Fashion MNIST in Keras using a Multilayer perceptron.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Fashion MNIST with Keras using a Multilayer perceptron"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "**Note**: This tutorial is based off of the [Keras MNIST tutorial](https://github.com/wxs/keras-mnist-tutorial) by [wxs](https://github.com/wxs).\n",
    "\n",
    "*Fashion MNIST* is a dataset from online fashion retailer *Zalando* consisting of a training set of 60,000 examples and a test set of 10,000 examples, where each example is a 28x28 grayscale image belonging to one of 10 possible classes (such as a t-shirt, dress, etc). It can be used as a drop-in replacement for the original MNIST digit dataset. \n",
    "\n",
    "In this tutorial, we'll train a type of **fully connected neural network** called a **multilayer perceptron** to predict the clothing category for each image. Because the image can belong to one of ten categories, this prediction problem is a multi-class classification problem. We will access the neural network library *Keras* through a *Tensorflow* backend."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The class labels are as follows:\n",
    "\n",
    "| Label | Description   | \n",
    "| ----- |:-------------:| \n",
    "| 0     | T-shirt/top   | \n",
    "| 1     | Trouser       |   \n",
    "| 2     | Pullover      |  \n",
    "| 3     | Dress         | \n",
    "| 4     | Coat          |   \n",
    "| 5     | Sandal        | \n",
    "| 6     | Shirt         | \n",
    "| 7     | Sneaker       |   \n",
    "| 8     | Bag           |  \n",
    "| 9     | Ankle boot    |  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Import packages"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "import pandas as pd\n",
    "import seaborn as sns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "//anaconda/envs/tensorflow/lib/python3.6/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": [
    "from keras.callbacks import ModelCheckpoint\n",
    "from keras.datasets import fashion_mnist\n",
    "from keras.layers.core import Dense, Dropout, Activation\n",
    "from keras.models import Sequential\n",
    "from keras.utils import np_utils"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Load data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "X_train original shape (60000, 28, 28)\n",
      "y_train original shape (60000,)\n"
     ]
    }
   ],
   "source": [
    "(X_train, y_train), (X_test, y_test) = fashion_mnist.load_data()\n",
    "print(\"X_train original shape\", X_train.shape)\n",
    "print(\"y_train original shape\", y_train.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's look at some examples of the training data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x288 with 10 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "class_labels_dict = {0:\"T-shirt/top\",1:\"Trouser\",2:\"Pullover\",3:\"Dress\",4:\"Coat\",5:\"Sandal\",6:\"Shirt\",7:\"Sneaker\",8:\"Bag\",9:\"Ankle boot\"}\n",
    "\n",
    "fig = plt.figure(figsize=(10,4))\n",
    "\n",
    "for i in range(10):\n",
    "    plt.subplot(2,5,i+1)\n",
    "    img = X_train[y_train == i][0]\n",
    "    plt.imshow(img, cmap='gray', interpolation='none')\n",
    "    plt.title(\"Class: {}\".format(class_labels_dict[i]))\n",
    "    plt.axis('off')\n",
    "    \n",
    "plt.tight_layout()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Format the data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The inputs to our neural network have to be single vectors rather than 2D images. As a result, we will have to reshape the input so that each 28x28 image becomes a single 784 dimensional vector. Unfortunately, this leads to some loss of information, as it becomes less clear which image pixels are next to each other."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Training matrix shape (60000, 784)\n",
      "Testing matrix shape (10000, 784)\n"
     ]
    }
   ],
   "source": [
    "X_train = X_train.reshape(X_train.shape[0], 784)\n",
    "X_test = X_test.reshape(X_test.shape[0], 784)\n",
    "X_train = X_train.astype('float32')\n",
    "X_test = X_test.astype('float32')\n",
    "\n",
    "print(\"Training matrix shape\", X_train.shape)\n",
    "print(\"Testing matrix shape\", X_test.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We also normalise the inputs so that pixel values are in the [0-1] range rather than [0-255]."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "X_train /= 255\n",
    "X_test /= 255"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "At the moment, the target vectors `y_train` and `y_test` are represented as arrays which can take the values 0, 1, 2, etc. This means our model would assume that predicting 1 instead of 0 is better than predicting 5. But our labels are actually categorical - the value 0 encodes t-shirts or tops, which are no more similar to trousers (encoded by 1) than they are to sandals (encoded by 5). To address this, we use **one-hot encoding**, a common technique for dealing with categorical variables.\n",
    "\n",
    "One-hot encoding maps the classes in the following way:\n",
    "\n",
    "`\n",
    "0 -> [1, 0, 0, 0, 0, 0, 0, 0, 0]\n",
    "1 -> [0, 1, 0, 0, 0, 0, 0, 0, 0]\n",
    "2 -> [0, 0, 1, 0, 0, 0, 0, 0, 0]\n",
    "`\n",
    "etc."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Convert 1-dimensional class arrays to 10-dimensional class matrices using one-hot encoding\n",
    "num_classes = 10 # There are 10 classes possible \n",
    "Y_train = np_utils.to_categorical(y_train, num_classes)\n",
    "Y_test = np_utils.to_categorical(y_test, num_classes)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Build the neural network"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this tutorial, we are building a **fully connected** network. This is a general purpose architecture where each node is connected to every node in the previous layer. \n",
    "\n",
    "![Neural Network Architecture](http://www.dspguide.com/graphics/F_26_5.gif)\n",
    "Figure from [The Scientist and Engineer's Guide to Digital Signal Processing](http://www.dspguide.com/ch26/2.htm)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "When defining a neural network, we need to decide on the **number of hidden layers**, the **number of nodes in each layer**, and the **activation functions** used between layers. We will build a simple network of three layers - an input layer, one hidden layer, and an output layer, with 512 units in the input and hidden layer. \n",
    "\n",
    "**Activation functions** propagate the output of one layer's nodes forward to the next layer. We use the *ReLU* activation function for the input and hidden layer, and the *softmax* activation function for the output layer. The softmax activation function normalizes output and assigns a probability to each class. For instance, if the model see a picture of a dress, it might assign a 80% probability that the item of clothing is a dress, as well as 10% probability that it's a pullover, and smaller probabilities to the other remaining classes. We then predict the class with the highest probability.\n",
    "\n",
    "We also add **Dropout** masks in between layers. These masks help prevent the model from overfitting on the training data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model = Sequential() # Declare a Sequential model \n",
    "model.add(Dense(512, input_shape=(784,))) # Add a dense input layer of 512 units\n",
    "                                        # The input shape is (784,) because the features are 784 dimensional vectors\n",
    "model.add(Activation('relu')) # Apply ReLu activation function to the layer output\n",
    "model.add(Dropout(0.2)) # Dropout helps prevent the model from overfitting on the training data\n",
    "model.add(Dense(512)) # Add a hidden layer of 512 units\n",
    "model.add(Activation('relu'))\n",
    "model.add(Dropout(0.2))\n",
    "model.add(Dense(num_classes)) # Add an output layer of 10 units, the number of classes \n",
    "model.add(Activation('softmax')) # The \"softmax\" activation ensures the output is a valid probability distribution"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Compile the model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "When compiling a model, Keras requires a **loss function** and **optimiser**. \n",
    "\n",
    "- The loss function we'll use here is **categorical cross entropy**, a popular loss function well-suited to comparing two probability distributions. It is given by the formula $H(p,q) = -\\sum\\limits_{x} p(x) \\text{log} (q(x))$, where $p$ is the predicted probability and $q$ is the actual probability. We sum over the 10 possible classes. Cross entropy loss increases as the predicted probability diverges from the actual label. Suppose the actual label of an observation is 1 but our predicted probability for class 1 is 0.01. This would give a higher loss value than a predicted probability of 0.40. A perfect model would have a loss value of 0.\n",
    "\n",
    "\n",
    "- The optimiser determines the **learning rate** of the neural network. The learning rate is the amount by which you adjust parameters to minimise the neural network's errors. A large learning rate means parameters take leaps - this can save time but can overshoot. A small learning rate inches along slowly, but can take longer. Here, we will use the \"Adam\" optimiser to determine the learning rates for us."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train the model\n",
