VAE_for_Regression.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<script>\n",
       "code_show=true; \n",
       "function code_toggle() {\n",
       " if (code_show){\n",
       " $('div.input').hide();\n",
       " } else {\n",
       " $('div.input').show();\n",
       " }\n",
       " code_show = !code_show\n",
       "} \n",
       "$( document ).ready(code_toggle);\n",
       "</script>\n",
       "<form action=\"javascript:code_toggle()\"><input type=\"submit\" value=\"Click here to toggle on/off the raw code.\"></form>"
      ],
      "text/plain": [
       "<IPython.core.display.HTML object>"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "from IPython.display import HTML\n",
    "\n",
    "HTML('''<script>\n",
    "code_show=true; \n",
    "function code_toggle() {\n",
    " if (code_show){\n",
    " $('div.input').hide();\n",
    " } else {\n",
    " $('div.input').show();\n",
    " }\n",
    " code_show = !code_show\n",
    "} \n",
    "$( document ).ready(code_toggle);\n",
    "</script>\n",
    "<form action=\"javascript:code_toggle()\"><input type=\"submit\" value=\"Click here to toggle on/off the raw code.\"></form>''')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/wuzhejun/anaconda3/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 __future__ import absolute_import\n",
    "from __future__ import division\n",
    "from __future__ import print_function\n",
    "\n",
    "from keras.layers import Lambda, Input, Dense, Reshape, RepeatVector, Dropout\n",
    "from keras.models import Model\n",
    "from keras.datasets import mnist\n",
    "from keras.losses import mse, binary_crossentropy\n",
    "from keras.utils import plot_model\n",
    "from keras import backend as K\n",
    "from keras.constraints import unit_norm, max_norm\n",
    "\n",
    "from scipy import stats\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib\n",
    "import matplotlib.pyplot as plt\n",
    "import argparse\n",
    "import os\n",
    "from sklearn.manifold import MDS\n",
    "from sklearn.model_selection import StratifiedKFold\n",
    "from sklearn.metrics import mean_squared_error, r2_score"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "from mpl_toolkits.mplot3d import Axes3D\n",
    "\n",
    "# reparameterization trick\n",
    "# instead of sampling from Q(z|X), sample eps = N(0,I)\n",
    "# z = z_mean   sqrt(var)*eps\n",
    "def sampling(args):\n",
    "    \"\"\"Reparameterization trick by sampling fr an isotropic unit Gaussian.\n",
    "    # Arguments:\n",
    "        args (tensor): mean and log of variance of Q(z|X)\n",
    "    # Returns:\n",
    "        z (tensor): sampled latent vector\n",
    "    \"\"\"\n",
    "\n",
    "    z_mean, z_log_var = args\n",
    "    batch = K.shape(z_mean)[0]\n",
    "    dim = K.int_shape(z_mean)[1]\n",
    "    # by default, random_normal has mean=0 and std=1.0\n",
    "    epsilon = K.random_normal(shape=(batch, dim))\n",
    "    thre = K.random_uniform(shape=(batch,1))\n",
    "    return z_mean   K.exp(0.5 * z_log_var) * epsilon"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Load Toy Example Data\n",
    "training_feature = np.loadtxt('X.txt')\n",
    "training_feature.shape\n",
    "\n",
    "Y = np.loadtxt('Y.txt')\n",
    "ground_truth_r = Y\n",
    "\n",
    "np.random.seed(seed=0)\n",
    "\n",
    "original_dim = training_feature.shape[1]\n",
    "num_train = training_feature.shape[0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "encoder_input (InputLayer)      (None, 301)          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "dropout_1 (Dropout)             (None, 301)          0           encoder_input[0][0]              \n",
      "__________________________________________________________________________________________________\n",
      "encoder_intermediate (Dense)    (None, 128)          38656       dropout_1[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "encoder_intermediate_2 (Dense)  (None, 32)           4128        encoder_intermediate[0][0]       \n",
      "__________________________________________________________________________________________________\n",
      "r_mean (Dense)                  (None, 1)            33          encoder_intermediate_2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "r_log_var (Dense)               (None, 1)            33          encoder_intermediate_2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "z_mean (Dense)                  (None, 8)            264         encoder_intermediate_2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "z_log_var (Dense)               (None, 8)            264         encoder_intermediate_2[0][0]     \n",
      "__________________________________________________________________________________________________\n",
      "r (Lambda)                      (None, 1)            0           r_mean[0][0]                     \n",
      "                                                                 r_log_var[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "z (Lambda)                      (None, 8)            0           z_mean[0][0]                     \n",
      "                                                                 z_log_var[0][0]                  \n",
      "__________________________________________________________________________________________________\n",
      "pz_mean (Dense)                 (None, 8)            16          r[0][0]                          \n",
      "==================================================================================================\n",
      "Total params: 43,394\n",
      "Trainable params: 43,394\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "z_sampling (InputLayer)      (None, 8)                 0         \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 32)                288       \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 128)               4224      \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 301)               38829     \n",
      "=================================================================\n",
      "Total params: 43,341\n",
      "Trainable params: 43,341\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "encoder_input (InputLayer)      (None, 301)          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "ground_truth (InputLayer)       (None, 1)            0                                            \n",
      "__________________________________________________________________________________________________\n",
      "encoder (Model)                 [(None, 8), (None, 8 43394       encoder_input[0][0]              \n",
      "                                                                 ground_truth[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "decoder (Model)                 (None, 301)          43341       encoder[1][2]                    \n",
      "==================================================================================================\n",
      "Total params: 86,735\n",
      "Trainable params: 86,735\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "## Build VAE-Regression Model\n",
    "input_shape_x = (original_dim, )\n",
    "input_shape_r = (1, )\n",
    "\n",
    "intermediate_dim = 32\n",
    "batch_size = 64\n",
    "latent_dim = 8\n",
    "epochs = 100\n",
    "\n",
    "# VAE model = encoder   decoder\n",
    "# build encoder model\n",
    "inputs_r = Input(shape=input_shape_r, name='ground_truth')\n",
    "inputs_x = Input(shape=input_shape_x, name='encoder_input')\n",
    "inputs_x_dropout = Dropout(0.25)(inputs_x)\n",
    "\n",
    "inter_x1 = Dense(128, activation='tanh', name='encoder_intermediate')(inputs_x_dropout)\n",
    "inter_x2 = Dense(intermediate_dim, activation='tanh', name='encoder_intermediate_2')(inter_x1)\n",
    "\n",
    "# posterior on Y; probablistic regressor\n",
