJ7p1giVv8hWbo7Raq9/29UZnlWRZyNjjRqQlrsQ4ooSW+5KKRMwEegFxAGRSqmlWut9BfZpDLwOdNFan1dK1S6vApeW1prJa2Jp4OXC3c0LzAqNDYfk43D3+MornCXkT7iyczEWDBNCCMxruXcAYrTWh7XWmcA8oP91+zwFTNRanwfQWidYtphldyIpjT0nL/JIpwBMBZfq3foDuHhD0/sqr3CWkH9RtV47MFWLXjYhhBnMCfd6wIkCj+PythXUBGiilPpLKbVJKdXbUgW8WZFHkwDoHOR5dePF08adkNoML9tyu7eS/OGQ0t8uhCjAnKZeYXem0IUcpzHQHfAD1imlWmitL1xzIKVGA6MBfHx8iIiIKG15AUhJSTH7tUv3ZOBkC6eitxK/36hK/WMLaKhz2JwVTFoZy1AZCqu3KTuVVjWDOZTmR0oVqktpleZnbk2k3tWLReuttS72A7gNWFHg8evA69ftMxl4rMDjP4H2xR03NDRUl1V4eLjZ+971WYR+bNrmqxtysrX+vIXW0/uW+fyVpTT1tjbVte5S7+rFnHoDUbqE3NZam9UtEwk0Vko1UErZA0OApdftsxjoAaCU8sLopjlMJUu6nElMQgphgQWGP+ZfSA17vPIKJoQQ5azEcNdaZwNjgBVANLBAa71XKTVeKdUvb7cVQKJSah8QDryitU4sr0Kba+ux8wB0aFAg3Lf+AM5eVf9CqhBCFMOs4RVa6+XA8uu2vVXgaw38Pe/jlhF5NAl7kw0t67kZG/IvpHYeU/UvpAohRDGsevmByKNJtPJzu3oP1B2zQedAu0crt2BCCFHOrDbc0zJz2HMy+Wp/e24ObJ0JDbpVjTstCSHETbDacN8Zd4GsHE37wFrGhvwLqaGPVWq5hBCiIlhtuEflTV4KDcgL9/wLqcH3V2KphBCiYlhtuEcePU8Tnxq4O9tfvZDa1gpmpAohhBmsMtxzcjXbjp2/2t8uF1KFENWMVYb7gfhLXMrINvrbM1PlQqoQotqxynCPOmb0t4fVd4fFz0DyCej6t0oulRBCVByrXCM28uh56tR0xG/757BvCfR6FxrdWdnFEkKICmN1LXetNZFHkni+1mbU+s+MfvbOL1R2sYQQokJZXcv95IU0Ai5tZ1jWZ0Y/+32fgSps1WIhhLBeVtdyj96znSn2n5PtFggPzwSTXWUXSQghKlyVC/fUzGx2JGQX8WQSbdeNJhcbbEcsBCf3ii2cEELcIqpcuE8Mj2HCtgzmRx6/9onMVJj/CDUz45lUZzwmr4aVU0AhhLgFVLlwf+HOxrTwMvGP/+1m3pa8gI/fDd92h2PreSVzNO5Nb6/UMgohRGWrchdUHe1MvNDWgR+Pu/D6TztpfHgmoYcmgJMH27v/wJLfHZhX8M5LQghRDVW5ljuAvUkxZUA9lrl/Qej+Tzjh2Rme3cDK9GbYmRSt/aSvXQhRvVXJcPc8F4nD1Ntpnr2PmR7juP34U8zanULU0SRa1HPDyd5U2UUUQohKVeW6ZdjwNS33vAc+LVEDv2dwrSDWztnGvxbvwUbBk7fLhVQhhKh6LffG93Dc/wF46k/wboqDrYlJw0PpGeJDroYO0t8uhBBVsOXu3YTDjR6jvq3DlU32tjZMGt6OyKNJdG7kWXllE0KIW0TVC/ci2Nva0CXIq7KLIYQQt4Sq1y0jhBCiRBLuQghhhSTchRDCCkm4CyGEFZJwF0IIKyThLoQQVkjCXQghrJCEuxBCWCEJdyGEsEIS7kIIYYUk3IUQwgpJuAshhBWScBdCCCtkVrgrpXorpQ4opWKUUq8Vs99ApZRWSoVZrohCCCFKq8RwV0qZgInAvUAzYKhSqlkh+7kCY4HNli6kEEKI0jGn5d4BiNFaH9ZaZwLzgP6F7Pcu8DGQbsHyCSGEKANzwr0ecKLA47i8bVcopdoC/lrrXyxYNiGEEGVkzp2YVCHb9JUnlbIBvgAeK/FASo0GRgP4+PgQERFhViGvl5KSUubXVmXVtd5Qfesu9a5eLFpvrXWxH8BtwIoCj18HXi/w2A04BxzN+0gHTgFhxR03NDRUl1V4eHiZX1uVVdd6a1196y71rl7MqTcQpUvIba21Wd0ykUBjpVQDpZQ9MARYWuDNIVlr7aW1DtRaBwKbgH5a6yhLvPkIIYQovRLDXWudDYwBVgDRwAKt9V6l1HilVL/yLqAQQojSM6fPHa31cmD5ddveKmLf7jdfLCGEEDdDZqgKIYQVknAXQggrJOEuhBBWSMJdCCGskIS7EEJYIQl3IYSwQhLuQghhhSTchRDCCkm4CyGEFZJwF0IIKyThLoQQVkjCXQghrJCEuxBCWCEJdyGEsEIS7kIIYYUk3IUQwgpJuAshhBWScBdCCCsk4S6EEFZIwl0IIayQhLsQQlghCXchhLBCEu5CCGGFJNyFEMIKSbgLIYQVknAXQggrJOEuhBBWSMJdCCGskIS7EEJYIQl3IYSwQhLuQghhhSTchRDCCkm4CyGEFZJwF0IIK2RWuCuleiulDiilYpRSrxXy/N+VUvuUUruUUn8qpQIsX1QhhBDmKjHclVImYCJwL9AMGKqUanbdbtuBMK11K2AR8LGlCyqEEMJ85rTcOwAxWuvDWutMYB7Qv+AOWutwrXVq3sNNgJ9liymEEKI0zAn3esCJAo/j8rYVZRTw280USgghxM2xNWMfVcg2XeiOSo0AwoBuRTw/GhgN4OPjQ0REhHmlvE5KSkqZX1uVVdd6Q/Wtu9S7erFkvc0J9zjAv8BjP+DU9TsppXoCbwDdtNYZhR1Ia/0t8C1AWFiY7t69e2nLC0BERARlfW1VVl3rDdW37lLv6sWS9TanWyYSaKyUaqCUsgeGAEsL7qCUagtMAfpprRMsUjIhhBBlVmK4a62zgTHACiAaWKC13quUGq+U6pe32ydADWChUmqHUmppEYcTQghRAczplkFrvRxYft22twp83dPC5RJCCHETZIaqEEJYIQl3IYSwQhLuQghhhSTchRDCCkm4CyGEFZJwF0IIKyThLoQQVkjCXQghrJCEuxBCWCEJdyGEsEIS7kIIYYUk3IUQwgpJuAshhBWScBdCCCsk4S6EEFZIwl0IIayQhLsQQlghCXchhLBCEu5CCGGFJNyFEMIKSbgLIYQVknAXQggrJOEuhBBWSMJdCCGskIS7EEJYIQl3IYSwQhLuQghhhSTchRDCCkm4CyGEFZJwF0IIKyThLoQQVkjCXQghrJCEuxBCWCEJdyGEsEIS7kIIYYXMCnelVG+l1AGlVIxS6rVCnndQSs3Pe36zUirQ0gUVQghhvhLDXSllAiYC9wLNgKFKqWbX7TYKOK+1DgK+AD6ydEGFEEKYz5yWewcgRmt9WGudCcwD+l+3T39gRt7Xi4C7lFLKcsUUQghRGuaEez3gRIHHcXnbCt1Ha50NJAOeliigEEKI0rM1Y5/CWuC6DPuglBoNjM57mKKUOmDG+QvjBZwr42ursupab6i+dZd6Vy/m1DvAnAOZE+5xgH+Bx37AqSL2iVNK2QJuQNL1B9Jafwt8a07BiqOUitJah93scaqa6lpvqL51l3pXL5astzndMpFAY6VUA6WUPTAEWHrdPkuBR/O+Hgis1lrf0HIXQghRMUpsuWuts5VSY4AVgAmYprXeq5QaD0RprZcC3wOzlFIxGC32IeVZaCGEEMUzp1sGrfVyYPl1294q8HU6MMiyRSvWTXftVFHVtd5Qfesu9a5eLFZvJb0nQghhfWT5ASGEsEJVLtxLWgrBWiilpimlEpRSewps81BKrVJKHcr7XKsyy1gelFL+SqlwpVS0UmqvUmpc3narrrtSylEptUUptTOv3u/kbW+Qt6THobwlPuwru6zlQSllUkptV0r9kvfY6uutlDqqlNqtlNqhlIrK22ax3/MqFe5mLoVgLaYDva/b9hrwp9a6MfBn3mNrkw28pLUOAToBz+f9jK297hnAnVrr1kAboLdSqhPGUh5f5NX7PMZSH9ZoHBBd4HF1qXcPrXWbAsMfLfZ7XqXCHfOWQrAKWuu13DhXoOAyDzOAARVaqAqgtT6ttd6W9/UljD/4elh53bUhJe+hXd6HBu7EWNIDrLDeAEopP+A+4Lu8x4pqUO8iWOz3vKqFuzlLIVgzH631aTBCEKhdyeUpV3mri7YFNlMN6p7XNbEDSABWAbHAhbwlPcB6f9+/BF4FcvMee1I96q2BlUqprXmz98GCv+dmDYW8hZi1zIGo+pRSNYD/AX/TWl+sDuvQaa1zgDZKKXfgZyCksN0qtlTlSyl1P5Cgtd6qlOqev7mQXa2q3nm6aK1PKaVqA6uUUvstefCq1nI3ZykEa3ZGKeULkPc5oZLLUy6UUnYYwT5Ha/1T3uZqUXcArfUFIALjmoN73pIeYJ2/712AfkqpoxjdrHditOStvd5orU/lfU7AeDPvgAV/z6tauJuzFII1K7jMw6PAkkosS7nI62/9HojWWn9e4CmrrrtSyjuvxY5SygnoiXG9IRxjSQ+wwnprrV/XWvtprQMx/p5Xa62HY+X1Vkq5KKVc878G7gb2YMHf8yo3iUkp1QfjnT1/KYT3K7lI5UIpNRfojrFK3Bng38BiYAFQHzgODNJa37BAW1WmlOoKrAN2c7UP9p8Y/e5WW3elVCuMC2gmjEbXAq31eKVUQ4wWrQewHRihtc6ovJKWn7xumZe11vdbe73z6vdz3kNb4Eet9ftKKU8s9Hte5cJdCCFEyapat4wQQggzSLgLIYQVknAXQggrJOEuhBBWSMJdCCGskIS7EEJYIQl3IYSwQhLuQghhhf4PQw/YXx9lPeAAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation/Test loss    : 0.36095644533634186\n",