    "\n",
    "Next, we fit our model on the training set and use it to make predictions on the test set. \n",
    "\n",
    "The hyperparameters we pass here are:\n",
    "- **batch size** - the number of training examples used in one forward/backward pass of the neural network model\n",
    "- **number of epochs** - the number of times the training algorithm iterates over the *entire* training set\n",
    "\n",
    "A greater batch size results in a more accurate calculation of the gradient, but comes at the cost of a greater memory requirement and [potentially overfitting](https://stats.stackexchange.com/questions/164876/tradeoff-batch-size-vs-number-of-iterations-to-train-a-neural-network). If we have 60k examples in our training dataset, and a batch size of 128 examples, then each epoch will involve 469 iterations. \n",
    "\n",
    "Increasing the number of epochs leads to greater accuracy on the training set, but it can also lead to overfitting, demonstrated by an increase in the error rate on the test set.  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We will use Keras' `ModelCheckpoint` to save the model as an HDF5 file for future use. This saved model will include the specified architecture and the weights that the model has learned over the training data. We could then load this model using `load_model`. By specifying `save_best_only` we only save the model at its best-performing epoch."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 60000 samples, validate on 10000 samples\n",
      "Epoch 1/25\n",
      "60000/60000 [==============================] - 10s 172us/step - loss: 0.5149 - acc: 0.8133 - val_loss: 0.4130 - val_acc: 0.8515\n",
      "Epoch 2/25\n",
      "60000/60000 [==============================] - 10s 163us/step - loss: 0.3799 - acc: 0.8618 - val_loss: 0.3806 - val_acc: 0.8639\n",
      "Epoch 3/25\n",
      "60000/60000 [==============================] - 10s 170us/step - loss: 0.3494 - acc: 0.8705 - val_loss: 0.3547 - val_acc: 0.8713\n",
      "Epoch 4/25\n",
      "60000/60000 [==============================] - 10s 173us/step - loss: 0.3253 - acc: 0.8799 - val_loss: 0.3602 - val_acc: 0.8686\n",
      "Epoch 5/25\n",
      "60000/60000 [==============================] - 10s 170us/step - loss: 0.3107 - acc: 0.8852 - val_loss: 0.3402 - val_acc: 0.8773\n",
      "Epoch 6/25\n",
      "60000/60000 [==============================] - 10s 171us/step - loss: 0.2952 - acc: 0.8891 - val_loss: 0.3318 - val_acc: 0.8828\n",
      "Epoch 7/25\n",
      "60000/60000 [==============================] - 11s 178us/step - loss: 0.2856 - acc: 0.8931 - val_loss: 0.3254 - val_acc: 0.8811\n",
      "Epoch 8/25\n",
      "60000/60000 [==============================] - 10s 173us/step - loss: 0.2736 - acc: 0.8964 - val_loss: 0.3251 - val_acc: 0.8828\n",
      "Epoch 9/25\n",
      "60000/60000 [==============================] - 10s 174us/step - loss: 0.2638 - acc: 0.9013 - val_loss: 0.3283 - val_acc: 0.8792\n",
      "Epoch 10/25\n",
      "60000/60000 [==============================] - 10s 172us/step - loss: 0.2589 - acc: 0.9022 - val_loss: 0.3290 - val_acc: 0.8819\n",
      "Epoch 11/25\n",
      "60000/60000 [==============================] - 11s 190us/step - loss: 0.2564 - acc: 0.9028 - val_loss: 0.3182 - val_acc: 0.8855\n",
      "Epoch 12/25\n",
      "60000/60000 [==============================] - 11s 182us/step - loss: 0.2479 - acc: 0.9063 - val_loss: 0.3303 - val_acc: 0.8827\n",
      "Epoch 13/25\n",
      "60000/60000 [==============================] - 11s 180us/step - loss: 0.2461 - acc: 0.9060 - val_loss: 0.3290 - val_acc: 0.8857\n",
      "Epoch 14/25\n",
      "60000/60000 [==============================] - 8s 141us/step - loss: 0.2352 - acc: 0.9116 - val_loss: 0.3080 - val_acc: 0.8890\n",
      "Epoch 15/25\n",
      "60000/60000 [==============================] - 8s 131us/step - loss: 0.2279 - acc: 0.9141 - val_loss: 0.3144 - val_acc: 0.8906\n",
      "Epoch 16/25\n",
      "60000/60000 [==============================] - 8s 129us/step - loss: 0.2220 - acc: 0.9147 - val_loss: 0.3145 - val_acc: 0.8913\n",
      "Epoch 17/25\n",
      "60000/60000 [==============================] - 10s 174us/step - loss: 0.2193 - acc: 0.9155 - val_loss: 0.3232 - val_acc: 0.8812\n",
      "Epoch 18/25\n",
      "60000/60000 [==============================] - 9s 157us/step - loss: 0.2151 - acc: 0.9176 - val_loss: 0.3022 - val_acc: 0.8965\n",
      "Epoch 19/25\n",
      "60000/60000 [==============================] - 9s 155us/step - loss: 0.2087 - acc: 0.9196 - val_loss: 0.3209 - val_acc: 0.8899\n",
      "Epoch 20/25\n",
      "60000/60000 [==============================] - 10s 161us/step - loss: 0.2076 - acc: 0.9201 - val_loss: 0.3161 - val_acc: 0.8885\n",
      "Epoch 21/25\n",
      "60000/60000 [==============================] - 11s 189us/step - loss: 0.2023 - acc: 0.9225 - val_loss: 0.3098 - val_acc: 0.8905\n",
      "Epoch 22/25\n",
      "60000/60000 [==============================] - 9s 145us/step - loss: 0.1997 - acc: 0.9237 - val_loss: 0.3161 - val_acc: 0.8932\n",
      "Epoch 23/25\n",
      "60000/60000 [==============================] - 10s 174us/step - loss: 0.1953 - acc: 0.9254 - val_loss: 0.3113 - val_acc: 0.8955\n",
      "Epoch 24/25\n",
      "60000/60000 [==============================] - 8s 134us/step - loss: 0.1912 - acc: 0.9264 - val_loss: 0.3124 - val_acc: 0.8931\n",
      "Epoch 25/25\n",
      "60000/60000 [==============================] - 8s 134us/step - loss: 0.1865 - acc: 0.9283 - val_loss: 0.3184 - val_acc: 0.8958\n"
     ]
    }
   ],
   "source": [
    "filepath=\"mlp-weights.{epoch:02d}-{val_loss:.2f}.hdf5\"\n",
    "checkpoint = ModelCheckpoint(filepath, monitor='val_acc', verbose=1, save_best_only=True, mode='max')\n",
    "\n",
    "history = model.fit(X_train, \n",
    "                    Y_train,\n",
    "                    batch_size=128, \n",
    "                    epochs=25,\n",
    "                    verbose=1,\n",
    "                    validation_data=(X_test,Y_test))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Evaluate Performance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.plot(range(1,1+len(history.history['acc'])), history.history['acc'])\n",
    "plt.plot(range(1,1+len(history.history['val_acc'])), history.history['val_acc'])\n",
    "plt.title('Model Accuracy vs Number of Epochs')\n",
    "plt.ylabel('Accuracy')\n",
    "plt.xlabel('Epoch Number')\n",
    "plt.legend(['train', 'test'], loc='best')\n",
    "plt.xticks(np.arange(0,28,2))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Best test accuracy was 0.896500 at epoch 18\n"
     ]
    }
   ],
   "source": [
    "best_score = max(history.history['val_acc'])\n",
    "best_epoch = np.array(history.history['val_acc']).argmax()+1\n",
    "print('Best test accuracy was %f at epoch %i' % (best_score, best_epoch))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We see that, as expected, the accuracy on the training set increases with the number of epochs. The test set accuracy also initially increases with the number of epochs. However, at some point it may actually start to decrease due to overfitting. In fact, we start to see this effect here, where the test set accuracy is actually highest at the 18th epoch rather than the 25th epoch. At the 18th epoch, we achieve our highest **test set accuracy of 89.65%**. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Inspecting the output"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "It's always a good idea to inspect the output. Here we'll look at some examples the model gets right, and some examples it gets wrong."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "y_pred = model.predict_classes(X_test)\n",
    "\n",
    "# Check which items we got right / wrong\n",
    "correct_indices = np.nonzero(y_pred == y_test)[0]\n",
    "incorrect_indices = np.nonzero(y_pred != y_test)[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let's look at some examples that the neural network predicted correctly."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x576 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(8,8))\n",
    "\n",
    "for i, correct in enumerate(correct_indices[:9]):\n",
    "    plt.subplot(3,3,i+1)\n",
    "    plt.imshow(X_test[correct].reshape(28,28), cmap='gray', interpolation='none')\n",
    "    plt.title(\"Predicted:{}\\nActual:{}\".format(class_labels_dict[y_pred[correct]],class_labels_dict[y_test[correct]]))\n",
    "\n",
