2-us8k-ffn-train-predict.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "## Workbook 2 - Training a Feed-Forward Neural Network using UrbanSound8K audio data "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This loads the features we extracted using librosa (see workbook 1), and uses them to train a Feed-Forward Network, a simple deep neural net with a few hidden layers, and then predict the class of a set of previously unheard audio files. You will need librosa, keras and tensorflow installed to run this code, and the latest version of scikit-learn (currently 0.18.1)."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "import glob\n",
    "import os\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "plt.style.use('ggplot')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This code below loads previously computed numpy features for each of the 10 folds, we'll use the first 8 for training, and keep one back for validation and one for testing."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "Adding fold1\n",
      "New Features:  (873, 193)\n",
      "\n",
      "Adding fold2\n",
      "New Features:  (888, 193)\n",
      "\n",
      "Adding fold3\n",
      "New Features:  (925, 193)\n",
      "\n",
      "Adding fold4\n",
      "New Features:  (990, 193)\n",
      "\n",
      "Adding fold5\n",
      "New Features:  (936, 193)\n",
      "\n",
      "Adding fold6\n",
      "New Features:  (823, 193)\n",
      "\n",
      "Adding fold7\n",
      "New Features:  (838, 193)\n",
      "\n",
      "Adding fold8\n",
      "New Features:  (806, 193)\n",
      "\n",
      "Training Set Features: (7079, 193) Labels: (7079, 10)\n",
      "Validation Set Features: (816, 193) Labels: (816, 10)\n",
      "Test Set Features: (837, 193) Labels: (837, 10)\n"
     ]
    }
   ],
   "source": [
    "data_dir = \"data/us8k-np-ffn\"\n",
    "\n",
    "def add_folds():\n",
    "    subsequent_fold = False\n",
    "    for k in range(1,9):\n",
    "        fold_name = 'fold' + str(k)\n",
    "        print \"\\nAdding \" + fold_name\n",
    "        feature_file = os.path.join(data_dir, fold_name + '_x.npy')\n",
    "        labels_file = os.path.join(data_dir, fold_name + '_y.npy')\n",
    "        loaded_features = np.load(feature_file)\n",
    "        loaded_labels = np.load(labels_file)\n",
    "        print \"New Features: \", loaded_features.shape\n",
    "\n",
    "        if subsequent_fold:\n",
    "            features = np.concatenate((features, loaded_features))\n",
    "            labels = np.concatenate((labels, loaded_labels))\n",
    "        else:\n",
    "            features = loaded_features\n",
    "            labels = loaded_labels\n",
    "            subsequent_fold = True\n",
    "        \n",
    "    return features, labels\n",
    "\n",
    "train_x, train_y = add_folds()\n",
    "\n",
    "# use a fold for validation\n",
    "valid_fold_name = 'fold9'\n",
    "feature_file = os.path.join(data_dir, valid_fold_name + '_x.npy')\n",
    "labels_file = os.path.join(data_dir, valid_fold_name + '_y.npy')\n",
    "valid_x = np.load(feature_file)\n",
    "valid_y = np.load(labels_file) \n",
    "\n",
    "# and a fold for testing\n",
    "test_fold_name = 'fold10'\n",
    "feature_file = os.path.join(data_dir, test_fold_name + '_x.npy')\n",
    "labels_file = os.path.join(data_dir, test_fold_name + '_y.npy')\n",
    "test_x = np.load(feature_file)\n",
    "test_y = np.load(labels_file)\n",
    "\n",
    "print \"\\nTraining Set Features:\", train_x.shape, \"Labels:\", train_y.shape\n",
    "print \"Validation Set Features:\", valid_x.shape, \"Labels:\", valid_y.shape\n",
    "print \"Test Set Features:\", test_x.shape, \"Labels:\", test_y.shape\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "--------------------------------------------------------------------------------------------------------------------------"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training Neural Network with Keras and TensorFlow"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "from sklearn.metrics import precision_recall_fscore_support, roc_auc_score\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Dropout\n",
    "from keras.optimizers import Adam\n",
    "from keras.callbacks import EarlyStopping"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 414,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Features: 193 Classes: 10\n"
     ]
    }
   ],
   "source": [
    "def assure_path_exists(path):\n",
    "    mydir = os.path.join(os.getcwd(), path)\n",
    "    if not os.path.exists(mydir):\n",
    "        os.makedirs(mydir)\n",
    "\n",
    "# neural network dimensions\n",
    "n_dim = train_x.shape[1]\n",
    "n_classes = train_y.shape[1]\n",
    "n_hidden_units_1 = n_dim\n",
    "n_hidden_units_2 = 400 # approx n_dim * 2\n",
    "n_hidden_units_3 = 200 # half of layer 2\n",
    "\n",
    "checkpoint_dir = \"model\"\n",
    "assure_path_exists(checkpoint_dir)\n",
    "\n",
    "print \"Features:\", n_dim, \"Classes:\", n_classes"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following cell contains the code to create and train a feed-forward neural network with 3 hidden layers.\n",
    "\n",
    "In successive training epochs, notice how the training set loss decreases, whilst the accuracy increases - this shows the model is learning. But we don't want the model to simply memorise the training data and not have the capability to generalise to examples it hasn't seen before. So a stopping function is used to evaluate a separate measure of loss against an unseen validation set, which allows training to be halted at the point when the model begins to lose its ability to generalise.\n",
    "\n",
    "The neural network implemented below has 3 hidden (densely connected) layers, plus an output layer. In between layers 2 and 3, and 3 and 4, a Dropout policy of 0.5 has been added. This means there's a 50% that a neuron's activation output will be ignored, and not propagated to its downstream connections, and it will not have its weights updated on the backward pass.\n",
    "\n",
    "This random throwing away of information helps prevent the network from learning simple spurious dependencies, and promotes the creation of complex co-adaptations between neurons of the hidden layers. The idea is to encourage other neurons to become involved, if they learn on behalf of the missing neurons, it will result in the creation of multiple independent internal representations by different groupings of neurons across the network. \n",
    "\n",
    "Another parameter that affects performance is the batch size: the number of feature rows to be shown to the network before the backpropagation process updates the weights of the network. \n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 446,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 7079 samples, validate on 816 samples\n",