      "Validation/Test accuracy: 0.8828125\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_28 (Conv2D)           (None, 148, 148, 64)      1792      \n",
      "_________________________________________________________________\n",
      "activation_70 (Activation)   (None, 148, 148, 64)      0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_28 (MaxPooling (None, 74, 74, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_29 (Conv2D)           (None, 72, 72, 64)        36928     \n",
      "_________________________________________________________________\n",
      "activation_71 (Activation)   (None, 72, 72, 64)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_29 (MaxPooling (None, 36, 36, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_30 (Conv2D)           (None, 34, 34, 128)       73856     \n",
      "_________________________________________________________________\n",
      "activation_72 (Activation)   (None, 34, 34, 128)       0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_30 (MaxPooling (None, 17, 17, 128)       0         \n",
      "_________________________________________________________________\n",
      "flatten_22 (Flatten)         (None, 36992)             0         \n",
      "_________________________________________________________________\n",
      "dense_43 (Dense)             (None, 128)               4735104   \n",
      "_________________________________________________________________\n",
      "activation_73 (Activation)   (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dropout_22 (Dropout)         (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_44 (Dense)             (None, 1)                 129       \n",
      "_________________________________________________________________\n",
      "activation_74 (Activation)   (None, 1)                 0         \n",
      "=================================================================\n",
      "Total params: 4,847,809\n",
      "Trainable params: 4,847,809\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "---> Save model ...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "print('--> Training Conv-Conv-Conv-FC ver 2 ...')\n",
    "model = CNN_3Conv_1FC(convs=64, dims=3, epochs=50, batch_size=64, input_shape= input_shape)\n",
    "\n",
    "print(model.summary())\n",
    "\n",
    "# Save the weights and the full model\n",
    "print('---> Save model ...')\n",
    "model.save_weights(os.path.join(modelFolder,'model3_2_Conv-Conv-Conv-FC_weigths.h5'))\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Good! It is possible to obtain 88% accuracy in only 50 iterations! So, let's see what we can have in 100 iterations:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation/Test loss    : 0.2198791243135929\n",
      "Validation/Test accuracy: 0.921875\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_31 (Conv2D)           (None, 148, 148, 64)      1792      \n",
      "_________________________________________________________________\n",
      "activation_75 (Activation)   (None, 148, 148, 64)      0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_31 (MaxPooling (None, 74, 74, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_32 (Conv2D)           (None, 72, 72, 64)        36928     \n",
      "_________________________________________________________________\n",
      "activation_76 (Activation)   (None, 72, 72, 64)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_32 (MaxPooling (None, 36, 36, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_33 (Conv2D)           (None, 34, 34, 128)       73856     \n",
      "_________________________________________________________________\n",