    "plt.tight_layout() "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now let's look at some examples that the neural network predicted incorrectly."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAiAAAAI4CAYAAABTDFLqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAIABJREFUeJzsvXm8XePZ///5iCEjMpAmJEEipqiYqSlFzWM9SqiGUrRPp99XW6oDHVSfB+2jRSuKaM1TJC1KTNGaKkgQFI2QyDwn5sT1+2Otw173fZ2zV/Y5Z+29z/68X6/zOvu+1rXWutfa11r73uv+7OuimUEIIYQQokjWqHYHhBBCCNF4aAAihBBCiMLRAEQIIYQQhaMBiBBCCCEKRwMQIYQQQhSOBiBCCCGEKJyGHoCQ3ISkkVwzbd9LclQB+z2f5PXtvZ/VheTJJP+Z07cmj0Fk6egxXrqf8FhF46F4ry9qfgBCcjrJ90iuIDmX5LUku7fHvszsYDO7Lmef9m+PPjSzv3NJvpGeg5kkbylq36L9afQYJzmC5Mfp8S8n+W+SpxSxb1E8infFexM1PwBJOdzMugPYAcDOAH4cOjChXo4nN+no/SQA+6fnYCcAD1a3V6IdaNgYT5mVHv+6AM4GcBXJravcp7LU87fPKqN4V7zXzQAEAGBmbwO4F8AwACD5CMkLSD4G4F0Am5Fcj+TVJGeTfJvkL0l2Sv07kbyY5AKS0wAcWrr9dHunlbS/RvLldJT6EskdSP4FwEAAf01HsD9IfXcj+TjJJSSnkBxRsp1NSU5MtzMBQJ/VOOydAdxnZv9Jz8EcMxtdsu1TSvo4jeQZJctGpE9MziI5Lz0np5Qs701yPMllJP8FYHBwPi4lOSNd/gzJvVaj36ICGjTGS4/fzOwuAIsBbN0Uw8Ex5Pq2SrJ/Gt+LSL5O8msl9vdI9irx3T49Z2ul7a+m52UxyftIDirxNZL/TfI1AK9VcpwiQfHe4PFuZjX9B2A6km//ADAAwFQAv0jbjwB4C8A2ANYEsBaAuwBcCaAbgA0B/AvAGan/mQBeSbfTC8DDAAzAmiXbOy19fSyAt5EMAAhgCIBBYZ/S9kYAFgI4BMmg7gtpe4N0+RMAfgNgHQB7A1gO4PqS9Z8HcEIzx/9lAIsAfB/J049OwfJDkQwcCGAfJBftDumyEQBWAvh5em4OSZf3TJffDODW9FwNS4/3n8G+e6fn9iwAcwB0TpedX3oM+lOMtyLGRwCYmb5eA8DRAD4CsEXpsmbO1ydxCGCT4FgnArgCQGcAwwHMB7BfuuwhAF8r2eZFAP6Yvj4KwOsAtkrP+Y8BPF7iawAmpOe3S7Xjp97+FO+K90+2Xe1gzBmsKwAsAfBmeoK7lATXz0t8+wL4oPQkARgJ4OGSN+HMkmUHtBCs9wH4TrkLKG2fDeAvgc99AEYhGVmvBNCtZNmNWI0PbwAnAngAwDvpRXBOC753NfU7Deb3mo4vtc0DsBuATmnQb1my7FcoGYA4214MYLvwQtCfYrw1MZ7G6cfp8S8CMBnA8SXLVvuGjOQDaRWAHiXrXQhgTPr6NAAPpa8JYAaAvdP2vQBOLVlvDSQD90Fp2wDsW+24qdc/xbvivemvXuYvjzKzB5pZNqPk9SAkI+bZJJtsa5T49A/832xhnwMA/Cdn/wYBOJbk4SW2tZCMxvsDWGxm7wT7HZBz2zCzGwDckD4uOyp9/ZyZ3UfyYADnARiK5Fi7AnihZPWFZraypP0ugO4ANkASuM2eD5JnIQnc/kiCcF1U+KhRlKWhYxzJnPjGq+Ffjv4AFpnZ8qBPO6Wvbwfwe5L9AWyOJL7/kS4bBOBSkpeUrEsk34qbzmfpORarj+Jd8V43A5CWsJLXM5CMlvsEH7pNzEY2SAa2sN0ZCDQRzeyzyfcvZva10DGdS+tJsltJwA50tlEWM/sIwG0kzwYwjOQjAO4A8BUA48zsI5J3IQmecsxHMoofgOQRZlO/mvq9F5JvAfsBmGpmH5NcnHPbom1pmBh3eAfJoLppX52QDJ7LMQtAL5I9Sm7KA5E8goeZLSF5P4AvIXn0fJOlX/eQHOsF6cC/Odri2ISP4v3TfXXoeK8rEWo5zGw2gPsBXEJyXZJrkBxMcp/U5VYA3ya5McmeAM5pYXN/AvA9kjsyYUiJMGcugM1KfK8HcDjJA1NRVOdUTLSxmb0JYBKAn5Fcm+SeAA5HTpjk5jiUZI/0eA5GMj/6FIC1kcxBzgewMl12QJ7tmtkqAHcCOJ9kVyYK7FElLj2QDFDmA1iT5E+RPAERVaQjxngZXgXQOb0G1kIyP71OuZXMbAaAxwFcmPb1swBOBVB6k70RyeD9mPR1E38E8EOS2wAAExHksW1yNGK1ULx37HjvUAOQlK8g+WB+CYlm4XYA/dJlVyGZx5sC4FkkH8AuZnYbgAuQvFHLkWgrmlTEFwL4MRN19PfSN/9IAOci+cCegUQ02nR+TwCwK5L5vvMA/Ll0XySnkjyxma4sS7f7FpI5w/8F8HUz+2c60v02kotwcbqf8S2cm5BvIpmOmQNgDIBrS5bdh2Ru8FUkj+Hehx471wodLcabxcyWAvgGkg+Pt5F8Q5zZ4kqfMhLJPPksAGMBnGdmE0qWj0fyOHqumU0p2edYAP8D4GaSywC8CODg1e27aDMU7/mou3jnp09hhBBCCCGKoSM+ARFCCCFEjaMBiBBCCCEKRwMQIYQQQhSOBiBCCCGEKBwNQNoArkYZ+2rBDlC6WdQf9XBtCNGWKObzU/cDkPT33NNIvrQa60QFf4qE5KkkX2FSyGguybtJ9qhWf0THpE6vjbVJnk/yNZLvMCnEdQ3JTdpg20ZySOt7KWoVxXy07ZqO+bofgCApBLQhkqqJO1e7M+VIE+j8CsBIM+uBJCvdrdXtVX7SC7wjxE0jUFfXRsrtAI5AkmdhPQDbAXgGSUZeIcqhmK8jOsIHySgA4wDcg2wmT5DsRfJakrOYlBm+i2Q3JAm2+jMpvbyCSbniMSR/WbJuZlRM8hyS/+GnZZyPrrC/OwN4wsyeAwAzW2Rm1zWlz037cXn6VGQ5yadIfpI+mOSWJCcwKbn8b5JfKll2KMnnSC4jOYPk+c11guQx6Ui7qQx2S6WnoxLZFR67KJa6ujaYlBz/AoAjzexpM1tpZkvN7HIzuzr1cUuOp8t2IflEGsOzSV5Gcu102aOp25T0uI6rpI+i5lHM11PMt0eFu6L+kOTMX4akZPIxABYAWLtk+d0AbgHQE0khoX2s+YqDYwD8sqSd8UFSyrk/kkHbcUgy1PVLl52MbBn7v6GZirUA9kJSofZnAPYAsI7Tj0UAdkFSq+cGADeny7ohych3Srpsh/SYtynp87ZpHz+LJL3wURZXTjwFSfnlIemycqWnH0FQIrva773+OuS18WsAE8scV0slx3dEUul5zTTeXwbw3ZJ1rSnm9dfx/hTz9Rfz9f4E5ItIChXdj+RNXhPAoQBAsh+SdLJnmtliM/vIzCZWuiMzu83MZpnZx2Z2C4DXkAwSPN/DzOzXzSz7R9rvHZBcEAtJ/oZJ0aEm7jSzf1lSfOkGJEEHAIcBmG5m11oyUn4WSTG6/0q3/YiZvZD28XkANwHYB1m+iySl8Agzez21fRnAPWZ2T7ruBCS1Dg4pWW+MmU1N9/tR2RMmqk3dXRsAeiMpLuZCcgCAPQGcbWbvm9lkJCmrT0q3/YyZPZnG6HQAVyKOf9FxUczXWczX+wBkFIBb05P/AZI6AE2P3QYgKU+8uC12RPIrJCenj7qWABiGCkvTm9m9ZnY4kjoERyIZMZ9W4jKn5PW7SOq1AEnZ5F2b+pD240QAn0n7uCvJh0nOJ7kUwJlOH78P4HIzKxVdNZWeLt3unvi03gKgOjD1Rj1eGwuRjbmQ5kqOb5T2YyjJv5Gcw6Smxa8q7IeoTxTzdRbzdfuTTJIbA9gXwC4kj0nNXZFUEuyD5AOzF8n1zWxJsLpXACdTBhnph3q6r0FIih7th0S/sYrkZLSyNL2ZfQzgQZIPIQngcsxA8rjuC80svxHAZQAONrP3Sf4f4mA8AMDfSc4xsztKtuuWni7tbo7+iRqgjq+NBwB8h0nFUe9XCS2WHAfwBwDPIRF4Lyf5XaRPB0XHRjFfnzFfz09ATkJSqXULJFMUwwEMRVI5cKQlZZzvBXAFyZ4k1yK5d7ruXAC9Sa5Xsr3JAA5JhUqfQTJV0UQ3JEE6HwBInoJ8A4YIkkeSPD7tE0nuguSR2ZM5Vv8bgKEkT0qPZy2SO5PcKl3eA8lo+f10uyc425gK4CAAl5M8IrU1W3q6kmMUVacurw0zewDABABjmZRMX5NkD5JnkvyqlS853gOJBmAFyS0BfD3YRVhyXXQcFPN1GPP1PAAZBeAKM5tT+gfgj/j0sdtJAD4C8AqAeUiDyMxeQaKPmJY+QusP4C9ISjpPRzKHeEvTjszsJQCXAHgCyRu6LYDHmusYyXtJntvM4sUAvoZkznAZkg//i8zshmb8PyEdAR8A4HgkI+M5SMoor5O6fAPAz0kuB/BTNPPzXktKMR8G4CqSB1v50tOivqjXawNIvr3dk+5jKZLS4Dsh+aYItFxy/HtIBt3LkXxD/aSfKecDuC49ri9BdCQU83UY8zTTk3UhhBBCFIu+4QohhBCicDQAEUIIIUThaAAihBBCiMLRAEQIIYQQhdOqPCAkDwJwKYBOAP7UQra3Jn8pXgUALDCzDardiUpYnZjv6PHep082xUynTp0iHzJOjfDxxx9n2h9++GHks2RJmKqhrmmIeE/9O3TMr7/++pn2O++8E/l069YtsoXXwRprxN/9w+sCABYvbpO8aYVjZrlyolQ8AGGSOvxyJHVDZgJ4muT49CdKQrTEm9XuQCUo5rMcfXS2/lZ4cwaANdeMbzErVqzItGfOjPMvjR07tpW9qykU7x2EESNGZNr/+te/Ip9ddokzsq+99tqZdufOnSOfDz74ILLdckv4q9qORWumYHYB8LqZTTOzDwHcjCSXhBAdFcW8aCQU76Jdac0AZCNk64PMTG0ZSJ5OchLJSa3YlxC1QNmYV7yLDoTu8aJdaY0GxJvjieb/zGw0gNFAx58fFB2esjGveBcdCN3jRbvSmgHITCQVBpvYGEmqWCE6KlWL+VDElieDsScA9TQZH330UabtiUlXrlwZ2cI5a09El0eE2rVr18jn73//e2Q7+OCDI1tIeHxev0Vuauoe78VSKOb0rgsvLkP+8Y9/RLa33norsu21116ZdnjtNNfPvn37ZtpezN9zzz2R7YQTsuW8fvrTn0Y+U6ZMybTXWmutyGfVqlWRLc95aW9aMwXzNIDNSW5Kcm0k9UnGt023hKhJFPOikVC8i3al4icgZraS5DcB3IfkJ1rXmNnUNuuZEDWGYl40Eop30d60Kg+Imd2DpIqfEA2BYl40Eop30Z60agAihCiGPBqQcD7cm+P15qxDLrvsssjm5SiYNSsrB/C27eU7CHMihHlBAGD48OFl++kRaj48PYs3Hy5qHy/mK30vw2RhXu6O999/P7KFOo099tgj8tloo+iHQpg6NfvgyLuePN3TnnvumWlvsskmkU+oAclzjdcKSsUuhBBCiMLRAEQIIYQQhaMBiBBCCCEKRwMQIYQQQhQO8yQ0arOdKUueSHjGzHaqdifam/aMdy/ZUB7xmZfM6/vf/36mPXjw4MjHS2AWivSmT58e+XiCvFAA6FUG9ZI5hVx00UWR7dJLL820ayHZEhok3oHKYz5vddiQfv36Zdpbb7115ONVrB05cmSm/fjjj0c+b7/9dmRbsGBBpn3UUUdFPhdeeGFkC7e/dOnSyMe77vbdd99Me8CAAZHP0KFDM+3nn38+8pk8eXJka0/yVsPVExAhhBBCFI4GIEIIIYQoHA1AhBBCCFE4GoAIIYQQonCUCVWIOiAU6eURnN50002R7dhjj41soUjv3XffjXw8QWCPHj0y7VAQ2BxdunTJtN97773Ix8tgus4662TaF1xwQeTzgx/8INP+1re+FfncfvvtkU1VdKtLHsHp//3f/0W2MKvuY489FvkcdNBBkW3RokWZtid6njNnTmR7/fXXM23vRxye4HP58uWZticGnz17dmQLr7Hw2gGA6667rsV9AcDDDz8c2UaNGpVpL1myJPJpb/QERAghhBCFowGIEEIIIQpHAxAhhBBCFE6rEpGRnA5gOYBVAFaWS7ajRGQipW4TM61OzLdlvOepdPv5z38+0w4rdwJxBVsgnkf3ko55+wsTioXbAYCXXnopsvXu3TvTXm+99SIfrxJpeK/y+hQmaOvatWvks8MOO0S2cG7f0wS04l7ZEPGe+rdZzB955JGZ9qmnnhr5/O53v8u0veR0c+fOjWwTJ07MtBcuXBj5jB49OrJtuOGGmbYX8y+++GJkC3VW48aNi3yee+65yHbttddm2htvvHHks/vuu2faI0aMiHy8a3qrrbbKtA8//PDIxzsvecibiKwtRKifN7MF5d2E6DAo5kUjoXgX7YKmYIQQQghROK0dgBiA+0k+Q/J0z4Hk6SQnkZzUyn0JUQu0GPOKd9HB0D1etButnYLZw8xmkdwQwASSr5jZo6UOZjYawGhAGhDRIWgx5hXvooOhe7xoN1o1ADGzWen/eSTHAtgFwKMtryVE/VKtmM+TqOmMM87ItL1kWl4V3VDg6gkwPRFbmAzNS47mVcP94IMPWtw/4As+Qz9vvbAP3nn77W9/G9lCAV6RVcJrmWre49dff/1Mu2fPnpHP2WefnWl7ybw8Ueimm26aaQ8bNizyeeihhyJbGCdeLHkJzMI49ISxt956a2QLRa/e9fTAAw9k2p7I+qSTTopsoRj8iCOOiHxCEWxbU/EUDMluJHs0vQZwAID4nRaig6CYF42E4l20N615AtIXwNj029KaAG40s7+3Sa+EqE0U86KRULyLdqXiAYiZTQOwXRv2RYiaRjEvGgnFu2hvVIxOiDrE02nssccembZX5M3TcoR6h7wakA8//DDT9jQnnk4jTGDm9dNbz+tXuX4uW7Ys8tl7770jW6gB8HQDolg8PUfIoEGDMm2vEJynm9hggw0y7f333z/yufPOOyPbvvvum2lfdtllkY/XhxUrVmTa559/fuQTFsgD4qKMXjHJz33uc5EtJEy8BsTJ94YMGVJ2O22N8oAIIYQQonA0ABFCCCFE4WgAIoQQQojC0QBECCGEEIUjEaoQdchxxx0X2Xr16pVpL1++PPIJRW1ALEJdunRp5NOlS5fIFlYCXWeddSKfsAqo1wevoqhXDTckjyjV8/Fs3/ve9zLtk08+uey2RfsyePDgTLtz586RTyhgHjhwYOTjJaMLq8q+9dZbkc8XvvCFyHbYYYdl2l7yPe/6CatQv/rqq5HPT37yk8gWimU97rjjjkx70qQ4I/6f//znyHb33Xdn2uPHjy+7r7ZGT0CEEEIIUTgagAghhBCicDQAEUIIIUThaAAihBBCiMKRCLVCvv71r0e2bbfdNtP+xje+UdG2PZGcqnOKUrzsh6HYzste6hEK6fIIToE4E6pXaXfJkiVl958382p4fHmuE287XsbWPffcs2w/RbHstddembaXHfeee+7JtI855pjIx8u027Vr10zbE0t/8YtfjGxPP/2039kSvGvlBz/4QabtCcT79OkT2R5++OFM+4orroh8br/99hb3BfgC1+HDh2faBxxwQOTT3ugJiBBCCCEKRwMQIYQQQhSOBiBCCCGEKJyyk8QkrwFwGIB5ZjYstfUCcAuATQBMB/AlM1tcSQfC+WZvvi7PeuF8dF5WrVpV1ufQQw+NbP37949s8+bNy7S95C8/+tGPItuMGTMy7bx6Dy+pVEie4xNZ2jvm24Ltt98+suXRSHiJk8L1vIRPXmKwMP68uO3Xr1/Z/eW9dsO5dW+9sE+ebsBbL+99pyNSq/E+cuTITHvcuHGRz1VXXZVpe7qNzTffPLK98cYbmXaoCQGAhQsXRrYLLrgg0w6rKAPArrvuGtnC7YeJyQDgtttui2yXXHJJpu3pRI4//vhMe7vttot8PF3X888/n2nPnz8/8mlv8jwBGQPgoMB2DoAHzWxzAA+mbSE6CmOgmBeNwxgo3kUVKDsAMbNHASwKzEcCuC59fR2Ao9q4X0JUDcW8aCQU76JaVPoz3L5mNhsAzGw2yQ2bcyR5OoDTK9yPELVCrphXvIsOgu7xot1p9zwgZjYawGgAIKlkFqJDo3gXjYZiXlRKpQOQuST7pSPjfgDmlV2jGUKh5mWXXRb5TJw4MbIVKRrzEop5CWk++OCDTHvmzJmRj1fFNBSvjh07NvLxEteEAlNPlJqnYmgelAit7WK+LQgrhQJxgi3vvfdEmWEceUJVL7bCbXnreXETJizLI6YGYvFqnvU8Hy8RWbdu3XL1oYGoqXgHgKeeeiqyTZs2LdPu27dv5PPuu+9GtldeeSXT9hLRXX/99ZEtFKaGydIAoEePHpEtTJj25JNPRj69e/eObFtssUWmvcMOO0Q+u+++e6Y9dOjQyCdMaAbEFYB32223yMfrZ1tS6c9wxwMYlb4eBSCWJwvRsVDMi0ZC8S7anbIDEJI3AXgCwBYkZ5I8FcCvAXyB5GsAvpC2hegQKOZFI6F4F9Wi7BSMmY1sZtF+bdwXIWoCxbxoJBTvolooE6oQQgghCqfQarhrrLFGlJFt4403zrQPP/zwaD0vS92LL76YaS9aFP6MPRYfeQK8AQMGRLavfvWrmfbcuXMjnwULFkS2sO933XVX5LP++utHtkMOOSTT3nLLLSOfUGwFABMmTMi033zzzcinUvJklAxFgYAyrxaFJ7YLBXJeddo8wk3vffXEpKHN258XN+F6Xsx4/QwzmHpVR/OIpT2fTTfdNNNed911I59ly5aV3bZoO8L7mfdjhJ122inT7t69e+Rz+eWXR7Ywk7B373z99dcj2wknnJBpP/HEE5GPl1E0zI7qCU6XLl0a2cLPsJNPPjnyyVPZ17umw0yzXgbkWhWhCiGEEEJUjAYgQgghhCgcDUCEEEIIUTiFakC6dOniVg8sxZvDC+fdAOCFF17ItL0Kl2FiMK8q4rbbbhvZwrnlRx99NPLxEsL8/e9/z7S9ZGlesqZwvQ03jLMee8llwsQxYXIdAJg6dWqmPWnSpMjHm7MM5+Wl7agt8iQU8zQSebblaSS8OeQw0ZnXJy+2wqRfnsbLSxbmbT8k1I54ydjyJOfzrjfv2hFtg3fff+ihhzLtKVOmRD6hvsOLtzlz5kS28LPh3nvvjXx69eoV2cLY9T4bjjjiiMi22WabZdpDhgyJfG699dbIFvp5fRozZkym7SUUmz59emT7z3/+k2mfccYZkc9FF11Udr3WoCcgQgghhCgcDUCEEEIIUTgagAghhBCicDQAEUIIIUThFCpC7dSpU5SIKxTVeCJNL3nX0UcfnWl7ichCEZ6XSMhLJPPqq69m2p4AMxR3AkCfPn0y7TDpGgAsXrw4svXs2TPT9qo3hpULPZtXhTGs1rjzzjtHPt7+lixZkml7iXrCKr5ALIQNxV6iMjbaaKOyPqFQtK0qITdHKFb19uclFAtta64Z34Y80atnCwmToXni9DznJRQNAhKhtieeKDO873oJtkaNGpVp33jjjZGPl/Trtttuy7Q9YaXXp/AHA17yMO8z7PHHH8+0Dz744MjH+5wJf5Dg/bDht7/9bab9pz/9KfLZaqutItvvf//7TNsTXh900EGRzUvsVil6AiKEEEKIwtEARAghhBCFowGIEEIIIQqn7ACE5DUk55F8scR2Psm3SU5O/w5paRtC1BOKedFIKN5FtcgjQh0D4DIAfw7svzWzi1dnZ6tWrcKKFSsytlD4cu2110breQLIULzqVfILhZTvv/9+5OOt99nPfjayhYTHAcRiTq86aL9+/SJbKJTzxLJeVsuwYqdXodcTvYZ452CDDTbItPv37x/5eOLVH//4x5n29ddfH/mMGzeubJ+qzBi0Ucy3FZ5ArBxe5tC8VY3zEIo5PRGdJwAM/bzr0hOvhmJVr995qv3myajqXacdmDGocrx71Z29quDlGDt2bGTzfsQQxpyXYdTLCBzem8PPGCDOlgrEWVy9z49Zs2ZFtj322CPT9n5sEX7O3HnnnZGPl501JMwuDsQ/kGhryl6JZvYogPioheigKOZFI6F4F9WiNRqQb5J8Pn181+wwieTpJCeRnOTVdxCijigb86XxXnTnhGhjVvseX2TnRP1T6QDkDwAGAxgOYDaAS5pzNLPRZraTme3k/d5fiDohV8yXxnuRnROijanoHl9U50THoKIRgZnNbXpN8ioAf8uz3vrrr48jjzwyYwvnvTytQah1AIBp06Zl2l6irnDA423b01bkGSh584phP70nPnPnzo1sYeIaTwPiESbm8RKDhXPi3vF6c/ChzXsPvPMZzpv+v//3/yKfOtCARFQa823F4MGDV3sdL/7yJOHytBWebsKbIw/xYiu8vjzdhtfPcH+e5iT08fqYRwMSJhVsNIqOd0/nF75PO+64Y+Tz7LPPZtp//etfI5+jjjoqsoWaE69Supf4zEs8lme98PNq9uzZkU+YABMAdt1110zbi90TTzwx077iiisiHy+p2jrrrJNpe58fXgXitqSiJyAkSxVaRwN4sTlfIToCinnRSCjeRRGU/apP8iYAIwD0ITkTwHkARpAcDsAATAdwRjv2UYhCUcyLRkLxLqpF2QGImY10zFe3Q1+EqAkU86KRULyLaqFMqEIIIYQonEJ/ltK5c+dITBeKc158MZ5qHDZsWGTbeOONM+08ibLyiM88P0+46dlC8Z5XidMTt4V+XoKvPD9hDpOHAfGxdO/ePfLJI6j1Eud4CcxCMZe3bbH6eO9tOTwxqZccr61+Hu8JR71rLvTzhKKeLRSreiLU8Jg9Qbl3vOH+vARqov0Iq74CwOc+97lMe5dddol8wnuQV317vfXWi2xf/OIXM+284vx///vfmfaIESMiHy852V133VW2T55QdPLkyZm2dx8OY3XGjBmRT/h5CQAvvfRSph1W+gWAN954I7K1JXoCIoQQQojC0QBECCGEEIWjAYgQQghO8H7wAAAgAElEQVQhCkcDECGEEEIUTqEi1I8//jgSWO62226Zdl4xUOjXtWvXyCfMKOpVi/VEmXnEqp7ALxS3eWI3bz1PrBoSZksFYkGSJ+QKz0uPHj0iH09MGp5f71g8gV+4rfPOOy/yEatPnkyoYWx5sea9Z3nizyMUbnrXjZf5NxRie/HniVDDa8C7V4TVQj3RtxfLoaC1EtGvqJw878mpp54a+eSpWuzFSXg9eYJmr/LsZpttlml7Vao9gWmYHXX+/PmRjycUfeWVVzLtUJQKAIcddlhkC/HE5+F63/nOdyIf7zOzLdETECGEEEIUjgYgQgghhCgcDUCEEEIIUTiFakDeeustfPvb387YwqQp3rybN48bzut589ih/qFnz56Rz/LlyyNbOE/uzQ96tnAOvEuXLpGPN7cdJs8JqxQ2t79w3jSPLsU7v17inFB3460XJuUBgAkTJkQ20XpCTYI3Zx5eE15iMC+O8lTI9XzCxGBe/HkJAivVWIV99/RbDz/8cKZ96KGHRj5e9d3w/tGrV6+yfRRth3f/Diulb7/99pGPV0E2xLvvh/d4L5aef/75yBYmR/Pi+/HHH49s4fXjVRfv1q1bZAvPwXvvvRf5hLoUj8985jORLdRIbrvttpHP+PHjy267NegJiBBCCCEKRwMQIYQQQhSOBiBCCCGEKJyyAxCSA0g+TPJlklNJfie19yI5geRr6f94ok2IOkPxLhoNxbyoFnlEqCsBnGVmz5LsAeAZkhMAnAzgQTP7NclzAJwD4OyWNrRq1SosXrw4Yzv33HMr6rgQ7USbxXtbEiaQyyO6nj59euTjVdMMq4yGwjfAF0Z7gupKfDw8oWooTPVEeyHh/QbwBaahwNVL2NaBqXrMe/EVCt/DZF4A8Nhjj5XdtpcYLPxhw8yZMyMfLwFmmDRv0003jXyWLl0a2cJEYF4CM88WimxPOumkyCdMKPaNb3wj8vGug/Ac7L777pHPQw89lGtblVL2CYiZzTazZ9PXywG8DGAjAEcCuC51uw7AUW3WKyGqhOJdNBqKeVEtVmuYT3ITANsDeApAXzObDSQBTDLOA56sczqA01vXTSGKR/EuGg3FvCiS3AMQkt0B3AHgu2a2LE/uAAAws9EARqfbqOx5rBAFo3gXjYZiXhRNrgEIybWQBOYNZnZnap5Lsl86Mu4HYF57dVKIIqnFeA81IF5Cot69e2faXuEqLxFZWBDSI4+Ww/vAyqOl8Ladx+bpRELdgJeoav/9949sYdGtPMnSOhLVjnnv/Q4Ll3oF3P70pz+V3fbw4cMj2+uvv55pe3Hy9ttvR7Ywxr0kjp4uZODAgZn2gQceGPl4hd9CncYRRxwR+YTnyeO5556LbEcdlZ1R84rhtbcWKs+vYAjgagAvm9lvShaNBzAqfT0KwLi2754QxaJ4F42GYl5UizzDmz0AnATgBZJNX6nOBfBrALeSPBXAWwCObZ8uClEoinfRaCjmRVUoOwAxs38CaG4ycL+27Y4Q1UXxLhoNxbyoFo010SmEEEKImqChsu0IUa+EokivGm7II488Etm22WabsutVmjzMWy+suAnECcTaUugWilCfeOKJyMcToXrVd0VxeALMsILrmDFjIp9JkyaV3XYo4AbiBGYvvvhi5PPmm29Gtr333jvTfvLJJyOfwYMHR7b99ss+SNpww/gXzV6snnrqqZm2J/QOExB6eJXLQ4Gr16ett946sk2cOLHs/vKiJyBCCCGEKBwNQIQQQghROBqACCGEEKJwNAARQgghROFIhCpEHfDBBx9k2p06dSq7zrhxcd4oLytkSFi5E8hXndbz8URz4fY98aq33tprrx13NiCsEvzoo49GPj/84Q8jW9gHTzwr2g+vguycOXMybS8TqpfZt9x2AOCNN97ItF955ZXIJ7zmgDi+pk6dGvlsttlmkW3bbbfNtL0qs/37949sW221VabtXfddunSJbCFTpkyJbOE15vV7gw02KLvt1qAnIEIIIYQoHA1AhBBCCFE4GoAIIYQQonCkARGiDvjoo4/K+qxYsSLT9pI7hUnAgMq1HHkqxq677rqRLU9V27wVckNC7YY3j+8lPgv70N5VQEUWL3nX8uXLM+1evXpFPnliwtOJzJo1K9P24uTzn/982X4++OCDkY93rUybNi3TXrx4ceTzrW99K7KFSdS6d+8e+XhJ1EJmzpwZ2cL7hZeMb9iwYZHt9ttvL7u/vOgJiBBCCCEKRwMQIYQQQhSOBiBCCCGEKJyyAxCSA0g+TPJlklNJfie1n0/ybZKT079D2r+7QrQvinfRaCjmRbXIo7RaCeAsM3uWZA8Az5CckC77rZld3H7dE6JwajLe33nnnUzbE6N17ty57HY8MWuYXMkT7XlC0dDPEwTmEZh663mCuHB/7733XuQTil49EaxHKEL0BLwdmKrH/NChQyNbHnHlgAEDMu0ZM2ZEPl7MH3JIdiw1e/bsyGfgwIGRLRSPesnRPKH3M888k2kPGTIk8tl0000j2/vvv1/WZ8SIEZEtJBTdAvH107dv38jnpZdeKrvt1lB2AGJmswHMTl8vJ/kygI3atVdCVAnFu2g0FPOiWqyWBoTkJgC2B/BUavomyedJXkOyZzPrnE5yEslJreqpEAWjeBeNhmJeFEnuAQjJ7gDuAPBdM1sG4A8ABgMYjmT0fIm3npmNNrOdzGynNuivEIWgeBeNhmJeFE2uAQjJtZAE5g1mdicAmNlcM1tlZh8DuArALu3XTSGKQ/EuGg3FvKgGZTUgTNK6XQ3gZTP7TYm9Xzp3CABHAyivGBKixqnVeA9Fc3369Il88lTF9DJOhuTNehravPU8QWsoOvV8vEyk4fa9/X32s5/NtH/xi1+U3U6jUwsx37NnPLsTVoL1spVee+21Zbc9atSoyPaTn/wk095xxx0jn7BiLgBMnz49095nn30in969e0e2UCwbtgFgyZIlkS3MtDp+/PjI56qrropseVi5cmWm/fLLL0c+XpXitiTPr2D2AHASgBdITk5t5wIYSXI4AAMwHcAZ7dJDIYpF8S4aDcW8qAp5fgXzTwDeV4Z72r47QlQXxbtoNBTzolooE6oQQgghCod5qgm22c7I4nYmaplnGkExX4vxfu+990a23XbbLdMOq5A2x1prrZVpe9qKMJESAHTq1Knstj2f8F4V7h8Arrzyykw7nOuvEg0R70BtxrwoHjPLJbTSExAhhBBCFI4GIEIIIYQoHA1AhBBCCFE4GoAIIYQQonCKFqHOB/AmgD4A6rXcZL32vZb6PcjMNqh2J9qbkngHauv8rw7qd+tpiHgHOsQ9vl77DdRO33PHe6EDkE92Sk6qV1V4vfa9XvvdUajX869+i0qo1/Nfr/0G6rPvmoIRQgghROFoACKEEEKIwqnWAGR0lfbbFtRr3+u13x2Fej3/6reohHo9//Xab6AO+14VDYgQQgghGhtNwQghhBCicDQAEUIIIUThFD4AIXkQyX+TfJ3kOUXvPy8kryE5j+SLJbZeJCeQfC3937OaffQgOYDkwyRfJjmV5HdSe833vSNSL/EOKOZF21AvMa94rz6FDkBIdgJwOYCDAWwNYCTJrYvsw2owBsBBge0cAA+a2eYAHkzbtcZKAGeZ2VYAdgPw3+k5roe+dyjqLN4BxbxoJXUW82OgeK8qRT8B2QXA62Y2zcw+BHAzgCML7kMuzOxRAIsC85EArktfXwfgqEI7lQMzm21mz6avlwN4GcBGqIO+d0DqJt4BxbxoE+om5hXv1afoAchGAGaUtGemtnqhr5nNBpIgALBhlfvTIiQ3AbA9gKdQZ33vINR7vAN1FjeK+apT7zFfVzFT7/Fe9ACEjk2/A24HSHYHcAeA75rZsmr3p0FRvBeIYr4mUMwXREeI96IHIDMBDChpbwxgVsF9aA1zSfYDgPT/vCr3x4XkWkgC8wYzuzM110XfOxj1Hu9AncSNYr5mqPeYr4uY6SjxXvQA5GkAm5PclOTaAI4HML7gPrSG8QBGpa9HARhXxb64kCSAqwG8bGa/KVlU833vgNR7vAN1EDeK+Zqi3mO+5mOmI8V74ZlQSR4C4P8AdAJwjZldUGgHckLyJgAjkJQ4ngvgPAB3AbgVwEAAbwE41sxCEVNVIbkngH8AeAHAx6n5XCRzhDXd945IvcQ7oJgXbUO9xLzivfooFbsQQgghCkeZUIUQQghROBqACCGEEKJwNAARQgghROFoACKEEEKIwtEARAghhBCFowGIEEIIIQpHAxAhhBBCFI4GIEIIIYQoHA1AhBBCCFE4GoAIIYQQonA0ABFCCCFE4WgAIoQQQojCaegBCMlNSBrJNdP2vSRHlVuvDfZ7Psnr23s/qwvJk0n+M6dvTR6DyNLRY7x0P+Gxio5JR4/pYJ9DSDZbMZbkT0j+scg+tSU1PwAhOZ3keyRXkJxL8lqS3dtjX2Z2sJldl7NP+7dHH5rZ37kk30jPwUyStxS1b9H+NHqMkxxB8uP0+JeT/DfJU4rYt2gfGjmmSZ6YHveK9Bx8XNJe0Zb7MrNfmNmZLfSl2QEMyWtIfpXkaSQfact+5aXmByAph5tZdwA7ANgZwI9DBybUy/HkJh3ZnwRg//Qc7ATgwer2SrQDDRvjKbPS418XwNkAriK5dZX7VBY9bWmRhoxpM7vBzLqnx34w0tgusRVCS7FJkgAOBHBvUf3xqKs33szeRnLChgEAyUdIXkDyMQDvAtiM5HokryY5m+TbJH9JslPq34nkxSQXkJwG4NDS7afbO62k/TWSL6ffyl4iuQPJvwAYCOCv6Yj2B6nvbiQfJ7mE5BSSI0q2synJiel2JgDosxqHvTOA+8zsP+k5mGNmo0u2fUpJH6eRPKNk2Yj0iclZJOel5+SUkuW9SY4nuYzkvwAMDs7HpSRnpMufIbnXavRbVECDxnjp8ZuZ3QVgMYCtm2I4OIZc32RJ9k/jexHJ10l+rcT+HsleJb7bp+dsrbT91fS8LCZ5H8lBJb5G8r9JvgbgtUqOs5Fo9JjOQ9qPZ9N77VySFwXLv5Ley+eTPKfE/kuSY9LXQ9LYPIXkWwDuB/BouqzpCczO6arbA5ibHtNlAPZKly9I/dcneX26v+kkf0iS6bLTSD5K8gqSS9Nz/fmKDtzMavoPwHQk3/4BYACAqQB+kbYfAfAWgG0ArAlgLQB3AbgSQDcAGwL4F4AzUv8zAbySbqcXgIcBGIA1S7Z3Wvr6WABvIxkAEMAQAIPCPqXtjQAsBHAIkkHdF9L2BunyJwD8BsA6APYGsBzA9SXrPw/ghGaO/8sAFgH4PpKnH52C5YciGTgQwD5ILugd0mUjAKwE8PP03BySLu+ZLr8ZwK3puRqWHu8/g333Ts/tWQDmAOicLju/9Bj0pxhvRYyPADAzfb0GgKMBfARgi9JlzZyvT+IQwCbBsU4EcAWAzgCGA5gPYL902UMAvlayzYsA/DF9fRSA1wFslZ7zHwN4vMTXAExIz2+XasdPLf41ekx7sV3G72kAI9PXPQDsmr4ekh7rH9M43gHABwA2T5f/EsCYwPdaAF0BdGmyOfv7ccn7cRqAR4LlNwK4M+3LZun1MKrEfyWAb6fv3QkAlgBYf7XjpNqBmjOQV6QH+CaSG0qXksD7eYlv3/TN6VJiGwng4fT1QwDOLFl2QAuBfB+A75S7uNL22QD+EvjcB2AUklH3SgDdgjc394c3gBMBPADgnfQCOacF37ua+p0G/3tNx5fa5gHYDUAnJDf5LUuW/QolAxBn24sBbJe+Pn91jkF/ivEWjn8EgI/T418EYDKA40uWrfYABMmH1SoAPUrWuxCf3qxPA/BQ+poAZgDYO23fC+DUkvXWQDJwH5S2DcC+1Y6bWv5r9JgOYjvPAORxAD8F0DuwNw0qPlNiexbAf6WvvQHIwHB9Z39PANg9fZ0ZgCAZVKwEMLTE9t8AHijxnwGAQZ9Grm6c1Mv85VFm9kAzy2aUvB6E5OTNTp8WAcnNo8mnf+D/Zgv7HADgPzn7NwjAsSQPL7GthWSk3h/AYjN7J9jvgJzbhpndAOCG9PHwUenr58zsPpIHAzgPwFAkx9oVwAslqy80s5Ul7XcBdAewAZIbdbPng+RZSIKtP5LAXhft+BiywWnoGEcyT77xaviXoz+ARWa2POjTTunr2wH8nmR/AJsjie9/pMsGAbiU5CUl6xLJN+am81l6joVPo8e0CxNd3+Vp82EzOxzAKQB+BuDf6TTT+WZ2T9M6ZjanZBNN9/DmaDE2SfZG8lTjqWZcNkTyBbX0PL+JJP6bmGnpyKNkef+W9utRLwOQlig9CTOQjKT7BB+6TcxGNoAGtrDdGQg0Ec3ss8n3L2b2tdAxnTvuSbJbSTAPdLZRFjP7CMBtJM8GMIyJcvkOAF8BMM7MPiJ5F5KbZTnmIxnlDkDyeLOpX0393gvJN4T9AEw1s49JLs65bdG2NEyMO7yDZFDdtK9OSAbP5ZgFoBfJHiWDkIFIHs/DzJaQvB/Al5BMtdxUckOdAeCCdODfHG1xbI1Mw8a0Jb/YuS6w/RvA8UwEuccCuINkzwq3X9pHr78HAZhgZh834zMPydPDQQBeTW2fXDsp4ZeFgUiuudWirkSo5TCz2UiEN5eQXJfkGiQHk9wndbkVwLdJbpy+uec0uzHgTwC+R3JHJgwpEaLNRTKCbOJ6AIeTPDAVTHVmIp7b2MzeBDAJwM9Irk1yTwCHIydMcnMcSrJHejwHI5k7fQrA2kjmJ+cDWJkuOyDPds1sFZI5vvNJdmXyi4NRJS49kAxQ5gNYk+RPkTwBEVWkI8Z4GV4F0Dm9BtZCMne9TrmVzGwGksfaF6Z9/SyAUwGUDipuRDJ4PyZ93cQfAfyQ5DYAwEQgeWybHI2IaMCYjiB5Esk+6aBgKZJBwcdlVsvDPABGsvS4DwVwT0l7LoCN0+ur6Yvu7QB+RbI7yU0B/H9IzlcT/Uh+k+SaJI9HMuj7++p2rkMNQFK+guSD+SUkmoXbAfRLl12FZI5vCpI5qzub24iZ3QbgAiQ3puVItBVNqvkLAfyYiXL6e+nN7kgA5yL5wJ6BRDTadH5PALArkvnt8wD8uXRfJKeSPLGZrixLt/sWkvnU/wXwdTP7Z/rN7ttILtDF6X7Gt3BuQr6J5FHeHABjkIiXmrgPyVz4q0ger70PPXauFTpajDeLmS0F8A0kHyxvI3kiMrPFlT5lJBJdyCwAYwGcZ2YTSpaPRzL9MtfMppTscyyA/wFwM8llAF5E8nNK0X40TEw3wyEAXia5HMDFAI4zsw9bu9H0M+JCAE+lx70Lkqfa95W4TUDya665JJumer4B4EMAbyARc1+H7PE/juSL8CIkOqxjzGzx6vaP2ac1QgghhOiIkPwcgIvN7HOt2MZpAL5sZiNa25+O+ARECCGEEDEfIxG71gQdQYQqhBBCiDKY2ZPV7kMpmoIRQgghROFoCkYIIYQQhaMBSE64GqXq22BfUf2LPMtybHcMyV+2rndC5KfI60aIWqAtY77cPZtJ/ZbNmlte69TFACT9Pfc0ki+txjoVf1C3FSS7pQFyT3nv+oXk+SSvL+8piqQer5s058L5JF8j+Q6TQljXkNykDbZtJIe0vpeiVqnTmD+V5CtMit7NJXk3yR551rWkwu60FrZd018662IAgqQQ0IZIqibuXM65hvgvJBn+DiDZr5yzEG1MPV43twM4AkkOhvUAbAfgGSS5C4QoR13FfJps7VdI6qj0QJKV99Y22nantthOe1IvA5BRAMYhyd5Wmq0TJHuRvJbkLCals+8i2Q1JEq3+/LQMcf9wNBiOfEmeQ/I//LSM89Ft0O8/IqmamElYk36z+x7J55mUNL6FZGdvIyS/nfYnqpWRHtcdTMomv0Hy22X61IfkhPQYJzJbZvxzJJ9O+/N0+pvx0v14pc0PQpLI57j0PE+JdymqRF1dNyT3R1KR9Egze9rMVprZUjO73MyuTn3cOEyX7ULyCSYJl2aTvIzk2umyR1O3KelxHVdJH0XNU1cxj6Rq7xNm9hwAmNkiM7suqGHUM30qspzkUyQ/STXPkqd6aZ//QPIeku8gyfx7IoAfpMf11wr72G7U/ACEZFckTxJuSP+Ob7qppPwFSa2IbZCMfH+b5u4/GEmBq+7pX5489f8BsBeSb14/A3B9c08uSP6NZLMpgUkORFIJsanfX3HcvoQkL/+mAD4L4GRnOz9J7fuY2cxg2RoA/ookQ+BGSL4lfpfkgS0c44kAfoGkqNzktG8g2QvA3QB+B6A3kjLUdzMpXAQANyHJQNkfyfvxK5L7mdnfkYzgb0nP83Yt7FsURJ1eN/sD+FeaobI53DhMl61CkjK6D4DdkVwP3wAAM9s79dkuPa5bchyXqCPqNOafAnAgyZ+R3IOkV2ZgZLqPngBeR5LptTlOSJf3QJK59AYA/5seV7ulkq+Umh+AAPgikmmM+wH8DUnukkMBIH3DD0ZSqnmxmX1kZhMr3ZGZ3WZms8zs4/QG9RqAXZrxPczMft3C5r4C4HkzewnJTXMbktsHPr9L97cIyUBieMkykvwNgAMBfN7M5jv72BnABmb2czP7MJ0LvArA8S30624ze9TMPgDwIwC7kxyA5Jy+ZmZ/Sb953oSkSN3h6fI9AZxtZu+b2WQkqbFPamE/orrU43XTG0nhMZdycWhmz5jZk2n8TgdwJYB9mtue6HDUXcyb2T/Sfu+A5