      "Epoch 1/30\n",
      "7079/7079 [==============================] - 17s - loss: 2.2883 - acc: 0.3007 - val_loss: 1.5413 - val_acc: 0.4841\n",
      "Epoch 2/30\n",
      "7079/7079 [==============================] - 5s - loss: 1.5947 - acc: 0.4681 - val_loss: 1.3609 - val_acc: 0.5098\n",
      "Epoch 3/30\n",
      "7079/7079 [==============================] - 5s - loss: 1.3750 - acc: 0.5447 - val_loss: 1.3245 - val_acc: 0.5490\n",
      "Epoch 4/30\n",
      "7079/7079 [==============================] - 6s - loss: 1.1998 - acc: 0.6067 - val_loss: 1.2851 - val_acc: 0.5833\n",
      "Epoch 5/30\n",
      "7079/7079 [==============================] - 6s - loss: 1.0744 - acc: 0.6533 - val_loss: 1.3241 - val_acc: 0.5404\n",
      "Epoch 6/30\n",
      "7079/7079 [==============================] - 6s - loss: 0.9533 - acc: 0.6966 - val_loss: 1.3954 - val_acc: 0.5784\n",
      "Epoch 7/30\n",
      "7079/7079 [==============================] - 6s - loss: 0.8557 - acc: 0.7237 - val_loss: 1.4544 - val_acc: 0.5846\n",
      "Epoch 8/30\n",
      "7079/7079 [==============================] - 6s - loss: 0.7756 - acc: 0.7466 - val_loss: 1.5494 - val_acc: 0.5735\n",
      "Epoch 00007: early stopping\n"
     ]
    }
   ],
   "source": [
    "tf.set_random_seed(0)\n",
    "np.random.seed(0)\n",
    "\n",
    "def create_model(activation_function='relu', init_type='normal', optimiser='Adamax', dropout_rate=0.5):\n",
    "    model = Sequential()\n",
    "    # layer 1\n",
    "    model.add(Dense(n_hidden_units_1, input_dim=n_dim, init=init_type, activation=activation_function))\n",
    "    # layer 2\n",
    "    model.add(Dense(n_hidden_units_2, init=init_type, activation=activation_function))\n",
    "    model.add(Dropout(dropout_rate))\n",
    "    # layer 3\n",
    "    model.add(Dense(n_hidden_units_3, init=init_type, activation=activation_function))\n",
    "    model.add(Dropout(dropout_rate))\n",
    "    # output layer\n",
    "    model.add(Dense(n_classes, init=init_type, activation='softmax'))\n",
    "    \n",
    "    model.compile(loss='categorical_crossentropy', optimizer=optimiser, metrics=['accuracy'])\n",
    "    return model\n",
    "\n",
    "# a stopping function to stop training before we excessively overfit to the training set\n",
    "earlystop = EarlyStopping(monitor='val_loss', patience=3, verbose=1, mode='auto')\n",
    "\n",
    "model = create_model()\n",
    "\n",
    "history = model.fit(train_x, train_y, validation_data=(valid_x, valid_y), callbacks=[earlystop], nb_epoch=30, batch_size=24)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Measuring Accuracy"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 445,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ROC: 0.91\n",
      "837/837 [==============================] - 0s     \n",
      "\n",
      "Accuracy = 0.61\n",
      "F-Score: 0.61\n"
     ]
    }
   ],
   "source": [
    "from sklearn import metrics \n",
    "from keras.utils import np_utils\n",
    "\n",
    "# obtain the prediction probabilities\n",
    "y_prob = model.predict_proba(test_x, verbose=0)\n",
    "y_pred = np_utils.probas_to_classes(y_prob)\n",
    "y_true = np.argmax(test_y, 1)\n",
    "\n",
    "roc = metrics.roc_auc_score(test_y, y_prob)\n",
    "print \"ROC:\",  round(roc,2)\n",
    "\n",
    "# evaluate the model\n",
    "score, accuracy = model.evaluate(test_x, test_y, batch_size=32)\n",
    "print(\"\\nAccuracy = {:.2f}\".format(accuracy))\n",
    "\n",
    "p,r,f,s = precision_recall_fscore_support(y_true, y_pred, average='micro')\n",
    "print \"F-Score:\", round(f,2)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "I used GridSearch (see section below) to investigate the best combinations of parameters, such as optimisers, dropout rates, hidden layers and activation functions. With 2 hidden layers the best accuracy I observed was an F-score of 0.59 (ROC = 0.89)\n",
    "\n",
    "With 3 hidden layers the best accuracy seen was an F-score of **0.645** (ROC=0.92), using the Adamax optimiser and a dropout rate of 0.5. There did not seem to be any benefit in using 4 hidden layers, the accuracy seemed equivalent to that achieved with just 2 hidden layers. (As the gradient computation is stochastic, you might get slightly different values when you run this code). "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "ROC curves are typically used in binary classification, so to apply this measure to multi-class classification, it's first necessary to convert the predictions into boolean values, each representing positive or negative predictions for each possible class. This enables us to draw one ROC curve per class, or draw a composite ROC curve by using micro-averaging."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 365,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAY8AAAEhCAYAAACHjCx5AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzs3XlYVGX7B/DvGfZlQEZANk0UWRxXQEERBSNzyaI0ilzS\nXFLR113TFHHN3UxFyjAtTUPNJV9fTVNQFAUFTFAQUkgEERwQkJ05vz/4MTiyDcvs9+e6vC7OmbPc\n88w493nOeRaGZVkWhBBCSDNw5B0AIYQQ5UPJgxBCSLNR8iCEENJslDwIIYQ0GyUPQgghzUbJgxBC\nSLNR8iB1pKeng8Ph4MaNG/IORelERERAQ0MDmZmZ8g4Fq1evRrdu3eQdhkxNnjwZw4YNa9UxFOkz\nVGSUPBTE5MmTweFwwOFwoKmpiY4dO+Lzzz+Xyxe4U6dOePbsGdzc3GR+bmWipaWFn3/+WWydh4cH\nsrKyYGVlJaeoxDEMI+8QGvT06VNwOBxcvXq1zY753Xff4dixYxJvrwyfoaKi5KFABg8ejOzsbDx5\n8gRHjhxBXFwc/Pz8ZB4HwzAwNzeHhoaGVM9TUVEh1eO3RGVlZav219TUhLm5eRtFo5haW0Y1WJZt\ns+RWExOXy4WxsXGrjqUOn2FboOShQLS1tWFmZgZLS0sMGjQI06dPR1RUFIqKikTbVFZWIigoCF26\ndIGenh569uyJH374Qew4r169wrx589CpUyfo6uqiS5cu2Lhxo+j158+fY9KkSTA3N4eRkRE8PT1x\n7do10etv3rYaNGgQZsyYUSdeJycnBAYGipaPHj2Kvn37Qk9PD7a2tli4cCGKi4tFr3t7e2Pq1KkI\nDAyElZUV3nrrrQbL4ubNmxgyZAj09fXB4/Ewbtw45OTkiF6vuSVz5MgRdO3aFXp6ehg2bBjS09PF\njnPx4kUMGjQI+vr6sLGxwRdffAGBQCB6ffLkyXjnnXewe/du2NraQldXF2VlZbh06RK8vb3Rvn17\ntGvXDl5eXoiJiRHtZ2trC6FQKKox1iTa8PBwcDgcUY0xIiICHA4Hly5dwpAhQ2BgYAA+n4/z58+L\nxRkXF4cBAwZAT08PTk5OOHnyJGxtbbFhw4YGywgA7ty5gxEjRsDY2BhcLhfu7u5icQLAmTNn4OTk\nBENDQ3h7eyM1NVX0Wn5+PiZMmIC33noL+vr6cHR0xPbt28X2b2kZAY1/Fzt16gQA8PLyAofDQZcu\nXVr9uU2aNEnsttX9+/cxfPhwmJiYwNDQEHw+H4cPH27WZwgAjx49wtixY9G+fXsYGBigT58+OHfu\nXKOfjaqj5KGgMjMzcezYMWhoaIjVAKZOnYpTp05h3759SEpKQmBgIL766iv89NNPom1GjRqFs2fP\nYs+ePUhKSsKhQ4dEV1KlpaXw9vZGcXExLly4gPj4eIwcORLDhg1DcnKy6BivXxF+/vnnOHbsmFhN\nITo6Gg8fPsTnn38OADhw4AACAgKwePFiJCUl4ZdffsFff/2FmTNnir2vY8eOITc3F5cvX8bFixfr\nfe/Z2dl499130alTJ9y+fRtnz55FQkICPv74Y7HtsrKysHfvXhw/fhyRkZEoKCjAmDFjRK9fvnwZ\nvr6++Oyzz5CQkIDTp08jPT0dH330kdhxoqOjceXKFZw5cwZ3796FlpYWioqKEBAQgFu3biEqKgr2\n9vYYPnw48vLyAAAxMTHgcDjYuXMnnj17hqysLFG51Xc1vXjxYqxYsQJ///033Nzc8Omnn+Lly5cA\ngJKSEowaNQodOnTA7du3cfDgQWzdulUsWdYnMTERQ4YMQfv27REeHo67d+9i0aJFEAqFom0yMzMR\nEhKCI0eOICoqCoWFhZgyZYro9bKyMvTs2RNnzpzBgwcPEBgYiKCgIBw8eLDVZQQ0/l2MjY0Fy7I4\nefIknj17Jko8rfnc3