convolutional_autoencoder_mnist_with_bn.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Extracting MNIST_data/train-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/train-labels-idx1-ubyte.gz\n",
      "Extracting MNIST_data/t10k-images-idx3-ubyte.gz\n",
      "Extracting MNIST_data/t10k-labels-idx1-ubyte.gz\n"
     ]
    }
   ],
   "source": [
    "import tensorflow as tf\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "# import mnist dataset\n",
    "from tensorflow.examples.tutorials.mnist import input_data\n",
    "mnist = input_data.read_data_sets(\"MNIST_data/\", one_hot=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Parameters\n",
    "learning_rate = 0.001\n",
    "examples_to_show = 10\n",
    "\n",
    "# Small epsilon value for the BN transform\n",
    "epsilon = 1e-3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# Store weights and biases\n",
    "\n",
    "w1_initial = np.random.normal(size=(3, 3, 1, 32)).astype(np.float32)\n",
    "w2_initial = np.random.normal(size=(3, 3, 32, 64)).astype(np.float32)\n",
    "w3_initial = np.random.normal(size=(7*7*64, 1024)).astype(np.float32)\n",
    "w4_initial = np.random.normal(size=(1024, 256)).astype(np.float32)\n",
    "w5_initial = np.random.normal(size=(256, 1024)).astype(np.float32)\n",
    "w6_initial = np.random.normal(size=(1024, 7*7*64)).astype(np.float32)\n",
    "w7_initial = np.random.normal(size=(3, 3, 32, 64)).astype(np.float32)\n",
    "w8_initial = np.random.normal(size=(3, 3, 1, 32)).astype(np.float32)\n",
    "\n",
    "b1_initial = np.random.normal(size=(32)).astype(np.float32)\n",
    "b2_initial = np.random.normal(size=(64)).astype(np.float32)\n",
    "b3_initial = np.random.normal(size=(1024)).astype(np.float32)\n",
    "b4_initial = np.random.normal(size=(256)).astype(np.float32)\n",
    "b5_initial = np.random.normal(size=(1024)).astype(np.float32)\n",
    "b6_initial = np.random.normal(size=(7*7*64)).astype(np.float32)\n",
    "b7_initial = np.random.normal(size=(32)).astype(np.float32)\n",
    "b8_initial = np.random.normal(size=(1)).astype(np.float32)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def activation(_input, activation='relu'):\n",
    "    if activation == 'relu':\n",
    "        return tf.nn.relu(_input)\n",
    "    elif activation == 'sigmoid':\n",
    "        return tf.nn.sigmoid(_input)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def batch_norm_wrapper(inputs, is_training, decay=0.999):\n",
    "    channels = inputs.get_shape().as_list()[-1]\n",
    "    offset = tf.get_variable('bn_offset', \n",
    "                             [channels], dtype=tf.float32,\n",
    "                             initializer=tf.zeros_initializer())\n",
    "    scale = tf.get_variable('bn_scale', \n",
    "                            [channels], \n",
    "                            dtype=tf.float32, \n",
    "                            initializer=tf.random_normal_initializer(1.0, 0.02))\n",
    "    pop_mean = tf.Variable(tf.zeros([channels]), trainable=False)\n",
    "    pop_var = tf.Variable(tf.ones([channels]), trainable=False)\n",
    "    \n",
    "    if is_training:\n",
    "        batch_mean, batch_var = tf.nn.moments(inputs,[0,1,2], keep_dims=False)\n",
    "        train_mean = tf.assign(pop_mean,\n",
    "                               pop_mean * decay + batch_mean * (1 - decay))\n",
    "        train_var = tf.assign(pop_var,\n",
    "                              pop_var * decay + batch_var * (1 - decay))\n",
    "        with tf.control_dependencies([train_mean, train_var]):\n",
    "            return tf.nn.batch_normalization(inputs, batch_mean, batch_var, offset, scale, epsilon)\n",
    "    else:\n",
    "        return tf.nn.batch_normalization(inputs, pop_mean, pop_var, offset, scale, epsilon)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create model\n",
    "\n",
    "def build_graph(is_training):\n",
    "    \n",
    "    # tf input graph\n",
    "    X = tf.placeholder(tf.float32, [None, 784])\n",
    "    batch_size = tf.shape(X)[0]\n",
    "    \n",
    "    reshaped_x = tf.reshape(X, (-1, 28, 28, 1))\n",
    "    print(reshaped_x.shape)\n",
    "\n",
    "    # Convolutional layer 1\n",
    "    with tf.variable_scope('conv_1'):\n",
    "        wec1 = tf.Variable(w1_initial)\n",
    "        bec1 = tf.Variable(b1_initial)\n",
    "        conv1 = tf.add(tf.nn.conv2d(reshaped_x, wec1, strides=[1, 2, 2, 1], padding='SAME'), bec1)\n",
    "        bn1 = batch_norm_wrapper(conv1, is_training=is_training)\n",
    "        conv1 = activation(bn1)\n",
    "        #print(conv1.shape)\n",
    "\n",
    "    with tf.variable_scope('conv_2'):\n",
    "        wec2 = tf.Variable(w2_initial)\n",
    "        bec2 = tf.Variable(b2_initial)\n",
    "        conv2 = tf.add(tf.nn.conv2d(conv1, wec2, strides=[1, 2, 2, 1], padding='SAME'), bec2)\n",
    "        bn2 = batch_norm_wrapper(conv2, is_training=is_training)\n",
    "        conv2 = activation(bn2)\n",
    "        #print(conv2.shape)\n",
    "\n",
    "    with tf.variable_scope('fc_1'):\n",
    "        wef1 = tf.Variable(w3_initial)\n",
    "        bef1 = tf.Variable(b3_initial)\n",
    "        fce1 = tf.reshape(conv2, (-1, 7*7*64))\n",
    "        fce1 = activation(tf.add(tf.matmul(fce1, wef1), bef1))\n",
    "        #print(fce1.shape)\n",
    "\n",
    "    with tf.variable_scope('fc_2'):\n",
    "        wef2 = tf.Variable(w4_initial)\n",
    "        bef2 = tf.Variable(b4_initial)\n",
    "        encode = activation(tf.add(tf.matmul(fce1, wef2), bef2))\n",
    "        #print(encode.shape)\n",
    "        \n",
    "    with tf.variable_scope('fc_3'):\n",
    "        wdf1 = tf.Variable(w5_initial)\n",
    "        bdf1 = tf.Variable(b5_initial)\n",
    "        fcd1 = activation(tf.add(tf.matmul(encode, wdf1), bdf1))\n",
    "        #print(fcd1.shape)\n",
    "\n",
    "    with tf.variable_scope('fc_4'):\n",
    "        wdf2 = tf.Variable(w6_initial)\n",
    "        bdf2 = tf.Variable(b6_initial)\n",
    "        fcd2 = activation(tf.add(tf.matmul(fcd1, wdf2), bdf2))\n",
    "        fcd2 = tf.reshape(fcd2, (-1, 7, 7, 64))\n",
    "        #print(fcd2.shape)\n",
    "\n",
    "    with tf.variable_scope('deconv_1'):\n",
    "        wdd1 = tf.Variable(w7_initial)\n",
    "        bdd1 = tf.Variable(b7_initial)\n",
    "        deconv1 = tf.nn.conv2d_transpose(fcd2, \n",
    "                                         wdd1, \n",
    "                                         output_shape=tf.stack([batch_size, 14, 14, 32]), \n",
    "                                         strides=[1, 2, 2, 1],\n",
    "                                         padding='SAME')\n",
    "        deconv1 = tf.add(deconv1, bdd1)\n",
    "        deconv1 = tf.reshape(deconv1, (batch_size, 14, 14, 32))\n",
    "        bn3 = batch_norm_wrapper(deconv1, is_training=is_training)\n",
    "        deconv1 = activation(bn3)\n",
    "        #print(deconv1.shape)\n",