    "r_mean = Dense(1, name='r_mean')(inter_x2)\n",
    "r_log_var = Dense(1, name='r_log_var')(inter_x2)\n",
    "\n",
    "# q(z|x)\n",
    "z_mean = Dense(latent_dim, name='z_mean')(inter_x2)\n",
    "z_log_var = Dense(latent_dim, name='z_log_var')(inter_x2)\n",
    "\n",
    "# use reparameterization trick to push the sampling out as input\n",
    "z = Lambda(sampling, output_shape=(latent_dim,), name='z')([z_mean, z_log_var])\n",
    "r = Lambda(sampling, output_shape=(1,), name='r')([r_mean, r_log_var])\n",
    "\n",
    "# latent generator (simplified)\n",
    "pz_mean = Dense(latent_dim, name='pz_mean',kernel_constraint=unit_norm())(r)\n",
    "# pz_log_var = Dense(1, name='pz_log_var')(r)\n",
    "\n",
    "# instantiate encoder model\n",
    "encoder = Model([inputs_x,inputs_r], [z_mean, z_log_var, z, r_mean, r_log_var, r, pz_mean], name='encoder')\n",
    "encoder.summary()\n",
    "\n",
    "# build decoder model\n",
    "latent_inputs = Input(shape=(latent_dim,), name='z_sampling')\n",
    "inter_y1 = Dense(intermediate_dim, activation='tanh')(latent_inputs)\n",
    "inter_y2 = Dense(128, activation='tanh')(inter_y1)\n",
    "outputs = Dense(original_dim)(inter_y2)\n",
    "\n",
    "# instantiate decoder model\n",
    "decoder = Model(latent_inputs, outputs, name='decoder')\n",
    "decoder.summary()\n",
    "#plot_model(decoder, to_file='vae_mlp_decoder.png', show_shapes=True)\n",
    "\n",
    "# instantiate VAE model\n",
    "outputs = decoder(encoder([inputs_x,inputs_r])[2])\n",
    "vae = Model([inputs_x,inputs_r], outputs, name='vae_mlp')\n",
    "vae.summary()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "__________________________________________________________________________________________________\n",
      "Layer (type)                    Output Shape         Param #     Connected to                     \n",
      "==================================================================================================\n",
      "encoder_input (InputLayer)      (None, 301)          0                                            \n",
      "__________________________________________________________________________________________________\n",
      "ground_truth (InputLayer)       (None, 1)            0                                            \n",
      "__________________________________________________________________________________________________\n",
      "encoder (Model)                 [(None, 8), (None, 8 43394       encoder_input[0][0]              \n",
      "                                                                 ground_truth[0][0]               \n",
      "__________________________________________________________________________________________________\n",
      "decoder (Model)                 (None, 301)          43341       encoder[1][2]                    \n",
      "==================================================================================================\n",
      "Total params: 86,735\n",
      "Trainable params: 86,735\n",
      "Non-trainable params: 0\n",
      "__________________________________________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "##Customize Lost Function of the VAE Model\n",
    "models = (encoder, decoder)\n",
    "\n",
    "#reconstruction_loss = K.tf.divide(0.5*K.sum(K.square(inputs_x-outputs), axis=-1), K.exp(outputs_var))   0.5*original_dim*outputs_var\n",
    "reconstruction_loss = mse(inputs_x,outputs)\n",
    "#kl_loss = 1   z_log_var - pz_log_var - K.tf.divide(K.square(z_mean-pz_mean),K.exp(pz_log_var)) - K.tf.divide(K.exp(z_log_var),K.exp(pz_log_var))\n",
    "kl_loss = 1   z_log_var - K.square(z_mean-pz_mean) - K.exp(z_log_var)\n",
    "kl_loss = -0.5*K.sum(kl_loss, axis=-1)\n",
    "label_loss = K.tf.divide(0.5*K.square(r_mean - inputs_r), K.exp(r_log_var))    0.5 * r_log_var\n",
    "\n",
    "vae_loss = K.mean(reconstruction_loss kl_loss label_loss)\n",
    "\n",
    "vae.add_loss(vae_loss)\n",
    "vae.compile(optimizer='adam')\n",
    "vae.summary()\n",
    "vae.save_weights('random_weights.h5')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "## Train the network\n",
    "np.random.seed(0)\n",
    "skf = StratifiedKFold(n_splits=10)\n",
    "pred = np.zeros((ground_truth_r.shape))\n",
    "fake = np.zeros((ground_truth_r.shape[0]))\n",
    "fake[:300] = 1\n",
    "\n",
    "# Run 5-fold CV\n",
    "for train_idx, test_idx in skf.split(training_feature,fake):\n",
    "    training_feature_sk = training_feature[train_idx,:]\n",
    "    training_score = ground_truth_r[train_idx]\n",
    "    \n",
    "    testing_feature_sk = training_feature[test_idx,:]\n",
    "    testing_score = ground_truth_r[test_idx]    \n",
    "    \n",
    "    vae.load_weights('random_weights.h5')\n",
    "    vae.fit([training_feature_sk,training_score],\n",
    "             epochs=epochs,\n",
    "             batch_size=batch_size,\n",
    "             verbose = 0)\n",
    "    \n",
    "    [z_mean, z_log_var, z, r_mean, r_log_var, r_vae, pz_mean] = encoder.predict([testing_feature_sk,testing_score],batch_size=batch_size)\n",
    "    pred[test_idx] = r_mean[:,0]\n",
    "    "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean squared error: 0.358\n",
      "R2 Variance score: 0.641\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAY4AAAEKCAYAAAAFJbKyAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1 /AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvNQv5yAAAIABJREFUeJztnX2UZGdd5z /qukm9Aw5MBVceUlXJR5XCKyGMCcmoIKOZ0 MaAQDKzQ4DuzppQEdVj2a0KvH5ZzeXdeXs0Eg2mCWkapVEPAQeTECC8uyvHYwLxNjMED3GOFIMihhaDST6Wf/qFuT6ur78txbt q 1PdzznOm6tat5z731vTv /ye3/P8HnPOIYQQQvjSKLoBQgghqoWEQwghRCokHEIIIVIh4RBCCJEKCYcQQohUSDiEEEKkQsIhhBAiFRIOIYQQqZBwCCGESMW ohswCS644ALX6XSKboYQQlSGW2 99QHn3ON9zq2lcHQ6HTY2NopuhhBCVAYz2/I9V0NVQgghUiHhEEIIkQoJhxBCiFRIOIQQQqRCwiGEECIVEg4hhBCpkHAIIWaOXq9Hp9Oh0WjQ6XTo9XpFN6lS1HIdhxBCRNHr9VheXmZ7exuAra0tlpeXAVhaWiqyaZVBHocQYqZYXV09JxoDtre3WV1dLahF1aMw4TCzC83so2Z2t5ndZWbHQs4xM3uDmd1rZneY2WVFtFUIUR9OnjyZ6rjYS5Eex8PALznnngpcAbzazC4ZOefHgO8OyjJw43SbKISoG4uLi6mOi70UJhzOua865z4fvP4mcDfwpJHTrgH yPX5NPBYM3vClJsqhKgRa2trLCws7Dq2sLDA2tpaQS2qHqWIcZhZB3gG8JmRj54E/N3Q /vYKy6DOpbNbMPMNu6///5JNFMIUQOWlpZYX1 n3W5jZrTbbdbX1xUYT0Hhs6rM7ADwbuC1zrkHRz8O YoLq8c5tw6sAxw6dCj0HCGEgL54SCiyU6jHYWZz9EWj55x7T8gp9wEXDr1/MvCVabRNCCFEOEXOqjLgD4G7nXO/G3HazcDPBrOrrgC 4Zz76tQaKYQQYg9FDlU9G3gZcKeZ3RYcex2wCOCc 33gA8DVwL3ANnC0gHYKIYQYojDhcM59gvAYxvA5Dnj1dFokhBDCh1LMqhJCiCSUX6o8FD6rSgghklB qXIhj0MIUXqi8ksdOXJEnkcBSDiEEKUnKo/U2bNnWV5elnhMGQmHEKL0xOWRUmbb6SPhEEKUnrD8UsMos 10UXBcCFF6BgHwI0eOcPbs2T2fK7PtdJHHIYQYm2lMlV1aWuL48ePKbFsCJBxCiLEYTJXd2trCOXduquykxEOZbYvH ouz68WhQ4fcxsZG0c0QYibodDpsbW3tOd5ut9nc3Jx gzzo9Xqsrq5y8uRJFhcXWVtbm3nxMbNbnXOHfM5VjEMIMRZV24pViwnHR0NVQoixqNpWrFGLCTWl1x8JhxAFUZfcS1XbirVqHlIZkXAIUQDTDChPmqoFrKM8oUajUXkRnxrOudqVZz7zmU6IMtNutx39bZB3lXa7XXTTSk2323XtdtuZmWu3267b7WaqY2FhIfT5D8rCwkKmuqsMsOE8baw8DiEKoCzDJVUaLsvipYXd36iH1Gw293xPMY8EfBWmSkUehyg7ZfA4wnreZe5pp31mvvdnZqH1mtkU7qo8kMLjKNTAAzcBXwNORHz XOAbwG1B XWfeiUcouyUwWiXQbzSkNbA 95f1Z7DpEgjHEUPVb0NuCrhnP/rnLs0KK fQpuEmDhFB5R7vV7ooj0o7 yitNN fYcDqzYrrBT4KsykCtAh3uN4X9o65XEIsZvhoHKr1XLz8/ORgeGy9rTTemlpPIk8gu5Vh6oMVTk/4TgF3A58EHiaT50SjnqiP 5s Mwi8jHEZSDN/4EyDAdWiToJx/nAgeD11cDfxtSzDGwAG4uLi3k T1ECZASyE9XzDit1e57qbPiTRjgKT3JoZh36w1FP9zh3EzjknHsg7jwlOawfVUykVxYajQY f fDz3I4CeDBgwcB PrXv66EgDWmNkkOzew7gX9wzjkzu5z SvdTBTdLFEBZ1j1UkcXFxchA IDhYPBoEsBTpx75k1NCQAEFpxwxsz8GPgV8j5ndZ2avMLNXmtkrg1OuBU6Y2e3AG4CfcUW7SKIQxk2kV6WFbnkTNmtobm6OVqsVOqMrLAngMFkWx Xx/Gf5NywdvmNaVSoKjtePcWIcdYuPdLtd12q1zt1Lq9VKvJc0Y/1R6yWGS5rFcXk8/7r9hmWEKgXHJ1EkHPUka6CzTgu8ut2um5ub23Mv8/Pzkc8j7XPzCaaneXZ5PP86/YZlRcIh4RBD1CmlRJxRj1qfkLan3u12c52ym8fzr9NvWFbSCEfRK8eFmDhV22gojrjJAGGfZdm0aGlpiVarFfpZs9lMvcI9j dfp9 wDkg4RO2pU0qJOEMZ9lnW2Wg33HBD6DM7fvx46tlUeTz/Ov2GtcDXNalS0VCVGKUuC8HSxjjGiQ3k