ix7f39/mJqa4ubNm0hISMD27dthYmLSrM8wOzsbAwcOREFBAc6ePYvExERs\n2LBB6jVzhccShTBp0iRWU1OTNTQ0ZPX19VmGYVgOh8MuWbJEtM3jx49ZDofDJicni+27Zs0atk+f\nPizLsuylS5dYDofDxsbG1nuen376ie3YsSNbVVUltn7o0KHs/PnzWZZl2bS0NJZhGPb69essy7Js\nfn4+q6enxx4/fly0fUBAADtw4EDRcufOndnvv/9e7JhXr15lGYZh8/PzWZZlWS8vL9bBwaHJslix\nYgXbsWNHtqKiQrTu7t27LMMw7LVr11iWZdmgoCCWw+Gwjx49Em3z8OFDlmEY9vLly6LzLVu2TOzY\n6enpLMMw7N27d1mWrS53ExMTtri4uNGYqqqqWBMTE/bXX38VrdPU1GQPHjwotl14eDjL4XDYp0+f\nipYZhmFPnTol2iY7O5tlGIb9888/WZZl2R9++IHlcrlsYWGhaJukpCSWYRh2/fr1DcY0fvx40ede\nn6CgIFZLS4t98eKFaN1vv/3GamhosGVlZQ3uN3fuXHbYsGGi5ZaWUVPfxYyMDJZhGDYiIkJsfWs+\nt0mTJrHvvPOOaNnY2LjOZ/Q6ST7DFStWsJaWlmxJSUljb1/tUM1Dgbi7u+Pvv/9GTEwMAgMDMWDA\nAKxdu1b0+u3bt8GyLFxdXcHlckX/NmzYgH/++QdA9dWciYkJ+vbtW+85bt++jaysLNFtjpp/kZGR\nSElJqXcfY2NjvP/++/jll18AVN86++2330S1jtzcXKSnp2PBggVixxwxYgQYhhG7TeLi4tJkOdy/\nfx/u7u7Q1NQUrevVqxeMjY2RmJgoWmdmZgZbW1vRcrdu3WBqairaJiYmBt9++61YTHw+HwzDiL1X\nJycn6OnpicWQlpaGCRMmoFu3bjA2NoaxsTEKCgrq3BaTBMMw6N27t2i55nlSdnY2AODBgwei20o1\nHBwc0K5du0aPGxsbi7fffrvRbaysrMDj8cSWWZbF8+fPAVQ/d9i4cSP69u0LMzMzcLlchISE1Hmf\nLSmjpr6LDWnN5/amRYsWYcqUKfD29sbq1asRFxfXrFhq3sfAgQOhq6vb7H1VmWbTmxBZqXlWAABB\nQUFITU3F7NmzRc80hEIhGIZBVFRUnf80kj54FAqF6N69O06dOgX2jQGV9fX1G9xv4sSJ+Oijj/Di\nxQtcu3YNr169wieffCI6JlDd0sXLy6vOvjY2NqK/DQwMJIqzLQiFQixduhQTJkyo85qFhUWjMY0a\nNQrm5uYIDg5Gx44doa2tDQ8PD5SXl7coFm1t7XrjqyGtVlFvnrfmPDXn3rp1KzZt2oRvv/0Wffr0\nAZfLxfbt2+vcz5dFGdVozef2phUrVmD8+PE4f/48Ll++jA0bNmDp0qVYs2ZNq2IklDwUWlBQEJyc\nnDBjxgw4OzuLrtrT09MxcuTIevdxcXFBXl4eYmNj4ezsXOd1V1dX/PLLL+ByuTA1NZU4lnfffRc8\nHg9HjhzBlStX8N5774latZibm6Njx45ISkrCF1980YJ3Ko7P5+PAgQOorKwU1T7u3r2Lly9fomfP\nnqLtcnJy8PjxY1HCffjwIXJzc8Hn80XvNTExUexBrCQEAgEePHiA7du345133gEAZGRkiK7Wa2hr\na6OqqqrF77NG9+7dERoaisLCQnC5XABAcnIy8vPzG93PxcUFf/31V6vOfe3aNQwfPlxUiwSqy7Ep\nkpRRU9/FmsT2Zhm29HNrSOfOnTFjxgzMmDEDmzZtwtatW0XJQ5LP0MXFBT/++CNKSkqarOmoE7pt\npcDs7OwwevRoLF++HADQtWtXTJ48GdOmTcOhQ4fwzz//4O+//8ZPP/2EzZs3AwCGDh2KQYMG4ZNP\nPsGZM2eQlpaGGzduIDQ0FAAwbtw42NraYtSoUbh48SLS09MRHR2NjRs34syZMw3GoqGhAX9/f+zd\nuxfnzp0T+7EBgPXr1+O7777Dhg0bkJiYiIcPH+LUqVP1ttJqyuzZs1FQUIBJkyYhMTERkZGRmDhx\nIoYMGYKBAweKttPT08PkyZNx584d3L59G5MmTYKzszO8vb0BAGvWrMHp06excOFC3L17F48ePcL5\n8+cxdepUlJWVNXh+ExMTmJmZYd++fUhJSUFUVBQ+++yzOjUzW1tbXLlyBVlZWXjx4oVo/Zs1ujeX\n3zRu3DgYGBhgwoQJuHfvHm7duoWpU6dCX1+/0RrJkiVLkJKSgs8++wx37tzBo0ePcPz4cdy6davR\n870ej4ODA8LDwxEeHo6UlBSsXLkS0dHRje4PSFZGTX0XTU1NYWhoiD///BPZ2dmiZNnSz+1Nr169\nwuzZs3HlyhWkpaUhLi4O58+fF11cAJJ9hrNmzYJQKMQHH3yAGzduIC0tDf/973/rtJhTN5Q8FNzi\nxYtx8eJFUUeqffv2Yf78+diwYQP4fD58fHzw888/o2vXrqJ9zp07h5EjR2LmzJlwdHTEhAkTRP8x\ndHR0EBERAVdXV3zxxRdwcHDAmDFjEBMTI9Z0tr4frc8//xxJSUlo164dRowYIfba+PHjERYWhv/+\n979wc3ND//79sWbNGrFbVpLemjE3N8eff/6JjIwM9O/fH++//z569epVp/OXlZUVpk+fjrFjx2Lw\n4MEwNDTEiRMnRK97eXnh8uXLuHfvHgYPHozevXtj4cKFMDIygpaWVoPnZxgGx48fxz///IPevXvj\niy++wPz582FpaSm23bZt23Dnzh107txZrF/Am++zvvf9+jo9PT3873//w/Pnz9G/f39MnDgR8+bN\ng4GBQaP32Xv06IHw8HDk5ubCy8sLffv2xfbt25tsBfT6uVeuXIkhQ4bA19cXAwcORH5+PubOndvo\n/jXHkKSMGvsuMgyD4OBghIWFoWPHjqLaSUs/tzdpamoiLy8PU6dORffu3TFixAhYWFiImuoCkn2G\nFhYWiIyMBJfLxahRo9CjRw+sWLFC4jhUFcM2dVlEiAJavXo1Dh8+LNEtFmWUnp4OW1tb/PHHHxg1\napS8wyGkDpWoebzeAkfdUVnUUqayOHz4MMLDw5Geno6IiAh88sknsLW1bfU4TTWUqSykjcqiVmvK\ngpKHiqGyqKVMZfHixQtMnToVTk5OGDduHDp37oyIiIhm3aZpjDKVhbRRWdRS++RB1M+qVatU6pbV\nf/7zH6SmpqK4uBgZGRk4evSo2PMiQhQNJQ9CCCHNRg/MCSGENJvKdBKkiVuqcblcFBYWyjsMhUBl\nUYvKohaVRa3WzFlCt60IIYQ0GyUPQgghzUbJgxBCSLNR8iCEENJsMntgvnfvXsTGxsLY2Bhbt26t\nd5v9+/cjPj4eOjo6CAgIQOfOnWUVHiGEkGaQWc3D29sbX3/9dYOvx8XFITs7G9999x2mT5+Offv2\nySo0QgghzSSz5OHo6Njo5C0xMTEYMmQIgOoZ4YqLi5ucz4AQQoh8KEw/D4FAgPbt24uWeTweBAJB\nk1NxEqKoeBEToJt5Wd5hiHDlHYACUfeyYFngZIITPjpwv8XHUJjk0RyJiYliA3r5+fmJZmBTd9ra\n2lQW/0/eZaFIiYOQ1zEMcDezAz4CEBYWJlrP5/PFJstqjMIkDx6PJzaT14sXL8Dj8erdtr43SD1G\nq1Hv2VqyKoumahiZ/k+lHkNT6HtRi8qi2kTfSgDVF98tIdOmuizLNjglp6urKyIiIgBUz6FsYGBA\nt6yIUmgscZRaDZVhJITUVVUlrHe9nl7r6g4yq3ns3LkT9+/fR2FhIWbOnAk/Pz9UVlaCYRj4+PjA\n2dkZcXFxmDNnDnR1dTFz5kxZhUZIszRU01CEGgYhr4uLe44lS65h+fL+8Pbu2KbHllnykGRe5ClT\npsggEkJap77EQTUMokiKisqxefMd7N+fAJYFgoPvKm/yIETVUE2DKKJLl/7FsmWRyMx8BQ0NBtOn\n98SCBc5tfh5KHkRtKVpTWkJaq7S0EsuXX0dm5iv06mWKLVs80aOHqVTORcmDqK3WJA66TUUUka6u\nJjZuHISUlDxMmdIDmprSaxNFyYOonTdrHHT7iaiSoUM7YujQtn2+UR8aVZeondcTB9UgiDIqL69C\naGgCSksr5RYD1TyIWuFFTBD9TTUOooxu387GkiXXkJycB4GgFIsXu8olDkoeRK3U1DqoxkGUTWFh\nOTZujMHBg/fBskDnzkYYOLDlc5C3FiUPonJef6bR0MhWgiG/yC4gQlrp+fNijBhxCs+evYKmJoOZ\nM3tj7ty+re4l3hqUPIjKaaoVFdU6iLIxM9NDz57tYWmpj82bPdG9e/umd5IySh5Epbz+TKNwegEN\ngEdUAsMw2LnTC4aGWtDQUIx2TooRBSFthJ5pEGVXUFBe73pjYx2FSRwAJQ+iAngRE2B1xBpWR6xF\n6+iZBlE2ZWVV2LbtDtzcjuDx45fyDqdJlDyI0nvzGQfVOoiyuXUrC8OG/Y7t22NRUFCOy5efyDuk\nJtEzD9IiijguFPXbIMrm5csyrF8fjcOHkwAAXbsaY/NmT7i7W8o5sqZR8iAtomiJg2obRBnl5ZXh\nxIkUaGlxMHt2H8ye3Ru6usrxs6wcUaowaVzBy3LWbrraJ6TlOnc2wrZtg+HkxIODQ/3TbisqSh5y\npmhX8M1BV/uEtJ6vr528Q2gRSh4Koq2u4LlcLvVtIETBPHggwKlTqfjqq35gGEbe4bQJSh6EECIl\npaWV2LkzDsHBd1FZyaJXLzOMGmUr77DaBCUPQgiRghs3MrFkyTU8flwAAJg40QmentZN7KU8KHnI\ngCI2ayWESM/582mYMuUiAMDevh02b/ZEv34Wco6qbVHykAEaqI8Q9eLlZYPu3XkYOdIWs2b1ho6O\nhrxDanOUPKSIpjslRD3p6mrif//7UKpziMub6r4zBUDTnRKi2qqqhEhLK6j3NVVOHADVPKSCahyE\nqL7ExBdYsuQanj8vxpUrY2FoqC3vkGRKtVOjnFCNgxDVVVJSiQ0bojFixEnEx+eAZYH0dPXrW0U1\nDymiGgchquXmzSwsXHgVaWkFYBjgiy/4WLrUVe1qHQAlD0IIkVhpaSXS0grg6GiCzZs94eLSQd4h\nyQ0lD0IIkZCXV0eEhr6Dt9/uBC0t9b7rT8mDEEKaYfjwzvIOQSGod+okhJA3VFYKERLyN3bsiJV3\nKAqNah6EEPL/7t3LxaJFV5GQ8AKamgw+/rgbbGxkOUOO8qDk0YZoDCtClFNxcQW2br2DffsSIBSy\nsLY2xDffeFDiaIRMk0d8fDwOHDgAlmXh7e0NX19fsdeLi4uxa9cu5ObmQigUYvTo0fDy8pJliK1C\n/TsIUU5r1tzCL788AIfDYNq0Hli82BUGBlryDkuhySx5CIVChIaGIjAwECYmJli2bBn69esHa+va\nIYovXLiAjh07YunSpSgoKMC8efPg6ekJDQ3lGlSM+ncQolzmzu2L1NR8rFzpht69zeQdjlKQWfJI\nTU2FpaUlzMyqPxgPDw/ExMSIJQ+GYVBSUgIAKC0tBZfLVYrEQberCFFulpYGOH78PXmHoVRk1tpK\nIBCgffv2omUejweBQCC2zfDhw5GRkYEvv/wSixcvxqRJk2QVXqvQ7SpClMPjxy/x6FG+vMNQCQrV\nVDc+Ph62trb4/vvvsWnTJoSGhqK0tFTeYUks0/8pBEN+kXcYhJA3VFQIsXt3PHx8TiAg4AKEQlbe\nISk9md224vF4yM3NFS0LBALweDyxbcLDw0UP0S0sLGBubo6nT5+ia9euYtslJiYiMTFRtOzn5wcu\nV/6tIhQhBm1tbYWIQxFQWdRS57K4c+cZ/vOfi7h3LwcA8NZb7aClpQd9fXogDgBhYWGiv/l8Pvh8\nvkT7ySx52NnZ4dmzZ8jJyYGJiQmuX7+OuXPnim1jamqKe/fuwdHREfn5+cjKykKHDnXHjqnvDRYW\nym9Uy5r/kvKMoQaXy1WIOBQBlUUtdS2Lb76JQXDwXQiFLDp14mLjxkF47z0nFBYWorBQee5qSAuX\ny4Wfn1+L9pVZ8uBwOJgyZQrWrVsHlmUxdOhQ2NjY4OLFi2AYBj4+PhgzZgyCg4OxaNEiAMC4ceNg\naGgoqxAJISrGxEQHDAPMnNkLCxe6QE+Pura1FYZlWZW4+ZeZmSmX877e0koRmuiq6xVmfagsaqlr\nWVRWCpGSkg8np9pb5OpaFvWxsrJq8b4K9cBcGdUkDmplRYj8sCyL+q6DNTU5YomDtJ1mJ4+XL19K\nIw6lR62sCJGP1NR8fPzxf3Hq1D/yDkWtSHQDsLi4GPv370dUVBQ4HA5++eUX3L59G48ePWrxwxZC\nCGmN8vIqBAffxXffxaOsrAoCQSk++KArOBxG3qGpBYlqHvv27YOWlhZ27twJTc3qfNOtWzdcv35d\nqsERQkh9bt/OxvDhJ7Flyx2UlVXhk0/sceLEe5Q4ZEiimse9e/cQEhIiShwAYGxsjPx86qlJCJEt\noZDFV19FIjk5D507G2HTpkEYNMi66R1Jm5Ioeejp6aGoqAjt2rUTrcvNzRVbVke8iAnyDoEQtcPh\nMPjmm0G4dOlfzJvXl5rfyolEpe7t7Y3t27fD398fLMsiNTUVR44cgY+Pj7TjU2jU0ooQ+ejXrwP6\n9avbgZjIjkTJ48MPP4SWlhZCQkJQUVGB7777Dj4+Phg1apS041M49Y2gSy2tCGl7QiGLY8ceYuRI\nW3C52vIOh7xBouRRWFiI0aNHY/To0WLrCwoKYGRkJJXAFNWbiYNqHYS0vZSUPCxZcg3R0dn4++9c\nrF/vIe+QyBskam01Z86cete/OTaVOsn0f0qj6BLSxsrKqrBt2x28887viI7OhpmZHtzdLeUdFqmH\nRDWP+npulpaWgsNRrw7q9ICcEOl59aoCo0adQkpKdSvOceMcsXx5f7RrpyPnyEh9Gk0eAQEBYBgG\n5eXlmD17tthrhYWFcHNzk2pwioYekBMiPQYGWujTxwxVVSw2b/bEgAFU41BkjSaPGTNmgGVZbN68\nGV9++aVoPcMwMDY2RseOHaUeoCKiW1WESMfatQOhpcWBri41v1V0jX5CPXv2BAD88MMP0NfXl0lA\nhBDV9+pVBQwM6k7GRK2qlIdE6V1fXx///vsvkpKSUFBQIPba2LFjpRIYIUT1VFUJ8fPPD7B16x38\n9ttI9OhhKu+QSAtJlDwuX76M/fv3o0ePHrh37x569uyJhIQEuLi4SDs+QoiKePBAgCVLriE29jkA\n4I8/HlPyUGISJY9Tp05h2bJl4PP5mDx5Mr766ivcuXMHt27dknZ8hBAlV1paiZ074xAcfBeVlSws\nLPSxbt1AjBhhK+/QSCtI1Nb25cuXojnDGYaBUCiEs7MzYmJipBocIUT5FRSU4+DB+6isZDFxohOu\nXPmYEocKkKjmwePxkJOTAzMzM1haWiI2NhZGRkZio+yqOurjQUjLmJvrY8uWwTA310O/fhbyDoe0\nEYl+/UePHo0nT57AzMwMH330EbZv346qqipMnDhR2vEpDOrjQUjLjRpFNQ1Vw7D1dR9vQnl5OSor\nKxWq+W5mZqZUj291pHq+gEz/p1I9T2txuVwUFhbKOwyFQGVRSxZl8fRpEX7++T6WLu2n0JMy0fei\nlpWVVYv3bdH4Itra2qiqqsKvv/7a4hMTQlRDVZUQP/6YAC+vY9i9+y5+/TVJ3iERGWjytlV4eDjS\n0tJgaWkJHx8flJWV4cSJE7h48SIcHBxkESMhREElJr7A4sVXcfduLgDgvfds8c47b8k5KiILjSaP\nQ4cO4erVq7C3t8f169eRkpKChw8fokuXLlizZg06d+4sozAJIYomJuYZxow5i6oqFpaWBtiwwQPD\nhlHiUBeNJo/r169j9erVsLS0REZGBhYuXIi5c+di4MCBsoqPEKKgnJ3N0auXGfr2NcPSpa4wNKSh\nRdRJo8mjuLgYlpbVI1va2NhAW1tb7RJHfTMHEkIADQ0Ofv/9PWhra8g7FCIHjSYPlmWRm5srWtbQ\n0BBbBgBTU9UeXuD1xEHNdIk6YlkWGRlF6NiRW+c1Shzqq9HkUVZWhoCAALF1by7/9ttvbR+VAlL0\nJrqESMO//xZg2bLruHs3BxERH6N9ez15h0QURKPJ48iRI7KKgxCiQCorq5vfbtlyG6WlVWjXTgcP\nH+ZjwABKHqRao8lD3aaZfRMNSULU0f37L7BgwVXcu1d9i9rXtyuCgtxhZqY4nYKJ/KnP4FQtQEOS\nEHVUWlqFhIRcWFsb4ptvPPD2253kHRJRQJQ8GvB6rYOmnSXqxNnZHPv2+WDwYJt6Z/sjBKDk0SCq\ndRB1RkOmk6ZI/FCjqqoKDx8+xM2bNwFUD45YXl4utcAUBdU6iCpiWRZhYQ+xYUO0vEMhSkqimseT\nJ0+wefNmAEB+fj7c3d1x7949XLt2DfPmzZP4ZPHx8Thw4ABYloW3tzd8fX3rbJOYmIiDBw+iqqoK\nRkZGWLVqlcTHJ4Q07fHjl1i6NBLXr1ePRP3++11oOljSbBIljx9//BFjxoyBl5cXJk+eDADg8/nY\nt2+fxCcSCoUIDQ1FYGAgTExMsGzZMvTr1w/W1taibYqLixEaGooVK1aAx+OhoKCgmW+nbVArK6KK\nKiqE+P77v7FjRyxKS6tgYqKDoKAB4PPbyzs0ooQkSh7//vsvhgwZIrZOV1cXZWVlEp8oNTUVlpaW\nMDMzAwB4eHggJiZGLHlERkbCzc0NPB4PAGBkZCTx8dsSPe8gqmjnzjjs2BELABgzxg6rVrlTpz/S\nYhIlD1NTUzx+/BhdunQRrfvnn39gYSH5lJICgQDt29de4fB4PKSmpoptk5mZiaqqKqxevRqlpaUY\nMWIEBg8eLPE5WuvNcazoeQdRJVOm8HH1agYWLnTBkCE28g6HKDmJkscnn3yCjRs3YtiwYaisrMSZ\nM2dw4cIFTJ06tU2DEQqFePz4MQIDA1FWVoYVK1bA3t6+TpJKTExEYmKiaNnPzw9cbt1xd5rr9cRR\n2XFYmxxT1rS1tZUybmmgsqilra2NTp3McPnyODCM4s7yJwv0vRAXFhYm+pvP54PP50u0n0TJw9XV\nFe3atcNff/0FR0dHZGZmYt68eejWrZvEAfJ4PLFBFQUCgej21OvbcLlcaGtrQ1tbG05OTkhLS6uT\nPOp7g202DyGbAAAgAElEQVQxrWTN10k0jpUSTlVJU2zWUteyyMkpxsuX5bCzaydap65lUR8qi1pc\nLhd+fn4t2lei5FFUVAQ7OzvY2dm16CQAYGdnh2fPniEnJwcmJia4fv065s6dK7ZNv379sH//fgiF\nQlRUVCAlJQXvvfdei89JiDphWRZHjyZj7dpbsLY2xLlzH0JLS72HGCLSI1HymDFjBnr27AlPT0+4\nurpCW7v5k75wOBxMmTIF69atA8uyGDp0KGxsbHDx4kUwDAMfHx9YW1ujd+/eWLRoETgcDnx8fGBj\nQ/dmCWnKP//kY+nSSERFZQEAnJ31UVhYDh5PV86REVXFsCzLNrVRfn4+bty4gevXryMjIwOurq4Y\nNGgQevfurTCDJ2ZmZrb6GFZHqlt+KfPw61Qlr6UuZREamoD166NRVlaF9u11sWbNAHzwQVexZxvq\nUhaSoLKoZWVl1eJ9Jap5tGvXDiNHjsTIkSORnZ2NyMhI/PLLLwgODm5WXw9CSNszNNRCWVkVPv3U\nHitWuMHEhGobRPqaPbZVcXExiouLUVJSAh0dHWnEJHM01SxRZn5+9nBw4KFPHzN5h0LUiETJIzMz\nE9evX0dkZCSKi4sxYMAAzJs3Dw4ODtKOTyZoqlmiLFiWrdPUlmEYShxE5iRKHsuWLUP//v0xefJk\n9OrVS2Gec7SF14ciUeZnHUS1PXv2CitX3oCHhxUmTZKsHT4h0iRR8ti3b1+LWlgpAxqKhCgyoZDF\noUMPsGFDNAoLKxAb+xz+/o7Q0dGQd2hEzTWYPCIjIzFo0CAAQFRUVIMHeHPMK2VBQ5EQRZeSkocl\nS64hOjobAODj0wkbNnhQ4iAKocHkERERIUoef/31V73bMAyjtMmDnnMQRbds2XVER2fDzEwPa9cO\nxHvv2ar90CJEcUjUz0MZNLefhyr06agPtWGvpexlkZQkwP79iVi+vD/atWtdy0ZlL4u2RGVRqzX9\nPCR68r1s2bJ613/99dctPjEhpHGOjjxs3uzZ6sRBiDRIlDyePq3/6rwtenXLA032RBQFy7I4d+4x\ncnKK5R0KIc3SaGur4OBgAEBlZaXo7xo5OTlKO+4UtbAiiiAzswhff30Df/6ZDl/frtizh76PRHk0\nmjxeHzL99b8ZhkGXLl0wcOBA6UUmA9TCishDVZUQP//8ABs3xqCoqAJcrhb697eotwMgIYqq0eTx\n6aefAgDs7e3h7Owsk4AIUWWVlUKMHXsWMTHVzW9HjOiMtWsHwtLSQM6REdI8DSaPpKQkODo6Aqie\nr/z+/fv1bte9e3fpREaICtLU5KBPHzM8eVKIdesGYsQIW3mHREiLNJg8QkJC8O233wIAdu3a1eAB\n9u7d2/ZREaLClixxxYIFLjAyUs1RG4h6aDB51CQOgBIEIS1RUlIJPb26/8X09bXkEA0hbatFIxw+\nePAAycnJbR0LISqBZVmcPv0PBg48ihs3lLM5OyFNkSh5BAUFISkpCQBw5swZbN26Fdu2bcOpU6ek\nGhwhyubp0yJ8/vkFzJp1Gc+fl+DYsRR5h0SIVEg0qu6///6Lbt26AQAuXbqEoKAg6OnpITAwEL6+\nvlINkBBlUFUlxE8/3cemTTEoLq6EkZE2Vqxwg7+/asx5Q8ibJEoeNe3Ps7OzUVVVhY4dOwIAioqK\npBocIcri1atKBAffRXFxJd57zxZr1gxEhw768g6LEKmRKHnY29vjwIEDyMvLQ//+/QEA2dnZ4HK5\nUg2OEGVhZKSNLVs8UVXFYtiwt+QdDiFSJ9Ezj4CAAGhra8PKygp+fn4AgIyMDAwfPlyqwRGiTN5+\nuxMlDqI2JKp5GBkZYfz48WLrXFxc4OLiIpWgCFFUAkEpfvjhHhYudIGWlupMx0xIc0mUPKqqqnDy\n5Elcu3YNAoEAPB4Pnp6e8PX1haamRIcgRKmxLIvff09FUNBNCASlMDDQwpw5feQdFiFyI9Ev/+HD\nh5GcnIzPP/8cZmZmyMnJwe+//47i4mJMnDhR2jESIlfp6QVYtiwSERHVUxMMHGiJUaNoWBGi3iRK\nHlFRUdi0aROMjIwAAB07doSdnR0WL15MyYOotJSUPAwffhKlpVVo104HgYFu8POzp9FvidqTKHkI\nhUJwOOL3dxmGgYrMYEtIg+zs2qF/fwuYmOhi9Wp3mJlR81tCAAmTh5ubGzZt2gQ/Pz+YmpoiJycH\nJ06cgLu7u7TjI0SuGIbBTz8Ng64uPdsj5HUS/Y+YMGECjh07hpCQENEDcw8PD4wdO1ba8REiM5mZ\nRbCyMqyznhIHIXVJ9L9CS0sLn332GT777DNpx0OIzL14UYKgoJs4fz4NV66MhY0NdX4lpCmNJo+s\nrCyEhITg33//RZcuXTBz5kyYmprKKjap4EVMkHcIREGwLItjx1KwevVN5OeXQVdXA3//nUvJgxAJ\nNNrLaf/+/TAxMUFAQAC4XC4OHDggo7CkRzfzMgCg1GqonCMh8vTvvwX49NNzmD8/Avn5ZRg82Bp/\n/TUWI0dSE1xCJNFozePRo0fYu3cvtLW1wefzMW/ePFnFJXWCIb/IOwQiR2VlVbh16xlMTHQQFDQA\nY8bYUfNbQpqh0eRRWVkJbe3qqTL19PRQXl7eqpPFx8fjwIEDYFkW3t7eDQ7nnpqaipUrV2LevHlw\nc3Nr1Tlr8CImiGodhHTrZoKQkLfRv78FeDxdeYdDiNJpNHlUVFTg+PHjouXy8nKxZQASt7gSCoUI\nDQ1FYGAgTExMsGzZMvTr1w/W1tZ1tvv111/Ru3dvSd+DRF5PHHTLigDA8OGd5R0CIUqr0eQxYMAA\nZGVliZbd3d3FlptTzU9NTYWlpSXMzMwAAB4eHoiJiamTPM6fPw93d3ekpqZKfOymvP6QPNP/aZsd\nlyi+S5f+xdWrT7FmzQB5h0KISmk0ecyZM6fNTiQQCNC+fXvRMo/Hq5MgBAIBYmJisGrVqjZNHvSQ\nXP08f16MOXMi8PvvDwEAPj6dMHiwdRN7EUIkpVC9nw4cOIBx48aJltt6+BN6SK76WJbFkSPJWLfu\nFl6+LIeeniYWL3bBwIGW8g6NEJUis+TB4/GQm5srWq7pqf66R48e4dtvvwXLsigsLERcXBw0NTXh\n6uoqtl1iYiISExNFy35+fhLNaqgOMx9qa2urxftsyL598Vi8+BoAYNiwLti2zRtvvWUs56jkT92/\nF6+jshAXFhYm+pvP54PP50u0n8ySh52dHZ49e4acnByYmJjg+vXrmDt3rtg2u3fvFv0dHBwMFxeX\nOokDqP8NFhYWNnhurgTbqAoul6sW77MhH3zwFg4fNseUKXyMH98HRUVFal0eNdT9e/E6KotaXC5X\nNDtsc8kseXA4HEyZMgXr1q0Dy7IYOnQobGxscPHiRTAMAx8fH1mFQlSYnp4mzpx5HwzDUL8NQqSI\nYSV8sJCQkIAbN24gPz8fS5YswaNHj1BaWoru3btLO0aJZGZm1ln3Zt8OdWhppS5XVYWF5cjMLIKD\nA6/BbdSlLCRBZVGLyqKWlZVVi/eVaBLmCxcuICQkBO3btxc9a9DU1MSRI0dafGJZoL4dqunChTR4\neR3H5Ml/oqSkUt7hEKKWJLptdfbsWaxcuRIdOnTA2bNnAQA2NjZ4+lQ5ruTVocahDp49e4WVK2/g\n3Lk0AEDfvmYQCEphbV13GHVCiHRJlDxKSkpEnftqVFVVQVNToVr6EhV28mQqli2LRGFhBQwMtPDV\nV674/PPu0NCQqPJMCGljEv3Pc3R0xJkzZ8TWXbhwQWGedxDVp6+vicLCCvj4dMKVK2PxxRc9KHEQ\nIkcSPTAXCATYuHEjSkpKkJubC0tLS2hqamLZsmUwMTGRRZxNqu+BudWR6h7F6nTbSpUfBkZHP0O/\nfh0kbkWlymXRXFQWtagsarXmgblE9514PB42bdqE5ORk5ObmwtTUFPb29uBw6MqPtD2WZetNEP37\nW8ghGkJIfSR+aMEwDBwdHaUZC1FzL1+WYcOGaFhZGWLu3L7yDocQ0giJkkdAQECDtwpe7xVOSEuw\nLItz59KwcuUNZGcXw8BAC5MmdYexsY68QyOENECi5DFjxgyx5by8PJw/fx4eHh5SCaot0FzlyiEz\nswgrVtzAhQvpAABX1w7YvHkQJQ5CFJxEyaNnz571rvvmm28watSoNg+qLdAw7MohKOgmLlxIh6Gh\nFpYv748JE5zA4dCwIoQouhZ31NDW1kZ2dnZbxiIVNAy7Ylu50g2amhysXOkGS0sDeYdDCJGQRMnj\