    "\n",
    "    with tf.variable_scope('deconv_2'):\n",
    "        wdd2 = tf.Variable(w8_initial)\n",
    "        bdd2 = tf.Variable(b8_initial)\n",
    "        deconv2 = tf.nn.conv2d_transpose(deconv1,\n",
    "                                         wdd2,\n",
    "                                         output_shape=tf.stack([batch_size, 28, 28, 1]),\n",
    "                                         strides=[1, 2, 2, 1],\n",
    "                                         padding='SAME')\n",
    "        deconv2 = tf.add(deconv2, bdd2)\n",
    "        deconv2 = tf.reshape(deconv2, (batch_size, 28, 28, 1))\n",
    "        bn4 = batch_norm_wrapper(deconv2, is_training=is_training)\n",
    "        deconv2 = activation(bn4)\n",
    "        #print(deconv2.shape)\n",
    "\n",
    "    decode = tf.reshape(deconv2, (-1, 784))\n",
    "\n",
    "    cost = tf.reduce_mean(tf.square(decode - X))\n",
    "    optimizer = tf.train.AdamOptimizer(learning_rate).minimize(cost)\n",
    "\n",
    "\n",
    "    return X, encode, decode, cost, optimizer, tf.train.Saver()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(?, 28, 28, 1)\n",
      "(?, 14, 14, 32)\n",
      "(?, 7, 7, 64)\n",
      "(?, 1024)\n",
      "(?, 256)\n",
      "(?, 1024)\n",
      "(?, 7, 7, 64)\n",
      "(?, 14, 14, 32)\n",
      "(?, 28, 28, 1)\n",
      "Iter:0, Minibatch-cost:0.460896015167\n",
      "Iter:100, Minibatch-cost:0.0708047151566\n",
      "Iter:200, Minibatch-cost:0.0600918531418\n",
      "Iter:300, Minibatch-cost:0.0511237643659\n",
      "Iter:400, Minibatch-cost:0.0431177765131\n",
      "Iter:500, Minibatch-cost:0.03758790344\n",
      "Iter:600, Minibatch-cost:0.0296867117286\n",
      "Iter:700, Minibatch-cost:0.0251566655934\n",
      "Iter:800, Minibatch-cost:0.0250085238367\n",
      "Iter:900, Minibatch-cost:0.0210404712707\n",
      "Iter:1000, Minibatch-cost:0.0194837357849\n",
      "Iter:1100, Minibatch-cost:0.0163002479821\n",
      "Iter:1200, Minibatch-cost:0.0148286977783\n",
      "Iter:1300, Minibatch-cost:0.014999166131\n",
      "Iter:1400, Minibatch-cost:0.0148470457643\n",
      "Iter:1500, Minibatch-cost:0.0128916036338\n",
      "Iter:1600, Minibatch-cost:0.0139916399494\n",
      "Iter:1700, Minibatch-cost:0.0111674442887\n",
      "Iter:1800, Minibatch-cost:0.0120390700176\n",
      "Iter:1900, Minibatch-cost:0.0116423396394\n",
      "Iter:2000, Minibatch-cost:0.0107246823609\n",
      "Iter:2100, Minibatch-cost:0.0106507809833\n",
      "Iter:2200, Minibatch-cost:0.0103559140116\n",
      "Iter:2300, Minibatch-cost:0.00927521195263\n",
      "Iter:2400, Minibatch-cost:0.01018205937\n",
      "Iter:2500, Minibatch-cost:0.0103256637231\n",
      "Iter:2600, Minibatch-cost:0.00923337973654\n",
      "Iter:2700, Minibatch-cost:0.00875254627317\n",
      "Iter:2800, Minibatch-cost:0.00874373316765\n",
      "Iter:2900, Minibatch-cost:0.0090249394998\n",
      "Iter:3000, Minibatch-cost:0.00904695317149\n",
      "Iter:3100, Minibatch-cost:0.00932120904326\n",
      "Iter:3200, Minibatch-cost:0.00865000393242\n",
      "Iter:3300, Minibatch-cost:0.00837577041239\n",
      "Iter:3400, Minibatch-cost:0.0082769850269\n",
      "Iter:3500, Minibatch-cost:0.00779626984149\n",
      "Iter:3600, Minibatch-cost:0.00751223554835\n",
      "Iter:3700, Minibatch-cost:0.0075622969307\n",
      "Iter:3800, Minibatch-cost:0.00765591440722\n",
      "Iter:3900, Minibatch-cost:0.00799025315791\n",
      "Iter:4000, Minibatch-cost:0.00731394765899\n",
      "Iter:4100, Minibatch-cost:0.00685548968613\n",
      "Iter:4200, Minibatch-cost:0.00688303168863\n",
      "Iter:4300, Minibatch-cost:0.00715461792424\n",
      "Iter:4400, Minibatch-cost:0.00663132034242\n",
      "Iter:4500, Minibatch-cost:0.00684204511344\n",
      "Iter:4600, Minibatch-cost:0.00665157148615\n",
      "Iter:4700, Minibatch-cost:0.00634515704587\n",
      "Iter:4800, Minibatch-cost:0.00654452573508\n",
      "Iter:4900, Minibatch-cost:0.00675962679088\n",
      "Optimization Done!\n"
     ]
    }
   ],
   "source": [
    "#Build training graph, train and save the trained model\n",
    "tf.reset_default_graph()\n",
    "X, _, _, cost, optimizer, saver = build_graph(is_training=True)\n",
    "\n",
    "with tf.Session() as sess:\n",
    "    sess.run(tf.global_variables_initializer())\n",
    "    for i in range(5000):\n",
    "        batch = mnist.train.next_batch(60)\n",
    "        optimizer.run(feed_dict={X: batch[0]})\n",
    "        if i % 100 is 0:\n",
    "            c = sess.run([cost],feed_dict={X: batch[0]})\n",
    "            print('Iter:{0}, Minibatch-cost:{1}'.format(i, c[0]))\n",
    "    saved_model = saver.save(sess, './save_cae/temp-bn-save')\n",
    "\n",
    "print(\"Optimization Done!\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(?, 28, 28, 1)\n",
      "(?, 14, 14, 32)\n",
      "(?, 7, 7, 64)\n",
      "(?, 1024)\n",
      "(?, 256)\n",
      "(?, 1024)\n",
      "(?, 7, 7, 64)\n",
      "(?, 14, 14, 32)\n",
      "(?, 28, 28, 1)\n",
      "INFO:tensorflow:Restoring parameters from ./save_cae/temp-bn-save\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAlMAAACNCAYAAACT6v+eAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzsvWlsXGd6oPuc2jduxaXIqiJZ3Is7qc3aLIryJrnltlvd\n6vSWpJEOkiBIcDO4F3Mz+TXAAIMezORiLmYubrqRHiS34XS33bJlSbZsWbIkarEkipRIcd/3fd9r\nPfcHXadFS7K4V7F0HqAg8qjq1PfyfN857/eugiiKyMjIyMjIyMjIrA9FsAcgIyMjIyMjI7OTkZUp\nGRkZGRkZGZkNICtTMjIyMjIyMjIbQFamZGRkZGRkZGQ2gKxMycjIyMjIyMhsAFmZkpGRkZGRkZHZ\nALIyJSMjIyMjIyOzATakTAmCcFwQhGZBENoEQfj7zRpUKCHLuPMJd/lAljFcCHcZw10+kGV8YRFF\ncV0vQAm0A+mABqgB8tZ7vlB8yTLu/Fe4yyfLGPyxyTLK8skyhpeM63mpNqCH7QPaRFHsABAE4bfA\n20DDsz4gCMJOLbfe/tW//8ALLmMYyDfPC34NQZYxxJHX4leEgYwv/DyFHS0jAKIoCs97z0bcfDag\n97Hf+746tgJBEP5CEIT7giDc38B3hQovpIxhJt8IL+A1BFnGHYi8Fnc+L+Q8hbCT8blsxDK1KkRR\n/CXwS9j52umzCHcZw10+kGUMF8JdxnCXD2QZw4UXQcbH2Ygy1Q8kP/a7/atj4cy2yKhWq1Gr1QAo\nFApUKhUajUY65na7WVpawuv1Sq+vfNmbQbhfRw3hLR+E/zUEWcZwQF6L4cGLIONz2YgyVQlkCYKQ\nxvIf8gfAjzZlVKHLtshYVFRESkoKAJGRkdjtdgoKCrDb7SgUCtra2rhz5w59fX0MDAzQ1NTE/Pz8\nZn19uF9HM3Au2IPYYsL9GoIsYzggr8Xw4EWQ8bmsW5kSRdErCMLfAJ+xHN3/v0RRrN+0kYUm722l\njHq9nqKiIk6ePInT6cTv9xMREUFycjIFBQXS+/bu3YvD4aCrq4tHjx4xOTlJR0fHZg1jS2VcK1FR\nUeTm5qLT6ejr62N4eJjZ2dmNnHIiGPIpFApiY2OxWCzExcWhVCqZnp6WZNpEyyKE2DXcImQZdz5B\nWYvbTEhfQ0EQUCgUiKKIIAj4fD4AtFot8fHxeL1e5ubmcLlc3+QBCWkZtwthk2/i3/xlO9xvupqI\n/o3I+MYbb/Cnf/qnHD58mMTERDweD4IgoFarUalW6r0ul4uFhQWqqqr4r//1v3L58mX8fv96v1ri\neTJu5zVUKBTs27eP//yf/zNWq5V/+Zd/4cKFC9TV1W3ktFWiKO75pjdshYwmk4nXXnuNb3/727z8\n8ssYDAYePXrEr3/9a86dO8fMzMymfddWz9NQQJZx58tHkNbidhLK8zQyMpKYmBhUKhU+nw+lUsn8\n/Dyjo6Pk5uZSXl7O3NwcnZ2djI6OMjIywvj4+BPPmVCWcbNYjYxbHoAu83wiIiLYs2cPp0+f5tix\nY1gsFgApRkoURXp6epiZmSEiIoL4+HgMBgNarRan08nu3bvp6emhtbVV2lmEA5GRkaSmplJQUIDZ\nbCYqKuoJpXKnoNFo2LdvH2+++SYJCQnAstWtpaWFpqYm6uvrWVxcDPIo14ZGo5HmX3t7OxUVFav+\nbEREBBEREczOzjI3N7fZlrltR61Wk5qaSllZGWq1mjt37tDZ2cn09HSwh/ZMVCoVBw8exGKxMDo6\nysDAAMPDw8zNza37PqJSqTCbzSwsLDA3N7fJI5bZCLGxsSQmJpKUlITVasVqtZKQkIBarcbv96NQ\nKJidnaW3t5eUlBReeuklfD4fY2NjuFwu7ty5w7vvvrtRz0DYsjOfTGFGUVERf/VXf0VZWZmkSPn9\nfrxeL5OTk/T09FBTU8PIyAgJCQkUFRVRUlKCRqMhOjqa4uJimpqa6O7u3nEP5G8iLi4Oq9WKy+Wi\np6eHzs5OxsfHgz2sNaNSqXA4HBQXF0uKFIDBYCAjI4OsrCy6urp21LUTBIHY2Fhee+01/vZv/5b3\n3ntvTcqUxWIhJyeHsbExenp6GB0dxev1buGItxabzcbx48f5h3/4B4xGI//0T//E+++/z/37oZkV\nrlAoKCgo4Gc/+xlOp5OHDx9y9epVbt68ua74S0EQiIiIICsri/T0dJqbm6mrq9sUa/lmIggCSqVS\ncm1tcvJOyCAIAoLwB2OK3W6ntLSUoqIiCgoKyM7OJiUlBbPZvOJzHo+HoaEhBEEgISEBjUYj/V9y\ncjJnz56VlalnICtTQSYiIoLCwkIOHDggKVIANTU1fPjhh3R0dDA3N8fg4CATExMoFAqOHz9OREQE\nubm5aDQaYmNjiY6ORqEIr1aLJpMJi8WCyWSivb2d6upqent7n//BEGP37t386Ec/Ii8v74n/EwQB\nURRxuVxBGNn6CVhisrKyiImJWXHTXQ0Wi4X9+/ej0+moq6vjwoULO1JRhmVluaysjD/6oz8iKSkJ\ngJ/+9KeMj4/z4MGDkLQWf/vb3+bP/uzPOHLkCBqNBpfLxb179xgfH1+XUmuxWPiTP/kTXn/9dSIj\nI3n33Xfp6+tjYmJiC0a/flJTUzl8+DAOh4Ph4WHOnTvH8PBwsIe1qQSsg+np6eTn50sJTWazmbi4\nOMxmM9HR0eh0uic+q1arSU5OfspZlzd/jytoMisJSWVKrVaj1+uJjIxEoVDg8XgApGC4paWlsNhN\nqFQqrFYrmZmZ0gReWlqiqamJs2fP8qtf/YqBgYEnPldYWMjCwgKw/DeZnp5mdnY25HaBG0GpVJKV\nlYXT6cTj8dDR0UF7e/vzPxgi6HQ69Ho9FouF48eP89Zbb2G1WhFFEb/fj1KpBJYtBEajkYSEBBYX\nFxEEYUfsllUqFVlZWaSlpeH3+1laWlrT55VKJWazmYKCAqKiorhx48aOVabMZjN79uzh8OHD0rGE\nhAQsFkvIXUe1Wk1WVhanT5/mrbfeAqC9vZ3W1tZ1W7bVajWlpaX8+Mc/pqioiLGxsZB68KrVakwm\nE0lJSbzyyiu8+eabFBQUUFtby40bN8JOmdJoNCQlJVFeXs7Jkyd56aWXpPvN8/D5fHi9Xvx+v3Sv\n0mg0aDQaxsbGQvYZIwgCWq0Wo9GIwWBAo9EgCAJLS0vMzMywsLCw5ZbvkFOmBEHAZrNRWlrK/v37\nMRgMjI2NIYoiU1NTVFVV0dnZydLSEpOTkyF3s1oLSqWSiIgIoqOjpWNtbW384he/4MKFC08oUlqt\nloMHD/Lqq6/icDiA5UD0hoYGmpqaJKUzHLDZbBw5coSDBw8yMDBAdXX1jpFPo9FQXFxMaWkp+fn5\nHDhwAJvNhlqtxufzrZizsbGx5Ofn4/f76enpYXh4mP7+fiYnJ4MowfPRaDRkZmaSlZWFTqdb8012\nZGSEtrY28vPzSUhIwGAwbNFItxaDwUBmZiY22x8KQLtcLhobG2lqagqph48gCOzbt4+f/OQnHD16\nFIC6ujrOnj3Lb3/7W5qbm9d13n379nHy5EmysrIYGxvj17/+NVevXg0J5VihUJCens6RI0c4ceIE\nRUVFkrV7bGzsqdaZnY5KpSIuLo6DBw9y8ODBVX/O7XYzODjIyMgIi4uLkqdDpVJhNBppa2sLuXuw\nWq0mMjISnU5HcnIyxcXFOJ1OkpKSUKlUdHZ2cvPmTZqbmxkYGNjURJ+vE3LKlFKp5MCBA3z/+99n\n165dGAwGSWmanp5m165d9PT0MDs7S19fHyMjI/j9flQqldRw8Gk7osCxQCzS9PQ0/f39TE1NBW2C\n+P1+JiYmuHfvHlFRUSiVSqqqqvj888/p6+t74v0mk4nDhw+zf/9+SQETBIG5uTlmZmZC6sa9EbRa\nLXFxcWRmZhIfH8+1a9eoqanZMTE1oihSVlbGO++8Q1JSEnFxcZIb7OvKv81m4+DBg9KDqK+vjy++\n+IKKioqQdA8FsFgsFBQUEB8fT19fH4ODg2v6/NLSEnNzc5hMJnQ6HSaTaYtGurVYrVbKysrIyMiQ\njo2OjnLz5k0aGxuDOLIniYqKorCwkNdeew2r1YrX66WiooKPPvqI+vr1ZbYbDAaOHDnCK6+8gtfr\n5fbt25w7d45Hjx5t8ujXR0REBEePHuX06dMcOHBghdIeGxvLgQMHGB4eXvP8DRXi4+OJjY1laWmJ\niYkJZmZmEEURnU5HYmLiE+8PGCECCQIzMzPMzc0xOzvLxMQEIyMjjIyMsLS0tMJ6rtFoaG9vD5m4\nzvj4eFJSUsjKysJms2EymbDb7WRnZ5Oenk5cXBx+v5/+/n7S0tLo7OxkYGCAwcFBxsfHJf3hac/Z\n9RKSytSePXt45ZVXiIiIAJYDkQPs3r0br9fL4uIig4ODDA0NIYriquKFFAoFfr+fxcVFmpqa+PTT\nT3nw4AEjIyNbJs83EXBfvffee3zxxRf4/X5mZmaemQEUERFBfn4+aWlpK5RDhUIRVvFSRqMRq9VK\nVFQUi4uLNDQ00NjYiNvtDvbQVkVMTAzHjh3jwIEDwLIr9uuKkd/vRxAE6WYYqPEyMjKCTqdjcHCQ\njo6OkJQ5Li6O0tJSKQbs5s2bNDQ8s8fpU9HpdFLdLZfLhV6v34qhbjlZWVmcOHGCzMxM6djU1BTV\n1dV0dXUFb2BfQ61WY7PZyMjIwG63AzAxMcH169fXHSQvCAJZWVkcPXqUnJwcHj58yLVr13j06NFm\nFhHeEA6HgzfffJNjx449scmOjo7mO9/5Dl6vl9///vchbw3+OpGRkRw5coSCggK6urqoqqqioaEB\nt9vNwsICo6Oj0jMFli1M7e3tNDY2Mjg4SG9vL11dXXR2djI8PIzL5cLj8eByufD5fE/8vURRDPoG\nT6lUYrfbeemllygvL2fv3r2kp6djNBqlTO/Hn4WZmZmkp6cjiiJLS0uMjY0xMDBAR0cHn3zyCR98\n8MGmxauGnDLl8/mora3l4sWLxMfH43K5mJycxGAwEB8fT0JCAomJiURFRREVFYXT6ZRiqJ51Q/b7\n/bjdbsly5ff7sdlsLC4uMjc3F3Rf8NTUFFNTU9/4HrPZzIEDB8jOziYqKko6PjQ0RFdXl2ShCwfs\ndjvHjh3DZrMxNDREc3Mz/f39O0K+7OxsTp06RX5+vnRMpVJJMQhKpXLFTSpQRyzwvuTkZL73ve8R\nFxfHf//v/52ampptl+GbMJlMnDhxgp/85Cekp6fT2NjIuXPnePDgwZrOExcXR15eHvHx8fT378xO\nFDqdDofDQW5u7gqLx+zsLA8ePNjUXe9GMRgMOJ1OnE4nWq2WiYkJPvjgA2pra9d1PqPRyOHDhzl9\n+jRHjhwBoLGxkYaGhpBRShwOBy+//DI5OTlP9Vbo9XpKS0sxGo3k5+dTWVlJTU0NnZ2dIaMMPguH\nw8H3vvc9vv3tbxMTE8PZs2epq6tDoVDgcrloa2vjN7/5DXfu3GF0dBSVSkVUVBT9/f10dXUxOzvL\nzMwMk5OTTE5OhrzVPyYmhuzsbAoLC6WMxKysLGljECDgrRIEAafTiV6vx+/3S3Fzer0evV5PXFwc\nU1NTtLe3U1dXJ8Ugb4SQVKauXr1KT08PdrtdymSLiooiJSUFq9VKeno62dnZ2O12TCaTZJZMTEyU\n/ngB/H4/s7OzTE1NYTabpT9+Wloa2dnZVFZWhkyg5LMwGAyUlZVx8uTJFZOnr6+PqqoqmpqaNmUy\nhAqpqamUl5eTlJREdXU1g4ODIeerfxbHjh3jRz/60YoSCAErVGDH5Ha7pTk5OzuLQqHAYrEQExOD\nWq3G4XDw05/+lIqKipBSpnQ6HU6nk7feeovXXnuNiYkJrl27xq1bt9actWW32ykpKUGr1eLxeHZc\n7KNKpSI1NZW0tLQn0svHx8dDKus0kDF88OBBcnJy8Hg8XLp0id/97nfrUvgEQSAxMZHXXnuNn/zk\nJ2i1Wjo6OqisrKS9vT1kNj2lpaW8+uqrK9aiz+eTwkICsUVxcXHs37+fW7du8eGHH/Lhhx+GtDKl\n1+spLy+XylqMjo7idrtX1Gvr7e3l7NmzCIIgWaaio6NZWFgISWv3szCZTCQmJpKXl8eRI0d4+eWX\nKS4uRqvVAsvWstnZWcnaFHgF4qji4uKoqanBYrGQl5cnJfmYzWZ2797NxMQEbrebhw8fbnisIadM\niaLI4OAgk5OT1NfX4/V6WVpaQq1WU1NTg06nIzIykvj4eMk3GthJZGZmEhUVJWnZASvUwMAA/f39\nnDx5kr/8y7+Ugg4D/uJQvpEbjUaKi4t5++23eeONN6Qb99LSEnfu3OGjjz4KqRv3ZmCz2UhPT0et\nVuPxeELGT/88kpOTSU1NJS4uTjI5Bx4sAYXd4/FQU1NDXV0dHR0djI+Po9FoSE9PZ9euXezdu1e6\nURQWFuJ0OmlqagqOQF/DZrPxrW99i7179+Jyufj973/P+++/v65AY4vFgtPpBNiRLmqDwUBeXh4Z\nGRkrxj8xMUFXV1fI7PSVSiWHDx/m1KlTlJeXY7fb6e/v5969e1RXV6+rsGbA1eJwONBqtczOznLj\nxg2uXr1KW1vbFkixdqKioigpKeHQoUPExMQAMD09zcTEBF6vl+Tk5BXB5wqFgt27dzM3N8ft27dD\nykX7OJGRkXz3u9/lxz/+MRkZGQwNDfHZZ59x8+ZNenp6Vrjhvl4P6nnej1DDarVK9xun00lqaio2\nmw2lUokoinR1dTExMUFdXR0ff/wxzc3NuFwuRkZGSE1NlSzG7777LkVFRcTExGA2m9HpdMzNzZGZ\nmUlSUhIjIyPU1dVteM2GnDIFyw8cj8fzxGT4eixRVVUV0dHRDA4O4nK5sFqtmEwmvF6vVL/H6/Uy\nMTGBy+WiuLhYUpxGR0epr6+nu7s7pJUpk8lETk4ORUVFxMbGAkiZX7dv3+bmzZshkTWzGQiCgNVq\nxWazSfFyPp8v6H761ZCTk8Orr75KaWnpirTwwIPW7/dLWV4ff/wxd+/epa+vj+npaTQaDSkpKXi9\nXvLz89Fqtbjdbvbu3cvo6Cjnzp2jpaUlqNY5pVJJUVER5eXlpKSk0Nvby9WrV/nyyy/XZIkQBIHU\n1FSys7Ml9+bk5OSaSysEm0D3gdTUVCnpZXFxkbt373L//v2QUaYUCgUOh4PS0lIpSL6vr4/W1tZ1\nZTYZjUYKCwspLy+X+oVeu3aNM2fOhEzAvVKpJCkpSdpsA7S2tkoKh1arlTJto6Oj0ev1KJVKDAYD\nJSUlvPrqq8zOztLR0RFS81Kn07F3715OnTolxYBVVlby/vvv8/Dhwx2z6VwNJpOJQ4cOcfr0afbs\n2SMpxLBs+a2urqayspLR0VGam5u5cuXKCoubx+Nhbm6OhYUFKisr6e3tJTExkVdffRWbzYbb7SYu\nLk4KZFer1eGpTK2WqampFVlsg4ODK5o2BuJU9Ho9xcXF5OfnS3FVLS0t3L17l9bW1pBWpqKjo0lO\nTl4RDzY6Osq9e/d4+PBhWNVI0ev1ZGRkYLPZVrQ3CBW3wbMwGo0cOHCAv/zLvyQrK+uJljcej4ex\nsTGam5u5evUqv//972lpaQGWg9MVCgWjo6MkJyczPj5OTEwMSqWSl156icjISFQqFRcvXqSmpiYo\nCpVWqyU3N5cDBw6Qm5uLQqGgv7+f3t7eNV+bpKQkTp48SWlpKQBzc3O0tLSEXHHH5xGIl0pKSpIU\n55mZGa5cucLVq1dD5sEmiqJk3fV6vZKLeT2uHr1ez9GjR/nWt75FeXm55DK8cOEC58+f34LRrw+T\nyURycjLx8fHA8ib88uXL/OM//iPt7e1YrVZ27dpFeXk5x44dIzMzE6PRKFX9/uM//mNMJhO/+MUv\nQsZCZTAYOHDgAO+88w579+5FEAQ8Hg/37t3j2rVrIe2WXCtms5mysjK+853vcODAgRWZvt3d3Vy/\nfp0LFy5w48YNZmZm8Pl8T8xnvV5PREQEer2e6Oho2tvbOX/+PJmZmaSkpKxIagvEU2+UHa1MAStu\n5s/KNrBarXz3u9+lrKwMgAcPHvDFF1/Q3t4esopUREQEu3bt4ujRo5SVlUmVlScmJrhx44YUcBhO\nBG5miYmJKJVKWlpauHfvXsg/aNVqNQkJCWRnZ0suugAzMzPcv3+fq1ev0traSnNzM62trSt2QYGM\nm4aGBm7duoVOp8Nut6NUKsnLy+O73/0uHo+H3t7eoCjPBoOB4uJiSkpKMJvN9PX1UV1dva4sWLvd\nTllZGYWFhfj9fiorK6moqAj5a/x1UlNTycnJwWq1Ssc8Hg+tra309PQEcWQrCbRPUSqV+P1+9Ho9\nDoeDV199FZ1Oh8/nY3JyUkp7D2wEAsk6JpOJhIQEYmJiyMjIoLy8nP3790t17gIxKqGEUqlEpVKh\n0+kQRZEvv/ySzz//XCr6OzAwwPT0NG63W+pPF7AmB6xaiYmJIdMHVK/X43Q6OXnyJG+++SZxcXF0\ndXVx/vx5Ll26FFaKlMlkoqioiLfffpuysjJJkRofH+f27dtUVFRw584dqqurvzFO2OPx8PDhQ/Ly\n8njnnXeYn5/HYrFICnaA3t5eBgYGNiWOLDRmyxaiUCgoKSnh1KlTpKen4/V6+fLLL/niiy9C+gae\nmprKn/3Zn0mB2CqVipmZGW7fvs358+f57LPPwq6RaGBHmZiYiCiKUs2tYJWuWAuB+DxghVWttraW\n3/3ud7z77rvPvel1dXVx8eJFzGYzVqsVhUKBSqWiuLiY7u5uzp49GxRlKpDtlJOTg0ql4tGjR1y7\ndm3N68doNJKVlUVhYSEmk4m5uTm+/PJLrl+/vqPiOQLxXmlpaSuOT05OhpzLPZASPjc3h9vtRqPR\nkJubK1lTvV6vZDGdnJyU4oj8fj8+n4/ExERyc3PJzs4mIyOD9PR0KXOxu7ubmzdvhty1CxSZ1Ol0\nLC4uUldXR2trKyqVStrEzM/PU19fz9jY2IoHqdfrZWRkhOHh4ZCxLiYkJLBnzx7Ky8tJT09nYWGB\nCxcu8F/+y3/ZsZmwTyOwmQ4kSwQ2KqIo0t/fz0cffcTHH38sZa4HwgQC74Hlja3RaMRms9HV1YXD\n4eBHP/oRVqsVQRBWtL0aHBzk2rVrm1ZcN+yVqcLCQvbu3SsVMBMEgfHxcUZGRkI6qyE6Ohqn07mi\nT9LIyAiff/45FRUVYadIBW7yR44cweFwMDQ0xMOHD6mvrw/5TMVAU9HHs0K9Xi91dXX8+te/XvXu\ncWpqigcPHpCbm0thYSF2u13aHQdaIwSDqKgo0tLSSElJAaC+vp4rV66s6SGq0WikrgaBHpSBwM+A\ny3MnIAgCFosFu92+wvXe19fH3bt3Q06Z8vv91NTUkJaWRn5+vrTTD2RGLy0tkZaWRlFRET6fb4Uy\nJYoikZGRJCQkoNVqUavVkuV1aWmJu3fvcubMmZCzTFmtVl5++WWSk5NRKBTY7XZSU1OfqPDucrmw\nWCxYLBYptjFQ+6+1tfWZ9f62m5iYGNLS0qSY2e7ubhoaGsJKkQqg0WgwGAwr1lYghlqj0ZCamip1\nTYiPj5fazXk8Hqnkgd1ux2w2IwgCSUlJ2O32JzwGAJ2dnZw/f37NNfKeRVgrU/Hx8Rw9epQjR45I\nAc1VVVW0tbVJlWJDEavVSmlp6YoKth6Ph7a2Nu7evRsyfvzNIhB4XlhYSHFxMTExMdy/f1+qhxLq\nBIpyBuaTy+WiqqpK2kmt9qbn9XqlYnojIyPExcVJDz+9Xk9mZqZU5Xg7iYqKkkpyzMzM0NzcvGZr\nRKAnXKD2iyiKDAwMMDQ0tBVD3lLi4+OxWq0r+p3V1tbyxRdfhKQy1dLSwtWrV0lPT5fqKhmNRkwm\nk/RKTU0F/hA/olAoVnRXeFyR8vv9DA8P8/DhQ27fvh0ySkcArVZLdHQ0arUajUaD0+kkNzeXiooK\nPB4POp0Oj8dDYWEhmZmZKx7cCoUCpVIZUoWQExIScDqdmEwmfD4fHo9HymgfHR0N9vA2jUCZg5GR\nEUZHR6V7jlqtJiYmhv3795ORkUFSUhJJSUmYzWYUCgVutxuXy4XBYMBisUghMQEl63FFKvAd/f39\nXLlyhdu3b2+atT9slSm9Xk9+fj7Hjx+XKlFXVFTwT//0T9y4cSPkTNMBDAYDP/zhD/mjP/qjFfVR\nmpubuXfvXtiVQYBlZSrQCsBsNuN2uxkdHd12pWG9BGrXPB6I/Itf/IL3339/zdV1Axaur9/Ms7Oz\n+cEPfsDi4iK3bt3a1PE/j8COD+D27dvriglSKBRSoV2lUsn09DSDg4Mh40pZLaIoEh8fT1JSkuQy\nWFpaora2llu3boWcMgWwsLBAVVWV9IBKTk6WFKtXX31Vcpe0tLRQU1MjWagUCgU1NTXU1tbywx/+\nkFOnTknn6+npobe3N+QUKVh239y+fRun00l8fDw2mw273U5sbCxqtVoq+Hj48GHJ2hpIWtJqtezd\nu5fBwUG+/PLLkIhLtdls7N69W2ohlpWVxSuvvMLExATV1dV0d3czNjYW5FFuDv39/dTW1lJfX09y\ncjJxcXEIgoDD4SA2Nhav14tWq0Wr1Uot5AIb2UDMWwC1Wv1E3Nvg4CA3b97kzJkzfPnll5u6mQtb\nZSrQOyw7OxtY3k1VV1dz5cqVkI7BcTgcHDp0iD179kgP54GBAb744guuXLkSkjfrzSCQomoymRgf\nH6e5uXlH7LoCNaLsdrt0vVwuF01NTWtWpBQKBUlJSaSkpJCQkLCiDs7S0hIjIyNBCTZ1uVzMzc1J\ncTfriS9QqVSkpaWRkZGBSqViZGSE9vb2HWF5fByr1SpVYA7EDnk8HgYHB0Mq8PzrzM3NSQ3RExIS\nsNlskos1MTGR6elpmpqaaGtrkx5YgiDQ2NhIf38/hw8fls7V09PD559/vu7q6VvN8PAwlZWVvPHG\nG8DyvWX37t389Kc/xeVykZ6ejtPplOpkPXr0iPr6erKzsykpKSEyMpKioiL279/P9PR00DewsbGx\nK8I99Ho9RUVFwHJ7tdbWVqqrq+ns7JSSCHQ6HUtLS8zPzzM3NyeVC4qIiECtVjM9PR0y5Tu+TkdH\nB+fOnaO9jnGSAAAgAElEQVSrq4vk5GQpVs9isaBUKqUCpYHOJwF59Xr9E9bEx0MvOjo6uHr1Kh9/\n/DFXrlzZ9M162CpTKSkp5OXlSe69R48e0djYGNIPaJvNxt69e3E4HNIkmJqaoq6ujk8//ZSKioog\nj3BrUCgUJCQkkJSUhFKppL+/n/v37zMwMBDsoT2XyMhIqSpv4JoJgrCuxr1ms5nS0lJKS0ux2+0r\nbgzNzc389re/DcpOeWpqio6ODqmYalpaGkqlck31v6xWK8XFxVLR2ZGRETo7O3eM9RH+UPMtsEYD\nBBrF7gT8fj9DQ0OSm+7MmTNSTb5AnFTAXR342Wq1SlaRQC2tM2fObFqsyWbjcrlob2+XaggKgsDB\ngwfZt2+f5MJUq9W4XC5aW1v5zW9+w0cffcTp06ex2+0kJCRgtVopLy+nr68v6MrU4wrB4z09y8vL\nee211xgfH+f69evcuXOH3t5e9Ho9ZrOZiYkJ+vv76e/vZ25uDrVaTXJyMkajkY6ODnp6ekIy9nZy\ncpIzZ87w4YcfEh0dLcmZk5ODXq9namqKsbExySoaGRmJzWYjLS2N5ORkaT5/XZH69NNPee+997hz\n586m9eN7nLBVpqxWq9TVfmhoiIsXL1JVVRWycVIARUVFvP7669hsNunY0tISg4ODYVVP6uvodDpS\nU1NJSUnB5/PR3NzM3bt3d0Qnd5PJRGlpKTk5OSuUn7W2KFIqlTidTr71rW9RUlLyxA7L7XavqKm2\nnSwuLjI5OYkoiqSnp3Pq1CkEQeDevXuMjo6iVqul8fp8PrRaLWazmejoaCIiItBqteTk5EhxOQG+\nHrQf6gQyhYxGo3RsYWGBrq6ukOlHt1oC7pHnKcQRERGUlZWRl5eHz+fjypUrfPLJJ3R2dm7TSNdH\noLaUxWLhxIkTREdHr8jkCmSSnjt3jgsXLtDV1UVTUxMtLS1Si5mSkhJu374tdWIIFj09PdTW1lJU\nVCSts8fbU8XGxnLo0CFSU1OZnp5GrVaj1+tZXFxkenqamZkZ3G43SqWS6OhoVCoVQ0ND9PT00NnZ\nyZdffhlySSB+vx+/38/Y2BjXr1+nv7+fuLg41Go1S0tLLCwssLi4yOzsLFarlTfffJOYmBjJQxC4\nr8zNzXHv3j0uX77M1atXqaur2xJFCsJUmYqMjCQ7O1vq5N7W1sbly5dDdicFy8UM9+/fz+HDh6Ws\nDVheSI2NjTtqB78WBEEgJSWFzMxM4uPjGR8fp6WlRaoJE+poNBpsNhsWi2VdikGg91RiYiKvv/46\nR44cwW634/P5UCqVeL1exsfHg5q5Mz8/T3t7O729vRQUFFBWVia1c2pvb0ej0Ug3dq/Xi8FgkOr3\nxMbGEh0dTWxs7IpCeS6XS7rJ7xQMBgNms3lFHMbIyAhVVVU7QvFfD7t27eI73/kOBQUFzM7OcuvW\nLe7duxdSlcGfxc2bN/H5fERERHD06FEpI3Fubk6q2n727Fkpfrazs5Pbt29L/V+Tk5PJy8sjOzs7\nqB0IGhsbOXPmDIuLixQWFq5oqh0gkJW4Wvx+PxMTE7S2tqLVaunu7t4yJWOjjIyMfGNozuzsLMeO\nHXsiPmpxcZHKykr+7d/+jcuXL9Pd3b2l4ww7ZSomJoajR49y8OBBEhMTpWKITU1NIRvsmpSUxLFj\nx9i/f79kpnS73XR1dXHp0iXOnj275RMhWKSkpHDo0CGys7MRBIHu7m4GBgZC2oK4mTgcDt566y1e\nfvll8vPzV2Rwut1u2traOH/+PBcvXgyadXJqaoorV66gVqsRRZH8/Hx2796Nw+GQGjUHFMmAeV0Q\nBCkDTKfTrbAKwHJNrZs3b+6obL6kpCRKS0tXFP7r7Ozkxo0bYbc+VSoVDodDai6r1+ulXqaBWJVQ\nZ3p6msrKSv71X/+Vrq4uCgoKiIqKoq+vj9///vdcunRpRSJSbW0tCoWC+Ph4IiIiiI2NZf/+/XR3\nd7O0tBS0DV5rayvvvvsulZWVlJeXc/r0aRwOBy6X66kp/6tBoVBI2cLDw8OMjo5y8+bNkA6DeRp2\nu50TJ05QXl5ORkbGirjVS5cu8eGHH3LlypV1NfReK2GnTBUXF3Pq1Cn27dsHQHV1NXfu3AnpbIfU\n1FROnjxJSUmJNBkCrsmLFy8+UR8lnIiLiyM7O5v4+Hh8Pp+U+r/WmJxgEShu6PP5VrjmVjN2p9PJ\na6+9xquvvsrhw4clhSMQq+J2u+nu7ubTTz/l1q1bQQsYXVpaoq6uDr/fz+zsLC+99BJZWVlS+vnj\nBUuVSiXz8/P09PQwOTkpPZTT0tIwmUzS32hsbCzosShrJSMjgwMHDkgKr9/vp7e3l+rq6h2lFK4G\ng8FAfn7+ihIt/f39DA8P76iK2+Pj41y7do3BwUEKCgok5eHGjRtPXLNAVuaXX34pZfsVFBQwMDDA\n7du3g6ZMLS4u0t7eTkdHB4ODg/h8PlJTU3G73SQlJUlZ3waDgYSEBEwm04qstm+ymOt0OgoKCjh0\n6BCtra07SplKTEzk5MmTfPe736WoqGiFYjk1NcXVq1f57LPPtm1tho0ypVKpyMjI4JVXXpH8p4OD\ng1y6dIkbN24E1ef9PJKTkzl27NiKUgh9fX28//773LlzJ4gj23p0Oh1RUVGo1Wp8Ph8ulytks0zW\nwvNcfsXFxfzt3/4tBQUFJCQkPFE9HZZvCH19fXR1dQX9b7KwsEB1dTXNzc1cunRJyq7R6/Ur4rg0\nGo3UO7K7u5uIiAjefvttfvCDH7Bnzx4pkHmnYTAYyM7OZs+ePdKDanp6mqGhoW3Z9W43er2e7Oxs\nKc5tdHSUBw8e0NzcvKOUKVhWqCorK6mpqUGhUOD1ep/ppvR4PDQ2NlJXV8eePXvQ6XRkZGSssBgH\nC1EUqauro6urS4qZysrKIi0tDa/XS2ZmJm+++abUwDkQiP08jEYjSUlJK2IBQ51AgsA777zDoUOH\nVtQK8/v9dHV1UVtbu62bnOcqU4IgJAP/H2ABROCXoij+34IgmIHfAQ6gC/i+KIpBi8KMiIjg+PHj\nHD9+XMoYCqSMtrW1bcp3CIIQsxUy6nS6FV2xYXk3sp5Gsk8jLi4Ot9stPZAtFgspKSm43W4ePHgQ\n1PgHs9mM0+kkLi5O8uH39vZupRtB+fy3rOFkSiVarXaFG0upVErX02QyUVBQQElJCUajEZ/Ph9fr\npaSkhDfeeENSLgK90wJyq1QqvvjiC37zm9+sOatxq+ZpwDJVX19Pb28vkZGRaDSaFXNUpVIxOzsr\nuSRdLhfNzc309vZKN3lYjmuMiYlZd+D2Vsn4NFQqFSkpKVK/REByw/f29u6I+KG1EkiIyMjIwO/3\n8/DhQ65du7bZsXubuhafRcDKu5r4PI/HQ319Pbdv32bfvn3s2rWLpKQkioqKePjw4Zr7uW72PPV6\nvStck9PT07S3t+N2u+nv78dqtRIfHy/df+bn5+nq6qKtrY3h4WEsFgs5OTkkJydLGccqlQq9Xr/u\nXoTbuRbhD02fv//977N3794VihQsZ+5/9NFHm/bcXy2r+et5gf9dFMVqQRAigCpBED4HfgpcEUXx\n54Ig/D3w98D/uXVD/Wby8/N5++232bNnD7BcgK6ysnKzTbNbImNgp/R4ryGTyYTT6ZSyMAIpvYC0\n41hYWGBhYUFSklQqFQaDQQpQFEURnU5HXl4eRUVF5OXlMTk5yX/6T/+JEydOUFFRgdPppLW1NWi7\nTZvNRn5+PoIgMD8/T3NzMz09PVuZtbapW0yfz8fi4iIul0uqC6XVaiksLGRsbAyDwcBbb73Fj3/8\n4ycU5gCP35wD13h2dpYLFy5w9erV9Qxry9fizMzMmpIiZmdnV1jXjEYj8fHxUtf3dbBt9xuj0YjD\n4ZA2abAsT3V1NS0tLSgUiqBkWW4lZrOZzMxMoqKicLvd1NTUcPPmzc3OWgy+uecpTE5OcvfuXa5d\nuya1Jzly5Ai9vb3r6b+4pfPU6/VKrrn29nYePnxIbm4uOTk5wB/KRASy2UtLS1GpVJI7EJbvYW63\neyNhFdu2FjUaDWlpaZSVlXH8+PEnYjGnp6e5dOkSH3zwwba73p+rTImiOAgMfvXzrCAIjYANeBs4\n+tXb/hW4RhCUKYVCwcmTJ/nBD35AcXExsOz7vnXrFufPn9/sWkXvsAUyBuqePE56ejr//t//eyYn\nJ1fUL4Jlpcnn89HS0kJzc7NU4yYiIoKcnByys7NRKBRSRlh8fDx2ux2dTocgCFy4cIHc3Fz++Z//\nmf/wH/4Dn3zyCdeuXdv2gqBxcXHYbDZJyQhU/d7idPmnazTrZG5ujsrKSux2OyUlJSiVSgwGA6dP\nn+bIkSMoFAocDsczFSl40iXo8/m4e/fuRmKKtmSebiYzMzNS/Mc62TYZlUrlE0H0g4ODXL16lerq\n6rBTpCwWC3v37pXaefj9fgYGBraijdWmrsXNpL+/n0uXLpGZmUlKSgplZWVMT09z7969td4nt22e\nTk1NUV9fvyJRxWg0UlJSgsfjIT4+nry8PKnR+ONsMLxi22Q0Go1kZ2eTlpb2hCLV2trKBx98wNmz\nZ2lvb9/20Ig12fUEQXAApcBdwPKVogUwxLIbcNvR6XQcO3aMU6dOodVq8Xq9kkn63r17m53BtyUy\nut1uZmdnV1S8jo+P55VXXvnGzw0MDNDY2Mjs7CyCIBAVFUVeXt6K2Kuv09LSQl1dHfv27WN0dJSs\nrCxSUlKIioraVmXKaDRKVYgfrwkyOzu71W6TTY0TnJ2d5fbt22RkZFBYWIhSqUSv11NYWEhhYaH0\nvkDdlACP10IJBHAvLS0xOjpKbW0tV69e3Uil/qCsxW/i6zWlFhYWNlrocttk1Gq1xMfHExUVJR0b\nHh6mrq4upBNb1oNSqWTXrl2Ul5dLpSyGh4e3qmtEyMbszs/Pc//+fcndl5iYyEsvvcS+ffukoqer\nZNvmqcvloru7m66uLpaWltDpdFINv+joaHJyckhMTFyRjerz+ejo6KC2tnYjVsdtkzEqKkrq0fc4\nbrebu3fv8m//9m9Sssx2s+rJLAiCCTgD/J0oijOP3xhFURQFQXiqI1kQhL8A/mKjA30WZrMZm80m\nRfK3trby29/+ltu3b29FKYQtkXFiYoKGhgb279+/plRXi8VCZGQksGytCryeRUdHBydPnuSdd97h\n7NmzuN1uLl68SENDwzdWwt2KaxgVFUVxcfGKStJtbW3U19cHpWXOemVcWFigoaGB5uZmFhYWntgt\nBfh6nEVAufB6vSwuLjI/P8/du3f53e9+R0NDAxMTExt5UAdlLX4Tj1fW3qxTPu3gVsgYERFBXl6e\nZKlxuVxMTEwEpdTKVl9DpVJJSUkJBw8eJDIykunpaR48eLDCwq9SqTAajXg8HhYWFjZ9DMGcp48z\nPj7OzZs3cTgcfP/73yc+Pp633nqL6elpzp07t9pN37auxdHRURoaGqiurqakpEQK+YiIiCArK2uF\nB8Tr9dLS0sKFCxd47733NhIPt20yWq1W3njjDXJzc6VjATf0vXv3aG1tDZqleFXKlCAIapYVqXdF\nUfzgq8PDgiAkiaI4KAhCEvDUrYsoir8EfvnVeTb1bup0Ojlx4gQFBQXAcnGvzz//nMuXL29Vhd4t\nkbGtrY13332XsbExCgsLUSgUUvZhoG6PXq8nJiZmxe5YqVRiMpmkm/vS0pLk/x4ZGWFsbEyKhfJ4\nPPz85z8nJiaG1tZWqa/Wxx9/zOTk5DfGv2zFNQxUGA70Tuzt7aW5uXk7aik91fa7XhlFUWR4eJiq\nqio+/fRTKZg+8OAFVpQN+LpLb3R0VMqSu3Xr1mb1jNr2tfhNBKxyjytT6w12fYxtkzEyMlLKXoTl\nMIJgZZ1u9TUUBIHo6GjJejE1NSXVfbNYLJI12efzUVdXx8OHDzfydZu6FreChoYGPvvsM4qKijh4\n8CCHDx+mubmZq1evrlaZ2ta16PV6efDgAR9++CGCILB7926pqG7A8+Hz+RgbG+PRo0dcuXKFixcv\nbrRMyZbLqNFoKCoq4vTp01I8WID+/n4pvjSYtSRXk80nAL8CGkVR/L8e+69zwJ8CP//q34+2ZITP\nICYmhhMnTvB3f/d3pKSkMDExwYcffsi5c+dobm7eqhpFWyJjd3c3v/rVr+ju7ubw4cNS5WtYnvgq\nlYrY2Fhyc3PZu3fvCn+32+3m1q1b1NXVMT09jSAIeDweampqePToEcPDwysmWFdXl/RQ8/v91NbW\nrjqFdjOJiYkhOztbKgtQX19PbW3tdiyGqee/Ze08ePCAX/ziFxQUFLBv3z5ef/116eEbsMoE+mo9\n7vKqq6vj/fff5/Lly5sZ37eta/F5aDSaJ5qQbkK15W2TUavVEhkZKWUNeb1ePB5P2MVKwR8y3wLX\nRxAEtFotGRkZpKam8q1vfYvMzEw6OztZWlraqDK1JWtxM5mcnKS6upra2lp2795NTEwM+fn5WK1W\nxsbGVnPf3Pa12NDQwMLCAkajEY1Gg9VqxWw2o9VqJVdgVVUVn3zyCefPn5d63G2ALZcxKSmJf/fv\n/h2nTp2SlEJRFJmfn+fhw4d88MEHQelb+jir2R4eAv4YeCQIQmDl/APLStR7giD8DOgGvr81Q3wS\nvV5PSUkJe/fuJSUlBViOXampqaG5uXkrd4w/36oT+/1+KisrGRoaWtF4FJYDsw0GgxRIHkixV6lU\nLC4u0tXVxcjICEtLSwiCgM/nY2hoaNVuomBUM1apVFIT6kA374qKiu1oGLslfT8mJiakzu319fV0\ndnayZ88eMjIyyMrKQqlUsrS0RFNTE42NjSwtLaFQKLh//z7V1dWbnSixZfN0PWRnZ3Po0CHsdjtu\nt5t79+5RX1+/0dNum4wul4vJyUk8Hg9KpVLqeRbKtevWiyAI6HQ6SXG0WCyUl5ezd+9e/H4/SqWS\n7u5ubt68uRn93HZED57BwUF+85vfoFAo+PM//3MKCwtxOp3U1NSs5uNBWYs9PT28//77dHR0sGfP\nHgoLCzGbzdTW1lJRUUFbWxstLS2boUjBFssYKPybk5OzIq440B/0/PnzW5EcsWZWk813E3hWetU3\nR0hvEcnJyZSXl5Ofny8d83q9TExMbOnDWBTFiS07Ocu7oJ3WMHW9TE9PS9a0hoYGrl+/vl0LYktM\nlqIoSuUCuru76ezspLa2lkOHDnHgwAFsNhttbW3cvHmTGzduMD4+jkajYXJykrGxMVQq1aZtArZ6\nnq6V5ORkMjMzpaajZ8+epaqqakPn3E4ZJycnefDgARkZGTidTsbGxhgZGQl6IdWtQBRFJiYmGBwc\nJCUlBa1WK7W4Gh0d5fbt23z88cdcvnx5M4qVhn6LA5ZDJCoqKhAEgbKyMnJycvjZz37GwsIC58+f\n/8bPBmst+v1+GhoaaGlpoaOjg/b2duLi4qioqOCLL77YVKvqVsuYkZHBnj17VoS4wB+qnF+5cmVL\nYvfWSshmUzyL6Ohodu/ezeuvv47T6ZSOb0NKvcwm0tLSwn/7b/8NnU7HzMxM2FWR7u3tZWZmhkeP\nHvHee++h1+uZnZ2VAstdLpfUMsfr9e6I1jnrxe1209HRwZ07d/jkk0+orKzcUY2BBwcHOX/+PBqN\nhrm5Ofr7+xkdHQ1LZcrr9XL9+nVMJhNvvvkmJpOJpqYmHjx4QG1tLT09PVJbmZ3Qn28z6e/vp7W1\nlczMTEpLSykoKKCiomKzrDtbQiC7vbu7G41Gw9jY2I5zT+/du5dvf/vbK6rQDw8Pc/36dW7evElP\nT08QR/cHdpQypVKpSE1NpbCwkJycnBWZU4/Ho8iEPjMzM1IQfDgS6DM4MRFSRqKg8PDhQ0RR5M6d\nO9y/f39FBeedwMLCAo8ePUKn0zEyMsLc3BwtLS1hWfnc7/fz6NEjfD4fc3Nz6HQ6GhoaqKyspKOj\nI9jDCyoTExN8/PHHUh2u7Oxs7HY7i4uLq6quHizm5+d3XAugx0lLS2PXrl0rYoWbmpo4f/78ZoQL\nbBo7SplSq9XY7XaSk5OfyAYK3OTCcbcoI7OTuXz5MhUVFVIW3E7E5/NRXV1NfX09oiji8XhC+gG6\nEQLV3RsbG1EoFLjd7rBUHNfKxMQEv/71rzGZTDgcDoxGI3FxcfT09ITtXAgFFArFiue9z+fj0aNH\nXL9+PaQaM+8oZUoQBCIjI4mLi5O01Lm5Oa5cucKFCxekwF4ZGZnQYXFxMagpy5vFavu7hQMejycs\nA+w3yvz8PBcvXkShUNDf309/f7/8d9piurq6qK+vZ/fu3QBUVVVx//79kAsN2VHKlCiKzM3NMTY2\nxuTkJAaDgaqqKn71q189NxBQRkZGRkZmo9TX14eUeyncqampkdyrPp+Pc+fOUVVVFXIxe8J2Dmij\nhbuUSiWJiYk4HA5sNhtKpZKRkREePny4LVWzRVF8bkBWsIvMbZTnybjT5QOqRFHc801v2OkyyvN0\nmXCXcafLh7wWAVnG5xEREYHNZiM3Nxe/309jYyP9/f3bGge2Khl3kjIVbOSFsfPlQ76BA7KMOwF5\nLe58GeV5usyLION2u/nGgPmv/g114lg5ztRVfm4OaN784WwJ65FxJ19DCH8ZVztPXwQZ5bUYOshr\n8dm8CDKG+1rcXssUgCAI95+3GwkF1jvOnSIfhL+MGxmnLGPoEO7zFMJfRnmebt1nt5Nwn6ew/rEq\nnv8WGRkZGRkZGRmZZyErUzIyMjIyMjIyGyAYytQvg/Cd62G949wp8kH4y7iRccoyhg7hPk8h/GWU\n5+nWfXY7Cfd5Cusc67bHTMnIyMjIyMjIhBOym09GRkZGRkZGZgPIypSMjIyMjIyMzAbYNmVKEITj\ngiA0C4LQJgjC32/X9z4PQRCSBUG4KghCgyAI9YIg/G9fHf+PgiD0C4Lw8KvXm6s4lyxjkNgsGUNV\nPgh/GeV5Ksv4tfOEtXxffUaWMUhspozAcr+7rX4BSqAdSAc0QA2Qtx3fvYqxJQG7vvo5AmgB8oD/\nCPwfsowvjoyhLN+LIKM8T2UZXxT5ZBnDR8bAa0OWqTVonPuANlEUO0RRdAO/Bd7eyHdvFqIoDoqi\nWP3Vz7NAI2AL/L8s4wp2qow5O10+CH8Z5Xn6Qsgoz9M/IMsYRJ4n41pZtzIlCIIS+H+AEyxrcz8U\nBCHvGW+3Ab2P/d7HBga9VQiC4ABKgbtfHfob4COgFjiILONOlFEAfgqIwE3gJztdPgh/GV/AeQrh\nL6M8T1ciyxgiPE1GQRBqBUH4X4IgxKzmHBuxTIWsxrkeBEEwAWeAvxNFcQb4f4EfA9dYNv/9HFnG\nkOcpMt4FvmBZ4e8HltjB8kH4y/iCztMXQUZ5nu4wXmAZM4ASYBD4x1WdaAP+xu8B//zY738M/M+n\nvO8vWDafTbO8I9nJrxdSxq/ku/+VjMEe30ZfvhfgGva+ADLKazH449voa1vWoiAIoiAIIXMNX4R5\nGm4ybnnM1GoQRfGXQCEwvtXfFSzCXUZRFH8pLjd+LAz2WDaBxacdDLNr2Pe0g2Em41MJdxnltbh2\nHnu4hwyhMk8FQdiyc4eKjNvFRpSpfiD5sd/tXx17AlEUvSz7WXc6sow7n1nC/xpeJfxllNfizudF\nWIshPU83SckMaRm3jQ24+VRAB5DGH1Ie85/zmaCb6zb42nYZFQqFqNVqRYPBIEZERIgRERGiXq8X\nlUplUGQMgWuw0dfCdl/DZ72USuVWuR/ktfgCyBgC49voK2TWYrCuoSzjznitRidSsU5EUfQKgvA3\nwGcs15L4X6Io1q/3fDuE97ZTRp1Oh8ViwW63Y7VaMZvNKBQK+vv7qa2tZXBwEJfLtdlfu60yBoGJ\nUJAvOjqayMhIJiYmmJub2+zTh/s1BFnGcCAk1uIWE+7XEF4MGZ/LtjY6/moXvmMRRfG5Dub1yKhQ\nKNDpdOh0OoxGI5GRkcTExBAXF0diYiLJycnk5ORgsVjwer00Nzdz584d+vv7GR0dpbOzk+np6fUJ\n9TWeJ+N2XkONRiP9PZRKJYuLi8zMzDA/P7+R01Z9FXPyTLZKRrPZjNVqxWazYTabEUWRhoYGmpub\nN1Up3qp5GkrIMu58+QjiWtwuwmGeCoLwje7AcJDxeaxGxnVbpmQ2D4PBQEpKCikpKaSmppKSkkJy\ncjJWqxWtVotGoyEjI4OYmBgmJyeJiooiISEBj8dDX18fH3zwAVVVVSwtLQVblE0lLi6OvLw8iouL\n0el0tLW18fDhQ5qbm4M9tDWj1Wo5dOgQJ06coKSkBIVCQVNTExqNhrm5Ofr6+nC73cEepoyMTBij\nUChQKBR4vU8PuVMqlahUKmlzp9VqMRgMzMzM4PP5tnOoOw5ZmQoBjEYjqampOJ1OsrKyyMnJISEh\nAYDFxUVcLhdjY2MsLCwwOTmJx+MhKSkJhUKBXq9n//79LC0tUVtbi8fjCbI0m0dSUhK7d+/m4MGD\n+P1+XC7XjlSk9Ho9hYWFlJWVUV5ejtPpZGZmhunpaSwWC3FxcQwNDe0oZUqhUBAdHY1er8fn87Gw\nsMDCwsIzb9JfR6vVotPp8Pl8eDwe3G53yGVcrYW4uDgcDgfJycmo1WpGR0dpa2ujt7f3+R8OAhER\nESQlJZGQkIBOp2NhYYGRkRFGRkaYn59f14NTrVaj0+lQKBQsLS1tRQiCzDowGAyIoojP5yMpKQmL\nxYLf70elUqHX64mMjJTeA8uWqPn5eSYmJlCr1fh8Pnp6ehgeHmZx8akJmDLIylRIEBMTg8PhID09\nHbvdTmRkJDMzM3R3dzM8PMzCwgIGgwGVSoXH4yEmJgabzYZWq0WpVJKdnc3Q0BBNTU1hpUwlJCTg\ndDpxOBxMTk4C4Pf7gzyqtZOamsrBgwcpLi4mKSkJAI/HgyiKGI1GoqOj0Wq1LCwsBHmkq0elUkkW\nVFEU6evro729nbm5uVUpRZGRkSQlJSEIAlNTUwwODu4oZfLrZGdnc/z4cfbv34/RaKSxsZEzZ86E\nrPR6Tm0AACAASURBVDKVkZHBgQMHyMvLw2QyMTQ0xP3797l37x6Li4trVqaUSiUajYbIyEgiIiIY\nHR0NOWVKEAQUCoVUDiCgYIQzBoNBUvC9Xi+5ublkZ2djNBqJiorCYrEQHR0tPV9mZ2fp6elhcXFR\nCq0YHh4mNjaWxsZG2tvbw+oZs5nIylQQUSgUmM1mkpOTsdvtOJ1OoqOjaWtro6amhoaGBvr7+3G5\nXERHR0sPXYfDAUBKSgoqlYqZmRkmJia2tGbIdqNUKjGZTBiNRgDm5+eZnp7eUQoHQGJiIllZWSQn\nJ2MwGCTrzczMDGq1msjISMnCE1AYdwo2m419+/ZhNpuprq5mfHycxcXFVVmnoqKiyM3NJT4+nv7+\nfi5fvrxjlamoqCgyMzMpLS2lpKQEvV6PWq2mqqoKo9G40Ri/TScxMZE9e/ZQXl6OxWLB4/Hg8/lQ\nq9UsLCys6WEpCAKpqakoFAo6OjpQq9U4HA4WFxcZHw+t8kJWqxWn00lKSgoAXV1dVFVVMTMzE+SR\nbQ0JCQmkp6eTlpZGYmIiUVFRREdHYzabiY+Px2azkZSUxNLSEouLi2RmZtLY2Mj/+B//g4iICP76\nr/8at9tNZ2cnNpuNqKgoJiYmGBkZCbZoIUlIKVOCIKDRaNDpdJLVBZatEYFdhNfrxev1rvh5LSgU\nipCxbiiVSiIjI4mKisJgMKDT6VhaWuLBgwdcunSJ9vZ2ZmdngeVg7Li4OMxmMxqNhpSUFFwuF263\nm6GhIUZHR0NGro0iCALx8fGYzWbUajXT09P09PTQ0dHBxMREsIe3KhQKBRaLhby8PJxOJ0lJSSiV\nSmZmZvD7/Xi9XjQaDVFRUcTHx5OSkoLH42FmZibkdvRPw2QykZKSQmFhIbGxsQwNDQGrtxwaDAYs\nFgsOhwO9Xk9tba0013cSGo2GpKQkUlNTSU5OJiYmZrkaskKBUqmU7mGhQmxsLEVFRbz00ksUFhbi\n9Xrp7e1lfHycsbGxNWeWiqKI3W4nPj5eunebzWaGh4e3SIK1o1QqsVqtlJSUcPToUYqLi/F6vVRW\nVjIyMkJDQ8OOdjE/jYiICLKysigtLSUnJweHw0FsbKwUTB4REUFcXBwREREsLi4yNTXF/Pw8MzMz\ndHV1ER8fL7kCA/epmJgYDAbDcwPStxulUvn/t/emsY2l573n73DfRYmUREqUqF0qVUlVqq6ubvdi\nJ05sDJzlzg2Qi3zIxQwywDUQDBDDkyBO5os/BghygYE/BIjtCTJBnJkL35vEcRAndpy47e6udler\nqlRaqYUUJZEUF5EUF3GReOZD9fumul3dpV2U+vyAQlerJPI84nvO+7zP8n8wGo0YDAYZITUYDBwc\nHMgyAuErqKpKo9E4k4hkUzlTZrMZn8/HyMgIAwMDtLa2otfrpedcLBbJ5XLkcjny+bz8ey6X+8jX\nNBqN6HQ6qtUqNpsNh8PB7u5uUxRrq6pKvV6nUqmQz+eJRCKUSiUeP35MJBL5wOZSq9XIZDLYbDaM\nRiMOh4NarUYqlSKfz1OpVI7sWDYr4qQfCAQwmUxkMhlCoRCzs7NHcqaeDuefJ2azmWAwyMjICKOj\no1y7do3u7m5ZXyQ+J4vFQmtrK8FgEL1eT2dnJ6urq0Sj0aY+LRsMBrq6uhgaGqKrq4v9/X1yuRzJ\nZPLQzlSj0aDRaLC3t4dOp2NwcJBqtcrW1lZTPaifh8PhIBgM4vP5MJvN1Ot1stksy8vLrK+vN52D\nGAwGef3115mYmMDlcrG9vU00GmV6epqVlZVjOfJOp5OXX36ZX/qlXyKbzTIzM9M0n6FOp2N0dJQX\nXniBu3fv8vLLLzM4OMje3h7lcpmxsTEymYw8DFwFXC4XAwMDjI2NMTExwc2bN/F6vTJY0Wg02NnZ\nIRKJEI1GpSMdDocpFot84QtfQFVV3nvvPRRFoVAosL29zcrKCqqqYrVaqVQqTXF4t1gsdHR00N7e\nLuVmurq6cLlcVCoVcrkc2WyWRCLBzs4OjUaDUqlEPB4/9WtpKmdKr9djt9vp7u7m+vXrDAwMYLVa\n5cIXztTu7i7FYpFCoUA+nyeRSMjNx2AwYDQaUVVVRn4ODg7IZrNSeiCXy8mUhHDKLoKDgwPy+Tyx\nWIyVlRV2dnbkg/hZ11StVrFYLHg8Hnw+HzqdjkKhQDqdJp/PXxlnymq10tnZKU9S8XicSCRyrKjU\neT/ULRYLgUCAl156ibGxMTweD36/H7fbDTyplTKZTBiNRsxms6x98/v99PT04Pf7efjwIQsLC6cm\nd3HadHd3c+vWLbq7u1EURT6Mj7IRm81mHA4HDoeDer3+gUj0ZUI4wx0dHZjNZnnAiUQiJJPJpnIq\nOjs7uXbtGmNjY5hMJjY2NlheXubRo0c8evToWBuMaJyZmpqiu7ubpaWlpmqE8Xg83Lx5k89+9rPc\nuXOHoaEhLBYLRqOR1tZWOjo6cDgcF32Zx8ZqtdLS0iL3gkKhIPUJOzo68Hq9tLa2AhCLxeQBRhSU\niz10d3eXer2OwWDA6XRSLpd5+PAhOp0OVVVJJpPkcjn29vYuvJzEaDTi8Xjo6uoiGAwSCATw+Xw4\nnU6sVisejwen0yltFYGKSCTC/v4+xWKRaDTK1tYWOzs7p1Ze0FTOVKPRoF6vyxvRarXS2tqKw+Gg\nWq1SrVbx+/00Gg0URUGv13NwcMDOzg75fJ6DgwOsVitOp1N+4CaTiVKpRCKRoNFoYDQaKZVKpNNp\nkskkc3NzPHr06EIeeqqqyrDqwcGBTAPFYrFnfr9wNEUOXGguxePxK5XmM5lMOBwOrFYrtVqNaDT6\nkb+Tj+MiPtOOjg5u3brFq6++Sl9fH7VaDbvdLqOj8OTkKLpkzGYzLS0teDweOjo6CAQCOJ1OgKZ1\nqIaHh7l79y6tra2kUimWlpaOfLIXDrPP56NarbKzs0Mmk2ka5+MwmEwm2tvb6enpoaOjA6PRyN7e\nHslkkmg0emGHtGdhs9kYGxvj+vXrtLa2srW1RSgUYmFhgQcPHhAOh48clRI1c7dv36azs5NisUg8\nHiedTjdF5N/hcNDf38/Nmze5desWo6OjshBb2OpwOGQB9mWrx4Qnzuzg4CAHBwdEIhFCoZDcB2q1\nGvl8nq2tLQqFAgsLC6TTabLZLNvb27I8pFgsyhSYSOtVKhUKhYLcZ8WeazAY2N/fv7C9xmg0EgwG\neemll2SUsb29HYfDIZ0nVVVl9sZqtdJoNOTvZn9/X2Z5Hj9+zE9/+lMSicSpOP9N5Uzt7e2RTqfl\nSTcajVIoFNDpdBSLRZmqc7lc2O12PB4PdrtdOhU6nU7m7S0Wi3S2MpmMjFBZLBYAdnZ2CIVCxGKx\nC88B7+7uEg6HAaTT+GHsdjsTExO8+uqrXLt2DZ1ORy6XI5VKkUqlmuLhdVpYLBZ6enpkSjYSibC1\ntXUpOm86Ojq4fv26TP2IXH2hUMBqtcqbu1QqyVOeGEcgitEVRaFWq8l0djPh8/kYHx9nZGQEh8PB\n4uIib775JqFQ6Eivo9frMZvNNBoNdnd3SSQSZ6EEf6ZYrVb8fj8jIyN0dXVJSYTl5WUWFhaaqm7I\narUyODhIMBjEYDCwubnJ/fv3mZmZYX19/UiOlJDFEGlsk8nE2toaCwsL/PSnP2V+fr4pHMnOzk4m\nJiZk12g2m5WyDSaTiba2Nvr6+nj99de5efMmm5ubhMNhwuFw00TWnoXD4cDn89Hf38/Q0BBOp5O1\ntTVUVUVRFPL5PEtLS5RKJTY3N+no6KBYLBKJRMhkMuzu7kopk1Kp9DN734f3w6czHgcHB+e+V4qO\nxM7OTnp6ehgaGuLOnTsyVX1wcECxWCSZTJJOp6nX69jtdhn9F4EKp9OJwWBAURQZoCmXy+j1emKx\n2IkjVE3lTMETJ2djYwObzcbW1hYOhwOTySQLc91utzzRlkolXC4XtVqNer0udWuq1arcqMrlMoVC\nQUatxCnEarWytbWFTneSWc+ng+ju+iiMRiPDw8O89NJLTE1N4fP5KBQKJBIJEolE09VlnJTOzk5Z\nM5fNZonH40eqx7kodDodgUCAgYEBHA6HTGWJtLTQVhKpadEVYzA8uQ2FQvrQ0BDlclnWM6RSqYs0\nSyJSJsPDwzidTorFIisrKzx48ICtrWfOOf1IWlpaaGlpwWQysb+/f+miAgaDAa/XS09PDz09PXi9\nXur1OplMhrW1NVZXV5vCoRB4PB56e3tpbW2lXC4TDod5/Pgxy8vLR95EbDYbIyMjTE5O0tvbS6VS\nYWFhgR/84AfMzs42hd1ms5mxsTEmJydpa2tjd3eXQqHA/v4+iqLg8XiAJ9G1rq4uHA4H0WiUt99+\nm0qlQjQavWALno3NZiMYDHL79m1u376N2+2WRfSFQkHW4UYiEeLxOEtLS9jtdllSIr7n4/i4fz/v\nA63VamV8fJwXXniBsbExgsGgdKpaW1spFApsbm6ytrZGJBJhe3sbVVXxeDyMjIzQ0dGBoihybJfT\n6cThcMivA1QqFXZ3d8nlcifaY5rOmQIIh8Mkk0kMBgMGgwGdTifbd8VDrL+/n5aWFqnYajKZsFqt\nmM1mWTSXy+WoVCq0tLQwODjI4OAgbrcbk8kEQC6Xa3pHxGw209PTw9TUFBMTE9jtdiqVCuVymZ2d\nnVPN+V40er1e1sv19fXh9XqJxWIUi8Wmt9FisdDX10d/fz9tbW2YzWb5UHK5XPJQUKlUSKVSzM3N\n8fjxY9lZI/TC/H4/LS0t9PT08KlPfYpKpcIPf/jDpujw8/v9vPLKKwSDQfL5PDMzM0xPTx+pVVqE\n6cfHxxkYGMBsNhOLxWhtbT2TotCzwmw2MzQ0xMDAAE6nE4vFgqqqFAoFMplMU0UURVOPiJ4lEglW\nV1cJh8PHuq9sNhtTU1NMTU3h8XhIp9MkEgnW19ebwpHyeDxMTU3x2muvMT4+TmtrK8Vika2tLYrF\nIoqiSGVvh8OB1+vF7XbT3d2Nw+EgnU5L0cpmSjuLSRlTU1O8+OKLDA0NUSgUyGazpFIp2SksqFar\nZDIZstnspdPUEvvezZs3eeGFFxgfH5ep9Eqlwvr6OouLi6ytrTE3N8fs7CyxWEw2tPh8PiKRCGNj\nY3R1dVEul8nn8+j1elnHKu4HoZH34d/fUWlKZ6pUKn2sNsvm5iaJRELWnYjUnslk+oCjlM1mqVar\nBAIBqtUqra2tdHZ2yhqqZisSfRYtLS2yI6yrq0vqSsXjccLhMBsbG1cmxWexWBgcHJQ3gHCiL4N9\nbrebiYkJrl27hs/nk1HScrmMwWBAVVVyuRwbGxvMzMzw4MEDlpaWUFVVOlrVapVgMEh3dzcul4up\nqSnZuj43N3dh61S0lr/44ovcunULr9dLOBxmenqamZmZI0WVHA4H169f59q1a3R0dFCpVGR64jLR\n0tJCf38/PT09WK1W2dEYi8XY3t5umjVrMpkIBAIMDg7i8Xhkd/Rxu3+NRiPXrl1jcnKSgYEBqtUq\n8Xic5eXlptGVcjqd9Pf3ywOZXq8nk8lIx99gMGCxWOjt7eXWrVs4nU75x2AwEAqF2NzclEXZzYJI\n73V1ddHS0sL+/j7pdJpYLEYqlXrmnim6Zi8TQqvslVde4TOf+Qyjo6O4XC45CWRra4toNEo4HGZ9\nfZ1oNEo0Gv3AZxWPx6U8SaFQkHIIQg7C4/HITJder6darZ7Y2WxKZ+owiNOweBCLdJ14KD+tJZFM\nJmVNxv7+Ptvb2ywtLbG0tNT0zpTX65URC6vVKnP/S0tLzM7Osrq6emW6+MRppK+vD7fbzfb2Nslk\nsimiMs/D7XZz48YN6fSKzVV0wIiT8ePHj/nJT35CNBqlVqvJaKvVapVOy8HBAT6fj2AwiKqqbGxs\nUK/XCYVCF7JW3W43n//85/nc5z5HMBiUdU6Li4tHjiY5nU5GR0elHMTm5iahUKhpUpmHwWQy0dHR\nQVdXF+3t7ZjNZgqFAuvr6ywsLBCNRptmAxOF/qKxwWg0YrPZ5AD1cDh8pE3kpZde4hd/8RelinYq\nlWJmZoZHjx41zagRoSUk/p7P51lZWeHHP/4x6+vrWK1WfD4fRqORGzduSF0i0Qk2PDxMMBhkbm6u\nqZwpm80mRYxTqRTxeJzNzU02NjbI5XJNvY8dBYfDwd27d/nc5z7Hiy++iN1uJ5PJEIlEWFtbY319\nnXfffZdHjx5RrVZRFOVn1vD+/j46nY69vT1WV1dRFAWv14vdbqe1tVXWYadSKWKx2Kn4AZfWmXr6\nYfX0zfMsrFYr7e3tMkUmojrxeLxpazUURaGjo4PBwUECgQCdnZ1YrVZSqRSrq6ssLCywubl5ZRwp\neFJz1NXVRWdnJzqdTorINdMD7Vno9Xr6+/u5deuWTPtks1kZAd3c3CQajRIKhVheXv6ZDUx0sEaj\nUebn5/F4PHi9XgwGA52dndy9e5etra0Lm0toNpsZHR1lYmKC9vZ2NjY2pBzHUeno6GB8fBy/30+5\nXCYUCrG0tHSpis8dDod0UMSDWbRcb21tsb293TQplf39fVkjIuQbKpUKIyMj5HI59Ho9qVQKRVFk\nl5bYtIUuWqPRwOl0MjQ0xKc//WkmJydxu91ks1m5waXT6Ys2VSKEG41GI7VajXg8zsrKirxOEZ0L\nBoNYLBZZu/d0DY2ImDYDBoMBt9stRZuF0ObGxgZLS0ssLy83tS7d0wiphY/73Yq1Nj4+Tm9vr5RF\n2t7eJhwOEwqFCIVCH+m82+12Oee2o6NDHlqdTid2u12WAoVCIf72b//21Lr5L60zdVhaWloYHx/n\nxRdfJBgMSmHAjY2Npqpr+DCtra1MTk4yNjZGW1ubbP2MRqPMzs6ytLTUFPUJp4ndbpc1Q3t7e8Tj\ncWKxWNOkTJ6FTqfD7/czODhIb28vbW1tcmPd3t5mfX2d2dlZZmdnCYVCH+s05HI5KQMxODgolfFv\n3LjB3NwcJpPpQmrHxMw1l8uF2Wwmm80Si8WO7OSKgd59fX04HA7W1tbkZtDsNXECRVHkXMHOzk7Z\nTSQ6g0TrebNsxJVKhYODA3Q6nYy+eL1exsfHpeBhIpGQwsjCcWpra8PtdrO/v0+9XpdaUsPDwzgc\nDjKZDLFYjNnZ2aardRPr1Ww2U61WicVibG5uyvrYg4MDqa+kqqqcuAFQKBRIJpNks9kjRxfPqivc\n6XQSCATo6uqSe4Fo1DqOLMlFITJIYqLJs9DpdHR0dODz+eS0D1H3VSqV5Dp9ultfr9fLw4Ber2dq\naorPfOYz9PT0yEieqKUSsjPlcpm5uTm++93vyqaET2xk6jDodDpu3rzJL//yL/Mrv/IrOBwOZmZm\niMVirK2tNU1Y+sOINEIgEKC7u1s+1ESnkFBXvkyn+edhsVgYGBigq6sLu91OPp/n/v37/OhHP2qq\nU++HMZvNshtoZ2eHSqWCwWCQEcTl5WXm5+dlq/LHIfRPtra2SCQSdHV14fP5ZFF+IBBgbW3tnCz7\nd0StTbFYZHd3l7m5Odk9dFj0ej19fX0MDw9LBySXy7G6usr6+nrTOB/PQ0SMA4GA1A9rNBokEglW\nVlaIx+NNE5UCZK3e2toaJpNJqkOPjIwwMjLC5z73OdLpNOvr6wAyZVmpVORmpSgKQ0ND9Pf3Y7fb\n5aYoZCCa7VAnxp+IOkQhZvl0uYAYGh+JRBgYGJCacIVCAbPZjMvlwmAwHNrJt1gsuN1uqbp9mhgM\nBhwOh4xKiTq1cDh8qeZ5iiL4j7rXFUWht7f3Z+pORSemwWDA5/NhMpno6+ujUqnIz1Q4VxaLhWvX\nrjE+Po7X65WNQOLgJ7o40+k0e3t7UivvNLjSzlRnZyevvvoqY2NjxGIxqaaeSCSIxWJN9dATiId1\nf3+/FOe0Wq1y41laWpL6W1cFRVHw+XwMDQ3JltVMJsPi4iKRSOSiL+9jESF4i8VCqVQik8nQaDRY\nXl5mbm6OUCjE+vr6oYbdCvXsWCxGNpuVk9wfP35MPB6nt7eXXC53rvMJhUq5qJMql8usrKywurp6\npBS52WxmeHgYm83G22+/LeU9mimKc1jcbjcdHR3Y7XYMBoOUsZidnW2aImyBqLl75513iEajjIyM\nyG7Zzs5O3G63FB0VXdFra2tEo1GZ6jKbzZhMJpnOTKfTPHz4kHfffZdIJNJ0pRIiUqgoitR2E9G2\nnZ0dqQbf0dEho4kej4eDgwNqtRput1uqhh8WoXF4FmUXdrsdn88n11y9XpdRmstW5vG8e13Urwm5\noIODAyksarVa6e3tpbe3V36fKBwXnfyipKe9vR2n00kikWBmZkZGY00mE7FYjLfeekvKRnzY0T4u\nV9aZcrlcTExM8Nprr7Gzs8Pv//7v09fXx2uvvUYymWza1JGY63bjxg2GhobkVPdEIsHi4iILCwuX\nJqx7WIQz1dvbS0tLC9Vq9QMjgpoZoRDsdDppaWkhnU4TiUR4+PAhMzMzR6pnUFVV6krV63Xa2tpI\nJBL84R/+IZFIRGqMnaczpSjKB4RFG40G5XJZPugOi+iwymQyfP3rX2dycpLXX3+9aQq1D4uYTSY2\nXIvFws7OjhStbLZDjqqqsj7UarUyOjoqRZE7Ojrk2C2r1YrBYKBUKskDQLlcZn9/Xxaut7W1YbPZ\nWFhY4I//+I+ZnZ0Fzl976HmI8SF37txhYGBAjrw5ODhgfn6evb09RkdHmZqaIhAIcHBwwObmpqyb\nam1txeVyHclRqVQqcsrGaSPGAIn5etVqlba2Nrq7u9nZ2bkUz8nDoKoq0WhURk5F3Z6Qx3G5XDid\nTjk1QkS46/W6VGcXjrPL5cJqtfLDH/6Qr371q7z66qt85StfoVar8eDBA/7iL/6CmZkZAKn+flKu\nrDPV1tZGf38/Ho9HFlWura1RLpcvJFVyGGw2GwMDA1y7do3BwUHa2tpQFIVcLifTRWd1w14kjUYD\nv9/P0NAQLpdLFm5fhjSmoigYjUYpy6EoitSSOo5DaLfbpU6VSKsZjUYAqX9znggFd3hSKF8sFqV6\n9GFtMxqN+Hw+Ojs7URSFYDBIvV6/lHV/LpdLpl5F+lM4U2tra015SKvVatRqNSl8XC6XWVpawul0\nUq/XsdlseL1eOQA3kUiQyWSoVCrY7XZ6e3vx+Xyy4ykajbK2tta0UZFSqcTW1hbpdBpVVenp6eHu\n3bv4/X76+/vZ3d1ldHRUjiZbX1+nVqvJNJrZbKa3t5eJiQnm5uYO9Zk2Go0zqfszmUzyoOZ2u2Va\nWegqut1u1tfXqVQqUkZGqJt/nJMrpg80W3NPpVIhEolw7949crmcnGmq0+mk+Kh4DtXrdYxGI3q9\nXupNirmLBwcH7O3tYTAYpAL+48ePSSQS/Ou//iuPHj06dduvrDPl9Xrp7OykVqvR09PDF7/4Rb79\n7W/z93//9xd9aR9JX18fL730EhMTE/h8PvR6PblcThYxLy8vN22d10kwGAz09vYyNDSE3W6/FA0C\nAp1Oh9VqlTl7MWdPVdUjh46FttTExAQtLS1sbW1hMpn4rd/6Lf75n/+Zf/iHfzj3Yl8xTaBer8uu\nGjGmY3d397lRMiF3MTo6SltbGyMjI9y6dYt33nmH733ve001cuV5GI1GAoEAQ0NDUr4jm82STCal\nLlGzk8vlmJ6eZnp6WurwWK1WbDYbe3t7UrhQnNS7u7tlR1QmkyGTyZBIJOSIkmY82FWrVVKpFGtr\nawwMDODz+WhtbWVgYIA7d+6wt7cn66PW19fZ3Nwkl8vR1dUlVbOHh4f5whe+gNFo5N69exdih06n\nw+VySRHqjo4O3G63HIw+MDDA9evXZcG8iGrHYjGpPSW6OYWDAU/WsdPplJ3tzfYZ5nI53nnnHcLh\nsNTVenoNqqoqU5xut1s6we3t7fLgWS6XqdVqBAIBvvSlLxGNRvnhD38oR1+dhRN5JZ2ppyM8breb\nfD7P+vr6kUdenBc6nQ6v18vo6Cg3btygv7+f1tZWqVUUiUQIh8OXSovnKIgNqqenB71ez9bWFtPT\n003XJfQsxGfX29tLIBCQqbD9/f0jpT8cDgfXrl3jzp07soNTaKWJDp5nzdE6axqNBtlslmw2y/7+\nPm1tbQwODqIoCna7nXg8TrValZ00QsfI5XIBT0L3gUCAa9euce3aNQKBAPv7+3LoaLNGN56F1+tl\nbGyMnp4eaZ+Iol4Gx18gNk+xTqvV6kdGCPf39+UBwWg0Eg6HWV5eJp/PN90mLCiXy1IhW7THi/ow\nMZsvHA6zubnJ/Pw8jx8/Jp1O09vbS61W486dO/h8Pm7fvk0ymZQTOc773tPpdFIrKZ1O4/V6ZQpS\nKO77fD45Tq1QKLC9vS0Po9vb29LhEM0EwhkTHXL9/f2Ew+Gm2hv39/dJJpNSnHptbU2O5xIlB2JI\nfF9fH729vVId3ePxoCgK9XqdfD5POp1me3ubBw8e8NZbbxGJRM4sLX3lnCmj0Uhvby8jIyMEAgEA\nlpeX+Zd/+RdWVlYu+OqejQilDw8P09fXR3t7O3q9nlKpRDweZ3V19co6Um63W2qCeDweqUy/sLBw\nKTYoi8WC1+vF5/Ph9XqlSGepVDp0ZMpqtTIyMsLt27e5efMmvb29WK1WjEYju7u7rK+vX2iXWD6f\nJxwO09PTQ29vrxSQDQaDZLNZcrkciqJgsVhkOL69vZ1Go0GhUMDj8RAIBOjo6MBqtcrP1WQyodfr\nL8Sm4yDmfT3dcp1KpYhGo5ciKnUcRASkt7cXRVHk5tbsXWS5XI75+XnZvdhoNDAajVitVqrVKtls\nlmg0KjtTy+Uy2WxWjgkaHBzE6XQyPj5OKBSiUqmc+/NIVVVqtZrsthQ1RKLF32AwYLPZaGlpwWKx\nsL+/TyAQIBgMkslkKJVKKIqC0+nk4OCASqVCZ2cnFouF5eVlqdQvRgw1U+2bqM0sl8sfufeJzlJR\nS9VoNOQIukqlQjqd5vHjx7z77rvcv3+fjY2NM73mK+dMtba2MjU1xcDAAKqqsrKywvT0NIuLE8VZ\nugAAGSNJREFUi037wLNYLHR2duL3+/F4PFitVnnKWFlZYWVlpekKW08LsUH19fVJSYGNjY2mf1gL\n3G63vKlLpZJ8ACSTyed2OZnNZqkhc+fOHW7fvs3g4CCdnZ1SiiCRSPDGG28wPT19YSleockiHthi\ngrvf72dvb4+9vT28Xi/t7e1ks1nZebO3t4fVapUnadFlJeZsfpzeTDPicDgIBoPS2S2Xy0QiERYX\nF8+1KeA8EIfSu3fvMjU1RTAYJB6Py07oyxBRFE5QNpvl+vXrtLa2SmdKRK7C4bC8T/P5PKFQiFKp\nhMViwWAw0N/fz+TkJGtra+fuTB0cHFAsFuXvW8x+FPVTHo8Hn88n9wzhfFksFrq7u2X6S9TdplIp\n6VBev36d7u5uNjY2CAaDDA4Oyg7Oy4KoH3W73VL3zW63YzQaZW30u+++y717984ly3GlnCm73c7N\nmzd5+eWX6e7uplQqsbS0xOLiYtO1LD+NmPPl9XqlLkw8HicUCrGyskIsFju14ka9Xi9b3ZvBQRNF\noe+99x6PHj2SJ5JmK4z8OERYuVAosLu7S6lU+pnPS4TZ4UkkqrW1lba2Nnp6ehgZGWFsbIzBwUFZ\n3yEepKlUiocPHxIKhS7CNOBJDcrq6iq1Wo1cLiejMy6XC71eL52iWq0mxz0I56ler+P3+zEYDDJM\nXyqVyOVyl0Ld/ml8Ph/t7e28++67PHjwgL6+PhKJBEtLS039fDkOFouFqakpXn75ZQYGBrBarezu\n7srC7vNEdMwedc5cvV5ndXVV6km1trbKQbm7u7tSguRpRBRoa2uLjo4OWltbGRoaIhgMsrm5ee7R\nGyERIFLi8XhcptLb29vp7u7G5/PJYepCXFev1zM7O4tOp2NiYkKmC/P5PDabDY/HQ0dHB9VqlcHB\nQXZ2dsjn85fGmTIajVIjanNzk89+9rMMDw+jKAp7e3skk0mmp6d57733zq1c5Mo4UzabjVdeeYXP\nf/7z3L59G5PJxOrqKmtra2xubjb1CViMjRFjSER30Pz8vGxRPg30ej1er5dgMEilUmFubu7CQ7uB\nQACfz8c3vvENQqEQv/mbv4ler5ddK82OeACJLpp8Pi8HaBqNRur1uozObG9vo9PpuH79uoxAiTlg\nYmaU0C4SXVfiRHmRiC6vfD7P6uoqfr+fnp4e/H4/TqcTnU5HW1sbqqry/e9/n6WlJZnSs1qtvPji\ni5jNZlnXILR9mlmi5MOIwbhOp5M/+ZM/YXV1lT/4gz8gk8mwvr5+4ffRaWO325mYmOD69es4nU4p\nGByPx8/9WSq6tGq12rH0gDY3N0mlUnI8TrValY7ZhyNs9XqdBw8e4Ha7uXv3rhzjcu3aNdLpNCsr\nKxei1l8ul9nY2GBra0uOvTGZTHg8Hvx+Px0dHVIUd3BwkL29Pb72ta8Rj8f5vd/7Pbq7u1leXqZe\nr+NyuaSelsfjoa+vj3Q6zfz8/KVpCBkdHWVsbIzvf//7/PSnP+XLX/4y7e3tJBIJecB59913iUaj\n53ZNz3WmFEXpAf4foBNQgT9TVfX/UhSlDfj/gD4gAvwnVVUvLDfT09PDnTt35NiYnZ0dMpkMoVCI\naDR6Kg87RVFaT9NGoWnS0tKCqqrs7e2xv78vC+disZiUzm80GrK+RMwa+ijMZrNULrZYLFitVorF\nIqqqUiwW2dnZoa2tjd7eXjnf76KcTSGs9+lPf1pqvmxvb5/l9Zx6kY7QXRIdMw6HQ7Zg7+zs4Ha7\n6e7ulgXct27doqenB5vNRmdnp/z8hWKzOFkJ0c+jSkSc9jqFJ/Ub+/v7FAoFSqUSqVRK1moI5eH9\n/X0WFhZkBEun08li9OHhYW7evInT6cThcKDT6U60KZ2FjR+FUF4WbfOvvfYafr+fZDIpdYuagdMc\nZyIKtq1WK41Gg42NDRYWFk5byuJQ96Jo4T/uM7zRaMh09POo1+vMz8/jdDrlMGSbzcbQ0BCJRIKN\njY0jrdvTXKcHBwcf+B3UajWKxSKZTAabzUYmk8FkMmGxWADo6urCaDQSi8XIZDJEo1EMBoOsyRXd\ngkKYV/zcUTmqjaKw/uPU0D8OocX40ksv0d7eLmumisUikUiE+/fv85Of/OTMa6Q+zGEiU/vA/6Gq\n6rSiKE7gPUVRvg/8r8C/qKr6R4qifAX4CvD7Z3epH013dzc3b95kamqK/v5+bDabbHl9Oid+Cpyq\njUIIz263s7e3R7Va/cCQTr1eT2trq+xQECeSXC7H9va21NkQNTuVSkUWJIqZRS6XC6/Xi9vt5ubN\nm5RKJb72ta/J7pyuri5KpRI7OzvnHglyOBzAk1P/7/7u77Kzs8PXv/51wuHwWV6L7zRfTAxNLRQK\nsuPE7/dLtd1sNktLSwt9fX10dXXJGVtWq5V6vY5OpwOQJ+VsNisLvu/du8fMzMxxTuNnei8+3eH3\nvO8rFoskk0kymYx0ND0eDy0tLbKN+Zic2/PGarXS1dWF2WzGYrHwO7/zO8Tjcf7yL/+S+fn5M5vJ\ndhSeFlc96esEAgEGBgawWCzs7e1xcHDA0tISs7Ozp639dqh78cNOxFmiqio7OzssLCzIOYSiW3do\naIj79+8ftTzizNdpsViUh2XhAHZ2dvJrv/Zrcp7r+vo62WwWu91OtVqlVCpRLBbl4GsxPujpcoQj\ncCQbxfscp2ZS6Eb19vYyOTnJr//6r2Oz2SgWizx8+JCFhQXeeOMN3n777XPX5HuuM6WqahyIv//3\ngqIoC0A38B+An3v/2/4C+DcuwJlSFIUbN27w8z//87LIMJfLsbCwwIMHD067MPR/5hRtVBRFzpHq\n7u6mv78fo9FIMpn8QDGz3