      "activation_77 (Activation)   (None, 34, 34, 128)       0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_33 (MaxPooling (None, 17, 17, 128)       0         \n",
      "_________________________________________________________________\n",
      "flatten_23 (Flatten)         (None, 36992)             0         \n",
      "_________________________________________________________________\n",
      "dense_45 (Dense)             (None, 128)               4735104   \n",
      "_________________________________________________________________\n",
      "activation_78 (Activation)   (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dropout_23 (Dropout)         (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_46 (Dense)             (None, 1)                 129       \n",
      "_________________________________________________________________\n",
      "activation_79 (Activation)   (None, 1)                 0         \n",
      "=================================================================\n",
      "Total params: 4,847,809\n",
      "Trainable params: 4,847,809\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "---> Save model ...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "print('--> Training Conv-Conv-Conv-FC ver 3 ...')\n",
    "model = CNN_3Conv_1FC(convs=64, dims=3, epochs=100, batch_size=64, input_shape= input_shape)\n",
    "\n",
    "print(model.summary())\n",
    "\n",
    "# Save the weights and the full model\n",
    "print('---> Save model ...')\n",
    "model.save_weights(os.path.join(modelFolder,'model3_3_Conv-Conv-Conv-FC_weigths.h5'))\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Conclusions**: Using `Conv-Conv-Conv-FC` topologies are improving the previous results up to `92.2% accuracy` (4,847,809 parameters) in 100 iterations! Even more: with only `1,212,513` parameters for the 32, 32, 64 Convs (model3_1) it is possible to have `87.9% accuracy`. Please do not forget about data augmentation! These results confirms the DL theory about the abstract patterns in lower layers to the more general ones in the next layers.\n",
    "\n",
    "### Summary 1Conv-FC, 2Conv-FC, 3Conv-FC\n",
    "\n",
    "Let's summarize the classifiers until this point (`only 50 iterations`). *Please remember that we didn't search for all optimal hyperparameters and these results are just to have a general idea about the power of basic CNNs*.\n",
    "\n",
    "| Model    | Topology  | Filters   | Dim   | Params     | Accuracy |\n",
    "| -------- | --------- | --------- | ----- | ---------- | -------- |\n",
    "| model1_1 | 1Conv-FC  | 32        | 3 x 3 |  5,608,385 | 0.824    |\n",
    "| model1_2 | 1Conv-FC  | 32        | 5 x 5 |  5,459,393 | 0.742    |\n",
    "| model1_3 | 1Conv-FC  | 64        | 3 x 3 | 22,431,617 | 0.824    |\n",
    "| model1_4 | 1Conv-FC  | 128       | 3 x 3 | 89,722,625 | 0.859    |\n",
    "| model2_1 | 2Conv-FC  | 32, 64    | 3 x 3 |  5,327,937 | 0.867    |\n",
    "| model2_2 | 2Conv-FC  | 64, 128   | 3 x 3 | 21,309,569 | 0.836    |\n",
    "| model3_1 | 3Conv-FC  | 32,32,64  | 3 x 3 |  1,212,513 | 0.879    |\n",
    "| model3_2 | 3Conv-FC  | 64,64,128 | 3 x 3 |  4,847,809 | 0.883*   |\n",
    "\n",
    "(* 92.2% accuracy in 100 iterations)\n",
    "\n",
    "## Hyperparameter optimization for 3 Conv - FC\n",
    "\n",
    "In the next section, I shall use the topology Conv-Conv-Conv-FC (64, 64, 128 filters) and try to find the best hyperparameters."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [],
   "source": [
    "def CNN_3Conv_1FC_optim(convs, dims, epochs, batch_size, drop_rate, activFunction, optimizer, input_shape):\n",
    "    # CNN with 3 Conv + 1 FC\n",
    "    # convs = no of filters\n",
    "    # dims = dimension of convolution dims x dims\n",
    "    # epochs = training epochs\n",
    "    # batch_size = training batch size\n",
    "    # drop_rate = dropout of neurons\n",
    "    # activFunction = 'relu', 'tanh', etc.\n",
    "    # input_shape = shape of the inputs\n",
    "    \n",
    "    # for reproductibility\n",
    "    #seed(1)            # numpy seed\n",
    "    #set_random_seed(2) # tensorflow seed\n",
    "    \n",
    "    # Start the computational graph for our CNN\n",
    "    model = Sequential()\n",
    "    \n",
    "    # Conv 1 filters\n",