nzalny8rKEBsb2+ZTxRL107EjF8HBVDskRNlIlDxen3oWAHR0dPDuu+/Cy8tLGjERFXTjRiZsbAzR\nqZORvEMhhLSBJpOHUChEr169MGDAAGhra8siJqJC8vJKsX59NI4cSYaXlw0OHRpOQ6UTogKa7OXH\n4XCwf/9+ShykWViWxenT/8DL6ziOHEmGtjYHrq4dIBS27dTChBD5kOi2lbOzM2JjY+Hs7CzteIgK\nYFkW06f/hXPnHgMA3NwssHmzJ+zs2sk5MkJIW5EoebAsi23btsHR0RHt27cXe23WrFlSCaw1qI+H\nfDEMAz6fh8jIp1ixwg3+/g7U/JYQFSNR8rCwsMDo0aOlHUuboT4e8jdrVm/4+zuiQwd9eYdCCJGC\nRpNHZGQkBg0ahE8//VRW8bQp6uMhfWVlVdDR0aizXltbgxIHISqs0Qfm+/btk1UcRAldvfoUQ4ce\nx//+91jeoRBCZKzR5CHBVB9EDQkEpZg3Lxz+/ueQllaAQ4eS5B0SIUTGGr1tJRQKkZCQ0OgBevTo\n0aYBEcXFsix+/z0VQUE3IRCUQkdHA/PnO2PGjF7yDo0QImONJo+KigqEhIQ0WANhGIaGZFcj5eVC\nbN8eC4GgFAMHWmLTJk906WIs77AIIXLQaPLQ1dWl5EBEdHQ0sGWLJ548KYSfnz31FCdEjbV4VF2i\nngYObPmcx4QQ1UEPzEkdxcUV2LEjFsXFFfIOhRCioBqtefz888+yioMoiPDwJ/jqq0g8eVKEwsJy\nBAa6yzskQogCottWBACQm1uCoKAonDz5DwCge3ce3n+/q5yjIoQoKkoeBNnZxRg69Djy88ugq6uB\nhQtdMG1aT2hpNTnoMiFETVHyIOjQQR+entbIzy/Dxo2D0LkzTdhECGkcJQ8CANixYwh0dTWo+S0h\nRCIqd1+ChmNv3PPnxfWu19PTpMRBCJGYyiUPGo69fkVF5QgMjMKAAUeRnCyQdziEECWnUsnj9VoH\nDcde69Klf+HtfRyhoQmoqBAiJiZb3iERQpScTJ95xMfH48CBA2BZFt7e3vD19RV7PTIyEqdPnwZQ\nPTTKtGnT0KlTJ4mPT7UOcbm5JVix4gb++OMRAKBXL1Ns2eKJHj1M5RwZIUTZySx5CIVChIaGIjAw\nECYmJli2bBn69esHa2tr0Tbm5uZYvXo19PX1ER8fj++//x7r16+X6PhU66irrKwKly8/gZ6eJhYv\ndsGUKT2gqalSlU1CiJzILHmkpqbC0tISZmZmAAAPDw/ExMSIJQ97e3vR3926dYNAIPm9eap11GVt\nbYjdu73h5MRDx45ceYdDCFEhMkseAoEA7du3Fy3zeDykpqY2uP1ff/2FPn36NP88VOsQM2zYW/IO\ngRCighSyn0dCQgLCw8OxZs2ael9PTExEYmKiaNnPz0/0N5erflfYt25l4rffHmDbtqHQ1tZWyzKo\nD5VFLSqLWlQW4sLCwkR/8/l88Pl8ifaTWfLg8XjIzc0VLQsEAvB4vDrbpaen44cffsDy5cthaGhY\n77Eae4OFhYVtE7ASKCwsx8aNMTh48D5YFnB2bo/x4/uoVRk0hsvlUln8PyqLWlQWtbhcrtjFd3PI\n7OmpnZ0dnj17hpycHFRWVuL69etwdXUV2yY3Nxfbtm3D7NmzYWFhIavQlNKFC2nw8jqOAwfuQ0OD\nwezZfegWFSFEZmRW8+BwOJgyZQrWrVsHlmUxdOhQ2NjY4OLFi2AYBj4+Pjh+/DiKiooQGhoKlmWh\noaGBb775RlYhKo1z5x5j2rRLAIC+fc2webMnundv38RehBDSdhhWVWZ82lY9tEam/1M5ByJ9FRVC\nfPzxWbz/fhd8/nl3aGjUViCpSl6LyqIWlUUtKotaVlYtnxlUIR+Yk8ZpaXFw8uRoGouKECI31GNM\ngZWVVSEpqf6+LpQ4CCHypFLJQ5U6CEZHP8OwYb/j00/PIT+/TN7hEEKIGJW6baUKHQRfvizDhg3R\nOHQoCQDQpYsxnj8vRrt2OnKOjBBCaqlU8lB24eFPsGDBVWRnF0NLi4OAgN6YM6cPdHXpYyKEKBb6\nVVIgurqayM4uhouLObZs8YSDQ91OlIQQoggoeSgQd3dLhIWNwoABluBw6IE4IURxUfJQMB4eLW93\nTQghskLJQ8ZKSyuxc2ccKiqEWLHCTd7hEEJIi1DykKEbNzKxZMk1PH5cAA0NBpMn82FtXf/gj6Rt\nGRoaqnXfGA0NDRpJ9v+pY1mwLIuioqI2PabKJA9F7uORl1eK9eujceRIMgDA3r4dNm/2pMQhQwzD\n0JAURG1JI1mqTPJQ5D4emzffxpEjydDW5uA//+mLWbN6Q0dHQ95hEUJIi6lM8lBkCxe6IDu7GMuX\n94edXTt5h0MIIa1GyUMGTE31sH//MHmHQQghbUalxraSt8TEF3j4ME/eYRBCiNRR8mgDJSWV2LAh\nGiNGnMSCBVdRVSWUd0iEKLWHDx9i5MiR8g5DKeTm5sLLywsVFRUyPS8lj1a6evUpfHxOYM+euxAK\nWfTta4byckoepHnc3NzQtWtXODg4wNnZGfPnz0dJSYnYNjExMfDz84ODgwO6d++OyZMnIyUlRWyb\noqIiBAYGon///nBwcICHhweCgoKQl6dcNeItW7Zg5syZ8g6jVcrLy7FgwQI4OjrC2dkZP/zwQ6Pb\n79y5E/3794eTkxMCAgLw6tUr0Wtr167FoEGD4OjoCC8vLxw/flz0mqmpKTw8PHDo0CGpvZf6UPJo\nhZUrb8Df/xzS0grg6GiC06ffx9q1A6GnR4+SSPMwDIOff/4ZycnJ+PPPP5GQkIBdu3aJXr99+zbG\njRuHESNGIC4uDlFRUXBycoKvry+ePHkCAKioqICfnx9SU1Px66+/Ijk5GX/88Qd4PB7i4+OlFntV\nVVWbHu/58+eIiorCu+++qxDxtNS2bduQnp6OmJgYhIWFYe/evYiIiKh327CwMJw8eRJnzpxBbGws\nSkpK8PXXX4teNzAwwM8//4ykpCTs2LEDq1atwp07d0Sv+/r6UvJQJl27toOOjgaWLnXF+fMfwcWl\ng7xDIkqsZkZoU1NTeHl5ITExUfTahg0b4Ofnh8mTJ0NfXx/GxsZYsmQJnJ2dsW3bNgDAsWPHkJWV\nhdDQUNjZ2QEAeDwe/vOf/8Db27vecyYnJ8Pf3x98Ph99+/bF7t27AQDz58/Hli1bRNtFRUXB1dVV\ntOzu7o7g4GD4+PjA3t4ewcHBmD59utixAwMDERgYCAAoLCzEokWL4OzsDFdXV2zevBkNzYB99epV\n9OzZE9ra2qJ1e/bsgYeHBxwcHDB06FCcP39e9FpYWBh8fX0RFBSEHj16YPv27QCAo0ePwsvLC3w+\nH+PHj8fTp7VTVAcGBqJfv35wdHTEyJEjER0dXW8srXH8+HHMmzcPXC4XdnZ2GDduHMLCwurd9tKl\nS/j0009hYWEBPT09zJo1C3/88QdKS0sBAAsWLECXLl0AAH379kX//v3FkoezszPS09PF3qO00SVy\nK0yc6IShQ23QqZORvEMhrWR1xLrNjpXp37r/wJmZmbhy5QoGDRoEACgpKcHt27exePHiOtu+9957\n2Lx5MwAgMjISXl5e0NPTk+g8r169gr+/P2bOnImDBw+isrISDx8+bHD7N3vonz59GocOHYKJiQly\nc3OxY8cOFBcXQ19fH0KhEGfPnsX+/fsBAPPmzYO5uTlu3LiB4uJiTJw4EdbW1hg3blyd8yQlJaFr\n165i6zp37oxTp07BzMwMf/zxB+bMmYMbN27AzMwMABAXFwdfX1/8/fffqKiowIULF7B7924cPHgQ\ntra22L17N2bNmoXTp08DqP4BXrhwIbhcLn788Ud8+eWXuHXrlljCqrFnzx7s2bMHDMOIEl7N3wzD\niCX5Gi9fvkR2dja6d+8uWte9e3expNcYlmVRXl6Ox48fw8nJSey1kpIS3L17F5MmTRKt09DQQOfO\nnXH//n1YW7fdd7kxVPOQQGWlEEJh3askDoehxEHazJQpU+Dg4ID+/fvD1NQUCxcuBADk5+dDKBTC\n3Ny8zj4dOnSAQFA9VXFeXl692zTk0qVLMDc3x7Rp06CtrQ19fX306dOnWfFaWFhAR0cH1tbW6Nmz\nJ/73v/8BqE5kenp66NOnD3JycnDlyhUEBQVBV1cXPB4P06ZNw6lTp+o9bkFBAQwMDMTWjRo1SpQo\nRo8eDVtbW8TFxYlet7CwwKRJk8DhcKCjo4NDhw5hzpw56Nq1KzgcDmbPno3ExETRlfmHH34IY2Nj\ncDgcTJ8+HeXl5fjnn3/qjScgIAD3799HYmIi7t+/L/Z3fYkDqE7MDMOI9ew2NDQUe47xOm9vb/z6\n66/IyMhAQUEBgoODAaDOcy8A+Oqrr8Dn8zFkyBCx9YaGhigoKKj3+NJANY8m3LuXi0WLrmLSpO7w\n93eUdzhESlpbW2gL+/fvh4eHB27duoWAgAAIBAJwuVy0a9cOHA4Hz58/r3NFnp2dDR6vet4XExMT\nPH/+XOLzZWZm4q233mpxvJaWlmLLH3zwAU6dOoUxY8bg1KlT+PDDDwEAT58+RUVFBZydnQFUX1Wz\nLNvgFbKxsXGdH9ljx45h3759yMjIAAAUFxeLNQKwshIfjTojIwOBgYFYs2aN6JwMw+DZs2ewtrZG\nSEgIjh49KiqvoqIiURJuCzXJr6ioSPT5FBYW1kmKNT799FNkZWVh7NixqKqqwpdffolLly7VKeO1\na9ciJSUFx44dq3OMoqIiGBnJ7mKWkkcDiosrsHXrHezblwChkMVPP93HJ5840DwbRGpqbom4ubnh\n448/xpo1axAaGgo9PT24uLjg7NmzGDBggNg+Z8+eFd3e8vT0xJYtW1BSUiLRrSsrKyvRbZw36evr\ni1311peU3ryNNXr0aKxduxZZWVk4f/48zpw5IzqPjo4OEhISJBqc0snJSaw10dOnT7F06VKEhYWJ\nnrsMGzZM7JnJm8e1trbG3Llz4evrW+f40dHR2Lt3L44dOwZ7e3sAAJ/Pb/AZzK5du7Br164656hJ\nSMnJyXX2MTY2hrm5ORITE+Hp6QkAuH//PhwcHOo9B8MwWLBgARYsWAAAiIiIgIWFhVjy2Lp1KyIi\nInDixIk6SaiqqgppaWlit8mkjW5b1SM8/AmGDj2O77+/BwCYPr0nTp4cTYmDyMy0adNw9epVPHjw\nAACwfPlyHDt2DD/99BNevXqF/Px8bNq0CbGxsZg/fz4AYMyYMbCyssL06dORmpoKlmUhEAiwa9cu\nXLlypc45fHx8kJOTg9DQUJSXl+PVq1eiW0F8Ph+XL19Gfn4+nj9/jh9//LHJmHk8HgYMGIAFCxag\nU6dOoof25ubmGDJkCFatWoWioiKwLIv09HTcvHmz3uMMHjwY9+7dQ3l5OYDqWgbDMODxeBAKhfjt\nt9/q/cF+3fjx47Fr1y7RM5yCggKcPXsWQPUVuqamJkxMTFBeXo4dO3Y0OuLsnDlz8PDhQyQnJ4v9\nq1nXkLFjx2Lnzp14+fIlUlJS8Ouvv+KTTz6pd9v8/Hykp6cDqO7jsmbNGlEiAaoT2KlTp3D06FEY\nGxvX2T8uLg6dOnWS2fMOgJJHHUIhi40bb+PJkyLw+e1x9uwHWLXKHQYGWvIOjaiwN69qeTwePv74\nY+zYsQMA0K9fPxw+fBj//e9/0bdvXwwYMAD379/HqVOn0LlzZwCAtrY2jh49iq5du8Lf3x+Ojo4Y\nPXo08vLy0Ldv3zrnNDAwwJEjR/Dnn3+ib9++8PT0RFRUFIDqROTk5AR3d3eMGzcOH3zwQaPx1vD1\n9UVkZKTollWNnTt3oqKiQtT66csvv2zwFltNv4Wah8vdunXDl19+idGjR6NPnz5ITk5Gv379Gi3P\n4cOHIyAgALNmzYKTkxN8fHwQHh4OAPDy8oKXlxc8PT0xYMAA6Onp1bnt1RYWLlyIt956C25ubvDz\n80NAQAAGDx4set3e3h4xMTEAAIFAgAkTJqBbt26YOHEi/P394e/vL9p206ZNyMrKgoeHB+zt7eHg\n4CBqGQcAJ0+exIQJE9r8PTSGYRuqqymZzMzMNjvW33/nIDIyE9Om9YSWlnLlVy6XS0OP/7/Xy4LK\nRbmkpKRg/vz5otoCadiLFy8wduxYXLhwod7WYkDD3//WJE1KHiqGfiRrUfIgpJo0kodyXVa3oYoK\nIb7//m/k5ZXKOxRCCFE6atnaKi7uORYvvoYHDwRITs7D9u1Dmt6JEEKIiFolj6KicmzefAf79yeA\nZYFOnbj44IOuTe9ICCFEjNokj6Kicrz99glkZBRBQ4PB9Ok9sXChCw1iSAghLaA2v5yGhtrw9u6I\nu3dzsGWLJ3r0MJV3SIQQorTUJnkAwKpV7tDS4kBTU23bCagtlmXFxhlSNxoaGgozVLm8qWNZSKNR\nrUyTR3x8PA4cOACWZeHt7V3v0AH79+9HfHw8dHR0EBAQIOoA1RwCQSl4PN066+kWlfpqrAexOqCm\nyrWoLNqGzC7BhUIhQkND8fXXX2Pbtm24fv16nbHn4+LikJ2dje+++w7Tp0/Hvn37mnWO8vIqfPtt\nLPr1+xW3b2e3ZfiEEEJeI7PkkZqaCktLS5iZmUFTUxMeHh6irvk1YmJiRMMMd+vWDcXFxcjPz5fo\n+LdvZ2P48JPYsuUOSkurcO2a/EdJJYQQVSWz+zgCgQDt27cXLfN4PKSmpja5jUAgQLt27