W6Xp2KhDh6LxYAnoV5VVeU4A6PRiMViwWKxYLfbsdlstLW14fV6ZVdH\nf38/169f5+/+7u+YmJjgrbfeOndHSnSZ+Hw+gsEgNpuNdDpNKpWSA0jPiKPNiXgO9XpdRpXcbjcO\nhwOr1UqtVsPlcqHT6XA4HLS3t+P3+3G5XPLzslgs1Ot16vW6nBUmBq1GIhHm5+dZWVk5TirsVNfp\nSRHt98IptNlsJzoJv8+52agoCmazWR5shP5bNBollUpduCMFSIX6kyLmo01OThIMBtHr9WxsbPD4\n8WMWFxdP25k61XvxNEmlUty/f192ora1tZHJZAgEAiSTyaPU+p3bOt3d3WVlZQWXyyXLJ2w2mxSy\nFIdXMVrHZrNRrVbZ2dmRz91jOq1HsvEkzScWi4X+/n4pbN3d3U29XieVSsnB8uFwmFgsdv7p6KN8\ns6IofcAU8A7Q+b6jBZDgSRrwWT/zX4D/cvxL/Hi6urp45ZVX5JToer1OOBxmenqa2dnZ054bdao2\nHhwcSHExvV4vF74QfBSzsFpbW+nr68PlcpFOp4lGo2xubsrOjFqtJtNlOp1OagCJSehCah+Qxe1T\nU1P81V/9FZVK5blh67P4DEXHyfDwMN3d3QBSbO6MVbGfueaPa2O5XCaRSLCzsyNP8+J0L4alihEH\ngUAAi8VCKpWiWq3K1G6xWKRarcrOo9XVVWZnZwmHw8c9XV3Ivfg8RJpa1KmI6OsxH+DnZqPBYMBu\nt+NyuTCbzVK3bmNj49wV3M/6M1QUhZ6eHsbHx+nu7qZWq8nB8GfQsXiq9+JpUiqVWFhYYHBwkMnJ\nSWw2mxT+FFpOh3Sozm2d1mo1NjY26OnpAZDK9mJmnRjELAY563Q6crkcGxsbrK+vn6Rb8Ug2nuTg\nLoR/+/v7cbvdH2hoEeLWF1UjfWhnSlEUB/DfgS+pqrorUk4AqqqqiqI88+pVVf0z4M/ef41TtbCz\ns5NPfepTfOpTn2JsbAxFUVhcXOT+/fs8fPhQRm9OkVO1UYiTLS8vEwwGgSc2iYJlr9crvzY5OYnH\n42Frawur1YrD4ZDTxHU6nZyMraoqiqLQaDRk+kkMqAX45je/yS/8wi/w3nvvSfHE5xp9Bp+hy+Xi\nxo0bjIyM4HQ6gSedNFtbWxfSZXhcG0ulEvfv36elpUWO5BCnQJ1OJ0d4OJ1OGTFsbW2V6TyhWByP\nx8nn87ID9fHjxySTyWObc5o2nhS9Xi8f4Hq9nnQ6TTabPWk09NxsNJlMeL1emfYSGnCpVOpMhvx+\nnIN51p+hqOP0+/20tLSwvb1NLpc719FOF7VOP4w42Ii6uP39fUZGRtjd3ZUjow5xEDi3dSpSlOl0\nmkKhQKPRkIdzr9cr6xRFl20+nyeRSLCyssLa2tqhPuOPSCWfi402m03qMQodO3EY3d7eZnZ2ljff\nfPPcu00Fh3KmFEUx8sSR+itVVf/H+1/eVhTFr6pqXFEUP3DsJ/9xGRwcZGJigunpaba2trhz5w6h\nUIgf/OAHLC0tncVbnrqNQg/j3/7t30gmkwwODsoBpLlcDqPRiE6n45/+6Z+kzo1I++3v71Ov12k0\nGlSrVYrFolTeLhQK5HI50uk0uVyOcrnMo0eP6OrqIpPJ8PDhQ9mldRFYrVb6+/vp7+8H4Pvf/z7f\n+ta3CIfDZ/3WpyqQU6/XiUajvPfee7jdbra3t+ns7KS9vR1FUdDpdFLf5rvf/S57e3u89tprWCwW\n2Qm2urrK6uoq29vbFAoF6Vyd4HR17vfix9HS0oLf75eztXZ3d6XjeIJamHOzUYzfaGlpYX9/nzfe\neIPvfOc7bG5unsn7XWQXq9Fo5MGDB3R2dvKrv/qrsiZzaGhIjhUSdX5ihMkJaHqxquXlZX784x9j\nMBjo6+vD7/cTCASIRCIfOKR+DOd+LwqxykwmQ29vL9evX6elpQWHwyHV7uPxuMxwLC4uHjV1+WHO\nxcauri6Gh4fp7e2lvb2der3Oj370I6anpykUCiwtLV2YIwWH6+ZTgG8CC6qq/ten/uk7wP8C/NH7\n//27M7nCj0CoRgeDQf70T/+UjY0NvvzlL5PJZHjzzTfPSqDzTGxMJpNyMG4sFqO9vV22x9vtdhqN\nBv/4j/9Io9Hgt3/7txkeHiaZTMrCQ6FtJAoJ9/b2KBQK8pQiOlnMZrPMq1/0EGGTySS7SMrlMt/6\n1rf467/+6/N461PPy9RqNVZWVlBVlfX1damg7PF4MJvNeDweAL7xjW8wNzfH1772NUZHRwmHwyws\nLBAKhWSeP5vNnkax7bnei8+jo6NDph7y+bxcnycUQTw3G0U9ooigfu973+M73/nOmb3fRdVgiRmT\n9+7dI5vNcuvWLTk+5/XXX2dwcFBGZSqVitywT0DTT7lOpVK89dZbjI2NMTo6KrtT3W73YYVLz/1e\n3NnZ4e2332ZjYwO/388Xv/hFbt++LcsKYrGY1F9cX18nkUgcusnlI9bmmdtosVgYHh5mYmJCdkDn\n83m+9a1v8Td/8zeyVOQiOUxk6lXgPwOPFUV5+P7X/pAnTtR/UxTlfwPWgf90Npf4s5jNZgYHB+nr\n6yMYDPIbv/EbhEIh4vE4MzMzZ6l0/kdn9cKqqrKxsUGpVJK1T6qqYjQaZUcGwLe//W06Ojpk95+o\nearVajIqVa/XqdVq0rESN4CYtn3RBbNiHIKobzMYDJhMJpn+OmPORMFNqCenUikikQgbGxv09/fL\n7hqLxcKNGzdwuVxsbm6SSCR48OABc3NzRCIR8vk85XL5tKISZ7ZOj4rb7ZaiguVyWXaOFovFk+r1\nnJuNouawUqlQrValIvbc3NyxdI+alXq9jsPh4NatW/zcz/0cgUAAp9OJ3++XY5NisRgbGxuk0+nT\nOJA1//BNnhR2r62tEQwGcbvdci7jIdfvqazTo3RqisN1vV5nb2+PxcVFKUOQy+VIJpNSvy4ej59G\n19uZ3osOh4PR0VHu3LnDtWvXZDlFqVTC5XJht9sPm3I9Uw7TzfcTQPmIf/6F072cw9HW1iZHb/j9\nfiYnJ1lcXOTP//zPWV5ePrP3VVX1THNi1Wr1uWqt09PTx379i3aiBDabTbZbi+J7eCLENjMzcyZ1\nKE9xJnecGPUgxgAJnbBAIEA4HMbhcNDV1UVbWxvz8/Nsbm4SjUYJh8On7vyf9To9LEJtXyg3F4tF\nstksi4uLJBKJE732edoo1OiF6OrExATZbJbNzc2T1LU1HUJl+4UXXuDTn/40ZrNZOr31el2KyG5s\nbLC8vHwaKZVLoXSq0+mIxWKsrq4yODiITqejp6eHvr4+IpEI8NHOzmms06PKXojP0el04nQ6WV5e\nlrM9xbgni8VCsVg8FY20s74XfT4ft27d4tatWwwPD2M2m2XBudPppK+vj1Ao1PzOVDPS2dnJCy+8\nwOTkJG1tbZRKJTY2Ni7VTLdPMpVKhbW1NVKplBwxIqJmZ3XSFw0T5+FQVqtVeXqfnZ2VkhVms1k6\nXSIte8aO44XSaDTY3d3l/v375HI5HA4H+XyetbU1uQkdhYvScyqVSqyuruLz+VBVlWg0yvLy8qVR\nbz8KpVKJ2dlZ7HY7yWQSu91ONpslHA6zurpKNBqVkY1mESs9a4Rg7e7uLvF4nHq9/oHnlF6vx2Aw\nUKvVzmR9Hlf2Ym9vj0QiQTqd/kCTktAgPOp4novC7XbLwvNgMCjrjO/du8e9e/dYW1trirFUl86Z\nMpvN9PT0MDk5KTv45ufnWV5evrBOMI2jIR5MzTp4+jQQukrn3TrfTKiqSqlUkil4IRCYSqWa4uF3\nWIQNiqKwvr5ONBplaWnpSjoTtVqNpaUlisUiq6ur2O12yuUym5ubbG5unlZN36VCaMDt7OxwcHBA\npVKRkVb494NasyEiVJcdj8fDwMCAnPG5v7/P1tYWs7OzrK6uNs19eKmcKZ1OR3t7O8FgkJ6eHkwm\nE/AkR7y9vU0ymbwUnrbG+dMsKc5PGqKztFgsysjSce/Ri/oMq9Uq6+vrxONx9Hq97KK9qk5FPp+X\nzpSIaIiN+ZN2H+l0OsxmM6VSiUQiQalUIpvNEo1GZQOFqFX9pP1uzouWlhY5AqdWq5FMJonFYnIK\nSLNwqZwpg8FAMBgkGAzKQbixWIy5uTkWFxe1qJSGRhNyWurcF8nBwUHTnIDPmqedp086IjVWKpVo\nNBqyizqbzX5gTZ/2+hajlprJWXia80q5G41G5ufn+e53v4uiKHR1dbG8vEwoFCIWizVVqv1SOVNi\n1pyqqiwuLsqZQ2+88QaLi4sXMs1bQ0NDQ+NqIqZRlEol9vb2KJVKZ+5EKIoinalmjAYK/bzjjoQ5\nLCaTCbPZLKdBqKrK+Pg44XCY+fl5kslkUzn8l8qZqtfrhEIharUa09PTVCoVNjc32drakoNULyM6\nnU6qHmtpSg0NDY3mQkRAzsOxUVVVBgaazZECZKr+rK5NdAEDMhpcKpV48803CYVCbG9vN11UCi6Z\nM7W/v08kEmF7exur1Uo2m20qz/S46HQ6DAbDiepJNDQ0NDTOhvN+Ljf7vnaWTp5wpmq1mqxL0+l0\nrK+vyxmCzbhPnrczlQZK7//3WAjNl3NQ8PbywesMHvLnisCRZtlcYH3CcWw88Wd4jnzYPrj6Nh52\nnX4SbDzyvXiBaPfis7nMNmr34r9z6HuxVquxtbX1ga81Go3zlD06lo3KeYcRFUW5r6rqnXN902Nw\n3Ou8LPbB1bfxJNep2dg8XPV1ClffRm2dnt3PnidXfZ3C8a9VdxYXo6GhoaGhoaHxSUFzpjQ0NDQ0\nNDQ0TsBFOFN/dgHveRyOe52XxT64+jae5Do1G5uHq75O4erbqK3Ts/vZ8+Sqr1M45rWee82UhoaG\nhoaGhsZVQkvzaWhoaGhoaGicgHNzphRF+Z8URVlSFGVFUZSvnNf7Pg9FUXoURflXRVHmFUWZUxTl\nd97/+lcVRdlSFOXh+3++cIjX0my8IE7Lxma1D66+jdo61Wz80Otcafve/xnNxgviNG0E/n1u1ln+\nAfTAKjAAmIBHwPh5vPchrs0P3H7/704gBIwDXwV+V7Pxk2NjM9v3SbBRW6eajZ8U+zQbr46N4s95\nRabuAiuqqq6pqloD/l/gP5zTe38sqqrGVVWdfv/vBWAB6D7GS2k2XiCnZGPT2gdX30ZtnR6Jq27j\nVbcPNBsvlFO0ETi/NF83sPHU/29ygos+KxRF6QOmgHfe/9L/rijKjKIo/7eiKK3P+XHNxibhBDZe\nCvvg6tuordNPvI1X3T7QbGwaTmgjoBWgSxRFcQD/HfiSqqq7wJ8Cg8AtIA78yQVe3qmg2ajZeBm4\n6vaBZiNXwMarbh9oNnIEG8/LmdoCep76/8D7X2sKFEUx8uSX+Veqqv4PAFVVt1VVPVBVtQF8nSfh\nyo9Ds/GCOQUbm9o+uPo2autUs/F9rrp9oNl44ZySjcD5OVPvAsOKovQrimICfgP4zjm998eiKIoC\nfBNYUFX1vz71df9T3/YfgdnnvJRm4wVySjY2rX1w9W3U1qlEs/Hq2weajRfKKdr4hKNWrB/3D/AF\nnlTLrwL/53m97yGu6zVABWaAh+//+QLwl8Dj97/+HcCv2Xj1bWxW+z4JNmrrVLPxk2SfZuPVsVFV\nVU0BXUNDQ0NDQ0PjJGgF6BoaGhoaGhoaJ0BzpjQ0NDQ0NDQ0ToDmTGloaGhoaGhonADNmdLQ0NDQ\n0NDQOAGaM6WhoaGhoaGhcQI0Z0pDQ0NDQ0ND4wRozpSGhoaGhoaGxgnQnCkNDQ0NDQ0NjRPw/wNl\nnHHK5YG+FgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7f8077058ed0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "tf.reset_default_graph()\n",
    "X, _, decode, _, _, saver = build_graph(is_training=False)\n",
    "\n",
    "with tf.Session() as sess:\n",
    "    sess.run(tf.global_variables_initializer())\n",
    "    saver.restore(sess, './save_cae/temp-bn-save')\n",
    "    \n",
    "    recons_out = sess.run(decode, feed_dict={X: mnist.test.images[:examples_to_show]})\n",
    "    \n",
    "# compare the original images with the reconstructions\n",
    "f, a = plt.subplots(2, 10, figsize=(10,2))\n",
    "for i in range(examples_to_show):\n",
    "    a[0][i].imshow(np.reshape(mnist.test.images[i], (28,28)), cmap='gray', interpolation='spline16')\n",
    "    a[1][i].imshow(np.reshape(recons_out[i], (28,28)), cmap='gray', interpolation='spline16')\n",
    "plt.draw()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "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": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}