czy2kejDr9hWUExDgmHqC9pZlXFxTjKZIjL1JZZRcIh4RDiHGFGuUzTW6sibnUnjXAUnnJkEijliBDxlCmFS1RbWq0W3/72t3cF9xcWFkq9n3mVSZNyRMFxMTPUfeVxmvsrUwqXqGueOnUq9YwwMR0kHGImyLJfdZXo9XocPXp01/0dPXo08v7KNL017TWVn6x4JBxiJsiynqFKHDt2jDNnzuw6dubMGY4dOxZ6fpmmt0a1JWotidZuFI EQ8wEZRqamQTDGWx9jk9y69q0Q4JRbYlaS6K1GyXAN4pepaJZVWKUuuc6Cru3QfElr7UWec7W0qyq6YGm40o4xG7KNP10Egyv6xgurVbL6/t5PZ 6C3SdSSMcGqoSM0HaoZmyzMCKasfo8Re96EXMz8/v u78/Dw33HCD13XyiAH1er3IDaPqMiQoAnwVpkpFHocYh7J4J1HtWFlZiTyedVhn3OyzYW3F0 PQcFQ5QENVEg6RnbIMt0S1o9lsRravqD1L4tK9x4luWURaVEg4gJuArwEnIj43 lvG3gvcAVzmU6 EQwzIYkinufdDVPuS9sSIM9JhRjjuOYzmvspiwON2DYzzhPIQaXks VAl4fgh4LIY4bga GAgIFcAn/GpV8IhnMvem52Wx5FmKMrH44g63mq1YgUl7FqDxIm RjnqmcVd27nJDJHJY8lGZYSj31Y6McLxB8CLh97fAzwhqU4Jh3AuuwBMyxilHYqKE5Y4oYkq7XY79hmleQ5R50bN9mo2m dEaRyRLsuwYh2ok3C8D/iBofcfAQ4l1SnhEM6N15udxvBH3PBOVBkeyhptX1ycIUuJMvpRRjmsTXH3GCeC4w6RaUvZ9NRJON4fIhzPjDh3GdgANhYXF/N8nqKilL03GtW RqORqd1pe/1ZSxqjnCRm4wT04 ovy29cJeokHBqqEnvwNTRlH//udrtufn4 VDhGjw961kmG1XfvjXFK2sB13LXH9QzK/htXiToJx4 zOzj WZ86JRz1Ja2hKPuMm7gV34Pe9OhwTBbDOPwcxhGNrNeOm0I8LmX/jatCZYQD GPgq8AZ4D7gFcArgVcGnxvwJuCLwJ14xDechKPW1G1oImmMfhL3myYWMhCwcY2yPIPyk7twAE8CnkV/ uwPAT/ke4EiioSjvtQhGDrcQ07qied5v8MBdB/Pw8xyNezyDMpNrsIB/CawCXwA POg3Ox7gSKKhKO xPWWq2CMfOINwz3xvDyOsOsOxOPAgQOhorGysjKBJyDKSt7CcQ/wKN8Ky1AkHPUlyfCWffgjaZio2WzuMtjTylorb0DkLRwfBA74VliGIuGoN0lrFsoa7/BNIzIqDHkY9WkP8UmIqkcuwgH8Hv08Ue mnyvqD4L3bwDe4HuBIoqEYzaoUrwj7ZTYUfEbNcRpM GOM SVVgQUCK8meQnHkZjys74XKKJIOGaDKs2wSjOTaVT80sZFwshqzLN8z/d3mZRXIm8nG3kPVR3zOVamIuGYDarUs027fmLYyPqKTpJgZjGoWcTZxxOc1G9Xpf8TZSNv4fh8yLG/8r1AEUXCMTtUpXcZZYD379 faOh8RWcSQ3RZhgN9xGZS3mKVvNCykYtwAC mP/X2H4Gbh8pHgQ/7XqCIIuEQZSMqvcjc3FxivCIvjyMLWQyxT69/UvGpKsW9ykZewtEGngt8CnjOULkM2Od7gSKKhGP2qILnkTbb7ACfGMfc3NxE7nmc2EgWMZTHURy5DlVVsUg4qkPSznQ YpD3uHZUosAip8QOX7/Varl9 /btqmN fn5i bomIcqKcZSPXIUD CbwYFD GTgLPOh7gSKKhKMaxP2RpzEAWYdTwoxh2HXn5ub2DDNNYhHeJOopsyHVrKpyMVGPA/gp4L k/d40i4SjGsQZwLhx/VFjkLYnH2dMfeMJWQx XNqPNAYuLljuOyTUbDZlWMUuJj5UBXw6y/emVSQc1SDO4CfNJBonn1Pc WmmzeadaHCwI15SbzlO3LLMyMoiXqJ 5CocwAuGyrXAfwM 5XuBIoqEoxpk9ThGhWFlZSX086gkfXGCNUmPw feR9sWlmwwKVg 3K409xMmPEloWKg 5C0c/3OovAVYBb7D9wJFFAlHNUgb44jq8efpcUwyxjFMWs9m9Fpxea/SrjrPKohRz6rVaklIKkhuwgE0gf/oW1lZioSjOvjMqkoycHnGOKLalHfPOq0nEBaT8BXM4bZH7f/h89yy3ENZAvEimbw9jo/6VlaWIuGoF0mGPs9ZVdMiiyfg45Ul7aPhc11fj8PXaxpnSE9Mj7yFYw14I/CD9Bf/XQZc5nuBIoqEo34keSZlnXIaR9pZXGHGeGVlJXFP8qjMugOhyfrcfNuuVdvVIHePI6T8b98LJNR9Ff2Nou4Frgv5/OeA 4HbgvLvfeqVcMweRXsQ45A2AeKwMU7ytrIMy/ni6zXJ46gGeQvHxT7H0hb68ZMvAhcD88DtwCUj5/wc8Ma0dUs4RJWI67mPuyd5lmG8NIyuaM97EoGYHmmEo0Ey7wo59qce30vicuBe59yXnHMPAX8CXJNDvUKUll6vR6fTwczYt28fZsbp06eZn5/fdd7CwgLdbpfjx4 zsLCw67P5 XlOnz5No9Gg0Qj/E15cXATg5MmToZ9vbW3R6XRoNBp0Oh16vV6m 1laWmJzc5OdnR0eeOABbrrpJtrtNmZGu91mfX2dpaWlTHWLEhOlKMBTgJ m7xUMr X4OeAuX2WKqf9a4K1D71/GiHcRXOurwB30BezCmPqWgQ1gY3FxcSKKLOrPpPIyRcUUBiVuGutor35ubi52aMhn4sA4sQ1RT8gpO 419NdunGL3Wo43AM/yvUBM/S8MEY7fGzmnBTwqeP1KPGMrGqoSWYiKB6TdpjWpzqjiM3wUJQRRKUTi0pxkub6oL7kIh3vEeF/pW1maAlwJ3DL0/nrg pjzm8A3fOqWcIgkwjyLSfTO086aivM24uqKm7mURx2i/qQRDuufP33MbB/wBeAw8PfA54CXOOfuGjrnCc65rwavnw/8qnPuiqS6Dx065DY2NibTcFF5er0ey8vLbG9vnzu2sLCw630S7Xabzc3NxPMajQZp/8YWFhZYX18H2NNOMwutz7c9AJ1Oh62trbHqEPXDzG51zh3yOXffpBsThXPuYTN7DXALfW/iJufcXWb2evrKdzPwC2b2k8DDwNfpxzyEGIvV1dU9IrG9vU2z2eTs2bNedUQFnUc5ePAgp06dStW 7e1tVldXz70exjm3RzwWFhZYW1vzrn9tbS1UONPUIWYcX9ekSkVDVSKOuHUTea6G7kZsFzsocek/kjIEjxvAzzIJoMprZUQy5BzjeBTwEuB1wK8Piu8FiigSjmozaQPlG8uIKr4xjqjrtFqtc fEbSc76TUYaajq6nzhT97C8RfAO4BfAX5pUHwvUESRcIRThR6jr4HKe8VzGk9j3A2XBkHobrcbOrV2sA1s1IysVqs19d uTCImJkPewnHCt7KyFAnHXqrSY/QxUEnp2H0ExXemUZjBz tefDySbrcb6pVM6reLen7j7JcuqkHewrEO/BvfCstQJBx7qUqP0cdAxRncrOLoIx5pn1WSWPt4JHHtimtP1hhG2i11y/b/R2Qnb H4a Ah skI7wDuBO7wvUARRcKxl6r0GH0MVNqkgD7G7fDhw7F1 AhQ2n084u41zUZWYe3IIqBp21NGj1VkJ2/haIcV3wsUUSQce6lKj9FnG9i0i p8xDFuhpNPjz3L8FmWHn7Ub ezWVPSb 3rAZU5Riayk6tw9Ovj 4DXBOX7fCsvqkg49lKVHuM4MY79 /eHfnc4ZhBFnIEep91Jw2dpYwpRdcRN /UV0Kp0LsRkyNvjOAacAF4flDuBn/e9QBFFwhFOFXqMvkNqYfcSNbXVRziieunNZnOsdvt4C2EkxTaGf7uo 057zap0LsRkyFs47gD2D73fj2IcYkKM0 sdJ47jM0SWpd1JHkCUgKcx4j7X8hWAKnQuxGTIWzjuBM4ben8ecKfvBYooEo7qMk6vd9yhlpWVlXOeR7PZ9BaNqHabWeTwmc/9 RrxrOIkxDB5C8cv0t d7zeCchvwWt8LFFEkHNUma6 36KGWsP2/8xq6imOcITohBuQqHP36uAz4Bfrxjmf4Vl5UkXDMLkUOtcQNVyUJyjjTosNWoM/NzcnLEKlIIxyRadXN7Hzn3INmdjDsc fc10O/WAKUVl0UQVIK9Xa7DTCRlOa9Xo/V1VVOnjzJ4uIia2tr2rJVpCJNWvU44Xifc 55ZvZl r2Ycx8Bzjl38fhNnQwSDlEEUftcDDAz3v72t4emNNfe3KJo0ghH E73gHPuecG/FznnLh4qF5VZNIQoirW1NRYWFiI/X1xcZGlpifX1ddrtNmZGu92WaIjKkbiRk5l9xDl3OOmYELPMYKhoe3ubRqPBzs7Ors HN0paWlqSUIhKE lxmNl5QXzjAjN7nJkdDEoHeOK0GigE9A1zp9Oh0WjQ6XTo9XpFN kcg61oB8NUOzs7zM/P02q15FWIWhIpHMB/AG4FnhL8OyjvBd6Ux8XN7Cozu8fM7jWz60I f5SZvSP4/DOBaIkZY9gwO fY2tripS99KRdccEFqAUkrQD7nh21F 9BDD3HgwAF2dnbY3NyUaIh6kTTtigmlF6G/z/gXgYuBefprRS4ZOedVwO8Hr38GeIdP3ZqOWx7ymB4bN801zTqNtOs8fBMXRrWtbJmHhYiDnBcAvhp47ND7xwGv8r1ATL1XArcMvb8euH7knFuAK4PX 4AHCGaCxRUJRznIa0Gez6I6H1FKu7I8TeLCNPVOCqULEeOQt3DcFnLsr3wvEFPvtcBbh96/DHjjyDkngCcPvf8icEFS3RKOcpBXtlXfPFBzc3Ou1WqdM5wrKyu7DGlazyDrKvCsAjkORa aF9UnjXDExTgGNMzMBm/MrEl/aGlcLOTY6KISn3P6J5otm9mGmW3cf//9YzduVskzCH3y5MlUx6NImuY64MyZM5w6dQrn nGQG2 8cVdcZOi/8S4WFxdTHY9jUsHwpN8lLM6yvb3N6upqbm0Q4hxJygL8FvCnwGHgR4B3Ar/jq0wx9WqoqkRMYm/rPPd3iEubnqaMehFZYhxR7ZjU0JSPN1GVHR5FeSHnoaoGsAK8C3g3/dlWTd8LxNS7D/gScBGPBMefNnLOq9kdHH nT90SjnQkbVOa1SDmOXwyvCte3G59PiVNHCBqO9hpDgv5CLA2YRLjkqtwTLIAVwNfoB 7WA2OvR74yeD1efS9nXuBzwIX 9Qr4UhHUgxh3AR84wZswwz1/Pz8nsR vqKRxz2Mk4I9LT7ehGIcYlxyEY5B757 fhx3jBbfCxRRJBzpSAoCF91rjZvdNDDorVYrcfvULIbUV7SK9jgGbdWsKpGVvITjCcG/7bDie4EiioQjHb7rJIoyTFm3kx2dVZX3GpJpCay8CTENKjNUNaki4UhHVIyj1WrtEo2ijNekx /jBDHNlNxJBqLlTYhJk5fH8U3gwajie4EiioQjPUmGqcjg6yRFK6nuMngcQkyDXD0O sHqVwGPAc6nP8PqV3wvUESRcORP0dM98 xxD9cVNUNrIAJJM840dCTqQt7C8RmfY2UqEo78qct0T18hGJ2xNLj/MAEdHtKrGhoCEwPyFo5PAkv0kxI2gtef9L1AEUXCkT91CdD6Dj2FCWJdxHNAXX5TkQ9phMMn5chLgBcB/xCUFwbHxAxRl53rfNKdzM/Pc/r06T3pPfJKoVIWlKZEZCVyz/Eqoz3HRRRR 4I3m012dnY4ePAgDz74IGfOnDn32WBP8NXV1dDvttttNjc3J9nsidBoNAj7 zezPTsYivqTy57jQ5X9azP7iJmdCN5/r5n9p3EbKcS06fV6nD59es/xhYUFjh8/zs7ODgcOHNglGvBILzws2eLwlrBVI8/kjmLGSBrLAv4PcDlDqdSBE75jYUUUxTjEKD5rVZyLX7cxmisrr2ByUQFqxTjEMOQcHP9c8O wcOzZo6NMRcIhRvENbEedlyarbhqKNt6aVSUG5C0cHwS C/h88P5a4IO FyiiSDiyU1dDkiZtyaghj/puHrOp6jZTS1SXNMLhM6vq1cAfAE8xs78HXgu80uN7omL0ej2Wl5d3bX60vLw81mZOZcF3PD9s9lj/b2ovecymiqpja2urFs9d1JQ4VaEfPH9R8Ho/8BhfRSqyyOPIRp17v MMCUU9l2azObZn5ptgUohJQ85DVR/3rawsRcKRjaLTikyC4aG3Vqu1a09yX6Pss9rczDLtyTGpTbSESEvewvFrwC8DFwIHB8X3AkUUCUc2pulx MZSxom5TGIHwrj8VmaWue44QRJiGuQtHF8OKV/yvUARRcKRjUkZ2lGD73udcdszKSFMmrKbhToPE4pqkKtwTKIEXsuHgL8N/n1cxHlngduCcrNv/RKO7Exqq1efVOW U2N9jemkht6S8l1leX5FT8sVIm P4zzgF4H3AO mP6vqPN8LRNT534HrgtfXAb8Zcd7pLPVLOIolyrAOtnr1HZYZ1/BPqhff7XZj25bV Nd1KrSoBnkLxzuBPwR OCjrwJ/6XiCiznt4ZGvaJwD3RJwn4SiQ4XTio6ul89o1b5IexyR78SsrK3vuc5LrPYSYNHkLx 0 x9IU4J9G3v9jxHkPAxvAp4Gf8q1fwjE cbN95ubm3Pz8fOqhqLgyMLrDIpSH4Z9kL360bl9PSogykrdwvA24Yuj99wNv9vjeh4ETIeWaFMLxxODfi4FN4LtirrcciMzG4uLihB7t7JDF A961r6bJUX11IfFoUrDNwpwiyqTt3DcDewEhnszeH0XcCdwh  FRur0Gqoa c7bgGt96pfHMT5ZhpvCds0bGPxWqxX6naStW6uEAtyiyqQRDp UI1cBFwHPCcpFwNXA84Cf8Ph GDcDR4LXR4D3jp5gZo8zs0cFry8Ang38dcbriZRkSa09/J2lpSU2NzfZ2dlhc3OTG264ITQl dmzZ0PrquLmSHXZ7EqIRHwVJs8CtICP0J O xGCBYXAIeCtwetn0fdqbg/ fYVv/fI4xiftcJNPzzps2EnDO0KUA8q jmPSRcKRD3GGHfLJ1ZR1eKdKsQ8hqoCEQ8KRK3FrFqIMeBrDnlYEFEsQIn/SCIf2HBe76PV65/bWbjabnD179ty/o zfvx/nHNvb2 eOLSwscOTIEY4fP77r IBms8ny8jJvfvObM7cxat/wdkX3/haiDOS657iYHYb34wDOiUVUAPtb3/rWHnHY3t5mfX09VDQGdd1444286lWvytzOqMB5moB6r9ej0 nQaDTodDql3PuiCm0UM4qva1KloqGqbGRZu5G1NJvN3NtZhhXleVGFNop6Qc7TccWMkNcU2EYj b9VlBczIK63vba2Fjq1d21tzat9q6uroZ7S6uqq1/enQRXaKGYYX4WpUpHHkY28PI79 /cnTuVtNBqR7fDpbY8zqypucWNZqOOmWqLcII9DZBkfD vJD5ibm/O 9iDO0W63I8959KMfHfmZT297dIFhmkV2UYsbzaw0cQTfPdKFKARfhalSmXWPY5zx8biMuGEZYcPKcKwhS8950r3tlZUVr7YXiWIcYtqgdRyzLRyTXI09uof33NxcrHHL0pZJtz9uGM1HnKa1 FCLHMU0kXDMuHDk0WP33Yuj1Wq5VqsVadyyxCtWVlb2CNLc3FwuhjMpjpMkTkV5AhIRMWkkHDMuHOP02LvdbmQm24EBj9uLI6rOKKMXZojn5ubcvn37dh2bn5/PxVjGDbX5CEARubU0bCWmgYRjxoUjLBbhMyspbWLD4dJsNjMZsjQzufIwzlHX821/EbOdlAhSTAMJxwwLR5jxNzO3srJy7pwoYdm/f38m0RinF5xm348k4 