    "    model.add(Conv2D(convs, (dims, dims), input_shape=input_shape))\n",
    "    model.add(Activation(activFunction))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    \n",
    "    # Conv 2 filters\n",
    "    model.add(Conv2D(convs, (dims, dims), input_shape=input_shape))\n",
    "    model.add(Activation(activFunction))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    \n",
    "    # Conv 3 filters\n",
    "    model.add(Conv2D(convs*2, (dims, dims))) # second Conv has 2x filters!\n",
    "    model.add(Activation(activFunction))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    # Convert convolution to a fully connected layer\n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(convs*2)) # FC has 2x neurons!\n",
    "    model.add(Activation(activFunction))\n",
    "    model.add(Dropout(drop_rate))\n",
    "    model.add(Dense(1))\n",
    "    model.add(Activation('sigmoid'))\n",
    "    \n",
    "    # Compile the model\n",
    "    model.compile(loss='binary_crossentropy',\n",
    "                  optimizer=optimizer,\n",
    "                  metrics=['accuracy'])\n",
    "\n",
    "    # Aaugmentation configuration for training set\n",
    "    train_datagen = ImageDataGenerator(\n",
    "        rescale=1. / 255,\n",
    "        shear_range=0.2,\n",
    "        zoom_range =0.2,\n",
    "        horizontal_flip=True,\n",
    "        vertical_flip=True,\n",
    "        rotation_range=90)\n",
    "\n",
    "    # Augmentation configuration for testing set: only rescaling!\n",
    "    test_datagen = ImageDataGenerator(rescale=1. / 255)\n",
    "    \n",
    "    train_generator = train_datagen.flow_from_directory(\n",
    "        train_data_dir,\n",
    "        target_size=(img_width, img_height),\n",
    "        batch_size=batch_size,\n",
    "        class_mode='binary')\n",
    "\n",
    "    validation_generator = test_datagen.flow_from_directory(\n",
    "        validation_data_dir,\n",
    "        target_size=(img_width, img_height),\n",
    "        batch_size=batch_size,\n",
    "        class_mode='binary')\n",
    "\n",
    "    model.fit_generator(\n",
    "        train_generator,\n",
    "        steps_per_epoch=nb_train_samples // batch_size,\n",
    "        epochs=epochs,\n",
    "        validation_data=validation_generator,\n",
    "        validation_steps=nb_validation_samples // batch_size,\n",
    "        verbose=0,\n",
    "        workers=7) # using multi-cpu!\n",
    "\n",
    "    # Evaluate final test loss and accuracy scores\n",
    "    score = model.evaluate_generator(validation_generator, nb_validation_samples//batch_size, workers=7)\n",
    "    return (score[0],score[1]) # test loss and accuracy"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Search**: Let's see what we can obtain if we search for the most accurate classifier with different `batch size` and `drop rate` values. I will use `workers=7` in order to use parallel CPU cores for fit and evaluate functions. *Please adapt this value to your device CPU cores!*\n",
    "\n",
    "**Notes**:\n",
    "* Topology: Conv-Conv-Conv-FC (64-64-128 filters).\n",
    "* Filter dimension `3 x 3`, 50 epochs.\n",
    "* Optimizers are using their default parameters.\n",
    "* `accuracy` is used for training and testing.\n",
    "* `binary crossentropy` is the loss function with the last activation function as `sigmoid` (to obtain polyps or non-polyps class).\n",
    "* `MaxPooling2D` is 2 x 2.\n",
    "* I will use the `for` loops without other SearchGrid functions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "--> Search for the best hyperparameters: batch size and drop rate ...\n",
      "Fixed params - Filters, Filter size, Epochs, Activations, Optimizer: 64 3 50 relu Adam\n",
      "Searching params - Batch size, Drop rate [8, 16, 32, 64, 128] [0.1, 0.5, 0.9]\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 8 0.1 relu Adam 0.423486386249597 0.8412162162162162\n",
      "--> Iter time: 1.9 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 8 0.5 relu Adam 0.3057052805214315 0.8783783783783784\n",
      "--> Iter time: 1.9 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 8 0.9 relu Adam 0.6931530849353688 0.5\n",