Zo8vq/v\nGbDaH5MAAAyISURBVLAs0LmzETZtGoRBg2Q3QBghhKgblXkIoKHBYMaM3pg3ry892yCEECmT2a8s\nj8dDbm6uaFkgEIgmSXl9mxcvXoiWX7x4UWcbAEhMTBSbwcvPzw8VFYFSiFo5qXOrojdRWdSisqhF\nZVHr9XnV+Xw++Hy+RPvJ7JmHnZ0dnj17hpycHFRWVuL69euiiV1quLq6IiIiAkD1mPYGBgb13rLi\n8/nw8/MT/WtoUnl1RGVRi8qiFpVFLSqLWmFhYWK/pZImDkCGNQ8Oh4MpU6Zg3bp1YFkWQ4cOhY2N\nDS5evAiGYeDj4wNnZ2fExcVhzpw50NXVxcyZM2UVHiGEkGaQ6cOBPn36YOfOnWLr3nnnHbHlKVOm\nyDIkQgghLaASXa2bU9VSdVQWtagsalFZ1KKyqNWaslCZyaAIIYTIjkrUPAghhMgWJQ9CCCHNplS9\n6WQ1sKIyaKosIiMjcfr0aQCArq4upk2bhk6dOskjVKmT5HsBVA+Rs3LlSsybNw9ubm4yjlI2JCmL\nxMREHDx4EFVVVTAyMsKqVavkEKn0NVUWxcXF2LVrF3JzcyEUCjF69Gh4eXnJJ1gp2rt3L2JjY2Fs\nbIytW7fWu02LfjdZJVFVVcXOnj2bff78OVtRUcEuWrSIzcjIENsmNjaW3bBhA8uyLPvw4UN2+fLl\n8ghV6iQpi+TkZPbVq1csy7JsXFycWpdFzXarV69mv/nmG/bmzZtyiFT6JCmLV69esfPnz2dfvHjB\nsizLvnz5Uh6hSp0kZfH777+zhw8fZlm2uhwmT57MVlZWyiNcqXrw4AH7+PFjduHChfW+3tLfTaW5\nbSXtgRWViSRlYW9vD319fQDVZSEQCOQRqtRJUhYAcP78ebi7u8PIyEgOUcqGJGURGRkJNzc30cgN\nqloekpQFwzAoKSkBAJSWloLL5UJDQ0Me4UqVo6MjDAwMGny9pb+bSpM8Gho0sbnbqILmvs+//voL\nffr0kUVoMifp9yImJgbDhg2TdXgyJUlZZGZmoqioCKtXr8ayZctw9epVWYcpE5KUxfDhw5GRkYEv\nv/wSixcvxqRJk2QcpWJo6e+m0iQP0jIJCQkIDw/HuHHj5B2K3Bw4cEDs/bNq3DpdKBTi8ePHWLZs\nGZYvX44TJ07g2bNn8g5LLuLj42Fra4vvv/8emzZtQmhoKEpLS+UdltJQmgfmbTmworKTpCwAID09\nHT/88AOWL18OQ0NDWYYoM5KUxaNHj/Dtt9+CZVkUFhYiLi4OmpqadcZWU3aS/h/hcrnQ1taGtrY2\nnJyckJaWBgsLC1mHK1WSlEV4eLjoIbqFhQXMzc3x9OlTdO3aVaaxyltLfzeVpubRlgMrKjtJyiI3\nNxfbtm3D7NmzVe6H4XWSlMXu3buxe/du7NmzB+7u7pg6darKJQ5AsrLo168fkpKSIBQKUVZWhpSU\nFNjY2MgpYumRpCxMTU1x7949AEB+fj6ysrLQoUMHeYQrdSzLNljjbunvplL1MI+Pj8dPP/0kGljR\n19dXbGBFAAgNDUV8fLxoYMUuXbrIOWrpaKosQkJCEB0dDTMzM7AsCw0NDXzzzTfyDlsqJPle1AgO\nDoaLi4tKN9VtqizOnDmD8PBwcDgcvP322xgxYoSco5aOpsoiLy8PwcHByMvLAwD4+vpi0KBBco66\n7e3cuRP3799HYWEhjI2N4efnh8rKylb/bipV8iCEEKIYlOa2FSGEEMVByYMQQkizUfIghBDSbJQ8\nCCGENBslD0IIIc1GyYMQQkizUfIgSmvXrl04fvy4vMNo0rx585CUlNTg6+vXr0dkZKQMIyKk9aif\nB5G7gIAAvHz5EhoaGmBZFgzDYOfOnU32ct21axcsLS0xduzYNotl165diIqKgpaWFjQ1NdGlSxd8\n8cUXsLS0bJPjHz16FAKBALNmzWqT4zVEKBTC398fOjo6YBgG+vr68PDwwPjx4yXa/969ewgJCcGe\nPXukGidRXkozthVRbV999RV69Ogh7zAAAB999BHGjh2L8vJyBAcHIyQkBKtXr5Z3WC2yfft2mJqa\nIisrC6tWrYKNjY1EEx7VJHFCGkLJgygslmWxY8cOJCX9X3v3FhLV9gdw/Ltnm6aMF3JEdFAsvIWZ\noF3QKcKUghBRvICFysBIiBaVhdVDSRLeSooMKkroQmahgtZD0IXKmTIckCjoqlHShRks3eCYTuN5\nkLM54u3M+T/8O531edysWb+1H2Z+s9dvZv1eMjExQUREBCaTCb1eP2PsyMgIp0+f5vXr10iSRHh4\nOFVVVcDUoXjNzc28fPkSb29vMjIy2Lx584LxPT09Wbdunfrte2JigsuXL9PT04NGoyE5OZlt27Yh\ny/K88UtLS9mxYwdjY2N0dnYC8OTJE/R6PTU1NRw6dIi0tDSSk5MpKSmhpqaG0NBQYOrMpfLycs6c\nOYNWq6W3t5fW1lbsdjvh4eGYTCbCwsIWvJeQkBCio6N5//69eu3evXt0dXUxNDSEv78/WVlZbNy4\nkdHRUerr63E6nRQVFSFJEqdOncLX15eOjg7u37+Pw+EgPj6ekpIStW+M8N8ikofwS0tKSqKsrAxZ\nlrl06RJNTU2zntHV2dlJcHAwlZWVTE5O8ubNG2AqAdXW1pKSksKePXuw2WxUV1ej1+sXfNJxOBx0\nd3er5/zcuHGDgYEBjh8/jsvloq6ujo6ODnJzc+eM/1eJiYlkZmbOuW3l6enJmjVrMJvN5OXlAWCx\nWIiPj0er1fL27VvOnTvH/v37Wbp0KQ8ePKC+vp4TJ04s2MRocHCQV69eER8fr14LCAjg4MGDBAUF\n8eLFC2pqaoiMjCQ8PJzKykrOnj1LU1OTOr6rq4u+vj6qq6vRarVcuHCB5uZmysvL540t/J5EwVz4\nJTQ0NGA0GjEajWqfZUmS2LBhA15eXnh4eJCbm0t/fz/j4+MzXi/LMt++fcNmsyHLMrGxscDUKaEO\nh4OsrCw0Gg3BwcGkpqZisVjmXEtHRwdGo5Fdu3bhdDopLS0FwGw2k5+fj1arxc/Pj5ycHB49ejRv\nfHcZDIZpxXOz2cz69euBqaZemzZtYtmyZUiSpG4/vXv3bs759u7dS2FhIRUVFSQkJEw7KDIxMZGg\noCAA4uLiWLFixbyF/Tt37lBQUEBAQAAeHh7k5OTw+PHjf3Sfwr+fePIQfgn79u2b8STgcrm4evUq\nPT09KIqi7sGPjIyg0+mmjc3Ozqa1tZUjR44gyzLp6elkZmZis9mw2+0YjcZp88bFxc25luzs7FmL\n8ENDQ9PiBgUFqR3XsrKyuH79+oz47lq5ciWjo6MMDAzg4+PD4OCgepS43W6nu7ubW7duqeOdTue8\nXd+OHTuGTqfDYrFw7do1fvz4oW4zWa1W2tvb+fz5M5OTk4yPjxMZGTnnXHa7nbq6umm1EI1Gw/Dw\nMP7+/m7fq/DvJpKH8Mt6+PAhfX19HD58GJ1Oh6IomEymWccuXryY4uJiiouL+fjxI1VVVURFRaHT\n6QgJCaGxsfF/Xs+SJUuw2WzqL69sNpvaNMfb23vW+MuXL582x0JF6D9rKd3d3fj4+LBq1So8PT0B\nCAwMJC8v7x8lpZSUFJ4+fUpbWxuFhYWMj4/T2NjI7t27SUxMRKPRUFtbO+86AwMD2blz57wJRvjv\nENtWwi/L4XCwaNEitFotY2NjtLS0zDnWarXy9etXYOqDXJZlJEkiOjoaDw8Pbt68ycTEBC6Xiw8f\nPtDf3+/2egwGA21tbSiKwsjICO3t7eqW0mzxNZqZby9/f39sNtuCcSwWC2azeVp/ibS0NG7fvq1u\nU42NjWG1WmfdxpvNn/0sFEXB6XTy8+dP/Pz8kCQJq9XK8+fPp61TUZRpbVnT09NpaWlRO/QNDw/T\n29v7t2ILvx/x5CH83831bTw1NZVnz56xfft2fH19yc/P5+7du7OO/fTpE83NzSiKglarZcuWLWrd\n4cCBA1y8eJHOzk6cTid6vZ6CggK31gKQm5vLlStXqKioQJIkDAaD2sZ0tvgxMTEz5khJScFsNmM0\nGgkNDeXo0aMzYsbExCDLMoqikJCQoF6PiorCZDJx/vx5vnz5gpeXF7GxsdOK4POJiIggJiaGrq4u\ntm7dSlFREQ0NDTidTlavXk1SUpI6NiwsjLVr11JWVobL5eLkyZNkZGQgSRLV1dV8//6dgIAADAbD\nb9mVUViY+JOgIAiC4DaxbSUIgiC4TSQPQRAEwW0ieQiCIAhuE8lDEARBcJtIHoIgCILbRPIQBEEQ\n3CaShyAIguA2kTwEQRAEt4nkIQiCILjtD/fZgkk+QZrEAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x20416ea50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "from sklearn.metrics import roc_curve, auc\n",
    "\n",
    "fpr = dict()\n",
    "tpr = dict()\n",
    "roc_auc = dict()\n",
    "for i in range(n_classes):\n",
    "    fpr[i], tpr[i], _ = roc_curve(test_y[:, i], y_prob[:, i])\n",
    "    roc_auc[i] = auc(fpr[i], tpr[i])\n",
    "\n",
    "# Compute micro-average ROC curve and ROC area\n",
    "fpr[\"micro\"], tpr[\"micro\"], _ = roc_curve(test_y.ravel(), y_prob.ravel())\n",
    "roc_auc[\"micro\"] = auc(fpr[\"micro\"], tpr[\"micro\"])\n",
    "\n",
    "plt.figure()\n",
    "lw = 2\n",
    "plt.plot(fpr[2], tpr[2], color='darkorange', lw=lw, label='ROC curve (area = %0.2f)' % roc_auc[2])\n",
    "plt.plot([0, 1], [0, 1], color='navy', lw=lw, linestyle='--')\n",
    "plt.xlim([0.0, 1.0])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('Receiver operating characteristic')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Showing Misclassification with a Confusion Matrix\n",
    "\n",
    "We can visualise the accuracy using a confusion matrix, which shows which categories tend to be correctly classified, and which classes are misclassified. Compare, for example, the consistent classification of air conditioning, with the more equivocal results for children playing."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 366,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Confusion_matrix\n",
      "           aircon   horn  child    dog  drill engine    gun hammer  siren  music\n",
      "    aircon     66      0      0      0      2     13      0     18      0      1\n",
      "      horn      2     23      0      0      2      0      0      1      0      5\n",
      "     child      4      1     73      9      0      2      0      0      1     10\n",
      "       dog      3      0     17     62      3      3      0      0      1     11\n",
      "     drill      3      1      9      3     45      1      0     17     14      7\n",
      "    engine      7      0     17      1      0     60      0      4      1      3\n",
      "       gun      0      0      6      8      0      0     17      0      1      0\n",
      "    hammer      1      0      6      0     17      2      0     70      0      0\n",
      "     siren     20      0     17      5      0      1      0      0     40      0\n",