wznDNu770II66puWIaSDhmWDiSDFu32828J/gkvIK8PI40gjBOvKCIYSN5HGIaSDhmWDjiRCGPBX7NZnNPjGOcXnBUjGN0Z8Ck4Pg0jeu0A9WKcYhpIOGYYeGIMqB5eRmtVst1u91ct3wNm1U1Kk5JwfGswzlVma1UlXaK6iLhmGHhiIpx5DUUNTDEUb3glZUVLwMXZwintfZDPXkhHkHCMcPC4dxeo5yXaIwa4jBPwUdMos4bGOws3kMWEVDsQIhHKL1wAC8E7gJ2gEMx510F3APcC1znW/ sC8coeYlHVkM8KgRRwjAw2FkNetrhHM1WEuIRqiAcTwW B/hYlHAATeCLwMXAPHA7cIlP/RKO3XS73T0rsbN4GlkNsW9JGgbLewhJHocQj5BGOArJjuucu9s5d0/CaZcD9zrnvuScewj4E CaybeufiwtLXH  edn/n472JI1KQPtuJlbB99fWlo6l13XzGi326yvr6fKgOvDuPt6pEU7 ona4KswkyjEexzXAm8dev8y4I0xdS0DG8DG4uJiXiJcG K8ATOLnCVlZok9/eG8Vr7DUkkr25OuldfsomkmLFQgXpQZyjBUBXwYOBFSrhk6J044XhgiHL/nc20NVfV agXVAAALT0lEQVQZNopJ02ezbm4Ul6YkTcDcVzQmZXwnLSAaFhNlpxTC4XXxeOG4Erhl6P31wPU 9Uo4ktOHjxrdrIYtLvA SHOSl1GOulbWPFkDpuENKBAvyk5dhGMf8CXgIh4Jjj/Np14JR7yR9c1S62M8kwLi0zC 4xr6aXgD8jhE2Sm9cADPB 4D/gX4h4FnATwR MDQeVcDX6A/u2rVt34JR7ahp7B1GUmeQtJU37h1H2kNfZprpWEa3oBiHKLslF44Jl1mUThGjXLUnho we5BfT6GrtvtxhrzPKfY5rF7XxjT8gaUNkSUGQnHjAlHmEGNS0ToYxDTGNO4dOx557bKO0/WoE55A2LWkXDMmHCkXRnu0zNPM3zT7XZdo9EIPX/fvn2xiw zeAmTMPTyBsSsI GYMeFIu2J7YBjjDGXa4Zs48UhqSxqG14yM7oUuhMiOhGPGhCPKyLdardCe eHDh0MX4K2srMTuN57Uq08rYHnEODSkJEQ SDhmTDjiDOqoZ3H48OHUXoFvrz7NkFmWtRea0irE5JBwzJhwOJc8Rt/tdmO9iTzEIypIPxrjyOolxLVNCDEeEo4ZFI44fFaR5zW8FCZgeQWeo2ZTNZvNrI9GCBGQRjisf369OHTokNvY2Ci6GaWh0 mwtbWVW33tIFvuML1ej9XVVU6ePMni4iJra2u5Z7M1s8jP6vj/WIhpYma3OucO ZxbSFp1MV1OnjyZeM7hw4dpNPz O4zW1 v1WF5eZmtrC ccW1tbLC8v5542vN1upzo 3D6lMxciPyQcM0DcPhlmxsrKCkePHvWu7 DBg7ver66usr29vevY9vY2q6ur6RqaQJb9M/IQNQmPECP4jmlVqSjGsTvW0Gq1IleSt1qt0NTncaXVau261jQzv6aNl4w7E0tTgMWsgGIcsx3jGPSyh72Aubk5zj//fE6dOrXnfDNLFSMwM3Z2ds69j4qhhMVCpk2j0Qi9t9F7iKLM9yZEnijGMeOEDR2dOXMGCA8wp 08jA59TXsL1jREDdP5bnMbFR/yiRsJUVckHDUkyqidOnVq7NlHYYIwrT3CszCuqI0rPELUEt8xrSqVWY9xpE16CNF7g0P05k9VYZx1JIpxiFkBLQCcbeGIMnZxe3RE5amSkVTmXDEbSDhmXDici17BPSooA9GI 54Qov6kEY5CZlWZ2QuB3wCeClzunAudAmVmm8A3gbPAw84z4j/rs6rimMYKbyFE9ajCrKoTwAuAj3uc 8POuUt9b0jEs7S0xObmJjs7O2xubrK0tBS6wE2L3oQQURQiHM65u51z9xRxbbGbsJXVR48e5eUvf/lEV1tLmISoML5jWpMowMeAQzGffxn4PHArsOxbr2Ic/qSZgZXXamvNVBKifFCGGIeZfRj4zpCPVp1z7w3O Rjwyy46xvFE59xXzOw7gA8BP  cCx3eMrNlYBlgcXHxmXlmg60zUSurw8hrtbVWYwtRPtLEOPZNqhHOuR/NoY6vBP9 zcz DLiciLiIc24dWId cHzca88Ki4uL3inX81ptrdXYQlSb0q4cN7P9ZvaYwWvg39IPqoscCVtZHUaeq621GluIalOIcJjZ883sPuBK4P1mdktw/Ilm9oHgtH8FfMLMbgc C7zfOfcXRbS3zgynC4kjTQqRpDQfeeW2UoBdiILwDYZUqSg4no1xU5AP47MH jgLDRVgFyJfKENwvEi0ADA9vV6PY8eO7Um7vrCwUJqEhcMowC5EvlRhAaAoEYO1HKOi0Wq1SikaoAC7EEUi4RCh 3cAHDhwoJSiAQqwC1EkEg5Ryd57mTePEqLuSDhEJXvvZd48Soi6I EQle29hyVsFEJMHgmHUO9dCJEKTccVQgih6bhCCCEmh4RDCCFEKiQcQgghUiHhEEIIkQoJhxBCiFRIOIQQQqSiltNxzex oOx7x14APFB0I3KkbvcDuqeqULd7Kup 2s65x/ucWEvhqAJmtuE7Z7oK1O1 QPdUFep2T1W4Hw1VCSGESIWEQwghRCokHMWxXnQDcqZu9wO6p6pQt3sq/f0oxiGEECIV8jiEEEKkQsJREGb2W2b2N2Z2h5n9mZk9tug2jYuZvdDM7jKzHTMr9ayQJMzsKjO7x8zuNbPrim7PuJjZTWb2NTM7UXRb8sDMLjSzj5rZ3cH/uWNFt2lczOw8M/usmd0e3NN/LrpNUUg4iuNDwNOdc98LfAG4vuD25MEJ4AXAx4tuyDiYWRN4E/BjwCXAi83skmJbNTZvA64quhE58jDwS865pwJXAK uwW/0L8CPOOe D7gUuMrMrii4TaFIOArCOfeXzrmHg7efBp5cZHvywDl3t3PunqLbkQOXA/c6577knHsI BPgmoLbNBbOuY8DXy 6HXnhnPuqc 7zwetvAncDTyq2VePh pwO3s4FpZRBaAlHOXg58MGiGyHO8STg74be30fFjVKdMbMO8AzgM8W2ZHzMrGlmtwFfAz7knCvlPe0rugF1xsw DHxnyEerzrn3Bues0ne7e9NsW1Z87qkGWMixUvb8Zh0zOwC8G3itc 7BotszLs65s8ClQczzz8zs6c650sWlJBwTxDn3o3Gfm9kR4HnAYVeRedFJ91QT7gMuHHr/ZOArBbVFRGBmc/RFo ece0/R7ckT59w/mdnH6MelSiccGqoqCDO7CvhV4Cedc9tFt0fs4nPAd5vZRWY2D/wMcHPBbRJDmJkBfwjc7Zz73aLbkwdm9vjB7EozezTwo8DfFNuqcCQcxfFG4DHAh8zsNjP7/aIbNC5m9nwzuw 4Eni/md1SdJuyEExaeA1wC/2g6zudc3cV26rxMLM/Bj4FfI Z3Wdmryi6TWPybOBlwI8Efz 3mdnVRTdqTJ4AfNTM7qDfefmQc 59BbcpFK0cF0IIkQp5HEIIIVIh4RBCCJEKCYcQQohUSDiEEEKkQsIhhBAiFRIOIaaMmT3XzPZMszSzS7NOKTWz1w297tQlC64oJxIOIUIwsyKyKlwKhAqHR3tel/C5ELmhlCNi5jCzXwOW6CcyfAC41Tn320GKh0/SX1x2s5m9C7gJeDxwP3DUOXfSzN4GvM85966gvtPOuQNm9lzgN4I6nw7cCrzUOeeCTAH/I/js8yFtmgdeDzzazH4A K/AU4EnAh3gATP7S CQc 41wXfeB/w2/bQUjw6S490FrAJNM3sL8Czg74FrnHPfzucJillHHoeYKYINpn6afjbVFwCjG0491jn3HOfc79Bf3f9HwZ4pPeANHpd4BvBa vt4XAw828zOA94C/ATwg4QkiQzSt/868A7n3KXOuXcEHz2TvtF/SdQFnXPXAd8OvrcUHP5u4E3OuacB/xTcsxC5IOEQs8YPAO91zn072Mfhz0c f8fQ6yuB/xW8fnvw3SQ 65y7zzm3A9xG31t4CvBl59zfBsksuynae3NGT HLzrnbgte3Bu0QIhckHGLWCEuZPsy3Yj4b5Od5mOBvJ0i2Nz90zr8MvT7LI8PBWXP7DLfn3HUDzov5XlQ7hBgbCYeYNT4B/ESwv/MB4Mdjzv0k/cy40I JfCJ4vUl/CAn6OwPOJVzzb4CLzOy7gvcvjjjvm/QTX0axSX vhoaZXUh/p8IBZ4I040JMHAmHmCmcc5 jnyL9duA9wAbwjYjTfwE4GmQrfRlwLDj FuA5ZvZZ4PuJ91Jwzv0zsEw/Y/AngK2IUz8KXBJkev13IZ//P DLwJ30g LDQfZ14A4zq8SGYKLaKDuumDnM7IBz7rSZLQAfB5YH 1cLIZLRuKeYRdbN7BL6MYLjEg0h0iGPQwghRCoU4xBCCJEKCYcQQohUSDiEEEKkQsIhhBAiFRIOIYQQqZBwCCGESMX/B6zqT4TVvzpWAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1198e07f0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "## Validation\n",
    "# The mean squared error\n",
    "print(\"Mean squared error: %.3f\" % mean_squared_error(ground_truth_r, pred))\n",
    "# Explained variance score: 1 is perfect prediction\n",
    "print('R2 Variance score: %.3f' % r2_score(ground_truth_r, pred))\n",
    "\n",
    "# Plot Prediction vs. Ground-truth Y\n",
    "fig = plt.figure()\n",
    "ax = fig.add_subplot(111)\n",
    "\n",
    "ax.scatter(ground_truth_r, pred,  color='black')\n",
    "plt.xlabel('ground truth')\n",
    "plt.ylabel('prediction truth')\n",
    "ax.axis('equal');"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(-4.876583864295159,\n",
       " 6.274853986551141,\n",
       " -0.8972894163755847,\n",
       " 0.9186161641912578)"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXYAAAEICAYAAABLdt/UAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1 /AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMS4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvNQv5yAAAIABJREFUeJzsnXecXUX5/9/POee27Zu2qaSHFEgCJLTQaxCQKqAUkSooiiJfpSmKKIqiWCmCSO9FIJSEFkoCJCGk103bJFuz/bZT5vfHubt77967SUhWw2 d9 t1X3v3lJk5c879zDPPPDNHlFJoNBqNpudg7OkCaDQajaZ70cKu0Wg0PQwt7BqNRtPD0MKu0Wg0PQwt7BqNRtPD0MKu0Wg0PQwt7JodIiL3iMgt/ E83hWRy1LfzxeRN/8DedwoIv/o7nR3It8zRGSTiLSIyH459isRGfXfLpem56KFfQ R pG3fTwRiaX9f76IlIjIgyJSKSLNIrJKRH6cdr4SkcUiYqRt 6WIPJT6Pix1TEunz7lftKxKqW8rpW7rlgvfufweU0qdsDtpiMhRIlLRKd1fKaUu273S7RK/A76rlCpQSn22q4mk3VOrOwolIheLyAfdkZbmy0W3PCCaL45SqqDtu4isBy5TSs1K2/ZPIB8YBzQCY4B9OiUzEDgPeHw7WZUopZxuKrZm1xgKLN3ThdD876At9i8vU4HHlVL1SilPKbVCKfVsp2N C/x8dy04ETlPROZ12vYDEfl36vtDIvLL1Pc IvKKiDSIyDYReb t19DZpdDpvNLUeTUiUp/6PriL8rRbkiLyf516HHZar RbIrI81aMpF5ErU9vzgdeAgWnnDRSRW0Xk0bR8vioiS1PX8q6IjEvbt15EfiQii0SkUUSeEpFwF U1RORmEdkgItUi8rCIFItISERaABP4XETW7sS9OFlEPhORppT75ta03bNTfxtS13RI6pxLUvVQLyJviMjQtPSUiHxbRFan9v9VfMYB9wCHpNJq2M69KE/V8ToROT9t 4ci8udU/awQkWPTzst5b9L2nyYiC1PXuVZEpqe2F4vIAyKyVUQ2i98LNXdUb5pMtLB/eZkL3J76gYzu4pjngSbg4t3M69/A3p3y Qa5ewLXARVAX6AMuBHYmXUpDOCf NbrXkAM MuOTlJK/TblwijA773UAE ndlcDpwBFwLeAP4jI/kqpVuAkYEvbuUqpLenpisgY4Ang2tS1zABeFpFg2mHnANOB4cBEuq7ni1Ofo4ERQAHwF6VUIq1nNkkpNXJH1wu0AhcBJcDJwFUicnpq3xGpvyWpa5qT2ncjcGbqOt5PXVc6p AbCpNS13SiUmo58G1gTiqtks4FSTWQfwJOUkoVAocCC9MOOQgoB/oAPwOeF5FeqX05700q3QOBh4HrU9d5BLA dd6/AAcYBewHnADsCffZ/9doYf/ycg3wGPBdYJmIrBGRkzodo4BbgJ KSKiLdGpTFmnbZ1znA5RSUeAl4OsAKYEfiy/4nbGBAcBQpZStlHpf7cSCQ0qpOqXUc0qpqFKqGbgdOHJH57UhIhHgReBupdSMVJqvKqXWKp/3gDeBw3cyyXOBV5VSM5VSNr4fPIIvXm38SSm1RSm1DXgZmNxFWucDdymlypVSLcANwHmyCz0ppdS7SqnFqV7aInyR3l49XQn8Wim1POVy xUwOd1qB 5QSjUopTYC72znOnLhAfuISEQptVUple5Sqgb mHoOngJW4jdGO7o3lwIPpureU0ptVkqtEJEy/Ab5WqVUq1KqGvgDvrtR8wXQwv4lRSkVSw32HQD0xrdSn0mziNqOmwFsBK7oIqk SqmStM/yLo57nJSw41vrL6YEvzN3AmuAN1Nd7J/szPWISJ6I3JtyVzThuxVKvkA3 wFgpVLqN2lpniQic8V3CTUAX8G3HneGgcCGtn UUh6wCRiUdkxl2vcoviW w7RS3y38Hs0XQkQOEpF3xHdZNeJb1du7pqHA3W0NN7ANEHbtOjJI9XzOTZVhq4i8KiJj0w7Z3KlR34BfFzu6N0OAXG6poUAglVfb9dwL9NuZ8mo60ML /wFKqSZ8Sywf3y3QmZuBm4C83cjmTaCPiEzGF/icA7JKqWal1HVKqRHAqcAP03yr0U5l6J/2/Tpgb AgpVQRHW4F2VHBUo3H3viWXtu2EPAcvqVdlnIlzEhLb0e9iC34QtKWnuALzuYdlWdHaeG7mhygahfSehy/pzREKVWM7wff3jVtAq7s1HhHlFIf7UReO9PTekMpdTx L20FcH/a7kGpemtjL2DLTtybTUAut9QmIEGmMVKklJqwE9eiSUML 5cUEblFRKaKSDA1aPd9oAG/u5uBUupdYDHwzV3NL9WNfxbfIu8FzOyiXKeIyKjUD7oJcFMf8P2v3xARMzUYlu5CKMT3qzekeh0/25lypdxP3wNOV0rF0nYFgRC