      "--> Iter time: 1.9 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 16 0.1 relu Adam 0.391617973231607 0.8923611111111112\n",
      "--> Iter time: 1.6 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 16 0.5 relu Adam 0.2892870720889833 0.9166666666666666\n",
      "--> Iter time: 1.6 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 16 0.9 relu Adam 0.5035901019970576 0.7430555555555556\n",
      "--> Iter time: 1.6 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 32 0.1 relu Adam 0.3220406182938152 0.9097222222222222\n",
      "--> Iter time: 1.5 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 32 0.5 relu Adam 0.32446981635358596 0.8888888888888888\n",
      "--> Iter time: 1.5 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 32 0.9 relu Adam 0.3362124347024494 0.8819444444444444\n",
      "--> Iter time: 1.5 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 64 0.1 relu Adam 0.2848696671426296 0.8984375\n",
      "--> Iter time: 1.5 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 64 0.5 relu Adam 0.269416157156229 0.92578125\n",
      "--> Iter time: 1.5 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 64 0.9 relu Adam 0.4051167443394661 0.84375\n",
      "--> Iter time: 1.5 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 128 0.1 relu Adam 0.4041638672351837 0.84765625\n",
      "--> Iter time: 1.6 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 128 0.5 relu Adam 0.47337886691093445 0.77734375\n",
      "--> Iter time: 1.6 mins ---\n",
      "Found 910 images belonging to 2 classes.\n",
      "Found 302 images belonging to 2 classes.\n",
      "64 3 50 128 0.9 relu Adam 0.37934036552906036 0.859375\n",
      "--> Iter time: 1.6 mins ---\n",
      "Done! Total time: 24.3 mins ---\n"
     ]
    }
   ],
   "source": [
    "# Search for the best hyperparameters:  batch size and drop rate\n",
    "print('--> Search for the best hyperparameters: batch size and drop rate ...')\n",
    "\n",
    "# Fixed params from the best model in Search 1\n",
    "convs = 64\n",
    "dims = 3\n",
    "epochs = 50\n",
    "activFunction = 'relu'\n",
    "optimizer = 'Adam'\n",
    "print('Fixed params - Filters, Filter size, Epochs, Activations, Optimizer:', convs, dims, epochs, activFunction, optimizer)\n",
    "\n",
    "# Grid search for parameters\n",
    "batch_sizeGrid = [8, 16, 32, 64, 128]\n",
    "drop_rateGrid  = [0.1, 0.5, 0.9]\n",
    "print('Searching params - Batch size, Drop rate', batch_sizeGrid, drop_rateGrid)\n",
    "\n",
    "# Start total timer\n",
    "start_time = time.time()\n",
    "\n",
    "# get all the results as list\n",
    "results = []\n",
    "for batch_size in batch_sizeGrid:\n",
    "    for drop_rate in drop_rateGrid:\n",
    "        # timer for each iteration\n",
    "        istart_time = time.time()\n",
    "        # get Loss and Accuracy for CNN Conv-Conv-Conv-FC with specific parameters\n",
    "        iLoss, iAccuracy = CNN_3Conv_1FC_optim(convs, dims, epochs, batch_size, drop_rate, activFunction, optimizer, input_shape)\n",
    "        # append the results to a list\n",
    "        results.append([convs, dims, epochs, batch_size, drop_rate, activFunction, optimizer, iLoss, iAccuracy])\n",
    "        # print for each iteration the results and the iteration time\n",
    "        print(convs, dims, epochs, batch_size, drop_rate, activFunction, optimizer, iLoss, iAccuracy)\n",
    "        print(\"--> Iter time: %0.1f mins ---\" % ((time.time() - istart_time) / 60))\n",
    "\n",
    "# Print total time for the parameter search \n",
    "print(\"Done! Total time: %0.1f mins ---\" % ((time.time() - start_time)/60))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's save these results on the disk as `ParamSearch.csv` in `results` subfolder:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "def WriteListOfList(sFile, header):\n",
    "    # Write search results to disk\n",
    "    # sFile = path + filename to write the list\n",
    "    # header = custom header for the list data\n",
    "\n",
    "    with open(sFile, 'w') as f:\n",
    "        f.write(header)\n",
    "        for _list in results:\n",
    "            sout = ''\n",
    "            for _string in _list:\n",
    "                sout = sout + str(_string) + ','\n",