      "     music      9      2      7      0      8      4      0      0      4     66\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import confusion_matrix\n",
    "\n",
    "labels = [\"aircon\",\"horn\",\"child\",\"dog\",\"drill\",\"engine\",\"gun\",\"hammer\",\"siren\",\"music\"]\n",
    "print \"Confusion_matrix\"\n",
    "cm = confusion_matrix(y_true, y_pred)\n",
    "\n",
    "def print_cm(cm, labels, hide_zeroes=False, hide_diagonal=False, hide_threshold=None):\n",
    "    \"\"\"pretty print for confusion matrixes\"\"\"\n",
    "    columnwidth = max([len(x) for x in labels]+[5]) # 5 is value length\n",
    "    empty_cell = \" \" * columnwidth\n",
    "    # Print header\n",
    "    print \"    \" + empty_cell,\n",
    "    for label in labels: \n",
    "        print \"%{0}s\".format(columnwidth) % label,\n",
    "    print\n",
    "    # Print rows\n",
    "    for i, label1 in enumerate(labels):\n",
    "        print \"    %{0}s\".format(columnwidth) % label1,\n",
    "        for j in range(len(labels)): \n",
    "            cell = \"%{0}s\".format(columnwidth) % cm[i, j]\n",
    "            if hide_zeroes:\n",
    "                cell = cell if float(cm[i, j]) != 0 else empty_cell\n",
    "            if hide_diagonal:\n",
    "                cell = cell if i != j else empty_cell\n",
    "            if hide_threshold:\n",
    "                cell = cell if cm[i, j] > hide_threshold else empty_cell\n",
    "            print cell,\n",
    "        print\n",
    "    \n",
    "print_cm(cm, labels)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Refinement and Optimisation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The performance of a deep neural network is highly dependent on many configurable hyper-parameters, values that govern how the model actually works, and which are not changed during the course of learning. Some of these parameters will have a greater influence than others. The classic machine learning solution to this challenge is GridSearch, to successively train separate instances of a model with distinct values of key parameters, and then compare the resulting accuracy scores to identify which combination of parameters has produced the best results. \n",
    "\n",
    "Scikit-Learn provides GridSearch functionality that can be adapted for use with a neural network implemented using Keras, as described here: http://machinelearningmastery.com/grid-search-hyperparameters-deep-learning-models-python-keras/. \n",
    "\n",
    "Because the training process is computationally intensive, and must be repeated for each combination of hyper-parameter values, optimisation can be a slow, time-consuming process. The code below shows how the 3-layer network was trained with different stochastic gradient optimisers (https://keras.io/optimizers/), and reveals which algorithms work best for this particular problem domain."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 267,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Running GridSearch...\n",
      "GridSearch complete\n",
      "Best: 0.500777 using {'optimiser': 'Adam'}\n",
      "0.374629 (0.026112) with: {'optimiser': 'SGD'}\n",
      "0.489900 (0.026818) with: {'optimiser': 'RMSprop'}\n",
      "0.485238 (0.028940) with: {'optimiser': 'Adagrad'}\n",
      "0.490747 (0.005791) with: {'optimiser': 'Adadelta'}\n",
      "0.500777 (0.017054) with: {'optimiser': 'Adam'}\n",
      "0.497669 (0.008901) with: {'optimiser': 'Adamax'}\n",
      "0.468852 (0.011152) with: {'optimiser': 'Nadam'}\n"
     ]
    }
   ],
   "source": [
    "from keras.wrappers.scikit_learn import KerasClassifier\n",
    "from sklearn.model_selection import GridSearchCV\n",
    "from keras.constraints import maxnorm\n",
    "\n",
    "# activations = ['relu', 'tanh', 'linear']\n",
    "# batch_sizes = [10, 20, 30, 40, 50, 60]\n",
    "# epochs = [10, 20, 30]\n",
    "# dropout_rates = [0.0, 0.1, 0.2, 0.3, 0.4, 0.5, 0.6, 0.7, 0.8, 0.9]\n",
    "\n",
    "optimisers = ['SGD', 'RMSprop', 'Adagrad', 'Adadelta', 'Adam', 'Adamax', 'Nadam']\n",
    "\n",
    "model = KerasClassifier(build_fn=create_model, nb_epoch=5, batch_size=24, verbose=0)\n",
    "\n",
    "# define the grid search parameters\n",
    "# just using one here for illustration, but add activation_function=activations \n",
    "# or dropout_rate=dropout_rates to grid search on other parameters\n",
    "param_grid = dict(optimiser=optimisers)\n",
    "grid = GridSearchCV(estimator=model, param_grid=param_grid, n_jobs=1)\n",
    "\n",
    "print \"Running GridSearch...\"\n",
    "grid_result = grid.fit(train_x, train_y)\n",
    "print \"GridSearch complete\"\n",
    "\n",
    "# summarize results\n",
    "print(\"Best: %f using %s\" % (grid_result.best_score_, grid_result.best_params_))\n",
    "means = grid_result.cv_results_['mean_test_score']\n",
    "stds = grid_result.cv_results_['std_test_score']\n",
    "params = grid_result.cv_results_['params']\n",
    "for mean, stdev, param in zip(means, stds, params):\n",
    "    print(\"%f (%f) with: %r\" % (mean, stdev, param))\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Generating Predictions ###\n",
    "Once we've trained a model, we can use the following process to pass an audio file into our model and generate predictions, which will be the activation value (between 0 and 1) at the output layer for each of the 10 possible classes. Collectively, all the activations will sum to 1, meaning we can use argmax to show what the network's best guess is, as a well as providing a measure of its confidence.\n",
    "\n",
    "The following fragment considers 10 different audio samples, one for each class, showing the model's best guess (highest activation value), as well as the second best guess, which can provide useful insights in those cases where the model misidentifies the sample. After all, as you can see from the results, the predictions are not perfect. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 101,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\n",
      "-----  air conditioner -----\n"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/jaron/anaconda/lib/python2.7/site-packages/numpy/core/numeric.py:482: VisibleDeprecationWarning: using a non-integer number instead of an integer will result in an error in the future\n",
      "  return array(a, dtype, copy=False, order=order)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Top guess:  air conditioner  ( 0.922 )\n",
      "2nd guess:  street music  ( 0.025 )\n",
      "\n",
      "-----  car horn -----\n",
      "Top guess:  car horn  ( 1.0 )\n",
      "2nd guess:  dog bark  ( 0.0 )\n",
      "\n",
      "-----  children playing -----\n",
      "Top guess:  children playing  ( 0.577 )\n",
      "2nd guess:  street music  ( 0.312 )\n",
      "\n",
      "-----  dog bark -----\n",