z91JHZceIlkF9BaR4i6Sfho4WUSOFZEAfsOTAHZGEDvzBPADERkuIgX4jfBTuxiNVAhsU0rFU77ob6Ttq8F3jYxI23YPcIOITID2wcev7WReVcDgTuMK7YhImfgDzPn4ddNCx70G35L nogEUnmOwxfwHd2bB4BvpereEJFBIjJWKbUV38D4vYgUpfaNFJGddtlpfLSwf3lR IONtfgW4fHAySkfbi5uxhfkzrRFULR9fridPB8HjgOe2Y4ojQZm4f/I5wB/SzUs4Dc p I3QOfj 8Tb CO D7sWf2D49e2UI51z8QcFl6ddwz0pP/338AW6Hl8A28cElFIr8AW3PNWtH5ieqFJqJXAB8OdUmU4FTlVKJXeyXOk8CDyC715aB8Txx0h2hauBX4hIM/BTOgaK28ZCbgc TF3TwUqpF4DfAE mXFxL8P3UO8Pb GGYlSJSm2O/gd/gbcF38RyZKl8bH M/D7Wpcp2dGkvZ0b35hNSAKn4o73t09Hguwm8YlqXOfRa/t6D5AshOjHtpNBpNBiJyMf7ci8P2dFk02WiLXaPRaHoYWtg1Go2mh6FdMRqNRtPD0Ba7RqPR9DD2yCJgffr0UcOGDdsTWWs0Gs3/t8yfP79WKdV3R8ftEWEfNmwY8 bN2/GBGo1Go2lHRDbs CjtitFoNJoehxZ2jUaj6WFoYddoNJoehhZ2jUaj6WFoYddoNJoehhZ2jUaj6WFoYddoNJoehhZ2jUaj6WFoYddoNJoehhZ2jUaj6WFoYddoNJoehhZ2jUaj6WFoYddoNJoehhZ2jUaj6WHstrCLSFhEPhGRz0VkqYj8vDsKptFoNJpdozvWY08AxyilWkQkAHwgIq8ppeZ2Q9oajUaj YLstrAr/6WpLal/A6mPfpGqRqPR7CG6xccuIqaILASqgZlKqY9zHHOFiMwTkXk1NTXdka1Go9FoctAtwq6UcpVSk4HBwIEisk OY 5TSk1RSk3p23eHr zTaDQazS7SrVExSqkG4F1genemq9FoNJqdpzuiYvqKSEnqewQ4Dlixu lqNBqNZtfojqiYAcC/RMTEbyieVkq90g3pajQajWYX6I6omEXAft1QFo1Go9F0A3rmqUaj0fQwtLBrNBpND0MLu0aj0fQwtLBrNBpND0MLu0aj0fQwtLBrNBpND0MLu0aj0fQwtLBrNBpND0MLu0aj0fQwtLBrNBpND0MLu0aj0fQwtLBrNBpND0MLu0aj0fQwtLBrNBpND0MLu0aj0fQwtLBrNBpND0MLu0aj0fQwtLBrNBpND0MLu0aj0fQwtLBrNBpND0MLu0aj0fQwdlvYRWSIiLwjIstFZKmIfL87CqbRaDSaXcPqhjQc4Dql1AIRKQTmi8hMpdSybkhbo9FoNF Q3bbYlVJblVILUt bgeXAoN1NV6PRaDS7Rrf62EVkGLAf8HGOfVeIyDwRmVdTU9Od2Wo0Go0mjW4TdhEpAJ4DrlVKNXXer5S6Tyk1RSk1pW/fvt2VrUaj0Wg60S3CLiIBfFF/TCn1fHekqdFoNJpdozuiYgR4AFiulLpr94uk0Wg0mt2hOyz2acCFwDEisjD1 Uo3pKvRaDSaXWC3wx2VUh8A0g1l0Wg0Gk03oGeeajQaTQ9DC7tGo9H0MLSwazQaTQ9DC7tGo9H0MLSwazQaTQ9DC7tGo9H0MLSwazQaTQ9DC7tGo9H0MLSwazQaTQ9DC7tGo9H0MLSwazQaTQ9DC7tGo9H0MLSwazQaTQ9DC7tGo9H0MLSwazQaTQ9DC7tGo9H0MLSwazQaTQ9DC7tGo9H0MLSwazQaTQ9DC7tGo9H0MLSwazQaTQ jW4RdRB4UkWoRWdId6Wk0Go1m1 kui/0hYHo3paXRaDSa3aBbhF0pNRvY1h1paTQajWb3 K/52EXkChGZJyLzampq/lvZajQazf8c/zVhV0rdp5SaopSa0rdv3/9WthqNRvM/h7WnC6DRaP43aExWsKDuEarjyykJ7sX vS gb3jsni5Wj0QLu0aj Y9Tl1jLSxu g6MSKDwakhupaP2E4wfdxl75B 3p4vU4uivc8QlgDrC3iFSIyKXdka5Go kZzK3 O7aKofBSWxSOSvBB1R9QSu3RsvVEusViV0p9vTvS0Wg0Xx7WNG/lnarPUcDRZRMZXThwl9Oqii/Nub3Frsb2ogTN/F1OW5ONdsVo/r gtrWVxZVV9M3PZ0JZP0RkTxfpS0NDPMZr5atpSsY5bPAwJvTpt8NzbM9hdfNWImaQYfnZ9flQ Vs8sv5tbM8B4KmN73PaoIMZWzScQivM1N7DsQxzp8sYNoqwvWjWdkNMTCO00 lodg4t7JovNUop7nzvAx6a/xlB08RTikFFRTx0zpmUFRbs6eLtcT7avJFLX3sBUNiexx/mfcRpo8Zxx5EndNn4vVu1hNuXPounFJ7yKIuUcOd 32RIXh8ANrbW8Mj6t0ikRB2gKenx4JqPiZifYYhB2Ahwz0HfZHRR/50q58Re5/Jxzb04Kt6 zZQQY4tPwpRsGXKVy9LG1bQ6McYXj6I4UPgFakWj14rRfKl5fdVqHvlsIUnXpSWZJGrblG/bxtUvvryni7bHSbouV77xEjHHJuY4OJ5H3HF4ec0K3tpQnvOc9S3V3Lr4KVqcOFE3Qdyz2dhayzXz7sdTvv/7g9pleGl b8cTbNcEhJhr0 okqEu2cPUnD7efsyNsBrEhMZTlrQPZGO9H0oswvOBwDun7naxjN7Ru4bJPb KOFffz5zWPcvmnN/NcxRtfvIL h9HCrvlS8895C4jZTsY2VylW1NSyubGp2/NbWVvLD197jVMfe5SbZs1kY0NDt fRXXy6tSLnwGPUsXlmxWIAXM/jg8pyni3/nFWNNTxfMRfHczOOVyia7Tif1a8DwBITocPa90U9m1Ynwef1m3ZYzveq3 DRDfdQb7fiYtDkBlmfGMj40m9hGsGMYz3l8Ytlf6HBbibmxom5cWzl8Mym11nSuHqHeWl8tCtG86XE9TweXvQZC u34oY9cMGwpV1wLENoTiS6Nc PKzZxyQsvkHBdPKVYWVPDv1es4Nnzvs7effp0a17dgbedaBJPKba0NnLe24/QkIzhKYVC0ScSwLE8cnlp6pMtABzZbx/uWTMD2pJXAmSfICJEneR2y h4Dq9seYakl3mc7SV5YO2jVMTyWddaRVm4hEtHnEi/cB4xN56VTtJL8nrl  xTPHq7 Wl8tMWu VJy/aw3uHPOB9jK8zXFBC/sixOA5yn6F3Svj/3mWW8Rc5x2wXSUotW2 fk773RrPt3F1AGDc27PswKctfcErpnzApWxJlqdJDHXJu46VEUT2Mlw1jmOcplYMgyAsnAJ1409g6BhETYCBE3oUPm0czyXyb322m4ZG xteGS7a5rtALOqalnWtJGYm2B9axW/XPoE71cvzpUVCmh1sgdfNbnRwq7pNjyve KRNzU2MmPNSmJOmgsmZTAqU4ECo8Fj i8fZMnGym7JM 7YrGuoz7lv7sZNzFrYfW6A7orbDlsWfz7uFMKWRcg0ESBiBThm6Aj26z AZfWVuJ3ysj2PRDJIyOjorFsiHN5vNH1DRe3bTh44lWem3cDE4rEkHd /nnYFmOIxpngLn9TNyLiuTa11VMUa27cVWIU5/fBbYkV4aWkqBTHH4ckNH S02ENGkGl99v8CtfO/jXbFaHabj5as43dPvMvGqnqK8kJcNH0qF02fimHsWkjikpoqAoZJws30BbfpQGgbeI5HK0mufehlZt5yWUYEiFKKT8oreGdFOfmhAF/dbzxDe5dsN8 AYabydLL2iQc3PvI6U0YPpiQ/stPXsaK hqpoM/v07k/vcB6rG2q55ZPX bS6gpBpctbIidy4/zFErMBOp9mZo4eOYPY3LuflNStoTsQ5fMhw9i8bwJZoE0YXUTG9gkXs36uYBdvWg3gUBpJsiM7n72vhqpEXtddl3HX4sLocN4cJPa6kkgH5TbxV9RwDI8OJOaXcuPBpGpJRPBQjCvrx2/3PY3BeL6b2Oox52z7EVh3umITrX7NS4HqCSt3cmGvixiOUhGIYBoAQNAIMzRvIkX2nZpTB8WJURt/H9looixxMXmDX4 x7GlrYNbvFglUVXP 3l0kkfUFsiib4xytziSaSXH3GYbuU5qDCIrwu uNWDIw07W2KJVhdWceYAb4P3PMUP3pqBu tXEcsaWMZBv cPZ fnX4sp 0/vss8TcPga/tM4MlFi3DSrVwPAo1giPD2orWcecg Oyx/XTzKxbOeYU1DLZZhkPRczhk5kRc3LKbVSaKAmOvwzJrPWde0jUePy57fp5RiSV0VjckEk/v0pyAYytq/sG4Ly orGVJQysX77IdpdHTAB YV0SecT0VrY8Z5QcPkyIFDWZd8k7J8u317UsHcuvkcV3Y4YwpHADC7amWX19jqBFPnJXi98hWeXucRczuEe1XTVi6b8w9eOfo6zhlyMYYYfFw3G0EIGiH6hAqpTrTgqTZR72iEkp5FwrEYVNCI65lMLh3Md0b9gEBaL6M29hkfbb0Gf jXQ ExuvibTOh9ddc3Zg/iOeuwo4 j3Eqs0NGYkZMR c/F72th/xLieYqFSzdRUdnAyKF9GT 6/5d2Qs69L33ULuptxJMOj89cwKWnHEwo8MUfsX37lTGsuIRV2 pwvLRuvIJAc2Y9CJmujdmr1rWLOoDjeTiex60vvsUx40dSGO76x3TTEUcyd81G1jTX 05dgUALBBvAcx1mfricAfn5HLTPsO32Rr7z7kss31aNo/xBX4An1nwOktlcJTyX TUVrGqoYUxJx4qnG5rqufCNZ6iJtmKKYHseNx90NBeO2w/w3Ubfeu9JltRX4imFKQa9w/k8deyF9Iv44w4iwu8POo1vzX4Cx/NIei55ZoB kUIm9wuxbnN2w5n0ksyvX9Qu7AHTzPncCQpDUmMdCubVNuOqzHr1ULQ4cebWruGwfntz3l6XcObgC4g5rRQGipldvYTblz1Ji uSPTArNNlhDghtImB49A0OzxB1VyWZU/l9HNWacdaaxkfol3cgfSNTcLwkgmAau94b6i6c EwS9d8FHMDBTcxCWu8n0ud5RHa B/hF0ML JaOhKcp3b36Kqtom2vRqzIh 3PXTswmH9vxD2pn1W7t6v4pQ19jKwD7FXzhNEeHh08/mh2  xpyKjRgi5JsB3M0OrpPprw2LyZ fmE1lXROTRw iUqLtop6OZRrMXbOR4/fpOqoiZFn87oTpXPK3Z4gpB8MGK H3EhQe85dsYtnqSvYdNYA//uAMLCs7DLA61sKCms2 qKfhN1AK6XSKZRisbaprF3alFBe8/jSbmhszGoHbP3mHoGnyzNpFLKjdjKcUYniICSIQb7X58cev8M jzms/Z0rfIcz6ylU8vXYhm1obOKjfUE7Zazwf1n6MIUbWIKUpJhGjY2D1mLLx3Ll0Bp0RoH kiapYARXREhQu4UArlmcQtYO0CbWHojreEZIaNIIEg76lf1TZRFqcGL9e kKOoVU/BU8JAQkxqmC/jH01sU9ROfz2rkqwZNuTrE88Q0V0JYIwqnAKpw76LvlW5nPYlFzJirrfU59YSNAsYXjRxQwt nq3G1BK2SQafgikjRuoKMopx259hGDBFd2aXxt68PRLxp1/n8mmrfXE4jbxhP9ZsaaS x//YE8XLScjB UOAxSBPsW7vv7H7KXrWLe4lvAWoaDa4lvD9 PAQYOJBP3GLRgwCSAktyX4eOkGNlTW88qHS5n/0XoMJ/fgZCCHEHdm4rABnDllHwpUANPzRT3dURBL2Cxas4XX5izPeX5zMoFldPWzyhYNx/MYXdxRh5/XVlIXi2Y5ouKOzY0fv8a8mor2qB3lCZ7rp kqxUdV64k5mY1aWaSQa/Y5nN8edCpnDZ9IyLQ4sNdkFL5ve2NNKQvKhzC/fC/Kq0uZWDyx/dxeoXxun3w2ISNAnhkkZBgYeOxdUo2jTCqiJXgYKBQiYBkeeYEOd4xSsG/JkC7qAk4ZdBBnDDkYS7LrK2zahEyPuJfk6YrHuHftr6lP1gLgqdwhlq6COQ1r2BRdgcLDw2VN8zweKr8hoyFotTcwZ8uF1MY/wlVRYs4WVtb/gRX1d3VZ1l3Fs5dBzqYrjhv/z02y08L JcJxXD6ctxbXzXwQkrbLa /kXkRpZ4glbd5euIbXPl3BtuauQ8Yc1yNpZw8ebo9vn34o4WBmxy8ctLho hSCu CGAXjig4Xc9vQstjVGcV1FLG7z0Pvz2W/kIH5 7vEUF0dwTUhaiqZij2Sef57rKVxXEY6mDaQKuAFoliR/ XAuM1es2WH N55zDPd95yxO2HcMASv7JxJPOMz4cFnOc4cVlhIys6/bMgwCpmRIe8gwmdpvCKNSwu54Hm tX4vtZQuBMlVWhAsIeJC 2d2JmaAFgXyuHX05KzcPpKqxCNu1cFyTmsYirnh7BnbaBKbjBkxg5nHXc9M p3LjPqfxz0PP4ZTBE6hP9sfrJB9t4i4owkaAw/qN2eGSA5ePPJ6 oWLCqYlKlhhYAvuWVpNwLVrcEB6Klc2L cOqm0l6SfpGpuKR/Zw2uiUoDNK7Ih4uzU4d61oXt29b03AfrsqMvHFVjA1Nj2N7zTusvy CSASU28XO/9zCZ9oV0418NnctT//jPWoqG5l04AjOu/xI g7IjMZoaYrxzD1v8cFriwhHgpxy0TROPOcgDMNIrd2R29p0nC4ejh3w6apNXHvPS4j4AuC4HteefjhfP7qje9vYEuM3D73Fe/PX4CnFhBH9ufHSExgxqPcO0584ciB//N7p3PXUe6zdXEdpYYSLTzqQc4 d/IXL6rgeN//zNd6cvwqlIAAoE wCiNkOD70/n0G9immMxf16EgDByVMYDlhJ35URci0cU4EhxEy/3pTA4i1VXP/ia1xzxCFceuiU9nyTrovtuuQHO2ZBTho EPMY4ZP563CcbLHM5YYBf5B1bKSMOfH1/gYBFOSZQR447kxun/c2CyurUJ7gmibDI31wPY E6/C1F56kvHEbSXGzjHtDcoZ3 6TGA1zlMb 2giMHjGzftam5kcU1lfTPL2S/fgM6Il4ShbhOPp7qsPAdpVjfXM9dn8/mR5OPxExZ0kWBCCcP7rifk0on8fqWXwO1WUURgb3ySzlv6GGcPfTArkrcTnEwn8en/YA3t37OwvpyhuT1ZXRRgFe2PkA8rYFReCS8OJ83zGVqryPYr8/NLKz9JZ5yUTiYEkGMAbgqW5g95bEtuYURTAKgMbGEXFa0QYCovYniUNeD7F8UsUYjZn Uu56MOyh5BPIu7LZ8OqOFHV8MWhqjRPLDWIGdW7HOsV0atrVSVBIhGArw nPz PuvXiER938oWzdt490Zi7jwmuOY/9EaAiGLY06exIO3vUDt1kbs1IDjfb94iWXz13PdnV8nGLCYMGYAS1ZuybDCTEM4dMrIrDIsXbaZB/41m3Xrahk4sIRLLjqcA/Yf1r4/lrS59p6XiCYyu d3vTCbeDTJecftTzhocfWvn2H9lm04qZ7CkjVbufy2J3nuzksoKdzx4M6UsXvx M yH9INNfX8/tX3 WTtJgrCIc6fNpmLjtgfQUg6DqGAleHTfPjNecxeVA4qTddcsKLg5EPMtllfU5/d Bm uFupHnrSczGNAP37FVJen3l8zHb48 w5fGPqJBzP4 aZs3hj5SpUAkqNMNP3HsXlRx7IoF7FjBteRiQUIBrPrL9IKMDpR 6bsy5mb1jP4opqxLNQIdfvE9uC3QJFZpjVNQ0oz7 