AvgQpK/YXb65E4z+1daqO8GZD8nQsaZ2WNpv96v5NiKpB4GIKMA3JreVD5AUgeg6dHaACQlt1c7B70Hya+QnJw+wl0CYBgqLz//FaRPF9IR9UQEjwSR1GBpIiyxvD6A0wFcmNbD8BiE5NHhkpI+nwugbwv9+uTbpZmtQJLLv3/6F5a5birB3Fxp840gapV6vG4W4tP6Hx4txiHJoem3zTlMarj8qsJ+iPqkHmMeZnZv+nSiF5LaNCcDOK3EpaXPiZC6qtdV05lQmWge9gWwC8ljUnNXJNUx+yA52b1Irm9mS4LVvQxrmdLeAD5Tsq9BSJ4e7IdkTm4VycmooPw8E+3E5kgqap6VmnsgeQryPfNLTocsBvBlALeSPNrMHnN8ZgB4w8w2X43ufVLWmmR3JEE/K/0bFPgORFLhsMXS5lBp8pqiXq8bAA8A+A6TaqTerxLKxeEfADyHRNC3nOR3kTySFx2cOo75TzuRVMJ9kORDSAY0FW2mTLumqPUnICchqca6BZLHTsMBDEUyBzzSkjLO9wK4gmRPkmuRbJrrnQugN8n1SrY3GcAhqRjpMwC+W7KsG5I3az4AkDwFlQfBKCQVBrcu6fcwJAGdu6qmmT2CRLMxluSujsu/ACwjeTbJLkxKSg9jy+rvQ0jumc6N/gLAU+mc+z0AhpI8gUmJ5ePS/v/Nypc2nwtgk1STIqpPXV43ZvYAkutmLJNy6muS7EHyTJJfzRGHPZBUj15BcksAXw92EZZjFx2Huox5kkeSPD7tE5lUq90HQFulTK/pmK/1D4xRAK4wszmlf0h+WdL0aO0kAB8h0SvMQxooZvYKEu3FtPQxWX8kIqQpAKYjmSf8RIiWajUuAfAEkjdtWwDeUwcAAMl7SZ7r2DsjEZf+Puj3G+n+w2mYFrGkfPgpAMaT3DFYtgrA4UgutjcALEAyJ75euJ0SbkRSWnoRgB2R/jrHzBYCOAzAWUgehf8AwGFmtiBdr6XS5rel/xeSfHZ1jk+0C3V33ZTwX0gGw7cAWArgRQA7IXk6ArQch99DIsJbjuQbaig0PR/AdelxfamFPoj6o15jfjGAryHRkCwDcD2Ai8zshmb8V5erAWydHtddbbTNNkO1YIQQQghROLX+BEQIIYQQHRANQIQQQghROBqACCGEEKJwWjUAIXkQyX8zSYncbFZQIToKinnRSCjeRXtSsQiVSaa2V5HUbpgJ4GkkP3dqtgohSSleBQAsMLMNqt2J1WV1Y75e43399dePbMuXL49sq1atKqI7n0Bm0yzUkYC+IeI9Xadu3pRK6Ns3m+NxvfXiHxx6cRnG7ooVKyKfWbPyZICvD8wsV06U1iQi2wXA62n6b5C8GUkWt9xlkEXDEmZcrRcaIuZHjBgR2f7xj39EtoULFxbQm09ZZ51smYz332+7TNPtPLhRvFeRNdaIH/R//PHHFW3ry1/+cqZ9+OFxeZX33nsvsq211lqZ9hNPPBH5/OQnPym7f+9YQio9tmrQmimYjZBN+zoTTmpukqeTnERyUiv2JUQtUDbmFe+iA6F7vGhXWvMExHvEEn1tMLPRAEYDHf/xnOjwlI15xbvoQOgeL9qV1gxAZqKkrgiAjZFkJxSio1JXMR9OKwDA1ltvXdZvyJAhkc/YsWMj29/+9rdM++mnn87Vr/Ax8o477hj57L///pHtwgsvzLTvv//+yGf27Gwx3TffzDf7UUd6kiKparznmRZbc83sR9jKlXGZrUqnJDp16hTZLrroorJ98vYXbmvfffeNfC6//PLINmfOnEy70mNZe+21I1t4rqoxddOaKZinAWxOctO0rsjxAMa3TbeEqEkU86KRULyLdqXiJyBmtpLkNwHcB6ATgGvMbGqb9UyIGkMxLxoJxbtob1ozBQMzuwdJ4SghGgLFvGgkFO+iPSm0GJ0ESiLlGTPbqdqdaG+KjvcwJ8HOO+8c+bz77ruRLZz7feaZZyKfkSNHRrbvf//7Le4fAF56Kf7F5iabbNLi/gHgmmuuiWyXXnpppu1pR3r27Jlpe7kVpkyZEtnamYaId6DymPf0SqGtLTUKBx54YKYd/rwWAA466KDIFmqMunXrFvlssEGc8iXUW8ydOzfyGTx4cGR76qmnMu2bb7458rnjjjsy7VA3Ug3y5gFRKnYhhBBCFI4GIEIIIYQoHA1AhBBCCFE4GoAIIYQQonAkQhXVoCFEeUXH+5577plpL168OPLx6qd07mFR5GgAACAASURBVNw50/aEda+99lpkW7JkSaa9zTbbRD6HHnpoZJs4cWKmPXny5MjHq3kxdOjQTNs7llDsFwpeAeDJJ5+MbF5xsDakIeIdqH7Mf/Ob34x8ttxyy8jWu3fvTNsTuHox0b1790z7nXfeiXxCITQQ1zHyePvttyNbeC2GNWUA4KOPPiq7nYsvvjiyjRs3LtNuy5o5EqEKIYQQombRAEQIIYQQhaMBiBBCCCEKRwMQIYQQQhSORKiiGjSEKK8ts0KG16mXdXTvvffOtN94442y2/H2F1YYBYCuXbtGtlDwGWaJBIB11103snXp0qXs/jxbKAr0Kp+GNm//XhbKvFVzK6Qh4h2oPOa9aq0ffvhhpn3wwQdHPmHGXC/TrycUDYWb3nXhCUdXrVpV1mfBggWRrVevXpn2Bx98EPl4hPHsXReheDS8vgCgR48eke2II47ItCdNmhT55LkXeUiEKoQQQoiaRQMQIYQQQhSOBiBCCCGEKJx4Qmk1IDkdwHIAqwCsbJR5TtG4FBHzeeZY+/XrF9nCeWVvO14io06dOmXa4Tw3EM/HA/Hcszcf7iULCxM1vffee7nWC+fDveML++T1u2/fvpGtnTUgdUtR93jvfQrxKtaGmg8v+Z6nYwg1J97+vRgMYy68dprrQ5hQLK/2MkwS6K0X2pYuXRr5eNf9j370o0z76KOPLrvttqZVA5CUz5tZrLoRouOimBeNhOJdtAuaghFCCCFE4bR2AGIA7if5DMnTPQeSp5OcRDL+jY8Q9UeLMa94Fx0M3eNFu9HaKZg9zGwWyQ0BTCD5ipk9WupgZqMBjAaUB0R0CFqMecW76GDoHi/ajVYNQMxsVvp/HsmxAHYB8GjLawlRv9RKzPfp0yeyhQJML2mRRyg6zSs8W7ZsWab9mc98JvIZNGhQZHvqqacybe9YvCRjecSyoeDQS3Dl2cL1ikzQWMvUSrwDwOabbx7ZwuR0XhKu5cuXR7awyqtXCdYTr+Zhiy22iGxhMjRvfx5hP73rIoxVL769ZGzbbbddpu2JyPMmTKuUiqdgSHYj2aPpNYADALzYVh0TotZQzItGQvEu2pvWPAHpC2BsOkpcE8CNZvb3NumVELWJYl40Eop30a5UPAAxs2kAtivrKEQHQTEvGgnFu2hv2iIPSIcjz1xzW3Lrrbdm2t5cXOgDAFOmTMm0p0+fHvmESZ7yHsuAAQMy7eHDh0c+Q4cOjWyXXHJJru2L1SOMie7du0c+YeE1b+47LMIFxEmYPB9vLjjUmHhFuMI5bCBOBDZ//vzIxyNMpuTFctinMAEU4B9L2Kc5c+bk6pNoP0IthacVCnVInm7DS8IVxqWXUMyL3TzaES+BWSX6pbz7CzUgXsx72pGwwKSX6O3qq6+ObG2J8oAIIYQQonA0ABFCCCFE4WgAIoQQQojC0QBECCGEEIUjEaqDJz4K8URLeQSeM2bMiGyLFi3KtN94443I56yzzopsoWjJEzGFgigvUY9XCTIUPXrCxIEDB0a2SZOy2ZgnTpwY+YjVJxSdrrfeepFPKNLzxJ2eQC5MXOTFtrdeWKlz7ty5kY8niAtjy4vJPEmgvGRhobDOOxaP8FhE9Tn22GMzbe+9zFOdNk8secLkPMnCvKRf4T0QALbddtvV7lNev/DzKrwGgHwVrg866KDIRyJUIYQQQnQ4NAARQgghROFoACKEEEKIwtEARAghhBCFIxGqQx5hUx7B6e233x7Zxo4dG9nefffdTPuQQw6JfHr06BHZQmGgl/EvZMMNN4xseSoshhlVm9ufJ44UrSd8/z3xWyjm9CpgeoLP8L31fLx4D4XJXmx54s4lS5aU9fGuufA6CTO/AsDgwYMz7TBTJuALaj3hnqguRxxxRKYdxg0QXwdeRebZs2dHtjBLsPfDg7zZUUN22GGHyBZeK962vftwuJ4nSg235V1P3rYXLlyYae+zzz6RT3ujJyBCCCGEKBwNQIQQQghROBqACCGEEKJwympASF4D4DAA88xsWGrrBeAWAJsAmA7gS2a2uP26WV3yVpANk7ZcfPHFkc+TTz5Zdju/+93vIluYyAYAhg0bVtZn/fXXz7RffvnlyMebEw/nSO+///7IZ8SIEWX7OX78+Min1qnFmO/du3emvWLFisgnPPe9evWKfLz1wjnkvMm7wm15iYzC+AOAm2++OdP2kpV5OpR58+Zl2htssEHkE1ZQffHFFyOfxYvjt83bX6NQi/EOxFoGL1lYqI3ykod5WqGQPNeFR97PhjwJxfLszzuWMHbzVMwFYo2YpxNpb/I8ARkDILyznAPgQTPbHMCDaVuIjsIYKOZF4zAGindRBcoOQMzsUQCLAvORAK5LX18H4Kg27pcQVUMxLxoJxbuoFpX+DLevmc0GADObTTL+/V0KydMBnF7hfoSoFXLFvOJddBB0jxftTrvnATGz0QBGAwDJeCJKiA6E4l00Gop5USmVDkDmkuyXjoz7AZhXdo0axRMHeYKdkN/85jeRLUwgduqpp1bUp1mzZuWy3XfffRVtv60YNGhQZDvxxBMz7QsuuKCo7rQ3VY35UGjmJWUKqxN7oj1vvVDYllfEFvbJE3x6QtgwgZ2XVC1PH7xEeKFY1xPBTp8+PbI1sgi1GQqN97DyLRALTL2K3KFw0osbLzHXggULMu08SRyBOBGZtz+vuni4/bzVcMPte/0M++TFcvijAm/b/fv3j3yOPPLIyDZu3Di/sxVQ6c9wxwMYlb4eBaDteiREbaKYF42E4l20O2UHICRvAvAEgC1IziR5KoBfA/gCydcAfCFtC9EhUMyLRkLxLqpF2SkYMxvZzKL92rgvQtQEinnRSCjeRbVQJlQhhBBCFE7NVcP1ssF5tlDE41Up9ERLIXkEpz/72c8iW5hxEfCz6eUhPL68AqUwC1+eY/HYb7/4i86Xv/zlTPu4446LfDxBY5hl0hMhLloUphwQpayzzjqRLRSfvf3225FPKLj0KoO+/vrrkS3MKOoJ3byMj2uumb19eBWTt9tuu8j23HPPRbY8hOI6T0wanjtPWOeJtzfbbLOK+iTahiFDhkS2+fPnZ9qeuLJPnz6ZtidonjZtWmQLt+Xdu73MoOHnjHeteHEZVmnOW2E9/AzzBLVhPJ900kmRj1el/M0338y0J02aFPlMnjw5srUlegIihBBCiMLRAEQIIYQQhaMBiBBCCCEKp+Y0IN48WN6Kg5Xgzf1eeOGFmXa/fv0iH28OPpxzf/fddyMfb1tLly4t289K2WmnnTLtO++8M/LxtDIPPPBApu1Vww3nNYFYh7LllltGPo8//rjfWQHAn+cN55qXLVsW+YT6h+233z7yufvuu8tu26tO62msQrx48DQYYfXlrl27Rj559FsLFy6MbOGx9OzZM/LxtCphHypNUCgqI7znejYvqdzGG2+caXsaJy8h5O9///tM+6mnnop8PH1HqHvyKol71513vYZ4MRduP0y0BwAzZswou52hQ4dGtldffbVsn9obPQERQgghROFoACKEEEKIwtEARAghhBCFowGIEEIIIQqn5kSonmhsm222iWyhQChMgAUAe++9d6a9+eabRz5bb711ZAsrzz7zzDORj1cJNkza4iUr85J3XXnllZn2mWeeGfnkwTtPoXjUq5DqJeE55phjyvp4wqowqVXfvn39zopmCYVuQJy4KKxgC8TvkXfu33nnncgWCky9pEWeCDXsU5g4CvCvgRBP9Or1MzwvXkyGiag84aInHAy37SW98kTloji8e6dnCwmTlQHA3LlzM20v6Zh3HYRx6VXD9baVJ3GmJ3IOheWzZ8+OfMJ7rkelgtO8CdMqRU9AhBBCCFE4GoAIIYQQonA0ABFCCCFE4ZQdgJC8huQ8ki+W2M4n+TbJyenfIe3bTSGKQzEvGgnFu6gWeUSoYwBcBuDPgf23ZnZxaztw4oknZtrf//73I5/HHnsssoUCNK+aYihU9aoiXnHFFZFt6tSpmfaBBx4Y+Xj7+/DDDzPt1157LfIJxU8AcPTRR2fap5xySuTzy1/+MrL9z//8T6Z97733Rj6hsMkTbXmZKBcsWJBpe+JVT8wXZrB84oknIp86YAzaMebL0b1798gWCn69rKNvvfVWpr3nnntGPp5oLhSVeSJYTzQXCuQ8Ibgnmgv70KNHj8jHi9NQEOf1Kay07An0PJFgKOr1zlMHZgyqGO/NEQo3vfcktHkZdD0hcujnCaE9sWV4P/XEymGVWSDOYOpt24vnMCuyJ1T1+h6SJ569/bdnFnIgxxMQM3sUgOqni4ZBMS8aCcW7qBatGeZ/k+Tz6eO7+LezQnQ8FPOikVC8i3al0gHIHwAMBjAcwGwAlzTnSPJ0kpNITqpwX0LUArliXvEuOgi6x4t2p6JEZGb2iZCB5FUA/taC72gAo1PfaALr8MMPz7S9OSevYm1YATCc/wbiqqteQhqvSmBo23nnnSMfryphOHfvzT16lQrDufOZM2dGPkcccURkC5OFeToNr2JoiKflCOcDwwRPgF8tMtSThBWCAWDOnDll+1Rr5I35cvGeh1BbAcRJvjx9RzgXfMcdd0Q+XmIhL+lXiKcTCvGqzN5+++2RLZw39zQgXpK7MN68Pr3yyiuZ9qabbhr5ePeY8FrNU423I9OW9/hW9CHTzqubCPHub+F92Lsvh5o+b3/e/r2Edd5nT7lt58Wrnt1W225vKnoCQrK0pvzRAF5szleIjoBiXjQSindRBGWfgJC8CcAIAH1IzgRwHoARJIcDMADTAZzRjn0UolAU86KRULyLalF2AGJmIx3z1e3QFyFqAsW8aCQU76JaNNSP3YUQQghRGxRaDXfDDTfECSeckLGF1TKXLl0arderV6/I1r9//0zbE0SGQj1PaOTZQlGcl+jFEzaFAj9PlOfZwr57VRi9BDThep5wL1zPO14vMVMobPLEi16yqDBBXB6BlMjiVXkNE2V54uJQ8Dtu3LjIxxNGh3hJzp5//vnIFiarO+iggyKfsLI0EMetJ+j24jRPEqhQeD5yZPzlfsstt4xsoUjPEwJ7Im9RXcI48e6TngAzTMzlrefdF70kfSHe9RPe9z1Rv0covPX6mXdbtYiegAghhBCicDQAEUIIIUThaAAihBBCiMLRAEQIIYQQhVOoCHXZsmV44IEHMrbhw4dn2ptvvnm0nieADMU5ntAoXM/bjifqCbMgesJYT6AUbssTsnnHF1bp9SrITpkyJbLdeOONmfaVV14Z+YRVez0hnSfgDc+vd7yebfny5Zm2J0IUq0+YSTEUWwLAz3/+80x7k002iXzCSs9AfO1svfXWkc/EiRMjWyiWfeGFFyKfUDwLxBmDvYq1Xt/DDLpbbbVV5BMKsS+99NLIx8tsHAqqvWtX1CeVZgH1BKeh4DOvqD/8bPD65H0WhTavqm1YAbqe0BOQ/7+9Ow+XoyrzB/79AmFJAmQP2UPCFhYFxbjgzIAMDqBsigxRMcxPBR104PnhjAyPjuiAMjMCLuj8DIuJgpEoILgghogCKgECkS0JCSRku9kIISEEQuD9/VF1oeuc96Yr93ZXd3V/P8+TJ7dOn64+t/NW9UnV2+8RERGRwmkCIiIiIoXTBEREREQKV2gOyMsvv4zHH8+uaXT22Wdntr2Vb0877bSobeLEiZlt7751WODIy3XwVuIMc0C8VRi9trBg2fTp06M+Rx11VNSWZ8XaPLyVZ4cOHZrZ9vJgvPcgvLfp/b5ecapDDz00s+2tRuqtXCxv8u4Ph20rVqyI+oT/juEqywCwevXqqC28r/3ggw9Gfbx73eHz8qxgC8TF6cKVfgHgkUceidrC4zkshAbE+SVe/tbb3/72qO3ee+/NbHvFB71xSmN5cRnKU0Aub9ExL08jtHjx4qgtzGnyxu29Xljw0nt9L4eqLHQFRERERAqnCYiIiIgUThMQERERKVzVCQjJUSTvJjmP5BMkz0/bB5CcSXJh+nf/+g9XpL4U79JuFPPSKHmSULcBuNDMHia5J4A5JGcCOBvALDO7nORFAC4C8MWeDigsygUAV1xxRU93CyAu5gQAo0ePjtrChD8vuc5r85LiinTJJZdEbeHvfP/990d9vCStkSNHZra9Yjfe88IEQ6/wVZMrNN493grCmzdvzmx7yWjz5s3LbC9fvjzq4yUch8XCRowYEfUJk8eBOKHZSwT3jucwaS5vEl04zvA9AeJj0Cto5iW633TTTZnt3r175xpTi2h4zNeTt7p0eI73kuy9BOowOdlLGM9bnCzkHfdhYmr4BQmg3LFa9QqImXWY2cPpz5sAzAMwAsApAKal3aYBOLVegxQpiuJd2o1iXhplh3JASI4FcASA2QCGmlkHkAQwgCG1HpxIIynepd0o5qVIueuAkOwL4GYAF5jZxjzfv06fdw6Ac7o3PJHGULxLu1HMS9FyTUBI9kISmDea2S1p82qSw8ysg+QwAGu855rZFABT0v1Ur+JSR16ORqPzNmrp17/+dc32tWTJkprtq2waHe/ePd08i6OFfbx7315ORJhP4uX7jBo1KmoLc0zGjBkT9ZkzZ07UFi7IuMcee0R9vHvyYSEyb2GusIia9/t670tYDDAsaNbqGh3z9RQW8wLimPcKNHrHYbhooZcn4hVf3Lp163ZfH/BzmrwYD3lF88oiz7dgCOA6APPM7MqKh24HMDn9eTKA22o/PJFiKd6l3SjmpVHyXAE5CsBZAB4jOTdtuxjA5QBmkPwkgKUAPlKfIYoUSvEu7UYxLw1RdQJiZvcB6Opm4LG1HY5IYynepd0o5qVRVAlVREREClfoargiUl2YsAbECWpeAaSwWNiCBQuiPhMmTIjawoRLbxVdr5BSWORuxowZUR+vOFm4iq6XYOsl5IVFmbwEwLBAYJi4CvjJhWFSYJ6kXykHL8E0z6q2Xsy/9NJLme2wYCPgJ1CHx6sXu16RsTBh2kvYzpOo2qzKO3IREREpLU1AREREpHCagIiIiEjhNAERERGRwikJVaTJeFVHwyQ2b2Xngw46KLM9efLkqI+XvHrggQdmtlevXh31WbRoUdQWroZ8zDHHRH08YYKnt6ryu971rqht/Pjxme21a9dGfT784Q9ntr1VfIcOHRq19e+fXWneSxKUcvKOlTBR1Esc3bJlS9QWxkWYGA34MRdW5PXK3HvJpHkqqHZXOIZa7jsvXQERERGRwmkCIiIiIoXTBEREREQKpxwQkSbj5R+E94fDfAgAeM973pPZPvHEE6M+3n3tJ598MrMdFlsCgJNOOilqC1e1nTt