1_vgg16_augmented.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Nature Conservancy Monitoring"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Set up training, validation and testing schemes\n",
    "Images are in Train and Test folders. The idea is to read all the training images in memory and dump into a numpy array on disk.\n",
    "So that we waste very little time when experimenting from scratch."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using Theano backend.\n",
      "Using gpu device 0: GeForce GTX 980 (CNMeM is disabled, cuDNN 5105)\n"
     ]
    }
   ],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import pickle\n",
    "import h5py\n",
    "import cv2\n",
    "import os\n",
    "import glob\n",
    "from keras.models import Sequential\n",
    "from keras.layers.core import Flatten, Dense, Dropout, Reshape, Lambda\n",
    "from keras.layers.convolutional import Convolution2D, MaxPooling2D, ZeroPadding2D\n",
    "from keras.optimizers import Adadelta, Adam\n",
    "from keras.utils.np_utils import to_categorical\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.cross_validation import train_test_split\n",
    "from keras.layers.normalization import BatchNormalization\n",
    "from keras.callbacks import EarlyStopping, History\n",
    "from keras.preprocessing.image import ImageDataGenerator\n",
    "import pandas as pd\n",
    "from __future__ import division, print_function\n",
    "from collections import Counter\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Create Validation directory"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "/home/trooper/Documents/Nature\n",
      "/home/trooper/Documents/Nature/train\n"
     ]
    }
   ],
   "source": [
    "% cd /home/trooper/Documents/Nature\n",
    "% rm -r train\n",
    "% rm -r valid\n",
    "% mkdir train valid\n",
    "% cp -r train_orig/* train\n",
    "% cd train\n",
    "\n",
    "# pool_counter = 1\n",
    "\n",
    "for fish_class in glob.glob('*'): os.mkdir('../valid/' + fish_class)\n",
    "\n",
    "rnd_ord = np.random.permutation(glob.glob('*/*.jpg'))\n",
    "for i in range(500): os.rename(rnd_ord[i], '../valid/' + rnd_ord[i])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## VGG Batchnorm"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "vgg_mean = np.array([123.68, 116.779, 103.939], dtype=np.float32).reshape((3,1,1))\n",
    "def vgg_preprocess(x):\n",
    "    x = x - vgg_mean\n",
    "    return x[:, ::-1]\n",
    "\n",
    "def VGG_16(size=(224, 224), weights_path='/home/trooper/Documents/Nature/vgg16.h5'):\n",
    "    model = Sequential()\n",
    "    model.add(Lambda(vgg_preprocess, input_shape=(3,)+size))\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(64, 3, 3, activation='relu'))\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(64, 3, 3, activation='relu'))\n",
    "    model.add(MaxPooling2D((2,2), strides=(2,2)))\n",
    "\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(128, 3, 3, activation='relu'))\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(128, 3, 3, activation='relu'))\n",
    "    model.add(MaxPooling2D((2,2), strides=(2,2)))\n",
    "\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(256, 3, 3, activation='relu'))\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(256, 3, 3, activation='relu'))\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(256, 3, 3, activation='relu'))\n",
    "    model.add(MaxPooling2D((2,2), strides=(2,2)))\n",
    "\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(512, 3, 3, activation='relu'))\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(512, 3, 3, activation='relu'))\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(512, 3, 3, activation='relu'))\n",
    "    model.add(MaxPooling2D((2,2), strides=(2,2)))\n",
    "\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(512, 3, 3, activation='relu'))\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(512, 3, 3, activation='relu'))\n",
    "    model.add(ZeroPadding2D((1,1)))\n",
    "    model.add(Convolution2D(512, 3, 3, activation='relu'))\n",
    "    model.add(MaxPooling2D((2,2), strides=(2,2)))\n",
    "\n",
    "    model.add(Flatten())\n",
    "    model.add(Dense(4096, activation='relu'))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(Dropout(0.5))\n",
    "    model.add(Dense(4096, activation='relu'))\n",
    "    model.add(BatchNormalization())\n",
    "    model.add(Dropout(0.5))\n",
    "    model.add(Dense(1000, activation='softmax'))\n",
    "\n",
    "    if weights_path:\n",
    "        model.load_weights(weights_path)\n",
    "\n",
    "    return model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Finetune"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "size = (224, 224)\n",
    "model = VGG_16(size=size)\n",