6pOvy0LLPeGXdSs4aPoE19XV iKdBxy8z1TAIBuSOFeqI8Qe 99GLfHL69/E8xbfefJZPKzcTMAxECYMLinnslHPol1fA3KqNWW4b8HsND6z4hE 3reUrw0Zhicmx/ScwKK9XxnH7loygIpZD2IF/HnIlJcHsRbs2Rat5cuPblDdvYXThYM7d62gG5fUlbAb56uCpnDpoCq5Xw/u172J7HWWzxKXAjGPSTHnzHA4oncbQolPoFd6XDc3/xvaaGZh/NJvjTVRvfQC7kzUuCGXh4e3/5weG0WKX07mp9FSSsLVzi5rtLCJCuNeDxOvOQ6lWP0/lYEXOxQx/pVvzysh3TyxyP2XKFDVv3rz/eD6u67FiXjmJeJLxU0cSzvNH7pMJG UpQpEg7728gHtve4nmhiimZXDy dO45McnY27HH/viYx/xyF/farfkvnL2FF5/dh6x1k6 P8vAtMx214oVMFHxJF5LDPCtSLt3Hm6vPPY/ZBRfP Mg vUu4Ns3PE7SdkkkHcKhAAX5Qf7x2wvp06tjpuXnizfx45ueJpHo6JKGQhY3/t8pHHHY3gDM mw1tz76Jq3xTuVSEPSEfGVxzbmHc/cTs4l1Ev9Q0OLyMw7hwpMzY4d3lqrGFk7/3cO0JpLtwhq2TEYX9WJLRQOxpE2/0gJ eO5RHHOAP6A5/Sf3U93QkpWWApLFUBQJ4SpFayLbxyq2IpIKZkmUgBsWvHBqNnwnCoJBHrnoa9z6ztssrqxEbfMwU0kKQsgy dW50zlh4mgeeH8ef3zrQzxRhJohv8lg sHjuOmS43MOtF33xuu8sDzbTVMQDLJ2zyfQAAAgAElEQVTfsP68v3V9DpcKhAyDZDy13QNlKDBTLZwrBA0DL J0iilXYIBpdWwrsIL8YspJ3PzBLJqTmUsuGK4wtf8Q7j/xdI548e80xBNk /09gmGbgoIEpgiGGJgifH/sSZw79OD2o16oeJs/r3o1NRO4LQ1FgQXfGX0m0wdkhro vv4tHlw3o33msIFByLS4a7/vMqZwMNH4HCq3/QDHrcLDpSJZwlstozHEpU gxR/nEDAIMihvAqcPuQOj00h00ovz11VX0eLUt7/Qw5QAAyOjuHj4He33qyGxhLlbL8ZLawAMCdE/7zgm9/tN1r1po9Vez8amx4nZm kdOZRBhadjGTvnTlHKxU1 CG4tRvBADCv3rOEdISLzlVJTdnRcj7XY1yzayC3n/Yl4NIkIeK7i0lvP5NM3FzP/7WWAYtCo/lTWtGKnBNqxXV597EMcx Wqn52RM913ZnzOP  e2W6ZA7z6zKe4du7JNOn csd2wTBS/WpFdEwfvLAFpsEnizfy cotHLj/ML5/xbHU1rZQsaWecaP6c/wR48iLZC6W9Pf73s4QdYBEwuHv973N4dPGICLEk3YXsxwVjqtojSf4w6Pv5rQDE0mH8oq6Lmp3xzz6/gLitpO5SmCTw6rq6nYZqKxr5pb7Z5AfOZ2Dxg/NalzSCVkmP5h GL977f1cl4OZTDUABb6ot23PMV5J0nVpSMRYVl2NE/UI2mQsepVwXG5 5k1mb9nAv5evJJ6XGqzMNxi6Tx9u OZxXUZPWCJtRnYWG5obcoo6KBLKD/trK4fhGRneAscFC4thZYWUN23DEMEVhWFmp3ffok9pthOdJ4viieKzqi3c8ek7tEZt8KS98cjPi1NSFMU0FZ4HSc/E9Uw85eIouHvFaxzRbywDIr5r0VUOhQGbmCs4noEhEDIdDENRn8yMnX o/E0e2fBGRq14 HHvf139AndOPJWK2gtQyjd4BBgYqGN6YZIFicGkR5Z6JNkaW87KprcZV3x8Rj5BI8zlI 9iZuWDrGz BFMsJpUcw9FlF2Tcr5LQPhzQ726W1N1G3NmKiMWQgjMZ2/v6HPfGpyY6m4XV16ZcVy518TlsaHqIgwc Q9Dc/uQ3ABETK3TEDo/rLnqksCcTNjec9Qea69PiXJXir9c/7vuyU2K7ccUWf19RQfuNT8RsXn9iLt/6v5MJdxJTgMfueSdD1P38cgw4ipBzpSUAy8QpCLSLehuJpMP7c9cwd9lGAgGTP994NnsPK8uZRPm6mpzbq2uasW2XYNDi4LFD290rnTFsBZ7fGGCQVVZDhHHDc e9MyzcsBU7feaoUpjxbJ1N2C53PzObx392IYeMH8qsBauzXC2BoME5R 4HQeHKYw/kLzPnkHRdlPJ7AWErQFQSRAs8lAiqbQa8A2bYwE0bjAyaJqP79ea2t98jYbtYcZBcZr0Bzy9ZljGQaXse5fX1vL12HcePznaNAZw1fgKvrOq0HAJ Y5IrDLP9Gg0DwzJJ2m5Gg6REQdgDS2ErOHnoeK7e9yDe3bKG6z9 haibmaaBsGJbjmcjFdrj2B5PLlqG17awl0CktJXS4lbagnkMA0LiklCCq/yNCni7cinD8gbwae1GbOVgSYCIlaA9WB8IGyHGFY1o/7/JjvLYhrdQSuX8OSxv2kBD8z9RKvM6TFEYIrnaZRwVZ2XTO1nCDlAQKOWMIdflOCuTvnnTOCryGq6KYkgII21NeMetJGkvwzKHEAyMRimXJTU3ZVj4nooTd6pZ3/ggY3r9cIf5/bfpkVExC95Z5lvHnVG0i3oGycyuvRjQWJftEgCoq8m9VKwIWIGdrE4Fdkk4Q9TTdhGPJWlqjXPdnS90uf5KSXFezu2RSJBAapygT3E 3z11GuH06BSlEAfEzgzjy3DqK4XyFBN2Q9hH9OuFKRmmVpeUb/F7Bt8783CK8kOEUuW3TINQwMIoC/D4Z4u4/Y13 d3sD4kHXAiABIW8/CC//cZJGAihBvBCCjeicMMKNw/G9OtNUThEXiBA0DQZWFLEqqY6VtfWoVTXA5JJ08tplUdtm1tmzWTOpo2sKa9mxaqtGb2yAwcP5vyJk7CCghWCcMDyLXhDUdXYmiNDf0MkEGB8aR/ES1ntKiXq Q5YHogCQ3Hf0rnc8vFMSiNhhpcUEg4oLBEiZoA8K8Ddh55Ozm4KgAvKa1sVMnX3FdgtYTpHaIpA0ExrnBQ8vOozvjPnKf6x6h2e3jCXtXX5JO2OuPegEWBkwRAmluzdvm1VcwWB7diPfYIu91Ss4t7ag3mifgrlCX/A3lEGC1qGZkXetBHohlmbIoJl5LeLulIetdt RMXWg6mpu4qt1SewtfoMWpLLcFV2NJnCpqr1zd0ux3 CHmmxtzRGv9hCS16mqLnxJDec/UcSMZuDT5zIBdefTGlff4GkkWMHsmT  qwkinsVMHrsABbOWYunlD RJoeb3gyYFIUiJF2F6/mRG10RjSdZtnYr 4zOXAOjoTFKc9z3n3oGKEsQVxFAOOesqRmCdOFxBzBlzGD 9Pz7zF xCZXw/dEZuXqktMCvB3GgwDDZWtXEhNG7tv7GN488gFc/W4HbFj7ZxWUqwEnNYhzYu4jnbr2Y595fxOdrtjBiQC9mVKxhU3MTnUOXFSBK0RiP85e35hDZqqgbCl6AjhASz19f/fJDpnDmvhNQKE556NGMNWi8sKCSis62oSkCppB0s9cur61v5frvP0lEmViGSSBg8rOffJUDJg9lfs1mnqlciFXq4XoeSeVgRA2SCYUoIbgVEgO8jPqwDOG2Q47n0QWfZ15k0M0yveKew7Pli3ir7jMSyiYQMLECHufstT/XjD SomCYA8sG83HVpk49HwWukaMhExzbxLENrEBm69sWSQUQt6HFaaIgr5GA5fqNVQhaHYOhBWUETZPJJeOpirfw7U/vYlThIM7b6xhcFSUv0IxrC7YyM6z2okCcfpEm6hwAk3o3n1nN4zhGrWBW/TjWxXtzaJ 1mJ2eHUvC7FN6csa2LbEKZlW9RmV8K6MKxnBMvxMpCZZmXe32aGp5gNbY80ACpfzfVyK5ALfpd6guQhYt48v5Fq8eabFPnDYma4Zilwhgdiiw4bko22Hr lq2VTXyxuMfcc0Jd9Da7Pv/LvvBiYTCmTNAQ EAV15/Etf/ muE84LtvQLXEBRkfMgLctEtp2MNLkJ1qv325a9T20Uka7o wPOvLCDpeSQKDRKlJskCg0SJSbTQ4Mgjx2YdP26vMq4  VCKCWK4nSz11HdRIK7CSCjMhAJPMbDsi88abWNEv178/dIzGNq31I/hDpioUG4Luai0I/KmtCDCZScdxJ vOYPTj55ITTx7sg6SsmYBx1Ms2lJJwvRww/ii7uJ/FDiu4t4PP WJBYtY39BAwMxsbb2A75NX O6nSNAiPxTgr9/4KoEc7/QUBcFGwc4TEgmXaCxJY1OMG3/xPBXV9XzzraeoT8SIOjYJx8WxFbblokRhtUJelUHJYotQpWBEIdAgFK8OcHT/ESyuqcrMLKjarzf9pikULUnfFZX0XBzl8vLmBYQt30777WEn0SsUSdkMChGFZXmEArlbV39Zhux9vr3jZ2xaQiDYQtByMVJeRv vh MVcNXI83lu01zervqMta1bmFk5n8s/vZN71v6dklAzA/MbybcSfumV32D0DzehyBRMB5MPWkexPtEbG4vPGvbC9gwcz8D1DASLSaWnMTS/Y/xwRdNS7ljxM bWfUB562rern6DXyy7gZpEdc7r7Yrmln 0 /k7SOIk3qUwuDdtlpqBYrjlcmjIZX rEq/lH6guJk3tKbrFYheR6cDd Ff D6XUHd2RbjrLPlnL479/hYo1VYyeNJTz/ 9Uho3NbU32G9ybM648ln//423iUb/CQ3lBDMPASTrYKZ 4aRqIZUAkiPKguFc zVX1GW4c13FpaWjlod 8SnVtC9uqmznqhH3YurWBDWur6T olAu fQxTDx/Dk/e9SyJu w uAGHL90C09QgMIel6PP/0J1glYVzHwUxzLSrACUObidJi21lvDwJYuGQTccPDC6Z YanfpGfCb/45i7/dcm7G8SvKq/je7c8QSzq FZar0tpaHgO8oJAwFC2xL/4ii4q6Rl76eCl1LVGmjR3GS9ddRNy2CVkWv3n HV6cvQQV93xXgwFWvsm3Tswd75xwnC4HKc0kWFEPu8jw69skowXNGDdU8MT8RewzsCw7VFLAKxSOnzSSaQP2oiAc5JgJI8kLBXn4vLO4/NmXqGppaTddzYSgQtA01CQ/6JFfmWrEXY8/v/1 h0crKeCktdwBhZEURAniQP4WC1JDPJGQxZzlGwhbVnsPQYnKboFJ 1/Acf30TdNf8XzRtgqm9BnGkMJijt27lJkba0k6ghVwCYVsYg15OPUR3E5VEAyAFXAyMlMKkm6HPHjKIT9kZ/nJRWBd61b sOpZ4mkv0/Dw8BRUxoIMyk9gCpTltZJ0YmxqLSFsOgSM3FZwVAWRlDQ1ORHeqx1Dn2ALluExoWgah/W7PK2cikc3PJDxIg9HObiuywsVT3HFyGty5pELL0cMfBv79rmVz2p RNypZEowSr4o/2eqqqHlT6jk 1D60JdmTafdFnYRMYG/AscDFcCnIvJvpVTuqXm7wMdvLuJXl95HIubfvKpNdXw6azF3vnw9oycNBfwf1qr55dgJh7EHjuSSn57JxMP2Zsa/3iPemuSosw7kkOmTePKuV5n55ByScdt3h1gWkrLum ubMUwDOvnnEy7MeG5 u7973cqt5BdGuG/GDyhOC0FcPG992kCqtFUQnfuSzU1xTjxpMs/OXEjScpDUL80zod0cwteoG/76Cn/84RkcMK5javbgAaV8sn5z9uCswKLVW0jaTsasz3ue/oB40mn3trRHbUibteZboqiORsLxPH78 5f41x0XsteAzBjmrnhnyVp /PAMXM/Ddj1mzF/B3oP6cv/VZxGwTK4/4ygaowneXbKWoGWSdFzOPGRfzj9iv5zpje7bm0ggQLTzoKNSSFIRqQa7UBH0DKyYQlTmUEE6tuuyub6R/oUFrK9vyBB4z1Ms3lzJVYcexPgB/dq3jy/rx/tXX8a v/0TTtxN b9TmELrAIO8ag/xwLZdaltbcSOphb8cg86qnOinCFd7GJ26aiJQk2gmEWmFROe1z7ui4xjXNVCBJI22HyzgKY8Pa5cRjHikD/ Hi6PEW4OEvDBR2yZkWliGsO8omyrPax8odT2D1niQuB3AMDyClodh JZ/7pIIa5s3545A8jIlxjI9 kSiOMrEVYKVI808swjDCIGbICAOluFRkywkYAQZnJ85aN3qtlBvZ6 jr1AsbVqAUh6SY8mCXERCR9Eae4nOA0KWOZC84FimDXqF1pYHCbf ESOjpxEHeyHYCyB4wE7l9Z mO1wxBwJrlFLlyu PPAmc1g3pAn6L/LcfP9Eu6gDKU8SjSf7xs2cBWP3ZOi4Y831 /NXfccP5f O0Mddz1Qm/IZQf5qf/uppfPXstJ3z9UApL87n8tnO47 NfYg0bgDugL6pvCV5BGJWKGLBzRbiEwxmDmK7j0VTfypUn3cXG8hoeeeA9rv/Ow9Q2RTHaBlBVbpUxTWHqwSO4/MxDGNq/lLxwAAmahPIC/mBqJ7FOJB3 9mzHOjFKKY48dEzmuEB6feGLTDor11VlHSf4ESLHHrw3h04eTjBgZv0wbdfl2TcX5synM7bjctNjrxO3HeyUKyqatFm uZqXPvHb FDA4s6LT a1n17K3799JrN fgU/OevoLldKNA2D359xEoaioy5TDZAbFqIDhGDSYGRBCQgUbCZVL7nrJhwI8K9zz2Js3z5pafljClsbW7jwkWdojMWpamrhR8/MYOrtf Ww39xLICEYXnaEhgBuagwvEg7wlb3H f84XfysBJzC7GtVCpZaFUh EiPfhrbVyb1cl6IwrGw3o6OS3L78MW5Z9AhJz8numeBHu/Qf0sJdx5zEJRP350cHHcZb37iE0aV9CRpCyHSxbYO65nziqcW8PM8gnrRwPSHpmlmPtO8OUqguGqPcXn1/W4tTgNFpICpgBDllwLn0Ceazd0EVk0sqGF 0lf1LN9I32MTxZQdlHB80Ql02g0ol LT2gS72ZlNafAOGUQztzaGJSB69S3 HiCAi5NGIQY6erLJ9cf S0B3CPghIf6NtRWpbBiJyhYjME5F5NTW5Q/VyEW9NULs195ttVi5YTzKe5PqTf0N9Y4JkOIJrmGAYrF9dxc3fvJf3XlnA/FmLWfrRSlzXI5lwuOab99Mcs33r2DAgHEIVF/ihcoCZvj6I2fVEpYbGGFdecA PP/g n89fz/oNtTghC5VaE0Sidoa4G6ZBQUGEA6aM4Hvn30vdrHLyljYwJa Ys4 YhNmFwK3f4q guGpNJRde QA3/uJ5gi2KQJNL5371oL7F5OdlRgwM6FtELgxDuOXK6Zx29L45F8hyXUVF5c69zHnJxsqcchpPOrw6P3PBrLxggMGlxRTl7Tiy4ZDhezHaK2r3m0vbxxCcCHiex8qGehKFflWbsVTvoxMiwtFjhrOktooDhw4iT5kYNhhJMFI66Xoezy1cytn3PMbrS1bRkkhSH42R7GI5ByX  aGQxZhRZZxx6CTOGTUxNRU/ 16GLYupE4YQtEzCAYu8UIBQwOKsUyaxurEOhSJQYhMcECXQJ04g4rs gilff9Aw/A5goLOwC54Skp7DBzXLuX/tG0zpPRKj84AwBof3G8f0kWP46WHHcMyIoZz17j28vrEC2wPPg8ZYXqey /4g2zaJJoJ4noGRko22qKKWZICYky36ngfKziyDALYyCRoBItYAvjrwfAqsIgSDAquI0wddyLS x3Jwb4fiQBxDwBKFJYqh fXUJddmpBc0gkwumZJV2wYefQONLK5/Fld1HWaajmUNYVDZexQXXk0oeCgF d9gQL/XiYSndZTf7A9kv1oQCYKx61Fk3U13 NhzqVG2naHUfcB94M883dnEg EAVtDCzfHS3OLeBfz uw