3btTn1FNPjdruvPPOzPbPf/7zqI+XqzJ//vyqfRYuXJjZXrx4cdRn4sSJUdvuu++e2fYKPknzyZO34OVbhLyiY95xGOYveTkgXo5RePx6r+ftK3w9L+bDwn5loisgIiIiUjhNQERERKRwmoCIiIhI4TQBERERkcJVTUIlOQrAjwDsA+B1AFPM7NskLwHwaQCd1YAuNrPf1GugIkVohnj3VsVcunRpZjssOgbECWq33nprzca0ZMmSqO0tb3lLZvu3v/1t1Ofaa6+N2sLf77nnnuvZ4HbQkCFDorZwxdR2KkTWDDHfXWEC5ssvvxz18RI+wxj0+ngrMofJyV7CtpcoGq7S7K3Q6xUiCwukeecG79gsizzfgtkG4EIze5jkngDmkJyZPnaVmX2zfsMTKZziXdqNYl4aouoExMw6AHSkP28iOQ/AiHoPTKQRFO/SbhTz0ig7lANCciyAIwDMTps+R/JRkteT7N/Fc84h+RDJh3o0UpGCKd6l3SjmpUi5JyAk+wK4GcAFZrYRwP8CGA/gcCSz5yu855nZFDM70syOrMF4RQqheJd2o5iXouWqhEqyF5LAvNHMbgEAM1td8fg1AH5VlxGKFKzR8e6tphlWc7z33nujPmFim5ew1l1eQt6jjz663dcHgFWrVkVt4Wq/3U1C9SpchpUxvT5XXXVV1LZy5crM9qZNm7o1prJqdMzXk1eZNExE9pJXwxVsAaBPnz6ZbS9xdJ999onawn7eseIJK/v269cv6rN+/fpc+2pGVa+AMDmCrwMwz8yurGgfVtHtNADxWVOkZBTv0m4U89Ioea6AHAXgLACPkexc7OFiAJNIHg7AACwBcG5dRihSLMW7tBvFvDREnm/B3AfAW82nqb4PLlILindpN4p5aRSthivSZNauXVu1z5o1awoYyZtefPHFqO2FF16o+jzvPnrv3r1rMaRcvNVSH3jggcJeX+rLKyAWmjZtWtS2bt26zHa4IjTgr04bFj7zCtaFxcO8Nu95Xl5ImIs0cODAqE+eFYE93X1eLakUu4iIiBROExAREREpnCYgIiIiUjhNQERERKRwLDIRheRaAM8CGARgXZXuzaqsY2+mcY8xs8GNHkS9VcQ70Fzv/47QuHuuLeIdaIlzfFnHDTTP2HPHe6ETkDdelHyorGV7yzr2so67VZT1/de4pTvK+v6XddxAOceuWzAiIiJSOE1AREREpHCNmoBMadDr1kJZx17WcbeKsr7/Grd0R1nf/7KOGyjh2BuSAyIiIiLtTbdgREREpHCagIiIiEjhCp+AkDye5AKSi0heVPTr50XyepJrSD5e0TaA5EySC9O/+zdyjB6So0jeTXIeySdInp+2N/3YW1FZ4h1QzEttlCXmFe+NV+gEhOTOAL4H4AQABwOYRPLgIsewA6YCOD5ouwjALDPbH8CsdLvZbANwoZlNAPAuAOel73EZxt5SShbvgGJeeqhkMT8ViveGKvoKyEQAi8zsGTPbCuCnAE4peAy5mNk9ANYHzacA6FzbeRqAUwsdVA5m1mFmD6c/bwIwD8AIlGDsLag08Q4o5qUmShPzivfGK3oCMgLAsort5WlbWQw1sw4gCQIAQxo8nu0iORbAEQBmo2RjbxFlj3egZHGjmG+4ssd8qWKm7PFe9ASETpu+B1wHJPsCuBnABWa2sdHjaVOK9wIp5puCYr4grRDvRU9AlgMYVbE9EsDKgsfQE6tJDgOA9O81DR6Pi2QvJIF5o5ndkjaXYuwtpuzxDpQkbhTzTaPsMV+KmGmVeC96AvIggP1J7ktyVwBnAri94DH0xO0AJqc/TwZwWwPH4iJJANcBmGdmV1Y81PRjb0Flj3egBHGjmG8qZY/5po+ZVor3wiuhkjwRwLcA7AzgejO7rNAB5ERyOoCjkSxxvBrAVwD8AsAMAKMBLAXwETMLk5gaiuR7AdwL4DEAr6fNFyO5R9jUY29FZYl3QDEvtVGWmFe8N55KsYuIiEjhVAlVRERECqcJiIiIiBROExAREREpnCYgIiIiUjhNQERERKRwmoCIiIhI4TQBERERkcJpAiIiIiKF0wRERERECqcJiIiIiBROExAREREpnCYgIiIiUjhNQBwkx5I0kruk23eQnFzteTV43UtI3lCjff2B5Ke6eGw0yRdJ7lyL15LGaYVYzfFaZ5O8b0cfy7HfLo8RaW2tcNy0wjm+tBMQkktIbknf5NUkf0iybz1ey8xOMLNpOcf09/UYQxevdzHJxel7sJzkTXmeZ2ZLzayvmb22nX3r5FwjitU3XnNfkq+T/H6Rr1s0klNJXtrocZSdjpvWP8eXdgKSOsnM+gJ4G4B3APhS2IGJsv+ekXS2fhaAv0/fgyMBzKrBflvy/WoCbRurFT4B4HkAZ5LcrdGDkVJo2+OmHc7xTTGInjKzFQDuAHAo8MbM7jKSfwLwEoBxJPcmeR3JDpIrSF7aeXmK5M4kv0lyHclnAHygcv/hTJHkp0nOI7mJ5JMk30byxwBGA/hlOlv9t7Tvu0j+meQGkn8leXTFfvYl+cd0PzMBDNqBX/sdAO40s6fT92CVmU0J+owh+ad0/78jOSh93fDyY/h+/RjA3wC4Ov1drt6Bccl2tGmsdvoEkg+QVwGcFIzbSH6G5EKSz5P8Hkl6OyH5PyTvI7m389hBJGeSXE9yAckzqoxpPMkHSL5A8jaSAyr2dTLJJ9L34w8kJ1Q8NiFt25D2OTltPwfAxwD8W/re/jLvmyNda9PjpvXP8WZWyj8AliCZGQLAKABPAPjPdPsPAJYCOATALgB6AfgFgB8A6ANgCIAHAJyb9v8MgPnpfgYAuBuAAdilYn+fSn/+CIAVSIKDAPYDMCYcU7o9AsBzAE5EMtk7Lt0enD7+FwBXAtgNwN8C2ATghornPwrgo138/h8HsB7AvyKZGe8cPP4HAE8DOADAHun25eljY53fL3y/3vid9Uex2pNYTR//GwCvAOgP4LsAbg8eNwC/AtAPyUl+LYDj08fOBnBfOq5rANwJoHflY+nPfQAsA/BP6Xv5NgDrABzSxZj+kL4/h6bPvbnzd0qPm83p+9ALwL8BWARg13R7EYCL0+33pe/HgelzpwK4tNFxV/Y/7X7coA3O8Q0Psh4G54sANgB4FsD3AexR8WZ/raLvUCQnvz0q2iYBuDv9+fcAPlPx2Pu3E5x3Aji/2gGTbn8RwI+DPncCmIzkJLsNQJ+Kx35SGZw53oOPAbgLyYnyOQAXBcH5pYrtfwbw2+0E59eCfTc8OFvlj2LVAOBaAL9If343kqsgQyoeNwDvrdie0RnPSCYZswHchGSSsGtFv7Px5gTkHwHcG7zuDwB8pYsx/QHpCTvdPhjAVgA7A/gygBkVj+2E5EPpaCSTqVUAdqp4fDqAS9Kfp0ITEB03OsdX/bMLyu1UM7uri8eWVfw8BsmMr6Piqu5OFX2GB/2f3c5rjkIy68xjDICPkKy83NwLyex7OIDnzWxz8Lqjcu4bZnYjgBtJ9gJwavrzI2Z2Z9plVUX3lwBsL4Fr2XYek55r21gluQeS/1V+CgDM7C8klwL4KIBvVXTdXrzuB+CtACaa2dbt/A7vJLmhom0XJJebuxK+l72QXCYfjor31sxeJ7kMyf94twFYZmavB88dsZ3Xke5p2+MGaP1zfNknINtjn0TBUgAAIABJREFUFT8vQzI7HmRm25y+HcgGxejt7HcZgPE5XrOz74/N7NNhR5JjAPQn2aciQEc7+6jKzF4F8DOSX0RyOfnOKk9xd1NlW+qn1WP1NAB7Afg+ye+mbf2Q5IR8q8tnZc0D8D0Ad5B8n5ktcPosA/BHMzsu5z6B+L18Fcltm5UADut8IM1HGYXkKshrAEaR3KliEjIawFPpzzp2itHqx82bL9qi5/iWSEKtxsw6APwOwBUk9yK5E8nxJP8u7TIDwL+QHEmyP4CLtrO7awF8geTbmdgvDTQAWA1gXEXfGwCcRPIf0iSo3UkeTXKkmT0L4CEAXyW5K8n3IkjM2x4m9Q8+QHLP9Pc5Acn9vdl591FF+LtIAVoxVpFcjr4eyQf64emfowAcTvKw7T2xkplNR5J3cRdJ7wPiVwAOIHkWyV7pn3ewInnU8XGSB5PsDeBrAH5uyVcXZwD4AMlj0/99XojkA+7PSI6xzUgSTXulSYcnAfhpuk8dOwVrxeOmHc7xbTEBSX0CScLYk0i+CvhzAMPSxzoT2/4K4GEAt3S1EzP7GYDLkNzL24Qk8akzc/4bAL7EJBv6C2a2DMApSE6aa5HMlv8Vb77vHwXwTiSJRl8B8KPK12KSXf+xLoayMd3vUiT3SP8bwGfNrFtFmRzfBnA6k28kfKdG+5R8WiZWSY4AcCyAb1mSxd/5Zw6A3yKZnORmSa2GrwH4PcmxwWObkNzbPxPJFYxVAP4LSQJgV36MJGdjFYDdAfxLuq8FSJIAv4vkishJSL4SujW9BXQygBPSx74P4BNmNj/d53UADk7f21/syO8nPdIyx02q5c/xTJNRRERERArTTldAREREpEloAiIiIiKF0wRERERECqcJiIiIiBROE5ACsAdLhos0ozLENIP1MER6oqiYD18njeH96v26jdDyE5D0e9zPkHxyB55zNMnl9RzXdl6786T5It9chvpXJHekuJK0sLLFdPr6nyQ5n8miWatJ/prkno0aj5RL2WLeOY8vIbm92iNtqeUnIEgWABqCZLXEdzR6MDugnyVLML8VwEwAt5I82+uo/+G1nVLFdFoM6usAJpnZngAmICkMVQrph187nCubWalivkLneXwSgP8geXyjB1RNkZ8n7XBQTQZwG4DfICh6RHIAyR+SXJkWY/kFyT5Iln0eXjF7HU5yKslLK56bmV2TvIjk03xz+ebTajH4tGjTtwFcAuC/Ok+E6Yz6iyQfBbCZ5C7pOG8muZbkYpL/UjG+iSQfIrkx/R/olWn77iRvIPlcWlznQZJDazF2qZuyxfQ7APzFzB4BADNbb2bT0sJhSMfxvfSqyCaSs1lR6ZTkQSRnklxPcgHJMyoe+wDJR9K4Xkbykq4GQfLD6XHTuaT79pZRj5Z77+bvLrVRtpjPMLO/IFnN91A6twbTePtUtf2Q3Jvkj9Jz/LMkv8SkSupuaRwfWtF3MMktJIek2x8kOTft92eSb6noG32e1OL3rqalJyBMyiufDuDG9M+ZJHet6PJjAL2RlLcdAuCqtGb/CQBWmlnf9M/KHC/3NJJVMvcG8FUAN5Ac5nVkcktlRy/H3ZKO8cCKtkkAPoBkXY3XAfwSSaW/zuqTF5D8h7TvtwF828z2QrLOQef/QCenYx4FYCCSZau37ODYpCAljenZAP6B5FdJHkXSq0w6KX2N/kiWur8s3W8fJFcAf5L+PpOQrClzSPq8zUgqYPZDcix8luSpzvj+CUlV1L83s8eZVGj9NYBLkVS5/AKAm0kOrnjaWQDOAbAntr94mdRRSWO+sh9JHpWO75EcY9ie76ZjGwfg75DE/j+Z2StIPiMmVfQ9A8naSGtIvg3JcgjnIjnP/wDA7cGx+MbnSRfr6dRcS09AAHwIyfoNv0OyTsQuSN5gpEF1ApIlmp83s1fN7I/dfSEz+5mZrTSz183sJgALAUzsou8HzezyHXyJzoNnQEXbd8xsmZltQfK/zMFm9rW0XPQzSMoPn5n2fRXAfiQHmdmLZnZ/RftAAPuZ2WtmNsfMNu7g2KQ4pYtpM7s3HffbkHzoP0fySpI7V3S7xcweSE98NyJZLwYAPghgiZn90My2mdnDAG5G8oEEM/uDmT2WjvFRANORnJgrXYCkPPbRZrYobfs4gN+Y2W/S585Esm7HiRXPm2pmT6Sv+2rVN0zqpXQxX2EdkjLs1wK4yMxmdXds6fHyjwD+3cw2mdkSAFcgmSgDySS9cgLy0bQNAD4N4AdmNjs9z09D8p6+q6J/5edJIVp9AjIZwIz0BNI5Q+y8fDcKwHoze74WL0TyExWXtzYgWbFwUC32nepc6nt9RVu4HPXwztdPx3AxgM7bKZ8EcACA+Uxus3wwbf8xkjUSfppewvxvJotvSXMqZUyb2R1mdhKSCfQpAM4GUHnJuatlxccAeGcQ1x8DsE86xneSvDu9JP0Ckit44Rj/FcD3zKwyIbFzGfXK/b4Xb64dAjTh8uVtqpQxnxpkZv3NbIKZ9XS9lUFI1rqpvBr3LN78bPg9gD3SY2IMkkn8reljYwBcGMT7KADDK/ZVeLy3bPIiyZEA3gdgIskPp829AexOchCSN3sAyX5mtiF4urdAzub0+Z32qXitMUiuNhyL5F73ayTnAmBtfhsAyZLmawBULkMeLke92Mz2955sZgsBTGKSQ/IhAD8nOTC9VPlVJCs2jkVyj3UBkgW1pIm0Qkxbsnz9LJK/R3Jyr2YZksvIXX0L7CcArgZwgpm9TPJbiD8w3g/gtyRXmdnNFft1l1GvHG6O8UkdtULMdzEGpOPovNq8Txd9K61DcsV6DJIF9wBgNIAVQHJskZyB5CrIagC/6syzQvI+XWZml21n/4XHeytfATkLwFNIciY6lwA/AMByJNn4HUiSlL5Psj+TZbX/Nn3uagADSe5dsb+5AE5kkvC0D5LLup36IPnHWwu8cb85z8m1KpJDSX4OyUqK/56ewD0PANiYJhLtwWRp6EOZZoyT/DjJwenzOw/U10geQ/Kw9PLeRiQB/lotxi41V8qYJnkKyTPTMZHkRCS3Se6v9lwkl9wPIHlW+vv0IvkOkhPSx/dE8j/gl9P9ftTZxxMAjgfwPZInp21dLqPend9R6qaUMb89ZrYWyaTh42ns/R8keXnVnvcakty9y0jumU6Y/i+SWO70EyS3aT6GN2+/AMnE6jPp1RGS7MMkgbuhX4Vv5QnIZADft+wS4KsA/D+8efnuLCQfuPORXF24AAAsWVZ7OoBn0stVw5HcqvgrgCVI7kXe1PlCZvYkkntxf0ES9IcB+FNXAyN5B8mLq4x/A8nNAB5Dcl/6I2Z2fVed0+A8CckBuhjJbPlaJAlLQHICfoLki0gSUs80s5eRzLx/jmTyMQ/AH5ENaGkeZY3p55Hcg16IJM5uAPA/ZnZjtV84/R/c+5HkMq1EcqvmvwB0Js/9M4CvkdwE4D/Qxdd7zeyvSPJJriF5glVfRl2aQ1ljvppPI4m355Akp/455/M+j+QKyjMA7kMyyXjjc8HMZqePD0cyMetsfyh9zauRHI+LkNwGbSia6SqjiIiIFEuzfRERESmcJiAiIiJSOE1AREREpHA9moCQPJ5JaeRF1EI70gYU89JOFO9ST91OQk2/tvkUgOOQfCXqQSRfi+pytUKSTZfx2qtXXHPrtdey30Lt7ntUywRfMv4q+s4775zZ3ratkOq5tbDOzAZX79ZcdjTmmzHepSHaIt7T55Qi5sPzqb6MUVtmlqt2Sk8KkU0EsMiSkt8g+VMkX2vLvVxyMxg8OD4vbNiQrWfz6qvdq8L8+utxyQ5vIhHyDoZwsgEAAwYMyGyvWrUq6hPaaaf4opc3zjor67oaLRHzUjjFewPtskv8MReeT1955ZWihtMQ3udO+DmTp0+t9eQWzAhkS7cux5slYUVakWJe2oniXeqqJ1dAvP/KR9MlkucgWVFSpOyqxrziXVqIzvFSVz2ZgCxHsphNp5F4c8XWN5jZFABTgPLcHxTpQtWYV7xLC9E5XuqqJxOQBwHsT3JfJHXtz4S/DkNTu+mmm6K23XffPbPt5Ujsu+++UVuepFBvX2HbM888E/Xx8lD69++f2b744rgq8B133JHZzpODIl1qiZgXyal08X7yySfnauvbt29m+09/iiuuz58/P2p74IEHMtsbN26M+nQ3b2L8+Hg5mDPOOCOzPXr06KjP3LlzM9s/+MEPco0pzAdsQC5g9ycgZrYtXSTtTgA7A7jezJ6o2chEmoxiXtqJ4l3qrSdXQGBmv0GyfLtIW1DMSztRvEs9qRKqiIiIFK5HV0DKxis65glzN7z8i8ceeyxq624OSLX9AMBLL70UtYXfb9+6dWvVfTfiPp+IyI7KUyzsuOOOy2x/9rOfjfp49ZHCc+f73ve+qM+xxx4btZ133nmZbe+zwTsP5znvhrmHQPyZtWzZsqjPu9/97sz2yJEjoz5f/vKXuzWmetMVEBERESmcJiAiIiJSOE1AREREpHCagIiIiEjh2ioJdcSIeBmD3XbbLWoLFybykp/CFXOBOLEpb0GacF9eEqqX0Br2O+CAA6I+s2bNymw3YsEhEZEdlee8FCahbt68OerjJYquXbs2s+0VAVuxYkXUFiZu7rXXXlGfXXfdNWoLz7vr16+P+njn+PCz4cUXX4z6hEaNGhW1jR07NmpbsmRJZrsRC5XqCoiIiIgUThMQERERKZwmICIiIlI4TUBERESkcG2VhDpo0KCozUsYChOZ8iahhm15Ez7Dft6+X3755agtTBAaPnx41CfP64uIlNGYMWMy2+EXCIB45VsAePrppzPbe++9d9THS+p//PHHq+7bWyE3/IKC97kzePDgqG327NlVxxl+kcL7EsPEiROjtjAJtRGfDboCIiIiIoXTBEREREQK16NbMCSXANgE4DUA28zsyFoMSqRZKealnSjepZ5qkQNyjJmtq8F+6m7o0KFRm1d8JQ8vvyNs8/rkKezi3YvzxhkW2FEOSGFKE/ON1t3CdwsWLIjapk6dmtm+6qqroj5erpT0WFPEe79+/aK28LzoxZtXnCwsSnnPPfdEfY455pio7bDDDstsL1++vOqYgHg183333Tfq440hXA3XK1YWtnnH1+GHHx61zZgxo+rz6k23YERERKRwPZ2AGIDfkZxD8pxaDEikySnmpZ0o3qVuenoL5igzW0lyCICZJOebWeY6Uhq0ClxpFduNecW7tBid46VuenQFxMxWpn+vAXArgOjLxmY2xcyOVPKStIJqMa94l1aic7zUU7evgJDsA2AnM9uU/vx+AF+r2cjqYPTo0VGbl3gTtnmJTd4Ki2FxGW/fXhJqmLTkFSLzxhAmH3mF1qR2yhjzjeYl5IXx/fa3vz3qM3DgwKjt85//fGb761//etTngQceyGzfddddUZ9bb701anvooYeitnbXbPGep5Dkli1boj5bt26N2sLCYIccckjU57777ovaTj/99Mz2uHHjoj5eIbCwgNidd94Z9Vm9enXUFn5xwjueBgwYkNnetGlT1f00i57cghkK4Nb0g3EXAD8xs9/WZFQizUkxL+1E8S511e0JiJk9A+CtNRyLSFNTzEs7UbxLvelruCIiIlK4tlqMzssB8XIywrZwsZ+uhHkaeRexC+8Z5lmwzhtneC9QpNG8eA9dc801UVtHR0fV53l5WOE9+c997nNRH69t5cqVme0nn3wy6nPDDTdEbV4+idTHkUfGOa5hoS7vPOnlaaxatSqz7S0Od+ihh0ZtYbGwSZMmRX28gpDh88LCZIBfLCwsoublwYR5L14xvrDwWrPQFRAREREpnCYgIiIiUjhNQERERKRwmoCIiIhI4doqCXXkyJFRW57Vabu7Yq6377wr3YbyJKHuvvvuOzA6kdoKizsB/uqdofHjx0dtzz33XNQWJmt7x1f4PO/Y8o6lMNH8lFNOifoMGzYsalMSanEmTJgQtYUJl96/t1cYLEyOzlPoEQCGDBmS2f7lL38Z9fH2Fb6el0zqxXPfvn0z23vttVfU55VXXslse58xXmLqAQcckNl+6qmnoj71pisgIiIiUjhNQERERKRwmoCIiIhI4TQBERERkcK1VRJqv379ojYv8SessOhVyfOE+/KSkfIkk+ZZMReIE5t69+6da5wi9ZAn4RSIVx4999xzoz7XX3991LZ06dLMtleheI899shs51ntGojH7q1MqiTvxhozZkzUFlYKDSujAv6XDxYtWlT19bzzaZgU6iW45klC9Sqheknc4cq2XsJ2eBx4Cdx77rln1BauQq0kVBEREWkLmoCIiIhI4TQBERERkcJVzQEheT2ADwJYY2aHpm0DANwEYCyAJQDOMLPn6zfM2vBWi/XuW4f34rZu3Rr18XIywn3lLYrj3ZMOefcVw+ctX7486hMWvFm3bl3V12p3rRTzzejKK6/MbL///e+P+ixevDhqC48n71gK86fyFv4L8768/JI+ffpEba2gLPG+7777Rm3z58/PbHt5G15eSP/+/TPb4eq4gH/O3bhx43b309XzwgJiYT4T4OeFvPDCC5ntMAcFiPNEvM8rb0wHH3xw1Fa0PFdApgI4Pmi7CMAsM9sfwKx0W6RVTIViXtrHVCjepQGqTkDM7B4A64PmUwBMS3+eBuDUGo9LpGEU89JOFO/SKN39Gu5QM+sAADPrIDmkq44kzwFwTjdfR6RZ5Ip5xbu0CJ3jpe7qXgfEzKYAmAIAJKsnO4iUmOJd2o1iXrqruxOQ1SSHpTPjYQDW1HJQ9RKuZAgAHR0dUVuYtBQWkfH6APlW1s1bnCzkJc6FCXZe4lxYhEdJqN1WypivFS/+8sT75MmTo7ZHH300s+0lF3rHV5gc7h2X4ZjyFPAD4mPHS073ViJtYQ2P9zAG8qy2HBaiA/wCcmHSsRcn3vP23nvvzLa3yqx3XGzZsiWz7SXLer9f+MWCoUOHVt239x54MT9ixIiorWjd/Rru7QA6zyyTAdxWm+GINC3FvLQTxbvUXdUJCMnpAP4C4ECSy0l+EsDlAI4juRDAcem2SEtQzEs7UbxLo1S9BWNmk7p46Ngaj0WkKSjmpZ0o3qVRVAlVRERECtdWq+F6KwKuWLEiaguTO71qil7C0CuvvFL1eXnkSTj12rzKkN4KwNK+vNjyYjnkVVfMw1tJ+gtf+EJm26sK6b1ennGGx4B3THj7DpMSw8Q+ABg7dmzV15faGTVqVGbbO5eFCfzev7dXwTZ8nhenXryFK816CafdrWztjTNMevWEn2tegqsX897qwkXTFRAREREpnCYgIiIiUjhNQERERKRwLZ0DEt4/84obeffEQ14Rlzwr3ebNHQn3n2dM3vO832/gwIG59iXlF95X9uIvb05EnueF8Tdnzpyoj7d651NPPZXZ9uLduycfFp3KU8AvzO3oqi28/x7mc3XVNmlS9gsk06dPrzomyScsHOnlW4Qx4eVReCvWhjH3/PPxQr9ezO+zzz6Zba+wo3eOD/NXwlV1Ab/4XZiL5K0SHRYU887569eHS/3ky6mqN10BERERkcJpAiIiIiKF0wRERERECqcJiIiIiBSu8VkodbT//vtntr3VDbubmOoJ9+Ul4HlefPHFzLaX/OQl3IUJSl7i3ujRo3ONQZpHnhWTvT5esnTIS8AMk7W9pD1vxdobbrghsz1r1qyoz3ve856oLc+qtnkTv0Ph++Il2nkFCcNVTb2kRG9Ml1+eXSJFSai1M2zYsMy2l4Qaxu6gQYOiPhs2bIjannnmmcy2V2TOS+IOP0PCMQLAkiVLqj7P+9zxjs1wBWavcOYRRxyR2V61alXUJyygBsTHireKrleQr5Z0BUREREQKpwmIiIiIFE4TEBERESlc1QkIyetJriH5eEXbJSRXkJyb/jmxvsMUKY5iXtqJ4l0aJU8S6lQAVwP4UdB+lZl9s+YjqqEwCdWr+OglNoXJbt7zvCSiMOHNW73RS2wKk1C918tTidJLkvOSB6WqqahRzOepThrKs/Jxd51xxhlR27hx4zLb3/jGN6I+1157bdQ2YMCAzPaECROiPs8++2zUFiZGe8nh3U1CDY9B77hZtGhR1BZWZ/XGFK7OCgDvfve7M9vDhw+P+qxcudIfbPOYiiY8x4fnTy9JMzy+wpgEgDVr1kRteaqAepVvw4TWl156KerjxVzYb7fddov6eG1hEqqXYBryYtdLXt1vv/0y2wceeGDUZ+7cuVVfryeqHtFmdg+AuI6rSItSzEs7UbxLo/QkB+RzJB9NL9/F39tLkTyH5EMkH+rBa4k0g6oxr3iXFqJzvNRVdycg/wtgPIDDAXQAuKKrjmY2xcyONLMju/laIs0gV8wr3qVF6BwvddetQmRmtrrzZ5LXAPhVzUZUQ2EhMO8espcDEuZgePcevXuIYTEj7/7gCy+8ELWF+8+7Ymn4el6fcEVJ6Z7uxnytcjfe+ta3ZrbD3AMAOO+88zLbXu7B+PHjo7Ywlh9//PGoz8knnxy1rV69OrPtxXvv3r2jtk2bNmW28+aAhO+ld888HMPChQujPps3b47awnODdwx6/5bh+WPt2rVRnzJqhnN8mP/grega/lt6xRjvvffeqC2MHS/evLgcPHhwZvvpp5+O+njCz6LwdwP83KSwaJ73euHx5OUnesUFw/cub+HMWurWFRCSleXfTgMQn7FEWohiXtqJ4l2KUPUKCMnpAI4GMIjkcgBfAXA0ycMBGIAlAM6t4xhFCqWYl3aieJdGqToBMbNJTvN1dRiLSFNQzEs7UbxLo6gSqoiIiBSupVfDDQseeYVl8ibOhbwktTCpZ926dVEfL3k1TGTzkuu8JKkw4cpbIXWfffaJ2qQYO+20U7TC5G233ZbZ9v7NvBU2w1U/vZUrw/jzijJ5q1uGycvhawF+Maewn/d63qqyeQr9eQlxeVbmDIv6ecebt2Lqxo0bM9thgjfgnyuWLl2a2f7MZz4T9fnud7/rD1a2Kzx3hcmWQHzu9BJHvfgKYyBMqAbihFMA6OjoyGx7CZ99+vSJ2sLY9b7Y4H2mhHHp/S7h8eTFvPflh/Bzxlslut50BUREREQKpwmIiIiIFE4TEBERESmcJiAiIiJSuJZOQj3ssMMy295Kgl7CaZgM5CU2ecKkUK9yopcgFL6e97w8K4F6iapDhw6t+jypj/333x/XXHNNpi1cUTVPkiYQJ5F5CXlhHHlVfvPGcshLTA0TaL3kaS+Ww7F7ibheLIe8RL7wd/YSCb3XCytFev8GXsJhmLgXrpYq3RcmNXvJnSNHjsxsL1iwIOoTJiYD8b+bt3L5wIEDo7bw+Fm8eHHUx/uyQ3hseHEZJpx64/KO37BysVf9Os+XGMaOHRv1qTddAREREZHCaQIiIiIihdMERERERArX0jkgYf6Dd0/cu0cd3i/z8jY84X1j716zN4awzbv/7N1fD++Bb9u2LerjrQ4pxdi4cSN+97vfZdpOPPHEzLaXV7BixYqoLfy39v5dwxU2vfjzinnlyUHyjoHwOPHi1luFM9yXVwDJi/fwvrl3Xzs8BrwCU9699nCl1bDAGODf73/kkUcy27feemvUR7onLAbn5fyE8ewV3PKOsTB2vDwRr9jfkiVLMtt5C1mGRQLzjAmIj0WvEFmYd3TIIYdEfbzfJTwOvAKI9aYrICIiIlI4TUBERESkcJqAiIiISOGqTkBIjiJ5N8l5JJ8geX7aPoDkTJIL07/jm70iJaN4l3ajmJdGyZNduQ3AhWb2MMk9AcwhORPA2QBmmdnlJC8CcBGAL9ZvqDsuTGLyEtu8VTfD53nJQV4yaZiE561u6AmT8rwkQG9fYdKhl4SadwzyhprFe0dHBy699NJM23e+853M9hlnnBE97/TTT4/axo0bl9n2YjlMfvMS8rxktDxJ0HmKhYVFoboSJnjec889UZ+77rqratvatWujPt4qtvXkvS8l1JTn+LDolpe4GfISVb0iY+G5MkxCBvxE7zCZ1DtXe0X7wud5RdW8VZrDY8w7n4e/s/ceeL9LOHbv3FBvVa+AmFmHmT2c/rwJwDwAIwCcAmBa2m0agFPrNUiRoijepd0o5qVRduhruCTHAjgCwGwAQ82sA0gCmGRc/zV5zjkAzunZMEWKp3iXdqOYlyLlnoCQ7AvgZgAXmNnGvJcezWwKgCnpPuLrVSJNSPEu7UYxL0XLNQEh2QtJYN5oZrekzatJDktnxsMArKnXILsrvKflLRLkHWThPXDvPp+XAxLuy9t3noXm8uac5FmwKxyDd3/f23c7q2e8h8V/rr322qiP1xbe+/UWnDrooIMy2+HCd4B/LzjMQfLi1lvI8dlnn81sewXUvEJgteLlweQpcuYdA+H74hVe8wpMhYWoli1b5g+2yTXjOT78d8pTSNL7t/UWCAzzSbwiXGHeBhAX1vPixCv2F/Li0htnePx4hf3C4mRecbQ8C0V2d6HKnsjzLRgCuA7APDO7suKh2wFMTn+eDOC22g9PpFiKd2k3inlplDxXQI4CcBaAx0jOTdsuBnA5gBkkPwlgKYCP1GeIIoVSvEu7UcxLQ1SdgJjZfQC6uhl4bG2HI9JYindpN4p5aRRVQhUREZHCtfRquGGCkJeE6iURhck5eRJVgThJyks4zZOEmmfFXG9cXlJquG+vmI/3vkgx8iYqh/+2HR0dUR+vrdG8wklhslt3C5/lOZa82PYKNYX7ypuYrQTu+gkTnw888MCoT7iKrXc+99rCRM2BAwdGfbyCYvPnz89se8XDvBVrwyTQPImqQFxs0jt/h8myXgFCL043bdqU2fYKWdabroCIiIhI4TQBERERkcJpAiIiIiKF0wRERERECtfSSahhUk13V4b1kuS8BLg88pQ3zpvYlqfKaXf3LcXobhyVhZcY7VWYFAmFCcR5jpW8iZRhXHrnZS9Ow88Qb/Vlb1XZcLVdr49XiTRc8dmrxhu+T94Kr7KyAAAGZklEQVR74CWvhomxeT4/ak1XQERERKRwmoCIiIhI4TQBERERkcK1dA5InhyJPPcVu3tvbAeWs85s5ymO5rV5r7dkyZJcYxARaSZr1mQX3/WKyoW5FN5qz97zwsJjXm7FokWLorawiJ1XrMwrdJenIGRYVA0A+vbtm9kOV9P2ePvp06dP1Ba+BzNnzqy671rTFRAREREpnCYgIiIiUjhNQERERKRwVScgJEeRvJvkPJJPkDw/bb+E5AqSc9M/J9Z/uCL1pXiXdqOYl0bJk4S6DcCFZvYwyT0BzCHZma1ylZl9s37D65kwKTPPKqMeL4mpu6t1djeZ1CsgFiY25Umo9ZK0vKSpNlbaeBfppqaM+Z/97GeZ7Q996ENRn/B85hXh8s55oeeeey5q874M0K9fv8y29wUFrzhZmATqnc+9QmThZ4r3WTRkyJDM9uDBg6M+3gq9ixcvzmz/8Ic/jPrUW9UJiJl1AOhIf95Ech6AEfUemEgjKN6l3SjmpVF2KAeE5FgARwCYnTZ9juSjJK8n2b+L55xD8iGSD/VopCIFU7xLu1HMS5FyT0BI9gVwM4ALzGwjgP8FMB7A4Uhmz1d4zzOzKWZ2pJkdWYPxihRC8S7tRjEvRcs1ASHZC0lg3mhmtwCAma02s9fM7HUA1wCYWL9hihRH8S7tRjEvjVA1B4RJRuR1AOaZ2ZUV7cPSe4cAcBqAx+szxO475JBDMtteUlGe1WG3bNkSteVZWddLGPKESadeEqrXFiaPelVP3/nOd2a2vVUYtTrpm8oc7yLd0awx//zzz2e2vQqfRx6Zvehy3XXXVd0PAJx//vmZ7RdeeCHqs2LFiqgtTDoNq4kC+b5EEK6OC/gJtAcccEBmO6yMCgA33XRTZnvcuHFRn/322y9qCz+fNmzYEPWptzzfgjkKwFkAHiM5N227GMAkkocDMABLAJxblxGKFEvxLu1GMS8NkedbMPcB8BY1+U3thyPSWIp3aTeKeWkUVUIVERGRwrX0arjh/bmwYAvgr4IY3mfzcke84ja77bbbdvfTlTCfxMsd8VZBXLt2bdU+4X3FdevW5RqTiEgz8fI77r///sz29OnToz5eLsfVV1+d2fbO5/37x986Dj8LvFwOL2cw/Czw8j28Ambd4a1867XNmTOnJq/XE7oCIiIiIoXTBEREREQKpwmIiIiIFE4TEBERESkc86ygWrMXI9cCeBbAIABlzYYs69ibadxjzCxesrHFVMQ70Fzv/47QuHuuLeIdaIlzfFnHDTTP2HPHe6ETkDdelHyorOsGlHXsZR13qyjr+69xS3eU9f0v67iBco5dt2BERESkcJqAiIiISOEaNQGZ0qDXrYWyjr2s424VZX3/NW7pjrK+/2UdN1DCsTckB0RERETam27BiIiISOE0AREREZHCFT4BIXk8yQUkF5G8qOjXz4vk9STXkHy8om0AyZkkF6Z/x6sVNRjJUSTvJjmP5BMkz0/bm37srags8Q4o5qU2yhLzivfGK3QCQnJnAN8DcAKAgwFMInlwkWPYAVMBHB+0XQRglpntD2BWut1stgG40MwmAHgXgPPS97gMY28pJYt3QDEvPVSymJ8KxXtDFX0FZCKARWb2jJltBfBTAKcUPIZczOweAOFay6cAmJb+PA3AqYUOKgcz6zCzh9OfNwGYB2AESjD2FlSaeAcU81ITpYl5xXvjFT0BGQFgWcX28rStLIaaWQeQBAGAIQ0ez3aRHAvgCACzUbKxt4iyxztQsrhRzDdc2WO+VDFT9ngvegJCp03fA64Dkn0B3AzgAjPb2OjxtCnFe4EU801BMV+QVoj3oicgywGMqtgeCWBlwWPoidUkhwFA+veaBo/HRbIXksC80cxuSZtLMfYWU/Z4B0oSN4r5plH2mC9FzLRKvBc9AXkQwP4k9yW5K4AzAdxe8Bh64nYAk9OfJwO4rYFjcZEkgOsAzDOzKyseavqxt6CyxztQgrhRzDeVssd808dMK8V74ZVQSZ4I4FsAdgZwvZldVugAciI5HcDRSJY4Xg3gKwB+AWAGgNEAlgL4iJmFSUwNRfK9AO4F8BiA19Pmi5HcI2zqsbeissQ7oJiX2ihLzCveG0+l2EVERKRwqoQqIiIihdMERERERAqnCYiIiIgUThMQERERKZwmICIiIlI4TUBERESkcJqAiIiISOH+P4E3Y3BySI6FAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<Figure size 576x576 with 9 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.figure(figsize=(8,8))\n",
    "for i, incorrect in enumerate(incorrect_indices[:9]):\n",
    "    plt.subplot(3,3,i+1)\n",
    "    plt.imshow(X_test[incorrect].reshape(28,28), cmap='gray', interpolation='none')\n",
    "    plt.title(\"Predicted: {}\\n Actual: {}\".format(class_labels_dict[y_pred[incorrect]], class_labels_dict[y_test[incorrect]]))\n",
    "    \n",
    "plt.tight_layout() "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Encouragingly, a human observer can understand why many of these items may have been misclassified. For instance, the shoe in the middle row appears to have some of the characteristics of a sandal rather than an ankle boot. Similarly, the shoe in the bottom row looks like it could easily be an ankle boot rather than a sneaker. We also see a few examples where the model has incorrectly predicted a coat to be a pullover. "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Our analysis above also suggests that some classes may get misclassified more frequently than others. We can use the `classification_report` method from `scikit-learn` to get the classification metrics on each class. This will allow us to see which classes are most frequently misclassified. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "              precision    recall  f1-score   support\n",
      "\n",
      " T-shirt/top       0.83      0.88      0.85      1000\n",
      "     Trouser       0.99      0.98      0.98      1000\n",
      "    Pullover       0.77      0.87      0.81      1000\n",
      "       Dress       0.90      0.90      0.90      1000\n",
      "        Coat       0.83      0.79      0.81      1000\n",
      "      Sandal       0.98      0.97      0.97      1000\n",
      "       Shirt       0.76      0.67      0.71      1000\n",
      "     Sneaker       0.96      0.96      0.96      1000\n",
      "         Bag       0.99      0.97      0.98      1000\n",
      "  Ankle boot       0.96      0.97      0.96      1000\n",
      "\n",
      "    accuracy                           0.90     10000\n",