      "Top guess:  dog bark  ( 0.978 )\n",
      "2nd guess:  gun shot  ( 0.022 )\n",
      "\n",
      "-----  drilling -----\n",
      "Top guess:  drilling  ( 1.0 )\n",
      "2nd guess:  jackhammer  ( 0.0 )\n",
      "\n",
      "-----  engine idling -----\n",
      "Top guess:  engine idling  ( 1.0 )\n",
      "2nd guess:  street music  ( 0.0 )\n",
      "\n",
      "-----  gun shot -----\n",
      "Top guess:  gun shot  ( 0.695 )\n",
      "2nd guess:  dog bark  ( 0.174 )\n",
      "\n",
      "-----  jackhammer -----\n",
      "Top guess:  jackhammer  ( 0.889 )\n",
      "2nd guess:  air conditioner  ( 0.031 )\n",
      "\n",
      "-----  siren -----\n",
      "Top guess:  siren  ( 1.0 )\n",
      "2nd guess:  dog bark  ( 0.0 )\n",
      "\n",
      "-----  street music -----\n",
      "Top guess:  street music  ( 1.0 )\n",
      "2nd guess:  children playing  ( 0.0 )\n"
     ]
    }
   ],
   "source": [
    "import librosa\n",
    "\n",
    "sound_file_paths = [\"aircon.wav\", \"carhorn.wav\", \"play.wav\", \"dogbark.wav\", \"drill.wav\",\n",
    "                    \"engine.wav\",\"gunshots.wav\",\"jackhammer.wav\",\"siren.wav\",\"music.wav\"]\n",
    "sound_names = [\"air conditioner\",\"car horn\",\"children playing\",\"dog bark\",\"drilling\",\"engine idling\",\n",
    "               \"gun shot\",\"jackhammer\",\"siren\",\"street music\"]\n",
    "parent_dir = 'samples/us8k/'\n",
    "\n",
    "\n",
    "# just extract the features, \n",
    "def extract_features_only(filename):\n",
    "    features = np.empty((0,193))\n",
    "    X, sample_rate = librosa.load(filename)\n",
    "    stft = np.abs(librosa.stft(X))\n",
    "    mfccs = np.mean(librosa.feature.mfcc(y=X, sr=sample_rate, n_mfcc=40).T,axis=0)\n",
    "    chroma = np.mean(librosa.feature.chroma_stft(S=stft, sr=sample_rate).T,axis=0)\n",
    "    mel = np.mean(librosa.feature.melspectrogram(X, sr=sample_rate).T,axis=0)\n",
    "    contrast = np.mean(librosa.feature.spectral_contrast(S=stft, sr=sample_rate).T,axis=0)\n",
    "    tonnetz = np.mean(librosa.feature.tonnetz(y=librosa.effects.harmonic(X), sr=sample_rate).T,axis=0)\n",
    "    ext_features = np.hstack([mfccs,chroma,mel,contrast,tonnetz])\n",
    "    features = np.vstack([features,ext_features])\n",
    "    return features\n",
    "\n",
    "# create predictions for each of the sound classes\n",
    "for s in range(len(sound_names)):\n",
    "\n",
    "    print \"\\n----- \", sound_names[s], \"-----\"\n",
    "    # load audio file and extract features\n",
    "    predict_file = parent_dir + sound_file_paths[s]\n",
    "    predict_x = extract_features_only(predict_file)\n",
    "    \n",
    "    # generate prediction, passing in just a single row of features\n",
    "    predictions = model.predict(predict_x)\n",
    "    \n",
    "    #for i in range(len(predictions[0])):\n",
    "    #    print sound_names[i], \"=\", round(predictions[0,i] * 100, 1)\n",
    "    \n",
    "    # get the indices of the top 2 predictions, invert into descending order\n",
    "    ind = np.argpartition(predictions[0], -2)[-2:]\n",
    "    ind[np.argsort(predictions[0][ind])]\n",
    "    ind = ind[::-1]\n",
    "    \n",
    "    print \"Top guess: \", sound_names[ind[0]], \" (\",round(predictions[0,ind[0]],3),\")\"\n",
    "    print \"2nd guess: \", sound_names[ind[1]], \" (\",round(predictions[0,ind[1]],3),\")\"\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As the feature extraction process relies on slicing the sound recordings into smaller, more manageable chunks, if we needed to predict the class of a longer recording, we might need to employ some kind of ensemble learning, where we establish a prediction for each slice of the recording, and then use a voting process to determine which source is declared the best guess. \n",
    "\n",
    "This voting approach could make use of the fact that the softmax activation function will return equivocal values if it has low confidence in its prediction. An example of that can be seen above in the gunshot example, where dogbark and gunshot have very similar probabilities. In such cases we could ignore weak predictions, ensuring only strong predictions contribute to the final voting decision. A system that uses this approach is described here: https://chatbotslife.com/finding-the-genre-of-a-song-with-deep-learning-da8f59a61194"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Evaluating Convergence"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "There is also valuable information about learning performance in the history object. This can, for instance, reveal the speed of convergence over epochs, initially a slope, which may ultimately plateau out.\n",
    "Whether the mode may be over-learning the training data (inflection for validation line).\n",
    "\n",
    "From the plot of accuracy indicates whether the model could be trained further (if the accuracy trend is still rising for the final epochs). It will also indicate if the model has over-learned the training dataset, it should show comparable performance on both train and test datasets.\n",
    "\n",
    "If the model is overfitting the accuracy graph will show excellent performance on the training data, and poor performance on the test data.\n",
    "\n",
    "Ideally, in the loss graph, we should see comparable performance in both the training and validation datasets. Should these parallel plots start to diverge, it might be a indication we should stop training earlier. For a more detailed explanation, see http://machinelearningmastery.com/display-deep-learning-model-training-history-in-keras/\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 371,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "History keys: ['acc', 'loss', 'val_acc', 'val_loss']\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA7gAAAIACAYAAACo18D3AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x203f983d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(16,8))\n",
    "\n",
    "print \"History keys:\", (history.history.keys())\n",
    "# summarise history for training and validation set accuracy\n",
    "plt.subplot(1,2,1)\n",
    "plt.plot(history.history['acc'])\n",
    "plt.plot(history.history['val_acc'])\n",
    "plt.title('Model Accuracy')\n",
    "plt.ylabel('accuracy')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper left')\n",
    "\n",
    "# summarise history for training and validation set loss\n",
    "plt.subplot(1,2,2)\n",
    "plt.plot(history.history['loss'])\n",
    "plt.plot(history.history['val_loss'])\n",
    "plt.title('Model Loss')\n",
    "plt.ylabel('loss')\n",
    "plt.xlabel('epoch')\n",
    "plt.legend(['train', 'validation'], loc='upper left')\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.11"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 0
}