    "            f.write(sout[:-1] + '\\n')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [],
   "source": [
    "sFile = './results/ParamSearch.csv'\n",
    "header= 'filterNo, filterSize, epochs, batch_size, drop_rate, activFunction, optimizer, Loss, Accuracy\\n'\n",
    "WriteListOfList(sFile, header)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the results for the second search parameters we can see that same set of hyperparameters gives the best accuracy results. The accuracy difference are generated by several random initializations such as weigths in layers, etc.\n",
    "\n",
    "| Batch size | Drop rate | Accuracy |\n",
    "| --------   | --------- | -------- |\n",
    "| 64 | 0.5 | 0.926 |\n",
    "| 16 | 0.5 | 0.917 |\n",
    "| 32 | 0.1 | 0.910 |\n",
    "| 64 | 0.1 | 0.898 |\n",
    "| 16 | 0.1 | 0.892 |\n",
    "| 32 | 0.5 | 0.889 |\n",
    "| 32 | 0.9 | 0.882 |\n",
    "| 8 | 0.5 | 0.878 |\n",
    "| 128 | 0.9 | 0.859 |\n",
    "| 128 | 0.1 | 0.848 |\n",
    "| 64 | 0.9 | 0.844 |\n",
    "| 8 | 0.1 | 0.841 |\n",
    "| 128 | 0.5 | 0.777 |\n",
    "| 16 | 0.9 | 0.743 |\n",
    "| 8 | 0.9 | 0.500 |\n",
    "\n",
    "**You can observe that in less than 2 minutes you can train a 90% accuracy classifier for colonoscopy polyps!**\n",
    "\n",
    "Let's train again the best model using the interactive plot of the loss during `500 epochs`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Validation/Test loss    : 0.15933184139430523\n",
      "Validation/Test accuracy: 0.96484375\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_148 (Conv2D)          (None, 148, 148, 64)      1792      \n",
      "_________________________________________________________________\n",
      "activation_270 (Activation)  (None, 148, 148, 64)      0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_148 (MaxPoolin (None, 74, 74, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_149 (Conv2D)          (None, 72, 72, 64)        36928     \n",
      "_________________________________________________________________\n",
      "activation_271 (Activation)  (None, 72, 72, 64)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_149 (MaxPoolin (None, 36, 36, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_150 (Conv2D)          (None, 34, 34, 128)       73856     \n",
      "_________________________________________________________________\n",
      "activation_272 (Activation)  (None, 34, 34, 128)       0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_150 (MaxPoolin (None, 17, 17, 128)       0         \n",
      "_________________________________________________________________\n",
      "flatten_62 (Flatten)         (None, 36992)             0         \n",
      "_________________________________________________________________\n",
      "dense_123 (Dense)            (None, 128)               4735104   \n",
      "_________________________________________________________________\n",
      "activation_273 (Activation)  (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dropout_62 (Dropout)         (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_124 (Dense)            (None, 1)                 129       \n",
      "_________________________________________________________________\n",
      "activation_274 (Activation)  (None, 1)                 0         \n",
      "=================================================================\n",
      "Total params: 4,847,809\n",
      "Trainable params: 4,847,809\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "---> Save model ...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "print('--> Training best model for Small CNNs: Conv-Conv-Conv-FC ...')\n",
    "model = CNN_3Conv_1FC(convs=64, dims=3, epochs=500, batch_size=64, input_shape= input_shape)\n",
    "print(model.summary())\n",
    "\n",
    "# Save the weights and the full model\n",
    "print('---> Save model ...')\n",
    "model.save_weights(os.path.join(modelFolder,'model_best_Conv-Conv-Conv-FC_weigths.h5'))\n",
    "model.save(os.path.join(modelFolder,'model_best_Conv-Conv-Conv-FC_full.h5'))\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can observe an overfitting during the training and therefore I will try extra regularization by increasing the drop rate to `0.9`:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 69,
   "metadata": {},