      "   macro avg       0.90      0.90      0.90     10000\n",
      "weighted avg       0.90      0.90      0.90     10000\n",
      "\n"
     ]
    }
   ],
   "source": [
    "from sklearn.metrics import classification_report, confusion_matrix\n",
    "target_names = [class_labels_dict[x] for x in range(0,10)]\n",
    "print(classification_report(y_test, y_pred, target_names=target_names))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Five of the ten classes (Trouser, Sandal, Sneaker, Bag, and Ankle boot) achieve an F1-score of 0.96 or higher. Our neural network is more successfully in correctly classifying these classes, suggesting that there are obvious image features that can be used to tell them apart.\n",
    "\n",
    "On the other hand, the classes with the lowest F1-scores are Shirt (0.71), Coat (0.81) and Pullover (0.81), followed by T-shirt/top (0.85) and Dress (0.90). This may suggest that it is more difficult to \n",
    "\n",
    "The classes with the lowest precision are Shirt (0.76) and Pull over (0.77). Low precision suggests that there was a substantial number of items that we classified as pullovers but which aren't actually pullovers. We saw this scenario three times in the examples above, where coats were misclassified as pullovers.\n",
    "\n",
    "The classes with the lowest recall are Shirt (0.67) and Coat (0.79). Low recall suggests that there was a substantial number of shirts which we didn't identify as shirts, and coats which we didn't identify as coats. Again, we saw examples of this above.  "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can also look at the **confusion matrix** to see how the classes have been misclassified. For instance, we can see that t-shirts/tops were commonly misclassified as shirts, while coats were misclassified as shirts or pullovers. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th>Actual</th>\n",
       "      <th>T-shirt/top</th>\n",
       "      <th>Trouser</th>\n",
       "      <th>Pullover</th>\n",
       "      <th>Dress</th>\n",
       "      <th>Coat</th>\n",
       "      <th>Sandal</th>\n",
       "      <th>Shirt</th>\n",
       "      <th>Sneaker</th>\n",
       "      <th>Bag</th>\n",
       "      <th>Ankle boot</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Predicted</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>T-shirt/top</th>\n",
       "      <td>877</td>\n",
       "      <td>0</td>\n",
       "      <td>18</td>\n",
       "      <td>18</td>\n",
       "      <td>2</td>\n",
       "      <td>1</td>\n",
       "      <td>82</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Trouser</th>\n",
       "      <td>3</td>\n",
       "      <td>978</td>\n",
       "      <td>2</td>\n",
       "      <td>9</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Pullover</th>\n",
       "      <td>18</td>\n",
       "      <td>1</td>\n",
       "      <td>867</td>\n",
       "      <td>9</td>\n",
       "      <td>55</td>\n",
       "      <td>0</td>\n",
       "      <td>50</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Dress</th>\n",
       "      <td>21</td>\n",
       "      <td>10</td>\n",
       "      <td>9</td>\n",
       "      <td>903</td>\n",
       "      <td>30</td>\n",
       "      <td>0</td>\n",
       "      <td>24</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Coat</th>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "      <td>131</td>\n",
       "      <td>29</td>\n",
       "      <td>791</td>\n",
       "      <td>0</td>\n",
       "      <td>47</td>\n",
       "      <td>0</td>\n",
       "      <td>1</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Sandal</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>972</td>\n",
       "      <td>0</td>\n",
       "      <td>15</td>\n",
       "      <td>0</td>\n",
       "      <td>13</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Shirt</th>\n",
       "      <td>125</td>\n",
       "      <td>1</td>\n",
       "      <td>102</td>\n",
       "      <td>26</td>\n",
       "      <td>70</td>\n",
       "      <td>0</td>\n",
       "      <td>673</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Sneaker</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>12</td>\n",
       "      <td>0</td>\n",
       "      <td>959</td>\n",
       "      <td>0</td>\n",
       "      <td>29</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Bag</th>\n",
       "      <td>7</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>4</td>\n",
       "      <td>3</td>\n",
       "      <td>2</td>\n",
       "      <td>8</td>\n",
       "      <td>4</td>\n",
       "      <td>969</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ankle boot</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>7</td>\n",
       "      <td>1</td>\n",
       "      <td>23</td>\n",
       "      <td>0</td>\n",
       "      <td>969</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "Actual       T-shirt/top  Trouser  Pullover  Dress  Coat  Sandal  Shirt  \\\n",
       "Predicted                                                                 \n",
       "T-shirt/top          877        0        18     18     2       1     82   \n",
       "Trouser                3      978         2      9     3       0      5   \n",
       "Pullover              18        1       867      9    55       0     50   \n",
       "Dress                 21       10         9    903    30       0     24   \n",
       "Coat                   1        0       131     29   791       0     47   \n",
       "Sandal                 0        0         0      0     0     972      0   \n",
       "Shirt                125        1       102     26    70       0    673   \n",
       "Sneaker                0        0         0      0     0      12      0   \n",
       "Bag                    7        0         3      4     3       2      8   \n",
       "Ankle boot             0        0         0      0     0       7      1   \n",
       "\n",
       "Actual       Sneaker  Bag  Ankle boot  \n",
       "Predicted                              \n",
       "T-shirt/top        0    2           0  \n",
       "Trouser            0    0           0  \n",
       "Pullover           0    0           0  \n",
       "Dress              0    3           0  \n",
       "Coat               0    1           0  \n",
       "Sandal            15    0          13  \n",
       "Shirt              0    3           0  \n",
       "Sneaker          959    0          29  \n",
       "Bag                4  969           0  \n",
       "Ankle boot        23    0         969  "
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "matrix = pd.DataFrame(confusion_matrix(y_test, y_pred), index=target_names, columns=target_names)\n",
    "matrix.index.name = 'Predicted'\n",
    "matrix.columns.name = 'Actual'\n",
    "matrix"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Conclusion and Areas for further investigation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "With our general purpose neural network architecture, we have managed to achieve a **test set accuracy of 89.65%**, nearly 90%. While this still isn't good enough for an industrial application, in comparison, a random classifier would achieve 10% accuracy. \n",
    "\n",
    "Recall that when we built the neural network, we had to specify the number of hidden layers (also known as the **network depth**), the number of nodes in each layer (the **layer size**), and the activation functions between layers. We also had to pass several hyperparameters when compiling and training the model, including the batch size. Experimenting with these settings could lead to different results. To see how varying the network depth and layer size affects performance, please refer to [this great tutorial on the original MNIST dataset](https://medium.com/tebs-lab/how-to-classify-mnist-digits-with-different-neural-network-architectures-39c75a0f03e3). \n",
    "\n",
    "In this tutorial, we have used the multilayer perceptron architecture. As a result, we had to reshape our images into single vectors which led to some information loss. However, there are many different architectures, for instance **convolutional neural networks** which are particularly well suited for image classification problems."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Resources\n",
    "This tutorial was written with the help of the following resources:\n",
    "- [Building a simple neural-network with Keras](https://github.com/wxs/keras-mnist-tutorial/blob/master/MNIST%20in%20Keras.ipynb) on *Github*\n",
    "- [Not another MNIST tutorial with TensorFlow](https://www.oreilly.com/learning/not-another-mnist-tutorial-with-tensorflow) on *O'Reilly*\n",
    "- [Train your first neural network: basic classification](https://www.tensorflow.org/tutorials/keras/basic_classification) on *TensorFlow*"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [conda env:tensorflow]",
   "language": "python",
   "name": "conda-env-tensorflow-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.8"
  },
  "toc": {
   "base_numbering": 1,
   "nav_menu": {
    "height": "318px",
    "width": "412px"
   },
   "number_sections": true,
   "sideBar": true,
   "skip_h1_title": false,
   "title_cell": "Table of Contents",
   "title_sidebar": "Contents",
   "toc_cell": false,
   "toc_position": {},
   "toc_section_display": true,
   "toc_window_display": false
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}