    "model.pop(); model.pop(); model.pop(); model.pop(); model.pop();\n",
    "\n",
    "for layer in model.layers:\n",
    "    layer.trainable = False\n",
    "\n",
    "model.add(Dropout(0.2))\n",
    "model.add(Dense(512, activation='relu'))\n",
    "model.add(BatchNormalization())\n",
    "model.add(Dropout(0.3))\n",
    "model.add(Dense(8, activation='softmax'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Train, Validate generator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Found 3277 images belonging to 8 classes.\n",
      "Found 500 images belonging to 8 classes.\n",
      "Epoch 1/50\n",
      "3277/3277 [==============================] - 135s - loss: 1.8742 - acc: 0.4544 - val_loss: 1.1314 - val_acc: 0.6640\n",
      "Epoch 2/50\n",
      "3277/3277 [==============================] - 127s - loss: 1.2246 - acc: 0.6204 - val_loss: 0.8689 - val_acc: 0.7300\n",
      "Epoch 3/50\n",
      "3277/3277 [==============================] - 129s - loss: 1.0054 - acc: 0.6677 - val_loss: 0.6905 - val_acc: 0.7960\n",
      "Epoch 4/50\n",
      "3277/3277 [==============================] - 127s - loss: 0.8529 - acc: 0.7202 - val_loss: 0.6134 - val_acc: 0.8180\n",
      "Epoch 5/50\n",
      "3277/3277 [==============================] - 127s - loss: 0.7680 - acc: 0.7479 - val_loss: 0.5316 - val_acc: 0.8340\n",
      "Epoch 6/50\n",
      "3277/3277 [==============================] - 127s - loss: 0.7163 - acc: 0.7531 - val_loss: 0.5715 - val_acc: 0.8480\n",
      "Epoch 7/50\n",
      "3277/3277 [==============================] - 125s - loss: 0.6584 - acc: 0.7711 - val_loss: 0.5128 - val_acc: 0.8400\n",
      "Epoch 8/50\n",
      "3277/3277 [==============================] - 132s - loss: 0.6474 - acc: 0.7836 - val_loss: 0.5544 - val_acc: 0.8580\n",
      "Epoch 9/50\n",
      "3277/3277 [==============================] - 127s - loss: 0.5790 - acc: 0.8041 - val_loss: 0.4358 - val_acc: 0.8820\n",
      "Epoch 10/50\n",
      "3277/3277 [==============================] - 127s - loss: 0.5966 - acc: 0.7974 - val_loss: 0.5069 - val_acc: 0.8720\n",
      "Epoch 11/50\n",
      "3277/3277 [==============================] - 125s - loss: 0.5415 - acc: 0.8056 - val_loss: 0.5170 - val_acc: 0.8620\n"
     ]
    }
   ],
   "source": [
    "nb_train_samples = len(glob.glob('*/*.jpg'))\n",
    "nb_validation_samples = len(glob.glob('../valid/*/*.jpg'))\n",
    "nb_epoch = 50\n",
    "bath_size = 16\n",
    "nb_test_samples = 1000\n",
    "\n",
    "classes = ['ALB', 'BET', 'DOL', 'LAG', 'NoF', 'OTHER', 'SHARK', 'YFT']\n",
    "\n",
    "train_datagen = ImageDataGenerator(\n",
    "        shear_range=0.2,\n",
    "        zoom_range=0.1,\n",
    "        rotation_range=10.,\n",
    "        width_shift_range=0.2,\n",
    "        height_shift_range=0.2,\n",
    "        horizontal_flip=True)\n",
    "\n",
    "train_generator = train_datagen.flow_from_directory(\n",
    "        '../train/',\n",
    "        target_size=size,\n",
    "        batch_size=bath_size,\n",
    "        shuffle = True,\n",
    "        classes=classes,\n",
    "        class_mode='categorical')\n",
    "\n",
    "valid_datagen = ImageDataGenerator()\n",
    "\n",
    "validation_generator = valid_datagen.flow_from_directory(\n",
    "        '../valid/',\n",
    "        target_size=size,\n",
    "        batch_size=bath_size,\n",
    "        shuffle = True,\n",
    "        classes=classes,\n",
    "        class_mode='categorical')\n",
    "\n",
    "callbacks = [EarlyStopping(monitor='val_loss', patience=1, verbose=0)]\n",
    "model.compile(loss='categorical_crossentropy', optimizer=\"adadelta\", \\\n",
    "              metrics=[\"accuracy\"])\n",
    "\n",
    "# fine-tune the model\n",
    "hist = model.fit_generator(\n",
    "        train_generator,\n",
    "        samples_per_epoch=nb_train_samples,\n",
    "        nb_epoch=nb_epoch,\n",
    "        validation_data=validation_generator,\n",
    "        nb_val_samples=nb_validation_samples,\n",
    "        callbacks=callbacks)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Plots"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAA4MAAAFRCAYAAADO/nj3AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzs3XeYlNX5//H3vfTepZdFQEBAQUUUkKUpioIgICAo2P1Z\nv4nRRKNiTEyMibEmFhAFBEFRQZAmuBQRGyjSUZalSpPe2T2/P57ZZYFlmVlm9pnZ+byuay5mnjb3\nPIlz9p5zn3PMOYeIiIiIiIjElwS/AxAREREREZG8p2RQREREREQkDikZFBERERERiUNKBkVERERE\nROKQkkEREREREZE4pGRQREREREQkDikZFIkQMxtuZn8J8tgUM+sQ6ZhERERiXbja11CuI5JfKRkU\nERERERGJQ0oGRSRHZlbA7xhEREREJPyUDEpcC5SPPGxmP5rZXjN7y8zOMbPPzGyPmU03szJZju9m\nZkvM7Dczm2VmDbPsa25m35vZbjN7Hyh60ntda2aLzGynmc0zs6ZBxniNmS0MXDfVzJ46aX8bM/sy\ncN1UM7s5sL2omf3bzNYG9s0xsyJm1s7M1mdzHzoEnj9lZh+Y2Ugz2wXcYmaXmNn8wHU2mtkrZlYw\ny/nnB+7VDjPbbGZ/NLPKZrbfzMplOa6FmW1Vgikikr/FQvuaTcx3mNlqM9tuZp+YWdUs+/5jZlsC\nMfxoZo0D268xs6WBz7TezH6Xm/cW8YuSQRHoCXQEGgDdgM+APwIVgQLAAwBm1gAYHXhdCZgCfGpm\nBc2sEPAx8C5QHvgAuCHjDcysOTAMuCOw/w1gYuC8M9kHDHTOlQG6AnebWbfAdWsH4n0pEO+FwA+B\n8/4NNAdaBd7zESA9sM+d4T27AeOcc2WB94BjwEOB61wGdAD+XyCGksCMQBxVgXrATOfcFuALoE+W\n6w4Axjjn0oL43CIiEtuivX3NFPhB9FmgF15btg54P7DvSqANUC/QFvcBdgROHQrc4ZwrDTQBZoXy\nviJ+UzIoAq8457Y75zYDc4GvnXOLnXNH8Bqg5oHj+gCTnHOzAsnMv/B+nbwcL+Eq6Jx72TmX5pwb\nD3yb5T3uAF53zn3nPCOBw4HzcuScm+OcWxp4vgSvcWoX2N0PmOGcGxd4353OucVmZsBg4AHn3K+B\n91zgnDsa5D35yjn3aeA9DzvnFjnnvglcZx3wZpYYrgU2O+dedM4dcc7td85lfPYRwEAAM0sIxDsy\nyBhERCS2RXX7epL+wDDn3I+BtvJPQCszqwUcBUoBjc3MnHMrAz94AhwBzjezUs653c65H7K/vEh0\nUjIoAluyPD+YzeuSgefVgNSMHc45B2wAqgf2bTzpuqlZntcGfh8of/nNzHYCNQLn5cjMWgZKZrYG\nyjbvwvtVFaAm8Es2p1UEigBrznT90zi5jLS+mX0aKAHdBfwtiBgAJgCNAj2YVwK7nHPf5TImERGJ\nLVHdvp7k5Bj2A78B1Z1zXwCvAq8BW8zs9UBVDHi9lF2BVDP7wsxCTUJFfKVkUCR4m/Aanaxq4jVS\nm/Ean6xqZXm+Hvibc6584FHOOVfSOTc2iPcdDXyC1yCVxSuBsSzXrZfNOduBQ8C52ezbDxTPeBEY\nv1fppGNOLiP9H7AcODcQw+MnxZDd++CcOwyMw+sdHIB6BUVE5FR+ta+njcHMSgAVAjHgnHvVOXcx\n0Bg4D/hDYPv3zrnr8drRCXhtnkjMUDIoErxxQFczax8Yx/AwXsI1H/gKOGpm9wf29QRaZjn3Lbyx\nfi3Ba2QCg85LBPG+JYGdzrmjgfP7Z9n3HtDRzHqZWQEzK29mFwR+VR0OvGBmVc0swcxaBcZQrAKK\nmtnVgUlg/gwUPkMMpYA9zrkDgUH992TZNwmoYmYPmFlhMyuZ8TkDRgKDgOtQMigiIqfyq33Nagww\n2MyamVkRvPGDXznn1pnZxYEqnYJ4PZqHgHQzK2Rm/c2sdKC8dS+gMfESU5QMSrw7uQfstBOrOOdW\n4fVuvQpswysLuc45dywwvqAn3ji9HUBvYHyWc7/HG9fwqpn9hpeQ3RLM++JN1PKMme3GS9wyf+10\nzq0HrgEexitnWQQ0C+x+GPgJb2zFDuAfQIJzbk/gmsPwynD2Bv7NycPATWa2B69n8v0sMewDOuNN\nDvBr4LMlZdk/H2/imoWBeEVEJP+LhfY1c59zbibwBPARXm9gIt44d4DSeEnnb0AKXvXN84F9A4GU\nwBCKOznxB1uRqGdeB0IE38CsC/AiXuI5zDn33En7ywJv45WZHQRudc4tC+ZcEYkNZjYTeM8597bf\nsYhEOzOrgTf5UmW8H1Lecs69nM1xLwNX45V+D9LEFSIiEqqI9gwGZg98FbgKOB/ol3XdmIDHgEXO\nuQvwfsl5OYRzRSTKmdkleDPGhTp+QyReHQN+55w7H28pl3tPbv/M7Gq8Mbz18SaVej3vwxQRkVgX\n6TLRlsBq51xqoJv/faD7Scc0JrAmi3NuJVDHzCoFea6IRDEzeweYDjwYmJlNRM4gsBzMD4Hn+/Am\nb6p+0mHd8XoPcc59DZQxs8p5GqiIiMS8SCeD1TlxivqMaYKz+hGvFpzA4N9aeLNGBXOuiEQx59yg\nwMxumjhGJBfMrA5wIfD1SbtObiM3ojZSRERCFA0TyPwDKGdmC4F78SbA0ExMIiIS1wLrmH2I17O+\nz+94REQk/ykY4etv5MS1YGpw0sKhzrm9wK0Zr80sBW+h7OJnOjfLOZGdBUdERKKGc87OfFRsC0xh\n/yEw0jk3IZtDNuKtw5Yh2zZS7aOISHwJtY2MdM/gt0A9M6ttZoWBvsDErAeYWZnA2meY2R3A7MAv\noGc8NyvnnB5BPp566infY4ilh+6X7pfuV/Q84sjbwDLn3Eun2T8RuBnAzFoBu5xzW7I70O//zWLp\nof8edb90v6LnofsV+iM3Itoz6JxLM7P78CaQyFgeYrmZ3eXtdm8CjYB3zSwdWArcltO5kYxXRETE\nb2bWGrgJ+MnMFuGthfYYUJtA2+mc+yywsPbPeEtLDPYvYhERiVWRLhPFOTcVOO+kbW9keb7g5P05\nnSsiIpKfOee+BAoEcdx9eRCOiIjkY9EwgYzksaSkJL9DiCm6X6HR/QqN7pdI9NB/j6HR/QqN7ldo\ndL/yhuW2vjSamJnLD59DRERyZma4OJhAJlzUPoqIxI/ctJERLxP1U506dUhNTfU7jKhVu3Zt1q5d\n63cYIiKSx9Q+5kzto4jEi3zdMxjIjn2IKDbo/ohIrFHPYGjUPuaO7o+IxKLctJEaMygiIiIiIhKH\nlAyKiIiIiIjEoXw9ZlBEJFrtPrSbsUvHsuvQLhpXakzjSo2pU7YOCabf6ERERCRvaMxgjLrnnnuo\nUaMGjz/+eK6vkZ/vj0g0cs7x5fovGbpwKJ+s+IROdTtRs3RNlm1fxvJty9lxcAcNKjTwksOKXoLY\nqFIjzi13LoUKFPI7/KigMYOhicf2Ec6+jczv90dE8qfctJFKBn2SmJjIsGHD6NChg28xRPP9EclP\ntu7fysgfRzJ00VAAbm9+OwMvGMg5Jc454bg9h/ewYvsKlm9bzrJtyzKTxI17N3JuuXO95LBio8ye\nxAYVGlCkYBE/PpJvlAyGJhbbR/C/jYz2+yMikh0tLZFPpKWlUaBAAb/DEJGzkJaexudrPmfooqHM\n+GUGPRr1YOh1Q7m85uWYZf89XbpIaVpWb0nL6i1P2H7w6EFW7ljJsm1ecjhu2TiWbVtGys4UapWp\nlZkcZiSLDSs2pEThEnnxMUXynNpIEZHw0eAUH9x8882sW7eOa6+9ltKlS/P888+TkJDA22+/Te3a\ntenYsSMAffr0oWrVqpQrV46kpCSWLVuWeY3Bgwfz5JNPAjB79mxq1qzJCy+8QOXKlalevTrvvPOO\nHx9NJO6t272Op5Ofpu7LdXl81uN0SuxE6kOpDO8+nNa1Wp82EcxJsULFuLDKhfRv2p9nOjzD+D7j\nWX7vcvb8aQ+f9P2EAc0GULRgUSavnsytE2+l4vMVSXwpka6ju/Lw9Id5e9HbLNiwgN2HdkfgE4uE\nl9pIEZG8o55BH4wYMYK5c+fy9ttv0759e1JTU3n00UeZM2cOK1asICHBy9GvueYa3nnnHQoVKsSj\njz7KTTfdxKJFi7K95q+//srevXvZtGkT06dPp1evXvTo0YMyZcrk5UcTiUtH0o4wadUk3lr4Ft9s\n/IZ+TfrxyY2f0Lxq84i+b+EChTN7BLNKS08jZVeKV2q6bRnJa5P533f/Y/m25ZQpWiZzTGKjSsdL\nTisWrxjRWEWCpTZSRCTvxHUymIsf6LOV22EFWccjmBlPP/00xYoVy9w2aNCgzOdPPvkkL774Inv3\n7qVUqVKnXKtw4cI88cQTJCQkcPXVV1OyZElWrlxJy5YtTzlWRMJj5faVDFs0jBE/jqBhxYbc3uJ2\nPurzEcUKFTvzyRFUIKEA9crXo175enQ7r1vm9nSXzvrd61m+3RuT+N2m7xi5eCTLti2jUEIhLznM\nMnFN40qNqVqyaq56MyX2heN/9rMZdqc2UkQk8uI6GYy2seE1atTIfJ6ens5jjz3Ghx9+yPbt2zEz\nzIzt27dn29BVqFAh89dSgOLFi7Nv3748iVsknhw4eoAPl33I0IVDWbVjFYMuHMScwXNoUKGB36Gd\nUYIlULtsbWqXrU2Xel0ytzvn+HXfr96YxO3LWbp1KeOXj2f59uUcPnb4lIlrGlVqRK0ytbQMRj6n\nNlJEJP+L62TQT9n90p512+jRo/n000+ZNWsWtWrVYvfu3ZQrV06zm4n4ZOHmhQxdOJSxS8dyWY3L\n+N1lv6Nr/a75YskHM6NqqapULVWVjnU7nrBv+4HtmbObLt++nGm/TGPZtmXsOrSLhhUbnpIoJpZL\npGCCmhY5O2ojRUTyhlpsn1SpUoU1a9bQoUMHnHOnNGB79+6lSJEilCtXjv379/OnP/1JpVoieWzX\noV2M+WkMQxcNZceBHdzW/DZ+uOsHapap6XdoeaZi8Yq0rd2WtrXbnrB996HdrNi+InNc4lsL32LZ\ntmVs3reZ+uXrn5Ik1q9Qn8IFCvv0KSTWqI0UEckbSgZ98sc//pH777+fRx55hMcff/yURuzmm29m\n2rRpVK9enQoVKvDMM8/wxhtvBH19NYoiueOcY966eQxdNJQJKyZwVb2r+EfHf9CxbkeVRWZRpmgZ\nLq1xKZfWuPSE7fuP7GfljpWZvYmjl4xm2bZlpO5KpU7ZOicsgdG4UmPOq3gexQsV9+lTSLRSGyki\nkje06Hwc0/0ROW7Lvi2M+HEEQxcNpYAV4PYWtzOw2UAqlajkd2j5wuFjh1n92+rMnsSMSWx+/u1n\nqpWqduKYxIqNaFSpEaWLlD7lOlp0PjRqH3NH90dEYlFu2kglg3FM90fiXVp6GjPWzGDowqHMTJlJ\nj4Y9uL3F7VxW4zL1HOSRY+nHWLNzTWaSmJEorti+gvLFyp+QJDau1Ji2tdsqGQyBmbm0NEdCwinb\n9f2fA90fEYlFSgZP3a4v8xzo/ki8St2VyvAfhvP2orepUrIKt7e4nb5N+mbbEyX+SHfppO5KPaEX\ncdm2ZXx9x9dKBkNgZm7OHEfbtqds1/d/DnR/RCQWKRk8dbu+zHOg+yPx5EjaESaunMjQhUP5btN3\n9G/an9ua38YFVS7wOzQJgcpEQ2Nm7r77HK+8csp2ff/nQPdHRGKRksFTt+vLPAe6P5LfHTp2iNlr\nZzNp1STGLRtH40qNub357fRs1NP3heEld5QMhsbMXJUqjg0boECBE7br+z8Huj8iEoty00ZGfDZR\nM+sCvAgkAMOcc8+dtL80MAqoBRQA/u2ceyewby2wG0gHjjrnWkY6XhGJbRv3bOSz1Z8xefVkvlj7\nBU3OaULX+l2ZN3ge9SvU9zs8kTxXpQrMmwft2vkdiYiIRJuI9gyaWQKwCugIbAK+Bfo651ZkOeZP\nQGnn3J/MrCKwEqjsnDtmZmuAi5xzO8/wPuoZzAXdH8kP0tLT+GbjN0xePZnJqyezbvc6rjr3KrrW\n70qXel2oULyC3yFKGKlnMDRm5p591usZfO21E7br+z8Huj8iEouisWewJbDaOZcKYGbvA92BFVmO\ncUCpwPNSwA7n3LHAa8PrURQRybTz4E6m/TKNyasnM/XnqVQtWZWu9bvyytWv0KpGKwomaAlVkQy9\ne0ObNvDyyyeWioqIiET6L6bqwPosrzfgJYhZvQpMNLNNQEngxiz7HDDDzNKAN51zb0UyWBGJTs45\nlm1bxuTVk5m0ahI//PoDV9S+gmsbXMvfOvyNWmVq+R2iSNSqVw+qV4c5c6B9e7+jERGRaBINvW5X\nAYucc9WA5sBrZlYysK+1c64FcA1wr5m18SvIaDB79mxq1qyZ+bpJkybMmTMnqGNFYs3Bowf5bPVn\n3Dv5XhJfSqTr6K6k7krlj23+yJaHtzCp/yTuvvhuJYKS75jZMDPbYmaLT7O/tJlNNLMfzOwnMxt0\npmv26QPjxoU91KiiNlJEJHSR7hnciDcxTIYagW1ZDQb+DuCc+8XMUoCGwHfOuc2B7dvM7GO8XsV5\n2b3RkCFDMp8nJSWRlJQUnk8QZbIuhL1kyZKgj5X8IS09jaXblrJgwwIKJRSibrm6JJZLpHqp6hRI\niP36r3W71zF5lTf2b07qHJpXbU7X+l2Z3H8yjSs11v+n41BycjLJycl+h5HXhgOvACNOs/9eYKlz\nrlvGWHszG5VliMUpeveGVq3glVegYD6uolYbKSISmkg3Cd8C9cysNrAZ6Av0O+mYVKAT8KWZVQYa\nAGvMrDiQ4JzbZ2YlgCuBp0/3RlmTQZH8Yu/hvXy98Wu+XPcl8zfMZ8GGBVQpWYVWNVrhnGPYomGk\n7Ephx4Ed1CxTk8SyiV6CmPFvuUQSyyZSvlj5qPzD51j6MRZsWJCZAG7et5ku9bowoNkARvYYSbli\n5fwOUXx28o97Tz992mYg33DOzQu0m6c9hNOPtc9W3bpQuzbMng0dO4YrUhERiXURTQadc2lmdh8w\nneNLSyw3s7u83e5N4K/AO1nKYR5xzv1mZonAx2bmAnG+55ybHsl488o///lPvv32Wz744IPMbQ89\n9BAAF154If/85z/ZsGED55xzDo888gh33nlnttdJTExk2LBhdOjQgUOHDnH33XczceJEqlWrxqBB\ng/Lio0gYOedI3Z3qJX7r5/Pl+i9Z/dtqWlRtQeuarbn3knsZ1WMUlUpUOuXcg0cPkro7lZSdKazZ\nuYaUXSl8s+kb1uxcw5qdawBILJtIYrlE6pb1ksSMpLFO2Tp5uubejgM7mPrzVCavnsy0X6ZRq0wt\nutbvyhvXvkHL6i3zRQ+nSITlNNb+tDJKRaM9GVQbKSKSdyJeLOKcmwqcd9K2N7I834w3bvDk81KA\nCyMdnx/69u3LX/7yF/bv30+JEiVIT09n3LhxfPLJJ+zYsYPJkyeTmJjI3Llz6dKlCy1btuTCC3O+\nFUOGDCElJYWUlBT27dtHly5d8ujTSG4dSTvCos2LMhO/+evn43C0rtmay2tezsALBtKiagsKFyh8\nxmsVK1SMhhUb0rBiw1P2OefYeWgnKTtTSNnlJYtLty5l0qpJrNm5hnW711G+WPkTEsSsPYtnW4Lq\nnOOnrT8xadUkJq+ezJKtS2hfpz1d63fl+c7PU7109VxfWyROZYy172Bm5+JNtNbMObcvp5N694aW\nLU9cYiIaqY0UEck7+XjkwJnZ0+Epm3NPhbYWUa1atWjRogUff/wxAwYMYObMmZQoUYKWLU+caLVt\n27ZceeWVzJ0794wN3QcffMDrr79OmTJlKFOmDA888ADPPPNMyJ9FImf7ge18tf6rzMRv4eaF1Ctf\nj8trXs71Da/n+c7PU6dsnbCXc5oZ5YuVp3yx8lxU7aJT9qelp7Fp7yZSdqVk9ix+sfaLsypB3X9k\nP7NSZmWu/Ve4QGG61u/Kk1c8Sbs67ShasGhYP6NInDntWPvsDs46jKJChSSSk5OCepNwtJGhto+g\nNlJEJFjhGFcf18lgbhqpcOnXrx9jxoxhwIABjBkzhv79+wMwZcoU/vKXv7Bq1SrS09M5ePAgzZo1\nO+P1Nm3aRI0aNTJf166d03ATibR0l87K7StP6PXbvG8zl1a/lMtrXs4TVzzBpTUupXSR0n6HSoGE\nAtQsU5OaZWpyRe0rTtl/6Ngh1u5ae0LPYkYJasrOFBwuM0GsXaY2K3as4Mt1X3JxtYvpWr8rMwbO\n4LwK50XlmEWRKGaBR3ayHWt/ugtlTQZLlQp+VlG1kSIi0S0c4+rjOhn0U+/evXn44YfZuHEjH3/8\nMV9//TVHjhyhV69ejBo1iu7du5OQkECPHj1w7swNctWqVVm/fj2NGjUCIDU1NdIfQbI4cPQA3278\nNjPx+2rDV5QpUobLa15O65qtefDSB2lyTpOYHA9XtGDR05aggrcAfMY4xbW71tK6Vmvev+F9yhQt\nk8eRiuQPZjYaSAIqmNk64CmgMGcYax/MtXv1gosvjkDQYaY2UkQkbygZ9EnFihVp164dgwcPpm7d\nujRo0IB9+/Zx5MgRKlasSEJCAlOmTGH69Ok0bdr0jNfr06cPf//732nZsiX79u3j1VdfzYNPEb82\n7tmYmfh9uf5Llm1bRtNzmnJ5zcsZfOFg3rzuTaqVquZ3mHmiXLFyXFTsomxLUEUkdM65/mfYn+1Y\n+2DUru0tQr99e65CyzNqI0VE8oaSQR/179+fW265heeffx6AkiVL8vLLL9O7d2+OHDnCddddR/fu\n3U97ftayu6eeeoq7776bxMREqlevzuDBg3nppZci/hniwbH0YyzesviEks/9R/Zn9vq9cOULXFzt\n4jydkVNEJLd694YFC/yO4szURoqIRJ4FU14R7czMZfc5zCyo8pF4pfuTvWPpx1i4eSEz18xk1tpZ\nLNiwgJqla2bO8tm6Vmvql6+vMXAiPgh8b+k/viBl1z6uWwe1a+v7PydqH0UkFuWmjVQyGMd0fzzO\nOZZtW8bMlJnMTJnJnNQ51Chdgw51OtCxbkfa1GpD+WLl/Q5TRFAyGCq1j7mj+yMisUjJ4Knb9WWe\ng3i+P2t3rWXmGi/5m5Uyi+KFitMhsQMdEzvSIbEDlUtW9jtEEcmGksHQqH3MHd0fEYlFSgZP3a4v\n8xzE0/3Zun8rs1JmZSaA+4/uz0z8OiZ2JLFcot8hikgQlAyGRu1j7uj+iEgsUjJ46nZ9mecgP9+f\nPYf3MHvt7Myev3W719GuTrvM0s/zK52vMX8iMUjJYGjUPuaO7o+IxCIlg6du15d5DvLT/Tl07BDz\n18/P7Plbum0pl1a/NLPn76JqF1EwQZPnisQ6JYOhUfuYO7o/IhKLlAyeul1f5jmI5ftzLP0Y32/6\nPnPSl282fkOTc5pk9vxdXvNyihYs6neYIhJmSgZDo/Yxd3R/RCQW5aaNzNddJbVr11YpYA5q167t\ndwhBc86xdNvSzJ6/OalzqFWmFh0SO/DQpQ9xRe0rKFO0jN9hiojEBLWPOYul9lFE5Gzk655BiW0p\nO1Mye/5mpcyiZOGSmZO+tK/TXjN+isQh9QyG5kzt46uvwjffwIgReRiUiIhEhMpEJaZt2beFWSmz\nvFk/U2Zy4OgBOtbtmFn6WadsHb9DFBGfKRkMzZnax82boXFj79+iqqwXEYlpSgYl5qzfvZ4xS8Yw\nZskY1u5aS7va7TInfWlcqbHKmETkBEoGQxNM+5iUBL/7HXTrljcxiYhIZCgZlJiw69Auxi8bz6if\nRrF4y2JuaHQDNzW9ida1WmvGTxHJkZLB0ATTPv73vzB/PowalUdBiYhIRCgZlKh1JO0IU1ZPYdRP\no5j+y3Q6JnZkQLMBXFP/Gs36KSJBUzIYmmDax19/hYYNvVLRYsXyKDAREQk7zSYqUSXdpTN//XxG\nLR7Fh8s+5PxzzmdA0wG8ee2blCtWzu/wREQEqFIFWrSAadPg+uv9jkZERPKSkkEJu+XbljNq8ShG\nLxlN8ULFGdhsIN/f+T21y2qqbhGRaNSnD4wbp2RQRCTeqExUwmLz3s28v+R9Rv00is17N9O/aX8G\nNBvABZUv0CQwIhI2KhMNTbDt49at0KCBSkVFRGKZykQlT+09vJePV3zMqMWj+HbTt3Q/rzvPdXqO\n9