   "outputs": [],
   "source": [
    "def CNN_3Conv_1FCv2(convs, dims, epochs, batch_size, drop_rate, input_shape):\n",
    "    # CNN with 3 Conv + 1 FC\n",
    "    # convs = no of filters\n",
    "    # dims = dimension of convolution dims x dims\n",
    "    # epochs = training epochs\n",
    "    # batch_size = training batch size\n",
    "    # drop_rate = drop rate (0, 1.0) <---- added\n",
    "    # input_shape = shape of the inputs\n",
    "    \n",
    "    # for reproductibility\n",
    "    seed(1)            # numpy seed\n",
    "    set_random_seed(2) # tensorflow seed\n",
    "    \n",
    "    # Start the computational graph for our CNN\n",
    "    model = Sequential()\n",
    "    \n",
    "    # Conv 1 filters\n",
    "    model.add(Conv2D(convs, (dims, dims), input_shape=input_shape))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    \n",
    "    # Conv 2 filters\n",
    "    model.add(Conv2D(convs, (dims, dims), input_shape=input_shape))\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "    \n",
    "    # Conv 3 filters\n",
    "    model.add(Conv2D(convs*2, (dims, dims))) # second Conv has 2x filters!\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(MaxPooling2D(pool_size=(2, 2)))\n",
    "\n",
    "    # Convert convolution to a fully connected layer\n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(convs*2)) # FC has 2x neurons!\n",
    "    model.add(Activation('relu'))\n",
    "    model.add(Dropout(drop_rate))\n",
    "    model.add(Dense(1))\n",
    "    model.add(Activation('sigmoid'))\n",
    "    \n",
    "    # Compile the model\n",
    "    model.compile(loss='binary_crossentropy',\n",
    "                  optimizer='adam',\n",
    "                  metrics=['accuracy'])\n",
    "\n",
    "    # Aaugmentation configuration for training set\n",
    "    train_datagen = ImageDataGenerator(\n",
    "        rescale=1. / 255,\n",
    "        shear_range=0.2,\n",
    "        zoom_range =0.2,\n",
    "        horizontal_flip=True,\n",
    "        vertical_flip=True,\n",
    "        rotation_range=90)\n",
    "\n",
    "    # Augmentation configuration for testing set: only rescaling!\n",
    "    test_datagen = ImageDataGenerator(rescale=1. / 255)\n",
    "    \n",
    "    train_generator = train_datagen.flow_from_directory(\n",
    "        train_data_dir,\n",
    "        target_size=(img_width, img_height),\n",
    "        batch_size=batch_size,\n",
    "        class_mode='binary')\n",
    "\n",
    "    validation_generator = test_datagen.flow_from_directory(\n",
    "        validation_data_dir,\n",
    "        target_size=(img_width, img_height),\n",
    "        batch_size=batch_size,\n",
    "        class_mode='binary')\n",
    "\n",
    "    model.fit_generator(\n",
    "        train_generator,\n",
    "        steps_per_epoch=nb_train_samples // batch_size,\n",
    "        epochs=epochs,\n",
    "        validation_data=validation_generator,\n",
    "        validation_steps=nb_validation_samples // batch_size,\n",
    "        callbacks=[plot_losses], # using the plotting function for train and validation loss\n",
    "        verbose=0)\n",
    "\n",
    "    # Evaluate final test loss and accuracy scores\n",
    "    score_val = model.evaluate_generator(validation_generator, nb_validation_samples//batch_size, workers=7)\n",
    "    score_tr  = model.evaluate_generator(train_generator, nb_train_samples//batch_size, workers=7)\n",
    "    print('Train loss    :', score_tr[0])\n",
    "    print('Train accuracy:', score_tr[1])\n",
    "    print('Validation loss    :', score_val[0])\n",
    "    print('Validation accuracy:', score_val[1])\n",
    "    \n",
    "    return model # return the model!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train loss    : 0.0937776898166963\n",
      "Train accuracy: 0.9654017857142857\n",
      "Validation loss    : 0.2025301423855126\n",
      "Validation accuracy: 0.94921875\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "conv2d_169 (Conv2D)          (None, 148, 148, 64)      1792      \n",
      "_________________________________________________________________\n",