RcIG8SJZ/K5lwuOPqfxGMtqCUIpIX4qs/OYvmxlimZSziC3gwQF5hmLETBrLoo9WIIeQX59Nkp8Ik26zwVIy6F7HA9TpcJ8pPS/KDSNTGKwzhmYJ4vpvAUR59BhVz160vtMfCe0mX1fM2surzzbgj83JGyQwpK6GpOca1NzxJa2rMoM2iC7Z4JIs6GqKDRg7mxTc p6xPIVMnD8MyDS4961Buvvtl4mkDseGgxenHTiQUtBg7ogw7h3iFghb7j 96db6M 2SZXUYihQP YHNLNMFt97/BhwvLAaFvaQE3XXY8U8Zv/72Z/QeXsHpjU05D3LVAPEgWCSqgfMvKTfnc0zj/oEmc/PSjiEDScUnmuZhRMJMddRezHT5YvZ7WRDJjIlGuqwoYBr2cAONGFXLisRM4 cRJGIbw86nHU2xG Nv8T3JORrr7nFNp/UqSdxav4aFln1FhNfPAhs/8NzeZYcziOGKABH3xDgddrhh5GDXxKH0jedy/5gPiWQa7Imj5989WDk9v JCHD/4hV8 7L/Wu0yQRM0iBFebasaf4ZyjF9z5 mvqkPzgdbSogL5hIxbdn4ykh6Vi0xCJcNX4K/1r3DransF0LhRBNBnFjFvkFcZQSRBQtTRG2VRdREozTp9B/3svCoykLBZjSax m9z cPCvCUf2m4ygbSwKICM12HVtiy7NcP65K8mHNc4woyFyH6IKhl7C66X2aXDPlZhRKrVb6BlpQBEm4zeRZO dONM2 lBb/uOsDwqdA8 86PRQChCB83E7l8d gO4S9Akj/9Q mfVho9zEtk6988whmPDQ7wx0Tygty4gWH8fhdMwBBulgqVRkG8biDhEIkYx6P/vxZvNIiKEpNBVYKkjY4LoZpcMZlR3HmZUfx F9mMePRD2moj0Ko00SlNoE30lwnaR4AD7CKwrhhsy1UONXlV6xeU43VyVWiPIVKOAQbHZLFVoa4h4IWV541jZnvLMuYbKQEEkUGdp5geIpgg0fQNXj1vaXI7GWYpkFhfoi//vI8pu03gv 79Dj /NhsWqIJTFM487jJXHXe4QCU9S5i uHjefPD5cRTrigrdf5px ReE6Uz4waXURgJZa1LEwkGOPtQP40f/v5FlpVvbZ/pu6Wmket /yIP//IChm7Hj3/I2KF8tGlTtlBKqiIAZdF LwzPj9duc2X/9owTuHHOW0TT10YRcPPAcBTiCcpUBCIG9bFYzgHroOW/Os/x/GiUUybuzU9POTbrRSIiwg/2P4w1dfW8t2kdUcf2Vyw0DG477DiKQ2GKQ2HmeZVsCDSR9FxoK5ZnINEgZr7/nEfMAF8bvj/XTPSn5yuleKd6JcsatqbNFlZYptsu7OAPXD6y4nOeO/x63tz6OeUtVYwpGsBx/ScSNv1neWNrPVtiHWu e8qgOb79dYEMDAbklXBM2SQeWvsxybQ3Jrm2SXVVMabhYQVcHNvETQmt27gvp 09lIN6H0KvYPZ9FhEC0vEba3UaMcXCyWFp1yUq dmS 1jXuoW98vpz/tDpjCsaxrReZaxvXULSMwkbNoFUN8ySEGFz16O70lFeE25yARTehNF6D7ip9 2aeyGldyOSw5LfQ3SHsH8KjBaR4cBm4DzgG92QbjuX/vRMYi1x3np6LlbAwnNdzrrqeApK8jBMww9tbLOkO6NAwuEOa16BUd/sD48URJCGFt/qBsQ0ePXRD3n31c pqqjHSa3pgutmu2SUwogm8IJWhqinsiCWEvWOMilUQCDp4RqCmaN3H6z3IxOSRSYYUNa7kOvOP5pD9h3GXz8ub19CwDOgZYjVPtjqego8MJrIGPiNJ2xuvesV7rnjfE46fAInThtPU0uMgrxQ1uqEP770eMYOL OZNz6jNZrk8ANGcMmZh1CYv3MPq2EIf7nsdC7727M4rh/DrVB89cDxHLvvKNZtrmPF qp2UW/Ddlz 9fInXHHONMqKC7J6XQCn7zeBv733MS3pa8S0hR611Xv2ODLiQV4gQK2T/XLkNpyggrCLV jhIKxoqiZX6JBlGPz9/NMY2qeEwlCofV34nHUhwt OP5WPNm9k5oY1FARCnDlmPCNKfFHzlOKVtStyLs8riSAD oQpCoT55uiDOWOvSe37bvr0NVbW1WErfxJdwPQIB5PkdYpWsW2T58oX89Opx3P6kNwRR7bnZo8ZpOLW2wfX0wiIxbCCXtw77Rv0CeVnnatcA0Hhur6gt28HwvTmpAEns7P0Dg3KucyFYLKmJU55dCkKqEnUs6RxDT bcDkH9r2cmvgPCRkd/m9LQkzpc2nWmjK7QqLlARJNd/guFzxESskrvQ/DGoaYA3Y7/e5mt4VdKeWIyHeBN/DDHR9USu364uE5sAIW1/7hIi679WxqtzbQf0hvwvkh3njsQ0zTwAaUbUMwmPlSY8 DpJ0tFkohjS3 k x2vCvdcz0a61porI i0gcnc/UGRKCtQen0mKug0UnUSfvVCCqQirxRChpboDnqh9OXFhCSUkLNLsGQxb13fI2BA/1JFvuMG8TLr31OLG6TKDE6Imjw/wYz3xLmX4 nWFleTX1jlNLiPAxDKCnKPWPVMIQzjpvEGcdNyto3d kGHn9zPtuaoxw5eSTnHbcfhXnZgr/3oL689fPL WD5eupbY0wdOYS9 vrraGytbco5SOp6ipfnLuPF1SspKYjw2wu wpSRmQskleZHePiSr/GT599gdVWtX VK2mfPhiyTPqUFbGpozEpfRAgGTGK2ndNpGI6YxAocP wRRazAJrwtkLHMQMA0GFxazJRhg3Y6nE1EmDZ4KNMGD83a56nU5LQcGBi8c9K1WcevaqzmxfVLiLsOYKA8/ UakYIOq7atI9nQlIfR1RrgKUYU9qEwECbWaUmCkGEyvmQgyxoqERFMEU7fayLnjZjCqMK 7df/3b2P5 6VbxJ3/cHecCRJ6/9r777joyrzxY9/nlOmpBNC6CUUBQFBiPQFRMCGqFhYO6jrVtu9e3VX1N1r2evVn2tZ13XVda9rWwu2VREEu4BuREC6ghA6CSGFJFPOnOf3x0yGTCbUBCLD9 3L14tkJmeegcz3POc53 f7Jfl3y2dZTCzouc xNGQbXk5tewXztj1DOLZRyMDE0YqNtVkJIT/ohnl87Uz WngrZ3V6gIUlj1MWXEua3YbC1lfSM6vpyyNO8EuCVfcBQYiNR saasp/TUb D7MBTrPksWut3wWSe2g1s4zsNDLqbaUffsYAHrv1pejabiAUrYrni17SKUDXBuJr4EkfyIgLwXDyzEODjrh7diLuY2doXfpgQ269lMUkRvSYGo3aXgaBYPxmn1FagaqsIXJcJ1rnZtC Xi33kUN70qFdDhs3l1GZse9xJYxRHVwt98rdtXyw8Fuqa4IU9u/Cl6s38tc358fX59dt2clbny3nxd9fTkYjNz89lsW4/skf5J6d8xot8qUBx47WYi pqOaqx17huRsu5sQuieVT 7TP581fXs6i9ZuZ9tdXEgKj47pcNnQgD384n9pwOH4Tz2db3HDqCNpZGY2df0FDumtRo plOZgQ7ODgLbEwQwaGUow9rjv/fc7eC3819N2undyz4CO 3LqJTI Xq/oP5poBhRixn7cMg4Ft27Noe KKpaEUozt3i39d44S4e/Fs3tr4DWE3gjZUtHBX7KTjRCx27MwkIy2EbTuEwya7a3zgWpzSsWCfYzSU4v dfD4/nf88EV3XE9VDQWZrnh4VLcG8K1RDni8jfvO2vosLhtEhLYcnv/2I72uKAU1mbjVVZenxXFqfadEhI4upfQ5sOa  Ia0n0crTnvklM6lyyuieMZAn1y0n1HBXH1Bcsx1Xu7RL68 5Xf980K/VGFc7bK9 l2273zd8w oAACAASURBVKWtu4R0ahtO39BuGW54Cabn4HsPHG5HdUmBrNwMbnn8Ku66/LHo71IoFP3fUDi1AQgE4lkpyufDyMpEpfmjM3A3thC7F9pjoYKxx siRb0PtmkZmIYi6Di4VuJlubGvcgGuxp/mIT/Lx YtpUTat0aneVA1IczSCoywgx0Ictsd5yQEEssy a9fTuSO371GWTiE22DZP wHz 7k2NW2TVZCud99KVpWzM33vQFoHMflqVcMgkQI2XuWuULhCGWV1bzy4WKmnzV03wesx7ZNMEE7e8ZYl0vv1FvadTVc9 ybfDzjp40e57V/L0s6mUZczZPzvuSFa6fyyLz5LN60lfzMDH4 ZigT /Zi1terydhtsjuj3rZbN1rn3Y1EoO786YIKGOAonFx4YPzpTCjohecAWurV2VRVwbkzn6M6HEYrqA6HebDoczZUlnPP6D1Fq/5n9ATOf NFwm6EYCSCz7TwWxa3Dx8bf86l855j2bYSItrA8GqU7WJamkh4T9aNoWwC1V4CShGMOPhNC7/Hw 2FyQWyGjo5ryvvTbyON4uXsK2mkqFtujGufe94az /te 16TFte7O eitPr9tEyHXIyAlgeyNUl/vANblh4Ciu6DeINDu5mN6B6JU5mF6Ze/LCn9twG4FGOhylm/5oD9dmonWEr7ddS3lwMa6uJc upfHDG2g3 Srxh CoDuwAg8b0QSkVr/MMoENhqE3cGqwDAVyPjZkVbQCgDAOtXXTYSdylCGBbaMtEhWKPuG50jb3uJGEY5OZngWWyY NOsMOQHotOSqHCbjRC1V OqcsN0 A4LgOHFLChLjtHKbTfi9MqA2vdViaM7EnvBrtqy8truOXml6iuDuIPG1T57ISNT5FWJnZEYWtFbSCMx2NiGgZ33HAmW7aW8/asJewoqaRwUAHjxvbB26CdXdiJcOsDbxGoN6uuu1mrjNjNyZhgOMKnS9cdcGBfWbydT5auI KJ/jvV1ZZyvNEGGvVPgAooq6zhsffms2ZTKZ1aZ3PRqAHxJZ35a4obzTapCYUJhSJk40GXR9hZVcXS77Ywqmc3 nTMx1dh4Fa5hDM1WoG9W5HuWPQ7sR2fu8WEwhFUpRUbg8KNwM2z5/D0uekM7XRgmUFbdlTwswdeImOjQ4aCQB5U9IRaHF5dvYybTh5Bnj960/743DZ8 OOreWHlUlbt3MGA/Pb8uHd/cnzR36NbPn6Pr9fvpC4j2Q2aGD4HM93BMF3c2Dq2ZRi8PPEKPtnyPat27eDE1u24sMeJZHujx9lRu5t7vp7LvM1rMJXB5K59uXngODLt6NVWvi Tnxw3ikP1yY7lhNx6PQH8Ybz MOmmlxHd2pF iEG9MRd0OpV/bHiXYL1dpl7DZkqnsc32GgAlNR9TEVyCGysvUBYxyVBuUs0arQNg9WnW124uR31gNy0Tw4h EOu4gcZvlumq3ei81vEMGu2xUa4bXYuvu1RXCjczDbW7QTW3SGRPWqPjULYr2o9T2TYm4NaGcP2e6Po9YO2sxrUVbk4s 8bVqIjG57UYM7YPi77ZmLh2H9uh6nbKo0uv5C4us95dEt945Klw8fkjBPLMaO0Sr0WnvBweue0cVqzYzJKVm jQNoczT nLmm 3ceOvXyAScXEcl88XfsdLM7/ksYcuT6jxvmTV5r225TPC0bTC FCBNtl7vwr4ZNk6Hnj9E4p37MJrmLiR2GnXjG7Hd2IXOKGMRlZHNJhB Pv7RYScCJZh8Mr8pTx8zWSGHd VnHQ/2yuTK29GXM2Nf3uLst018WWaf362hKUbtvHMdRcxtm8PPl6 jkBprJyEYZCV4ePWCeO4dN4/2VkZjNVC2zOiWsdhxry5zL1y l7fa53q2hBX3/485VWxS3at8JVG68OUDgaPabF2V1k8sAO0SUvnhsHDk4713a6dvLZmBQ3/dtyAheGLYJqKNOXBUIoHR5xDv9x29MtN/p2pdcKcO fvlAZ2x0 Gr36/lKVlW3lj4vRm2f6e42n89yCiXbLtxu/nHKrzOo2lwtnNm5s/xlAGEe1yZvuRTO0ysVlfp7Tmw4RqjqWuTVvt4CVaRqBujlaqDdZum0iftq/gt3s16xia6qgP7LbHYviZA1nw7uI9NV4i 1hTdhyI9bZUTgS3VSaEIyjHAdNEe6NRRzVa9jcWoIIhnG210DoXZShc2yScmxbfn2 vL8WMzXbdshoirTPRfpvsnDTOm3Iyl1w6gtMnNt6pxfV5GD81 cO fn0podg6twIytjmklTqobJtLLhjE1VNHopSi05hsJo45IfbX4PKH 99JaMgRCITZsnUXM98o4vKLR x5a 7e9g4m3y4wTMX84o1MuPMpzhval6vHnRxP 1uwagM3P/0OgbATrffuNMi UIAJXgzCyo1nYUDs3Bq7qqnbFOS4Lk7I5bbnZjPnv3/CVWMG8/uZcxNSEm3ToEebXDZtK09Yew85EVZvLmHJ q38z6Vn8NzHi3hp/hICIYex/brT1c7kpzc R5ZXUdo/uumpoQ3l5VSHQgn9UBsz 7MVBBqUPlBaYQY0nnII5Tl0ztqztFFaW43HNMny7LkJHY5EeGTRAp5cUtRoxgyACllM6X0CFxYMon9u 0YLldV5p3gFVeFAwhVOyI2wrmonX5YUMzQ/ cbuwZradSSLytYSqNfKzkDRMa013TLy9/GTB08pxfSCs7m4y0RKg W09ubgN5tevrch22xFNDTGstBQLAv7yTdc8swQLi7lrqKWMFDBdyW/oEebp6gKLcdvdSLT07/Fa8Yc9YEd4PoHL2fbhlLWr9xMOBiOLps4jZQGALD2vGWlwROJ4Gb4sCyDUNBBu7GzsqGSdnXGhZ3o8kxNDTojPVpbvV5QCHdshb2jEiMYRrmQ5bHo0b8zJw3pzlmTB2FaBn6/h rq5K3Jfr HrEbWxPv06cBnn64mUK/Jh GAtxpGn9SDDxas5rX3FlMbCDNhVG/OO20gmzbvItRITnYoFOHDj1clBPYTe3dsdCOOz2uR1Sad0poaTMOgJhgi7IWaUBBCQZ6eV0TRd5v42y8uQCnFo//6PBrUY3 /jf16 2yLKSP7M7J/ATfNfIfqimB0FuQBI0CjN6SraoNsLqvgrIG9Wbu9jGc /QrbMglHXE7q2p4OGZl8W5y8o9nVLmu2lDCwoAPTxhUybVy0KuDiFZv4z3tejVbPDIERVriNxAjLMPBa / YrN1YGt8DkPR AwbDOhfQISOLJaVb Y/P3mbj7nI0mpPzO/PQqLPJT8vg5o9nM2vdmljmS Ns0 CmfqeQ70/uKdrQsrJtibn7MRFXs7q8pFkC 5DWvbim5wSe G4OtjKJoGnvy GBQdOafOy98ZleOqUdvl2eHTLPo7jyWVy9599Bo9jhWgQIJdSe0VqzK1TMF5snYSgb0Pitzgxs9394zAPbFHU4pERgz8xJ55F5M1jz9XreeGwOH71RhLOrImE7P0qhWuUkbGTy m2m33ga488v5Pppf2PLxjLQOro8blsQ2UspzrrZVG0QsjKiNV7CEeyyalQs8GqvjdMq liVhsVfF7NixRZmvvgFj/7tKs6efBKvv1aU1LP0nPOSiwgVf19CbpYPn88mFHLidWs8HosT nTgX58s592PlsfXx9dv2snsT1by  vO3GujDr /Qelhj8XdN07itw 8hdaasBPB67EZOag7/339WWzcUc6sRat4 uMiwvVOFkHHYdnG7SzdsI0B3dqzYUfj5R/qM02DHw3oztDeXZjb42peWbSMT9eup31WJkuXbqa4JLmMgas1aZ7ozsQbTh/J9DGD XbbTtpmZ9ApN5vnP/6a2faa Ekl/lqGQZe85NZlM2ctSiiJnL5FU9WFhPsWXtPk3D4nxG8m7svxBfn4vXZy5o9SjOjdlYfGn8WOmt1cMudFquvtov5iezFTZz/PCxMu4d11qxNqxTdkKMXfx110QEEd4LjsNvhNOyml0TIMCjKbL