nXaUyChgN/hiYhIkM45By6+GKZMgZ49/Y5GRETySkKk38DMupjZCjNbZWaPZrO/tJlNNLMfzOwn\nMxsU7LmS946mHeWz1Z/Rf3x/avynBh8s+4Dbmt/Gxt9t5J3r36FT3U5KBEUk7NLSYN06v6PI3zJK\nRUVEJH5EtEzUzBKAVUBHYBPwLdDXObciyzF/Ako75/5kZhWBlUBlIP1M52a5hspEI8g5xzcbv2HU\n4lGMXTqWeuXrMaDZAPqc34eKxSv6HZ5IzFq3Dnbvhho1oGxZiPeK6n37YM0a7/HLLyc+X7cOKlSA\nTZtUJhqKUNrHbdugXj2vVLR48QgHJiIiYReNZaItgdXOuVQAM3sf6A5kTegcUCrwvBSwwzl3zMxa\nBXGuRNDqHat576f3eO+n9zCMAc0G8NVtX3Fu+XP9Dk0kZq1dCx9+6PXApKRA5cqwYQMcPeolhTk9\nKlaM7YQxPd1LNLJL9tasgb17ITER6tb1HuedB1df7T1PTPTGssXy5492lSrBpZfCZ59Br15+RyMi\nInkh0slgdWB9ltcb8BLErF4FJprZJqAkcGMI50qYbdu/jbFLxzJq8ShSdqXQ9/y+jO45mourXayJ\nYERy6eQEsEcP+NvfICkJChXyjtm7FzZu9BLDjMePP8Lkycdf79sH1aufmiRm3Va5MhTwsVL74EHv\nM2aX7KWkQJkyXnJ37rnev507w913e8+rVIGEiA9ekJxklIoqGRQRiQ/RMIHMVcAi51wHMzsXmGFm\nzfwOKp4cOHqACSsmMOqnUXy57ku6NujKU+2eolPdThQqUMjv8ERiUjAJYFalSnkLfzdsePprHjx4\nasK4ciXMnHn89W+/eUlVTj2MVatmH0MwnPNmnswu2fvlF+/9a9c+nuzVrQsdOhzv3StZMnfvK3mj\nRw/4/e8nNW8NAAAgAElEQVRh/34oUcLvaEREJNIinQxuBGpleV0jsC2rwcDfAZxzv5hZCtAwyHMz\nDRkyJPN5UlISSUlJZxF2/peWnsaslFmM+mkUE1ZMoFWNVgxoNoCxvcZSsrD+WhPJjVATwFAVK+aN\n6apX7/THHD7slWJmTRjXrYP584+/3rrVKznNKWE8ejT7ZG/NGi+OrMleu3YweLC3rVq18PZMJicn\nk5ycHL4LSo4qVIDLLvN6pPv08TsaERGJtEhPIFMAb0KYjsBm4Bugn3NueZZjXgO2OueeNrPKwHfA\nBcDuM52b5RqaQCZIS7Yu4e1FbzNmyRhqlK7BgKYDuLHJjVQpWcXv0ERi0tq18MEH3iMjAezdG9q3\nh4LRUHuRjWPH4NdfT0wYsz7Wr/diz0j4sv6bmOiVevpF6wyGJjft49tve+MGP/wwQkGJiEhE5KaN\njPii82bWBXgJbxmLYc65f5jZXYBzzr1pZlWBd4CqgVP+7pwbc7pzT/MeSgaD8EXKF9z44Y3cddFd\n3NTsJhpWzKEeTUROK7sEsE8frwcwWhPA/ELJYGhy0z7+9puX9G/cqLJeEZFYEpXJYF5QMnhm3278\nlq6juzKu9ziS6iT5HY5IzFECGB3iIRk0s2HAtcAW51y2Y+jNLAn4D1AI2Oaca3+a43LVPl5zDdx8\nM/TtG/KpIiLiEyWDkq1l25bR4d0OvHndm3Q7r5vf4YjEDCWA0SdOksE2wD5gRHbJoJmVAeYDVzrn\nNppZRefc9tNcK1ft4zvvwMSJ8NFHIZ8qIiI+UTIop1i7ay1th7fl2Q7PMvCCgX6HIxL1Tk4Ae/b0\nxgAqAYwO8ZAMAphZbeDT0ySD9wBVnXNPBnGdXLWPO3dCnTreGNJSpc54uIiIRIHctJFa0Skf27Jv\nC51HduYPl/9BiaBIDtauheefh5Yt4ZJL4Oef4dlnvVk533gDOnVSIihRpQFQ3sy+MLNvzSzsX/Dl\nykHbtvDpp+G+soiIRBP9eZNP7Tq0i6tGXcWApgN44NIH/A5HJOpk7QFcu9YrAX32WfUASkwoCLQA\nOgAlgK/M7Cvn3M/hfJOMBej79w/nVUVEJJroT5586MDRA1w7+lra1W7Hk+3OWEUkEjeUAEo+sQHY\n7pw7BBwyszl4SzJlmwzmdh3ebt3g/vthzx4oXfpsQxYRkXALx1q8GjOYzxxJO8L1719PxeIVeef6\nd0gwVQJL/Nq6Fb77Dr791ltEOyMB1BjA2BVHYwbr4I0ZbJrNvobAK0AXoAjwNXCjc25ZNseeVfvY\nrZvXQzhgQK4vISIieSQ3baT+FMpH0tLTuPnjmylUoBDDug1TIihxZedOL/HL+tizBy6+GC66SD2A\nEjvMbDSQBFQws3XAU0BhAuvzOudWmNk0YDGQBryZXSIYDhmlokoGRUTyJ/UM5hPOOf7f5P/Hih0r\nmHLTFIoWLOp3SCIRs2cPLFx4YuK3dSu0aOElfxmPc88Fy/d9SPElXnoGw+Vs28c9e6BmTUhNhbJl\nwxiYiIiEnZaWiGOPz3yc6WumM+vmWZQqonnAJf/Yvx9++OHExG/9erjgghMTvwYNoEABv6OVSFMy\nGJpwtI/XX+8tsXLzzWEKSkREIkLJYJz61/x/MWzRMOYOnkvF4hX9Dkck1w4dgsWLjyd9334Lv/wC\nTZqcmPg1bqxyz3ilZDA04WgfR4/2HpMmhSkoERGJCCWDcWjYwmE8M+cZ5t06jxqla/gdjkjQjh6F\nJUtOTPxWrIDzzjsx8WvSBIoU8TtaiRZKBkMTjvZx716oUcObgKlcufDEJSIi4adkMM6MXzae+6fc\nT/KgZBpUaOB3OCKndeyYl+hlJH3ffeclgnXqeIu8ZyR+F1wAxYr5Ha1EMyWDoQlX+9izpzez6KBB\nZx+TiIhEhpLBODLjlxnc9NFNTB84nQurXOh3OCKZ0tNh9eoTE78ffoDq1U/s8WveHEqW9DtaiTVK\nBkMTrvbx/fdhxAj47LMwBCUiIhGhZDBOLNiwgG5juvHRjR/RplYbv8MR4eefvXX8PvsMFiyA8uVP\n7PFr0UIzEUp4KBkMTbjax337vB90UlK8/75FRCT6KBmMA0u2LqHTiE683f1trql/jd/hSJw6cgTm\nzfMmlJg82Zt+/pproGtXuOIKqKh5jCRClAyGJpztY69e3n/nt94alsuJiEiYKRnM59bsXMMVw6/g\n+c7P069pP7/DkTizZYvX8zd5Mnz+uTfRS9eu3qN5c0hI8DtCiQdKBkMTzvZx3DgYPhymTAnL5URE\nJMyUDOZjm/dups3wNjx82cPcc8k9focjcSA9Hb7/3kv+Jk/2xgF27uwlf1dfDZUr+x2hxCMlg6EJ\nZ/u4fz9UqwZr1kCFCmG5pIiIhJGSwXzqt4O/0e6ddvRr0o/H2j7mdziSj+3ZAzNmHB//V67c8d6/\nNm2gUCG/I5R4p2QwNOFuH/v0gauugttuC9slRUQkTJQM5kP7j+yn08hOtK7Zmuc7P4+Z/gaS8HEO\nVq063vv3zTdw+eVw7bVeAli3rt8RipxIyWBowt0+fvABDB0K06aF7ZIiIhImSgbzmcPHDnPdmOuo\nWbomQ7sNVSIoYXH4MMyZ4yV/kybBwYPHe/86dtRyDxLdlAyGJtztY0ap6C+/aKIoEZFoo2QwH0lL\nT6Pv+L6ku3TG9hpLwYSCfockMWzTpuOTv8yaBeeffzwBvOAC0O8MEiuUDIYmEu3jjTdCp05wxx1h\nvayIiJwlJYP5hHOOOz+9k5RdKUzuP5kiBYv4HZLEmLQ0b8H3jPLPtWvhyiu98s8uXfSLvsQuJYOh\niUT7OH48vP66N75YRESiR1Qmg2bWBXgRSACGOeeeO2n/w8BNgAMKAY2Ais65XWa2FtgNpANHnXMt\nT/Me+SoZfHTGo8xOnc3nN39OycKq2ZPg7NoF06d7yd+UKXDOOcd7/y6/HAqqc1nyASWDoYlE+3jg\ngFcquno1VKoU1kuLiMhZiLpk0MwSgFVAR2AT8C3Q1zm34jTHXws85JzrFHi9BrjIObfzDO+Tb5LB\n5+Y9x8jFI5k9aDYVimvubjk952DFiuMLvy9cCG3besnfNddAnTp+RygSfkoGQxOp9rFfP0hKgrvu\nCvulRUQkl3LTRka6r6AlsNo5lwpgZu8D3YFsk0GgHzAmy2vD61GMC29+/yZvfP8G826dp0RQOHwY\ntm2DrVu9R8bzbdvg11+9SWCOHfOSv4cfhg4doHhxv6MWkXjQpw+8+qqSQRGRWBfpnsEbgKucc3cG\nXg8AWjrnHsjm2GLABuBc59yuwLY1wC4gDXjTOffWad4n5nsGxy4Zy++m/445g+Zwbvlz/Q5HIuDo\nUdi+PfsE7+Rkb+tWb5bPSpW8cs+Mf7M+b9kSmjTR5C8SX9QzGJpItY8HD0LVqrByJVSuHPbLi4hI\nLkRjz2AorgPmZSSCAa2dc5vNrBIww8yWO+fm+RRfxEz9eSoPTH2AGQNnKBGMIWlp8Ntvp0/qTk7w\n9uyBChWyT+wuvvjUbWXKKNETkehUrJhXlfDRR3DPPX5HIyIiuRXpZHAjUCvL6xqBbdnpy4klojjn\nNgf+3WZmH+OVnWabDA4ZMiTzeVJSEklJSbmNOU99ue5LBn48kAl9J9CscjO/w5Esjh6FTz6B5cuz\nT/B27vQStux67po2PXVb+fKQEDdFzyLhkZycTHJyst9hSDb69IEXX1QyKCISyyJdJloAWIk3gcxm\n4Bugn3Nu+UnHlQHWADWccwcD24oDCc65fWZWApgOPO2cm57N+8RkmeiPv/5I55GdGdljJFfVu8rv\ncCRg92546y146SWoXx9at86+N69CBc3QKZLXVCYamki2j4cOeaWiy5dDlSoReQsREQlB1JWJOufS\nzOw+vEQuY2mJ5WZ2l7fbvRk49HpgWkYiGFAZ+NjMXCDO97JLBGPV6h2rufq9q3ntmteUCEaJ9eu9\nBHD4cLj6apg4EZo39zsqEZHoVLSot3bp+PFw771+RyMiIrmhRed9sHHPRtoMb8NjbR7jjovu8Duc\nuLdwIfz73zB1KgweDA88ALVqnfk8Ecl76hkMTaTbx08/hX/9C2bPjthbiIhIkKJuncG8EkvJ4PYD\n27li+BUMunAQj7R+xO9w4pZzXvL3r3/BqlXw4INwxx3eGEARiV5KBkMT6fbx8GGvVHTJEm8hehER\n8Y+SwSi39/BeOo7oSIfEDvyj0z/8DicuHT4Mo0d7PYGFCnnr8/Xp4z0XkeinZDA0edE+3nKLNyPy\n/fdH9G1EROQMctNGam7DPHLo2CGuH3s9F1a5kL93/Lvf4cSd336DZ5+FxEQYN84bG7hwIdx0kxJB\nEZGz0aeP970qIiKxR8lgHjiWfox+4/tRoVgF/tf1f5gWj8sza9Z4YwDr1YPVq2HaNJgyBTp21Bp+\nIhKdzGyYmW0xs8VnOO4SMztqZj3zKrbsdO4MS5fCxtMtHCUiIlFLyWCEpbt07vj0Dg4cPcConqMo\nkFDA75Diwtdfe79Wt2wJJUt641mGD/fW/xMRiXLDgRynmTazBOAfwLQ8iSgHhQtD9+7w4Yd+RyIi\nIqFSMhhBzjl+P+33rNy+ko/6fEThAoX9DilfS0+HCROgbVvo2xfatIG1a73yUE1sICKxwjk3D9h5\nhsPuBz4EtkY+ojNTqaiISGzSktkR9Le5f2NmykxmD5pNicIl/A4n3zp4EEaMgBde8GYDffhh6NlT\nC8KLSP5kZtWA651z7c2spd/xgFd6P2CAt15rzZp+RyMiIsHSn8sR8to3r/HOD+8wd/BcyhUr53c4\n+dK2bfDf/3qPVq3grbe8XkGNBRSRfO5F4NEsr3P81hsyZEjm86SkJJKSksIeUOHCcP31Xqno//1f\n2C8vIiLZSE5OJjk5+ayuoaUlIuC9xe/x6OePMnfwXBLLJfodTr6zapXXCzhuHPTu7f3h0bCh31GJ\nSF6Il6UlzKw28Klzrlk2+9ZkPAUqAvuBO51zE7M5Ns/ax2nTYMgQ+OqrPHk7ERE5iZaWiAKTVk3i\n99N/z7QB05QIhpFzMG+e98tzmzZQuTKsWAFvvKFEUETyJeM0PX7OubqBRyLeuMH/l10imNc6dPBm\nbU5N9TsSEREJlpLBMJqTOofBEwYzoe8Ezj/nfL/DyRfS0ryyo8sug8GD4aqrvElhnn4azjnH7+hE\nRMLPzEYD84EGZrbOzAab2V1mdmc2h0dNWUyhQtCjh2YVFRGJJSoTDZOFmxfSZVQXRt8wmk51O/ka\nS36wb5+3FMR//uPNBPrww3DddVBAK3OIxLV4KRMNl7xuH2fMgD//2VveR0RE8pbKRH2ycvtKuo7u\nyuvXvq5E8Cxt3gyPPw6JiTB7Nrz33vHyUCWCIiLRrX17WLPGq+AQEZHop2TwLK3fvZ4rR13J3zr8\njZ6NevodTsxasQJuuw3OPx9274YFC46Xh4qISGwoWNBb2ueDD/yOREREgqFk8Cxs27+NziM78+Cl\nD3Jr81v9DidmjRsHV1wBdet6kw+8+iqce67fUYmISG5oAXoRkdihMYO5tOfwHtq/256r613NXzv8\nNU/fO79wDv7+d29G0IkT4YIL/I5IRKKdxgyGxo/28dgxb6z3ggXej3wiIpI3NGYwjxw8epBuY7px\nafVLeab9M36HE5OOHIFbb4Xx4701qZQIiojkDwULwg03qFRURCQWKBkM0dG0o9z44Y1UK1WNV695\nFTP9QB2qnTuhSxfv3zlzvF+QRUQk/1CpqIhIbFAyGIJ0l86tE28lzaXx7vXvkmC6faH65RdvUpgW\nLbxewRIl/I5IRETC7YorYONG+PlnvyMREZGcKJsJknOOh6Y+xNpda/mg9wcUKlDI75BizpdfQuvW\n8NBD8K9/aakIEZH8qkABlYqKiMQCJYNBenr208xdN5dP+31K8ULF/Q4n5owZAz16wLvvwt13+x2N\niIhEmkpFRUSiX0G/A4gFLy14idE/jWbu4LmULVrW73BiinPw17/CsGEwcyY0bep3RCIikhfatIFf\nf4VVq6BBA7+jERGR7ATVM2hmH5lZV7PQB8mZWRczW2Fmq8zs0Wz2P2xmi8xsoZn9ZGbHzKxsMOfm\nhRE/juDfX/2bGQNnULlkZT9CiFmHD8Mtt8Cnn3pTjCsRFBGJHwUKQK9eKhUVEYlmQa0zaGadgMFA\nK+ADYLhzbmUQ5yUAq4COwCbgW6Cvc27FaY6/FnjIOdcplHMjtY7ShBUTuGvSXXxxyxc0qtQo7NfP\nz3bsgJ49oWJFGDkSiquyVkTCQOsMhsaPdQazmjsX7rsPfvzRtxBEROJGxNYZdM597py7CWgBrAU+\nN7P5ZjbYzHKaSaUlsNo5l+qcOwq8D3TP4fh+wJhcnhtWX6R8wR2f3sGk/pOUCIZo9WpvxtBWrbxf\nhJUIiojEp9atYft2WJHtT8AiIuK3oMs+zawCMAi4HVgEvISXHM7I4bTqwPosrzcEtmV3/WJAF2B8\nqOeG23ebvuPGD29kbK+xXFzt4rx4y3xj7lxo2xb+8Ad47jlI0BRFIiJxKyFBpaIiItEs2DGDHwNz\ngeLAdc65bs65sc65+4GSYYrlOmCec25XmK6XK8u3Lefa0dfy1nVv0T6xvZ+hxJxRo7ypxEeNgjvu\n8DsaERGJBppVVEQkegU7m+jLzrkvstvhnMup62wjUCvL6xqBbdnpy/ES0VDPZciQIZnPk5KSSEpK\nyiGs7KXuSuWqUVfxz87/pHvDPKtIjXnOwdNPe8tGfPEFnH++3xGJSH6RnJxMcnKy32HIWbjsMti5\nE5Ytg8aN/Y5GRESyCnYCmXuB9zJ67cysHNDPOfffM5xXAFiJNwnMZuCbwHnLTzquDLAGqOGcOxjK\nuYFjz3qA/JZ9W2g7vC33XnIvD7Z68KyuFU8OH4bbboOff4YJE6CyJlwVkQjSBDKh8XsCmQz/939Q\ntiw89ZTfkYiI5F8Rm0AGuCNr+aZzbidwxkJA51wacB8wHVgKvO+cW25md5nZnVkOvR6YlpEI5nRu\nkPGGZNehXXR5rwv9m/ZXIhiC7duhUyc4csTrEVQiKCIi2VGpqIhIdAq2Z/AnoFnGz4uBXrvFzrmo\nKAg8m18+Dxw9wFWjrqJ5lea81OUlzPSDczBWroSuXb0G/q9/1UQxIpI31DMYmmjpGUxPhzp1YMoU\nDSUQEYmUSPYMTgXGmllHM+uIN7ZvaqgBRpsjaUfoNa4XdcrW4cUuLyoRDFJyMlxxBTz2GDz7rBJB\nERHJWUIC9O6tWUVFRKJNsD2DCcBdeOP3wFtOYmiglNN3ufnlMy09jQEfD2D/kf2M7zOeQgVyWi5R\nMowY4S0bMWYMdOjgdzQiEm/UMxiaaOkZBFiwAG69FZYuBf32KiISfrlpI4NKBqNdqI2dc457P7uX\nZduWMeWmKRQrVCyC0eUP6enewP/Ro2HyZGjY0O+IRCQeKRkMTTQlg855paKTJ0OTJn5HIyKS/+Sm\njQxqaQkzqw/8HWgMFM3Y7pyrG1KEUeKJL57gm43fMOuWWUoEg3DoEAwaBOvWeb/sVqrkd0QiIhJr\nzLxS0XHjlAyKiESLYEd7DQf+BxwD2gMjgFGRCiqS/j3/34xfPp4pN02hdJHSfocT9bZt88pBzWDW\nLCWCIiLBMrMHzay0eYaZ2UIzu9LvuPyUMatolHRWiojEvWCTwWLOuZl4ZaWpzrkhQNfIhRUZby96\nm1e+eYXpA6ZTqYSymjNZvhxatYKOHeG996Bo0TOfIyIimW51zu0BrgTKAQOBf/gbkr8uucSrNvnp\nJ78jERERCD4ZPByYRGa1md1nZj2AkhGMK+w+Wv4Rf571Z6YPnE7NMjX9DifqzZoFSUnw5JPwzDOa\nMVREJBcyxm1cA4x0zi3Nsi0umXm9g6NisrZIRCT/CfZP/AeB4sADwEXAAOCWSAUVbp+v+Zy7J93N\n5P6TaVChgd/hRL2334b+/WHsWLglZv5XFhGJOt+b2XS8ZHCamZUC0s90UqCkdIuZLT7N/v5m9mPg\nMc/MmoY57oi65x6v2mTkSL8jERGRM84mGlhg/jnn3MN5E1Locpot7esNX3PdmOsY32c8bWu3zePI\nYkt6Ovz5z946UJMmwXnn+R2RiMiJYmk20UBFzYXAGufcLjMrD9RwzmWb5GU5rw2wDxjhnGuWzf5W\nwHLn3G4z6wIMcc61Os21omY20ayWL4dOneC552DAAL+jERHJHyIym6hzLi3QMMWcJVuX0P397gzv\nPlyJ4BkcPOj1Am7eDF99BRUr+h2RiEjMuwz4wTm338wGAC2Al850knNunpnVzmH/giwvFwDVzzrS\nPNaoEcyY4SWEZnDTTX5HJCISn4JaWgJYZGYTgQ+A/RkbnXMfRSSqMEjZmUKXUV34z1X/oWuDmJvr\nJk9t2QLdu8O558Lnn0ORIn5HJCKSL/wPuMDMLgB+DwzFm427XRjf43ZgShivl2caN/banIyEsH9/\nvyMSEYk/wSaDRYEdQIcs2xwQlcng5r2b6TyyM4+3fZx+Tfv5HU5UW7YMunb