      "activation_305 (Activation)  (None, 148, 148, 64)      0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_169 (MaxPoolin (None, 74, 74, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_170 (Conv2D)          (None, 72, 72, 64)        36928     \n",
      "_________________________________________________________________\n",
      "activation_306 (Activation)  (None, 72, 72, 64)        0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_170 (MaxPoolin (None, 36, 36, 64)        0         \n",
      "_________________________________________________________________\n",
      "conv2d_171 (Conv2D)          (None, 34, 34, 128)       73856     \n",
      "_________________________________________________________________\n",
      "activation_307 (Activation)  (None, 34, 34, 128)       0         \n",
      "_________________________________________________________________\n",
      "max_pooling2d_171 (MaxPoolin (None, 17, 17, 128)       0         \n",
      "_________________________________________________________________\n",
      "flatten_69 (Flatten)         (None, 36992)             0         \n",
      "_________________________________________________________________\n",
      "dense_137 (Dense)            (None, 128)               4735104   \n",
      "_________________________________________________________________\n",
      "activation_308 (Activation)  (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dropout_69 (Dropout)         (None, 128)               0         \n",
      "_________________________________________________________________\n",
      "dense_138 (Dense)            (None, 1)                 129       \n",
      "_________________________________________________________________\n",
      "activation_309 (Activation)  (None, 1)                 0         \n",
      "=================================================================\n",
      "Total params: 4,847,809\n",
      "Trainable params: 4,847,809\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n",
      "---> Save model ...\n",
      "Done!\n"
     ]
    }
   ],
   "source": [
    "print('--> Training best model for Small CNNs: Conv-Conv-Conv-FC ...')\n",
    "model = CNN_3Conv_1FCv2(convs=64, dims=3, epochs=400, batch_size=64,\n",
    "                        drop_rate=0.9, input_shape= input_shape)\n",
    "print(model.summary())\n",
    "\n",
    "# Save the weights and the full model\n",
    "print('---> Save model ...')\n",
    "model.save_weights(os.path.join(modelFolder,'model_best2_Conv-Conv-Conv-FC_weigths.h5'))\n",
    "model.save(os.path.join(modelFolder,'model_best2_Conv-Conv-Conv-FC_full.h5'))\n",
    "print('Done!')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "With 0.9 dropout we can obtain less accuracy but less overfitting: 0.949 for validation vs. 0.965 for training.\n",
    "\n",
    "**Conclusion**: The best model for colonoscopy using a small CNN could have an test accuracy of `96.5%` using a limited number of images but using keras data augmentation."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Final conclusion\n",
    "\n",
    "If you want a classifier to predict a polyp into colonoscopy images, you can use this small dataset with keras augmented data and with only a `small CNN` (Conv-Conv-Conv-FC, similar with LeCun) you can obtain an accuracy `over 90% in only 2 minutes`. In the next script, Transfer Learing will be tested with well known already trained models such as VGG or Inception (see [4-TransferLearningVGG16.ipynb](4-TransferLearningVGG16.ipynb)).\n",
    "\n",
    "Hf with DL! @muntisa"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**References**\n",
    "\n",
    "* [Dataset on Dropbox](https://www.dropbox.com/s/p5qe9eotetjnbmq/CVC-ClinicDB.rar?dl=0)\n",
    "* [Keras blog with scripts](https://blog.keras.io/building-powerful-image-classification-models-using-very-little-data.html)\n",
    "* [Blog: Colorectal cancer, polyps and deep learning](http://site.uit.no/deephealthresearch/2017/07/19/polyp-detection-using-deep-learning/) (2017)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Acknowledgements\n",
    "\n",
    "I gratefully acknowledge the support of NVIDIA Corporation with the donation of the Titan Xp GPU used for this research ([https://developer.nvidia.com/academic_gpu_seeding](https://developer.nvidia.com/academic_gpu_seeding))."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}