hc2m0053QcworKjbTypNMzo32Lz1ibIt0u4ITWd7Fi5x0oomVDDeUh0zRw3cQl2pA2Y/P6EJFY6d7q8FpWlt7CgLZPHumhxx2VgT3iRJj93KfMefYztIaJl43ktMt/xPGDCrjlqZ S27MjM/8yj0hpKQSCYJqo3ByMzMyEDJcLrxnD2LMHUvTJarZvLsN19 S06zRvNAe /sYjraPHqxPrXYrr4tlWEV1Hr3ssGIb0xClgKOjghCP89U9zue2uKezYUcmnn67GY1uEwg4jRh7HtKvGxJ9fvTvA7/7jRdYs34wZa5uW3zGHiuogHo/F6WecyMQz zP95mejLdXqXjrksHHrLlZv2EGH9jms37AzYbu/bZtUhcLccs9rTJ44gBGF3VFKMWxgATP/dA3vz1/N7uoAQwZ0o1 v6Ie0a7tW7AoEGt2VGXFdlm/czoBu7emSn8OK4mgdbB3b6ZK0EQY4qUf05nCG18v04YOZPjx6QpuZ9w33vf5RQvkD01D07dKW1ll71qez/D4GF wpSXT2yX34y wFBMNOfMJvGQb52RkM6ZVcsmBXZU3CeT99M4TTINAm2mRDGYrhnTrzu7Fjk362MROG9 GJV YTDO858VqWSfdOrfnLj6eglOLJFV/iuImBO6I1O2qreWf9Kjym1WhgN4iu0d87ZiJD2x34LHtyt748uOwTghEnvuHHUgZt/RmMbLfvdMiDlWH7GNL6h7XO3BTtMyeRnz6OXYFFGMpDjm8Q2yofZ0vFo2j23MMLN9JNR NQVrsAx93dYo04jrrArrXm9xf/iaWfrSIYq5uyfuUmFrzzNXe9ehNKKdp2ysWbk0HI5234wxAMQSSCx 9h b/X8dLjH2CaBpGaEMoy0Rm eO0YnZ1BuqGpKduNjrjRoF6vI1CXvp2pCLhUVIfiN00TWMnVGl1X89WX67Askxm3n0tJSRWbN5XRsVMubdpk4rqaYCCM12fzyB/eZtU3mwiHI igQ8RnsXVnFcpQtMrLYHBhAavWbcdspNRwIBjmi8Xfc/cdU7jxlhep3h1Ea00g6OAozfebd/L9ljIWfbORsyf05/qrxwGQm5PO1DMHNfp33yUvB59tJc2KbdOgY2600NivJo3ipiffIhh29qQWxgKo17ZQCu79yVkJZYXrO29YP77ZsI13ilZim9HagG1zMrlv2r53Lmal Xj2hh9zxz/nsKx4GwrF4G4dyLP9XPb7ZzmuSz7TzhpCt1jpgtEn92LFt9vis3bHG22gYVWD69H4MNm6cxeh0yL4rL3vMq3j99k8fdelPPjsh8z/eh2maTBxRG uu2RMfPb6feUugm5y4Nax/0KNLMEYwLAOXfjLxMlkew9uy3qG7eXV8VdyW9EsFu7YgIFifMde3Fl4ejynXuydaaSRl7YnY6hd5jWU186jNrwGV1dj4GfvPR8Vrg4CLRPY1UG1kGsmhYWFuqio6JB dtmCNdx2/oMEGqxP 9K93PnyjbTv0Za3nvmU1/ xIKEXZXx6VhNLgzQNTL83odOSBvBY6Ng2ettjMuXiYUw8vR/Xj7uHmqp6KZS2jTc7nd/8 Uo mb2Mj95clDAeDUTysxrdqJSdk8ar7yQ2wI04Lk//5QP 9XoRoaBDfrtsSraU48YCtuO30LaZcDyv1 LaX4zjsZc o6Y2cdnIMg2mThrMLy4fQyTisnhpMQuK1vHm 0sINDgJeDwWzz48jQ7t9t1tvaI6wBn3/I2qwJ7XMpQiPzuDWTOuwor1cI1nxZTsIjfDz1mFfUj3esjy zit8Hhy97L7tb7NOytYsXE7 dkZnNjt4C7tg2GHVeu386sHZhJyIriuxjAUXtvisV9fQL8e7akNhLj6lufYVlJJIORQk28k7TvwWCZXnjqYX549ci vdHBeWPM1dxd9kLTu7TMt3jjzCh796gvmrl bsMbut2zeOf/yeDmCQ W4LoZSEtCbSGuXysAnVAUX4bHaUVq7mO3V76BJPCn7rW4M6/hesy9JKaW 0loX7u95R92MfdnnawgFkte g7Uh5s38gk9mLyfiuETCkWhrGOrykxoso9h2Uvs8BeiQA2kay7ZIz/Ax5ZLhZLdKw8rJBCd2HNNEGdGbrf/3v29z9lVjWDDnG4K14YRjEQiDz04Ixh6vxZmTkxtLPPrALOa t5RgIPoLsm1LOfVTRhoGdYBQyGFx0Xr8PpvaQChhacE0DSaPPzH 58EndeODhWuSgjpE49miZRv3G9iz0338/VcXcevz7/H9jjI0MKBre 699Ix4UAcY3a87o/t13/uBDkDH1tl0bH1oxZu8tsUf//lxwnKO62pqg2Huf EDnrn9Uvw D0/dexlvz/uGd avYFnVzqT8 JAT4YOla5stsJ/XvR PfbOQsFsVz3rxmRYj23ejd6t8/njKmdz/5ae8sHIJNeEw/du05c5R45sc1IEDuk8g9k8pg2z/WLL9YwHI9o2nLDAfx63C1QEUNoay6ZP3hxa9z3DUBfbsNll4vB4CNYkzdttns CDVfFWeAAEQxi2QWaGn5qqWsKxD67Ha MYqtEbi4ZS5HdsxdDRxzP1ylHk5KYTDISorqxFNZIdsa24jHHnDOL5R96PZ9UAoMAKhImkedFa4/XZaK05aXA3Lpseraq4ZUMptdVB8trnMOfdJYRDDYJurLHk3q6ptIbi4p386a6p3HLv65SU7cYwFKZhcPt1Z9CpfWIz3 wsP5ZpJFSJBDBiOd0H4vgObZj5X5dTXl2LYSiy/C1b0a4yEODe2Z8wa/lqXK0Ze1x3bj19LKvWJzcXAVhZ7/tpfg8XTRrMyOE9mXL3P/Y03q4nu5GaOIfKb9m8ddaV/HHxp8zeuAavaXFxrwFc2ze60ctjmswYPpYZw8fiai2z66OA18pnWMdZbKmaSXng36TZBXTMuhi/ldSS4og66gL76HMLeXLGP5O bxhGvMZ5fW7YxeOzmXzlKOa 8iUAp14whG9XbeHLD1Yl1RC3vRY/ dWpjBjfN37G9XhtMlulU7EzeWNM 66tScvw8fDr1/PXu97i3x vwrJNxp49kFHnDMKf5sU0DbZu2UVBj3y6dM1j 6Yy7vzp02xaVxJdH7dNVL0bg/Upw8DUDVv 1r1nRe/eHejaMZcXH7mK9Zt2EgiG6dUtP6l6I8BZ4/rz8ltfJQV201QMHXRwN9Ny0vdd4vVIcLXmsr /zPc7d8W7Ns1d R2LN20lze9hd01yOmm6PzmnsWPrbHp1yGPlxu0JefB j8WlpzTetu9Q5frSuHvYadw97LR9Pk C tHDMjLpkj2NLtnTWnoocUfdGjvA6q/Wcedlf6a6Ipp6lJ7l54ZHp3P3L/5BOJQ862rfpTVTrxlNTn4Wg0/th2VbbPq hBumPEKoNoRTf0nGMPCleRl79kBuuGtK/NvvPv85T9z5RsJyi9dn89vHpjF0fN8DHrvWmqvH/oHtm8riVwxagZufk7TUopRi8NDudGyfw2dffEdpVSDhvoHfb/P4X6bTufOBX6p/tGANf3hkFkasW5TPa3PfbedxfI/knYs/dJ v3cD1L/ L6gYbg9I8NqNzOvHVkuKk5iKXTBzMz6ckL63sKN/Nz//8Glt2VmIaipAT4YpTB/PLSSOO6tQ9kVoOdI39qAzsEG2Jtn7FZtCabn07YRgG/3n w6xaXBztdhRjGApdU4PtRqI30Pxe7pv1G7r07siOLbt4/O63WDh3eXQJpV7jDI/X4pGZ19G1156NEHNe/oLn/jiLndsqaNclj2tum8zwiQdXuW550ffcPu0Jahvc/NWZPlRmWsKM0euzeeTJ6RT0bIvral57rYhXXv2SyspaTjihIz//2Th69jz4jRrBkMPy1Vvw2BZ9erWLXjUchZ5ZuIgH5n5GqJEUwSuGnISzNcjshavw2NFU0YlDj2fGtIkJ9wPq01qzelMJpZXVnNClLbmZzbslXoimStmbp3UMw6B7v2hxpnDIwVWa3/zpCm6e igVZdVorXHCESK1ASJVNfGljNrdAX530UM8veQ 8ju0omNBG7QyaJiOqjV8Pf/bhMA 8aKhTLzowCsaNmZXSWXjFX2rAnTp2ZZqFyrKaziuTwd ev0ECmKB2zAUF1xwMhdccHKTXh9iLe/677sd3dGgoHUrbNNICuxpts3x7fI4/4x /OqCH7FpRzmd8nP2m42jlKJ35 bdBi9ESzhqAzvAlnU7eOimZ1n xVqUgsJT 3H/K79i49oSSreW8 afZvFt0Y6En9EayraWU7xqC137dCQzOw3bYyUt4ViWQUZ2868j9z6pa7yYV30 v83ZUwqZdFnzZGAcC0b26EqbzAyCuypwYlkmhlKkeWzO6Hs8ALlZaQeUXilEKjk6r8GBmt0BbjrjPpYt/A434hJxXIrmLePms//IicN6MuGCIXstBqZMFU ZHHfOSY129kEpRozv1 zjzmuXw6TLRuJL21NUyvZa5LXLZvyU/V5hiXpMw DFq6YyvncPLMPAVIqRPbry0jUXk7aP1nVCpLqjdsb 0cwvCQVCCS3sIo5LeWklX32wnCET nPKRcPZsGozodoGtTJsk 6xpYi8ttnc vCl/O9/vBhPebdsk989dgVpGc1fOQ7gJzMm03tgV9565lOqqwKMOnMA504fja RjkRi31ql XnowknoWOVNySYR4igO7Bu/3ZaYsx7jhCNs/T5a5W/ST07lw5cXsHHNVgLVQSyPiWmZ3PLUzzDrpQMOGdubFxfezoqv1mNaBiec1DXh8eamlGL0pIGMnvTDa6l1tFJK7bXssBDHmqM2sPc8sQu dG9SaQHTMinoG90c4PV7eOjDO/jsjSK mvsNrTu04vRpY2jXtU3S8Twei4HDD67prhBC/BAdtYH9R5MH84//eYtw0CESa8pgeyw692pH/xHHxZ9n2RZjLxzG2AuHtdRQhRDiiGrSzVOl1IVKqeVKKVcpdUTv/Hl8Ng/P Q1jpxTiz/CRkZPG6VeM4t7Xb5INJUKIY1pTZ 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 pNEKIYTYr/0Gdq31 H09rpS6EpgEnKq1loAthBAtbL BfV UUqcDtwBjtNY1zTMkIYQQTdHUNfZHgUzgfaXUYqXU480wJiGEEE3QpBm71rpncw1ECCFE85Cdp0IIkWIksAshRIqRwC6EEClGArsQQqQYCexCCJFiJLALIUSKkcAuhBApRgK7EEKkGAnsQgiRYiSwCyFEipHALoQQKUYCuxBCpBgJ7EIIkWIksAshRIqRwC6EEClGArsQQqQYCexCCJFiJLALIUSKkcAuhBApRgK7EEKkGAnsQgiRYpoU2JVSdymlliqlFiul5iilOjTXwIQQQhyaps7Y79dan6i1Hgi8DdzRDGMSQgjRBE0K7FrrynpfpgO6acMRQgjRVFZTD6CUuge4AqgATtnH864FrgXo0qVLU19WCCHEXiit9z3JVkrNBdo18tAMrfWb9Z73W8Cntf7d/l60sLBQFxUVHexYhRDimKaU kprXbi/5 13xq61Hn Ar/kC8A6w38AuhBDi8GlqVkyvel9OBlY1bThCCCGaqqlr7PcqpY4HXGAD8LOmD0kIIURTNCmwa63Pb66BCCGEaB6y81QIIVKMBHYhhEgxEtiFECLFSGAXQogUI4FdCCFSjAR2IYRIMRLYhRAixUhgF0KIFCOBXQghUowEdiGESDES2IUQIsVIYBdCiBQjgV0IIVLMfjsoHZYXVaqEaJnfwy0PKD0Cr/NDIe83tcn7TW0H8n67aq3b7O9ALRLYjxSlVNGBtJFKFfJ U5u839TWnO9XlmKEECLFSGAXQogUk qB/YmWHsARJu83tcn7TW3N9n5Teo1dCCGORak YxdCiGOOBHYhhEgxx0xgV0r9WimllVJ5LT2Ww0kpdb9SapVSaqlS6nWlVE5Lj lwUEqdrpRarZT6Tin1m5Yez GklOqslPpQKbVSKbVcKXVDS4/pSFBKmUqpr5VSb7f0WA43pVSOUurV2Gd3pVJqeFOOd0wEdqVUZ2ACUNzSYzkC3gf6aa1PBNYAv23h8TQ7pZQJ/Bk4AzgBuFgpdULLjuqwcoD/1Fr3AYYBv0zx91vnBmBlSw/iCHkYeE9r3RsYQBPf9zER2IEHgZuBlL9TrLWeo7V2Yl8uBDq15HgOkyHAd1rrdVrrEPBP4JwWHtNho7XeqrVeFPtzFdEPfceWHdXhpZTqBJwFPNXSYznclFJZwGjgbwBa65DWurwpx0z5wK6Umgxs1lovaemxtICrgFktPYjDoCOwsd7Xm0jxQFdHKdUNOAn4omVHctg9RHQy5rb0QI6A7kAJ8PfY0tNTSqn0phzQap5xtSyl1FygXSMPzQBuBSYe2REdXvt6v1rrN2PPmUH0Ev75Izm2I0Q18r2U7qqITQAAAX1JREFUvxpTSmUAM4EbtdaVLT2ew0UpNQnYobX Sik1tqXHcwRYwCDgOq31F0qph4HfALc35YBHPa31 Ma r5TqDxQAS5RSEF2WWKSUGqK13nYEh9is9vZ 6yilrgQmAafq1NyosAnoXO/rTsCWFhrLEaGUsokG9ee11q 19HgOs5HAZKXUmYAPyFJKPae1vqyFx3W4bAI2aa3rrsJeJRrYD9kxtUFJKbUeKNRap2zFOKXU6cAfgTFa65KWHs/hoJSyiN4YPhXYDPwbuERrvbxFB3aYqOis5BmgTGt9Y0uP50iKzdh/rbWe1NJjOZyUUp8C12itVyulfg ka63/61CPlxIzdpHgUcALvB 7Slmotf5Zyw6peWmtHaXUr4DZgAk8napBPWYkcDnwjVJqcex7t2qt323BMYnmdR3wvFLKA6wDpjflYMfUjF0IIY4FKZ8VI4QQxxoJ7EIIkWIksAshRIqRwC6EEClGArsQQqQYCexCCJFiJLALIUSK f ddQ5rT xVWwAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11a886860>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "## Visualize Latent Space\n",
    "vae.load_weights('random_weights.h5')\n",
    "vae.fit([training_feature,ground_truth_r],\n",
    "         epochs=epochs,\n",
    "         batch_size=batch_size,\n",
    "         verbose = 0)\n",
    " \n",
    "[z_mean, z_log_var, z, r_mean, r_log_var, r_vae, pz_mean] = encoder.predict([training_feature,ground_truth_r],batch_size=batch_size)\n",
    "\n",
    "tsne = MDS(n_components=2, random_state=0)\n",
    "X_2d = tsne.fit_transform(z_mean)\n",
    "\n",
    "#%matplotlib notebook\n",
    "fig = plt.figure()\n",
    "ax = fig.add_subplot(111)\n",
    "ax.scatter(X_2d[:, 0], X_2d[:, 1], c=ground_truth_r)\n",
    "plt.title('TSNE visualization of latent space')\n",
    "ax.axis('equal')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}