1egWfesprkEVEJCyO\nOeecmXUHXnXODTOz28J1cTNrDwwGcqzeGTJkSObzpKQkkpKSwhXCWWvc2Osh7NzZe62EUEQkeMnJ\nySQnJ5/VNc44ZjAWZB0TsfPgTtq9044bz7+Rx6943OfIotuMGV5pzgsvaMyGiMSGGBszOBuYCtwK\ntAW2Aj8658444UugTPTT7MYMBvY3A8YDXZxzv+RwnagcM3iyJUu8hPCFF6CffsMVEcmViIwZDFx4\nOF5P4Amcc7eG8maRtv/IfrqO7krnup15rO1jfocT1d56C554AsaPh7YaTikiEgk3Av3x1hv81cxq\nAc8Hea5xmmUoAtcZDwzMKRGMJU2aHO8hNIO+ff2OSEQkPgTVM2hmN2R5WRToAWxyzj0QqcBCYWbu\n0NFDdHu/G9VLVWdYt2GY6h2zlZ4Of/wjfPIJTJ4M9ev7HZGISPBiqWcQwMwqA5cEXn7jnNsaxDmj\ngSSgArAFeAooDDjn3Jtm9hbQE0jFSxiPOudanuZaMdEzmOGnn+DKK+HFF+HGG/2ORkQktuSmjcxV\nmWhguux5zrnLQz45AszM9R7Xm2PpxxjXexwFE4IdChlf9uyBQYNgxw746COoUMHviEREQhNLyaCZ\n9cHrCUzGS9raAn9wzn2YhzHEVDIIsHgxXHUVvPwy9O7tdzQiIrEjYmWi2agPnJPLcyPit4O/Man/\nJCWCp7FkCdxwA7RvD2PGaMZQEZE88DhwSUZvoJlVAj4H8iwZjEXNmsG0aV5CCEoIRUQiKdgxg3s5\ncczgr8CjEYkolz6+8WOKFizqdxhRadQo+L//8wbmDxzodzQiInEj4aSy0B1Agl/BxJJmzWDqVC8h\nNINevfyOSEQkfwoqGXTOlYp0IGerVJGoDzHPHT7sJYGffw6zZkHTM85fJyIiYTTVzKYBYwKvbwQ+\n8zGemHLBBV5C2KWLlxDecMOZzxERkdAE2zPYA5jlnNsdeF0WSHLOfRLJ4CT3UlO90pqaNeG776B0\nab8jEhGJL865PwQmYGsd2PSmc+5jP2OKNRdeCFOmwNVXewlhz55+RyQikr8EO5voD865C0/atsg5\n1zxikYUgFgfIR9LUqd5EMY884vUMamJVEckvYmkCmWiQX9rHRYu8HsLXX4cePfyORkQkOkVyApns\nxjhoppYok5YGf/kLDBsGH3yg9QNFRPyQzTj7zF14y0OoViNEzZsf7yEEJYQiIuESbEL3nZm9ALwW\neH0v8H0wJ5pZF+BFvIRymHPuuWyOSQL+AxQCtjnn2ge2rwV2A+nksI6SwPbtcNNNcOSIVxZapYrf\nEYmIxKdYGGcfi1q0OLFk9Prr/Y5IRCT2BTur2f3AEWAs8D5wCC8hzFFgPcJXgauA84F+ZtbwpGPK\n4CWZ1zrnmgBZJ5FOxxub2FyJ4Ol9/TVcdJH3y+mMGUoERUTkuC37tvgdQti0aAGffQZ33QUTJvgd\njYhI7At2NtH9wB9zcf2WwGrnXCqAmb0PdAdWZDmmPzDeObcx8F7bs+wzNA33aTkHr73mlYYOHQrd\nuvkdkYiIRJue43oy6+ZZFCmYPxaYvegimDwZunaFhAS47jq/IxIRiV1BJVpmNiMwg2jG63KB6bLP\npDqwPsvrDYFtWTUAypvZF2b2rZllXQnPATMC2+8IJtZ4sW8f9O/vjQ/86islgiIikr1qpapx56Q7\nyQ8TyWS4+GIvIbz9dvj0U7+jERGJXcH2ulV0zu3KeOGc2wmcE6YYCgItgKuBLsATZlYvsK+1c64F\ncA1wr5m1CdN7xrTly6FlSyhRAubPh3PP9TsiERGJVu9e/y5Lti7h+fnP+x1KWF18MUyaBLfd5v0r\nIiKhC3YCmXQzq+WcWwdgZnXIfqa0k20EamV5XSOwLasNwHbn3CHgkJnNAS4AfnbObQZwzm0zs4/x\nyk7nZfdGQ4YMyXyelJREUlJSEOHFnrFj4b774Lnn4NZb/Y5GRCSykpOTSU5O9juMmFa8UHEm9J3A\npUMvpVHFRlx3Xv6pq7zkEi8RvPZaGD7cKx0VEZHgBbvOYBfgTWA23ji+tsCdzrkcS0XNrACwEugI\nbAa+Afo555ZnOaYh8Aper2AR4GvgRmAtkOCc22dmJYDpwNPOuenZvE++WEcpJ0eOwB/+4DV6H37o\nTRYjIhJvtM5gaLK2j19v+JrrxlzHzJtn0rRyU58jC6+vv/bGDr7zDlxzjd/RiIj4IzdtZFBlos65\nqcDFeIndGOD3wMEgzksD7sNL5JYC7zvnlpvZXWZ2Z+CYFcA0YDGwAHjTObcMqAzMM7NFge2fZpcI\nxoMNGyApCdauhe+/VyIoIiKhu7TGpbzY5UW6v9+dbfu3+R1OWF16KUycCIMGebONiohIcILtGbwd\neBCvzPMHoBXwlXOuQ2TDC05+7hn8/HMYOBAeesjrGUzQ3KoiEsfUMxia7NrHP8/6M7NTZ/P5wM/z\nzfGbUzMAACAASURBVAyjGRYs8CZUe/fd4wvUi4jEi9y0kcEmgz8BlwALnHMXBko7n3XO9cxdqOGV\nH5PB9HR49ln473/hvfegfXu/IxIR8Z+SwdBk1z6mu3R6jetF2aJlGdZtGGb563Z+9RV07w4jRkCX\nLn5HIyKSdyJWJgocCkzwgpkVCZR2nhdqgBKc337zxj5MnQrffadEUEREwifBEhjRYwQLNy/kPwv+\n43c4YXfZZfDJJ3DzzTAtmEWwRETiWLDJ4IbAOoOf4K37NwFIjVxY8eu777wFdRs2hC++gGrV/I5I\nRETym5KFSzKx30T+/dW/+Wx1/htkd/nlXkI4cCBMj8vZBkREghNUmegJJ5i1A8oAU51zRyISVYjy\nQ5moc/Dmm/DEE/C//8ENN/gdkYhI9FGZaGjO1D5+tf4rur/fnS9u+YLzzzk/DyPLG19+CT16eMMt\nOnf2OxoRkciK2JjBaBfryeCBA3DPPbBwIYwfDw0a+B2RiEh0UjIYmmDax1GLR/FU8lN8ffvXVCxe\nMY8iyzvz5nkJ4Zgx0KmT39GIiEROJMcMSoSsXg2tWnk9gwsWKBEUEZG8NaDZAPo07sMN427gSFpU\nFPyEVZs28NFH0K+fN0P3/2/vvsOjKvP+j7+/EJBeBKR3BYKUgII0JQgorCIlFIFVcW2P+qjPWlZd\nG/roLmzx0XV1XZW1hqJ0+LkiIAEBBUSaEIqAofelS0jI/fvjDEnAAJkwkzPJfF7XNVdmTs6c+WYu\nyJ3P3E1ERLIoDPpo4kTo2BEefNBbBrt0ab8rEhGRaPRK11eoUKICD/6/BynII23O5dprvZE3gwfD\n7Nl+VyMiEjk0TNQHaWnw9NNew/TZZ3D11X5XJCJSMGiYaHCCaR+PpB6h4786cleru3ik3SNhrswf\n8+Z5c/LHjYPrI2KnZBGR0NEw0QJgxw6vAVqzxls5VEFQREQiQdlLyjJt8DRGLhjJFz9+4Xc5YXHd\ndTB+PAwaBF995Xc1IiL+UxjMR0lJXvi74QaYPh0qVfK7IhERkSx1K9TlswGfcfuk20nem+x3OWHR\nubM3KmfQIG8LJxGRaKYwmA+cg5Ej4dZbvbmBzz0HRfTOi4hIBOpYpyN/6v4neo3pxf7j+/0uJyzi\n471AOHCg90GtiEi0UiQJs4MHvSWtJ0+GJUu0z5GIiJyfmY0ys91mtvI85/zNzDaY2XIziwt1DcPi\nhtG3SV/6f9aftFNpob58RIiP9+YODhgAc+f6XY2IiD8UBsNo+XJvWGidOl5DU7u23xWJiEgB8D5w\n47m+aWY9gYbOuSuA+4C3w1HEiG4jKFO8DA/9+6FCucIoeHP4x46F/v29xWVERKKNwmCYfPyx1wv4\n8svwt79B8eJ+VyQiIgWBc24+8J/znNIb+Chw7iKgvJlVDXUdRYsUJbFfIvO3zOfNJW+G+vIRo2tX\nLxAmJKiHUESij8JgGBw4AA895DUqt97qdzUiIlLI1AS2Znu8PXAs5MpdUo5pg6fxytevMHPjzHC8\nREQ4HQgHDPA+xE1P97siEZH8EeN3AYXRuHHQowc0bep3JSIiEu2GDx+eeT8+Pp74+Pignl+/Yn3G\n9R/HgM8GMG/YPBpXbhzaAiNE167w/fcwbBh8/rk3wqdhQ7+rEhE5t6SkJJIuchUsbTofBu3awfPP\nw69+5XclIiKFS7RsOm9mdYFpzrkWOXzvbWCOc25c4PFaoLNzbncO54asfRz1/ShGLhjJt3d/y6Ul\nLw3JNSNRRoY3veOVV2DECPjNb8AK/b84ESkMtOl8BFi3DlJSvL0ERURE8sgCt5xMBW4HMLN2wMGc\ngmCo3dX6Lm5udDMDPxtYaFcYBW/rp//5H28PwjfegH79YO9ev6sSEQkPhcEQ++gjGDoUYjQAV0RE\n8sDMRgMLgUZmtsXM7jSz+8zsXgDn3OfAZjP7Efgn8EB+1fbn7n+meNHi/HbGb/PrJX3TrBksWgSN\nGkHLlt7QURGRwkbDREMoIwPq1YPp06HFLwb2iIjIxYqWYaKhEo728dCJQ7Qf1Z6H2j7E/W3uD+m1\nI9XcuXDHHdCzJ/zlL1C6tN8ViYj8koaJ+iwpCSpVUhAUEZHCq3yJ8kwbPI0X577I7E2z/S4nX3Tu\nDCtWwLFj0Lo1LFnid0UiIqGhMBhCH37ofXIoIiJSmDW8tCFj+49lyMQhbNi/we9y8kX58t5UkJdf\nhptvhv/9X21BISIFX9jDoJn1MLO1ZrbezJ48xznxZrbMzH4wsznBPDdSHD0KU6bA4MF+VyIiIhJ+\n8fXi+d8u/0uvMb04eOKg3+XkmwEDvC0ovv4arr0WNm70uyIRkbwLaxg0syLA34EbgSuBwWbW5Kxz\nygNvAjc755oBA3L73EgycaLXKFSt6nclIiIi+ePeq+7lxoY3Mmj8INIzoqebrGZN+OIL7wPgdu3g\n3XchApYuEBEJWrh7BtsCG5xzKc65NGAs0Pusc4YAE5xz2wGcc/uCeG7E+PBDuP12v6sQERHJX3+9\n8a8APDbjMZ8ryV9FisDDD3vrBbz1FvTpA3v2+F2ViEhwwh0GawJbsz3eFjiWXSPgUjObY2ZLzOy2\nIJ4bEbZsgeXLoVcvvysRERHJXzFFYhjXfxwzNs7gnaXv+F1OvrvySm8LithYbwuK6dP9rkhEJPci\nYQGZGKA10BPoATxnZpf7W1JwPvkEBg6EEiX8rkRERCT/VShRgWmDp/HcnOeYs3nOhZ9QyBQvDiNG\nwLhx8NBDcN993sqjIiKRLtxbo28H6mR7XCtwLLttwD7n3AnghJnNA1rm8rmZhg8fnnk/Pj6e+Pj4\ni6k715zzVhd7//18eTkRkaiSlJREUlKS32VILlxR6QrGJIxh8ITBLPjNAhpe2tDvkvLdddd5I4Ue\nfhji4rwPi6+5xu+qRETOLaybzptZUWAd0BXYCSwGBjvnkrOd0wR4A69X8BJgETAo8LzzPjfbNXzb\ndH7RIrjtNli3DkzbIIuIhJU2nQ+OH+3jP5b8gzcWv8E3d31D+RLl8/W1I8n48fDgg3D//fDssxAT\n7o/fRSTqRdym8865U8B/A18Cq4GxzrlkM7vPzO4NnLMWmAGsBL4F3nHOrTnXc8NZb16c3ltQQVBE\nRATub3M/19e/nsETBnMq45Tf5fimf39Ytgy++QY6doQN0bEdo4gUMGHtGcwvfvUMpqZ6y0svXQp1\n6+b7y4uIRB31DAbHr/Yx7VQaPRN70qJqC1698dV8f/1IkpEBb74JL70Er7wC99yjD5BFJDwirmew\nsJs+HVq0UBAUERHJrljRYnw24DOmr5/OqO9H+V2Or4oU8RaVmTsX3n4bbrkFdu/2uyoREY/C4EXQ\n3oIiIiI5q1iyItMGT+P3X/2eeSnz/C7Hd02bwrffQvPm3uIy06b5XZGIiIaJ5tmePdCoEWzdCmXL\n5utLi4hELQ0TDY6fC6ydNnPjTG6ffDsLf7OQ+hXr+1pLpJg/31t8rnt3ePVVKFPG74pEpDDQMNF8\nNGaMN9RDQVBEROTcujfszjPXPkOvMb04nHrY73IiQqdOsGIFpKVBq1Zej6GIiB8UBvNIQ0RFRERy\n58E2D9KpTieGTBgS1SuMZleunLdH8ciR0KcPvPCCFw5FRPKTwmAerFoFe/dCly5+VyIiIhL5zIw3\ner7B8bTjPDXrKb/LiSj9+nlbUCxe7G1BsX693xWJSDRRGMyDjz7yxvoXLep3JSIiIgXD6RVGJ6+b\nzDtL3/G7nIhSvTp8/rm3b3GHDt6qo4VgSQcRKQC0gEyQ0tOhTh346ito0iRfXlJERAK0gExwImEB\nmbOt27eOHok9GNxsMC9f/zJFTJ9LZ7d2Lfz611CtGrz3nvdVRCQ3tIBMPpg1C2rXVhAUERHJi8aV\nG7P47sUs3LqQ3mN7a1GZszRpAgsXettPtGoFU6b4XZGIFGYKg0HSwjEiIiIXp0rpKsy8bSa1y9Wm\n3Xvt+PHAj36XFFGKF4eXX4bx4+HRR2HIEPjmGw0dFZHQ0zDRIBw6BHXrwsaNUKlS2F9ORETOomGi\nwYnEYaJne/u7t3kh6QU+6fsJ3Rt297uciHPkCLz1FowaBTExcPfd3roFVar4XZmIRBoNEw2zzz6D\nrl0VBEVERELlv67+Lz7t/ym3T76d1759jUgPr/mtbFl48klYt85bWGbFCrjiChgwAGbMgFPaqUNE\nLoJ6BoNw7bXw+OPQu3fYX0pERHKgnsHgFISewdN+OvgTvcf2pnX11rx909tcEnOJ3yVFrEOHYMwY\nb4GZPXvgN7+BO+/0Ri+JSPRSz2AYbdzofSrXs6fflYiIiBQ+9SrUY8FvFnAk9QhdPuzCrqO7/C4p\nYpUvD//1X/DddzB1KuzfD61bw403wqefQmqq3xWKSEGhMJhLH38Mgwd7k7pFRETCycx6mNlaM1tv\nZk/m8P1yZjbVzJab2SozG+ZDmSFXpngZPh3wKT0u70Hbd9vy3Y7v/C4p4sXFwRtvwLZt3j6Fb7/t\nrXr+6KOwerXf1YlIpNMw0VzIyIDLL/fmDF51VdheRkRELiAahomaWRFgPdAV2AEsAW51zq3Nds7T\nQDnn3NNmVhlYB1R1zqWfda0CM0z0bJOSJ3Hv9Ht5vcfrDGk+xO9yCpSNG+Ff/4IPPvD2Rr7rLhg0\nyJt/KCKFl4aJhsmCBVCqlDcEQ0REJMzaAhuccynOuTRgLHD2bHUHnP7Tviyw/+wgWND1je3LV7d/\nxbNfPctTs57iVIZWSsmthg3hlVcgJQWeeQamT/dC4d13a4sKETmTwmAunN5b0Ar1Z9EiIhIhagJb\nsz3eFjiW3d+Bpma2A1gBPJJPteWr5lWbs/iexSzavojeY3tz6MQhv0sqUGJi4OabYfJkSE72ViG9\n4w5o1gxefRX27vW7QhHxm8LgBfz8M0ycCL/+td+ViIiIZLoRWOacqwG0At40szI+1xQWlUtV5stf\nf0nd8nVpN6odG/Zv8LukAqlatawtKv7xD1i+3AuHAwdqiwqRaBbjdwGRbvJkaNsWatTwuxIREYkS\n24E62R7XChzL7k7gjwDOuY1mthloAvxixZXhw4dn3o+Pjyc+Pj601eaDYkWL8eZNb/LO0nfo9H4n\nPu77MTc0vMHvsgokM7juOu928KC3RcUzz8C993rbU2iLCpGCIykpiaSkpIu6hhaQuYAePbwhokM0\nd11ExHdRsoBMUbwFYboCO4HFwGDnXHK2c94E9jjnXjSzqnghsKVz7sBZ1yqwC8icy7yUeQwaP4gn\nOjzBb9v9FtMcjpBYtgxGjfLC4dVXe/MLb7kFLtF2jyIFRl7aSIXB89ixwxtXv22bt4CMiIj4KxrC\nIHhbSwCv403nGOWcG2Fm9wHOOfeOmVUHPgCqB57yR+fcmByuU+jCIEDKwRR6j+1NXLU43r75bUrE\nlPC7pELj9PSYUaPghx+8aTJ33QVXXul3ZSJyIREZBgMN2mtkNWgjz/p+Z2AKsClwaKJz7uXA934C\nDgEZQJpzru05XiMsjd2f/+yNrX/vvZBfWkRE8iBawmCoFNYwCHDs5DHunHInWw5tYdKgSVQvW/3C\nT5Kg/Phj1hYVdet6vYWDBkGZQjk7VaTgi7gwmMu9kjoDjznnbsnh+ZuAq5xz/7nA64S8sXMOmjeH\nt97yxtWLiIj/FAaDU5jDIIBzjle+foV/Lv0nEwdOpE3NNn6XVCilp8O//+31Fs6dCwkJXjC85hqt\ntC4SSSJxn8Hc7JUEcK6iDZ9WPF22DI4fh06d/Hh1ERERuRAz49nrnuWNnm/wq9G/InFlot8lFUox\nMdCrl7eo3po1cPnlcNtt0Lixt/DMRx/B5s3av1CkIAp30MrNXkkA7c1suZn9PzNrmu24A2aa2RIz\nuyechZ7tww+9X3RFtPmGiIhIROvTpA9z7pjD80nP8+TMJ7VBfRhVrw5PPQXr18PYsd5cwmnToH17\nqF0bBg+GN9+ElSshI8PvakXkQsI9TDQBuNE5d2/g8a+Bts65h7OdUwbIcM4dN7OewOvOuUaB71V3\nzu00syrATOC/nXPzc3idkA6DSUuDmjXhm2+gYcOQXVZERC6ShokGp7APEz3b/uP7GfDZAErElGB0\nwmgqlKjgd0lRwzlvjuH8+fD1195t3z7o0AGuvda7XX21VicVCae8tJHh3mfwgnslOeeOZrv/bzN7\ny8wudc4dcM7tDBzfa2aT8Iad/iIMQmj3Ufr3v72hDwqCIiL+CsUeShI9KpWqxIxfz+DRGY/S7r12\nTB08lUaVGvldVlQw8zaxv+IKb69CgF27ssLhww97i/K1bu0Fw06dvKBYvry/dYtEu3D3DOZmr6Sq\nzrndgfttgU+dc/XMrBRQxDl31MxKA18CLzrnvszhdUL6yWdCgre/4D35OjBVREQuRD2DwYm2nsHs\n3l36Ls989Qwf9f2IHpf38LscAQ4f9kZdnQ6I333nhcdOnbJ6D6trUViRPIu41UQhV3slPQjcD6QB\nPwO/dc4tMrP6wCS8eYMxQKJzbsQ5XiNkjd2BA9CgAaSk6NMqEZFIozAYnGgOgwDzt8xn4GcDeaz9\nYzza/lFtUB9hUlPh+++zhpUuWAAVK2YFw06doFEjrVgqklsRGQbzQygbu7fe8n4hjfnF1r0iIuI3\nhcHgRHsYBNhyaAt9xvah2WXNeKfXO9qgPoJlZHirlZ4Oh19/DSdPZvUcduoEcXHe6qYi8ksKgyFw\nzTUwfDj07BmSy4mISAgpDAZHYdBzPO04d065k58O/sSkQZOoUbaG3yVJLqWknLkozdat0K5dVkC8\n5hooVcrvKkUig8LgRVq7Frp08X7R6FMnEZHIozAYHIXBLM45/jj/j7y15C0mDppI25pt/S5J8mD/\nfm846ddfeyFx5Upo0SIrHHbsCJUq+V2liD8UBi/S73/vDUf4y19CUJSIiIScwmBwFAZ/aeq6qdw9\n9W7+esNfua3lbX6XIxfp+HFYtCir9/Dbb6FOHejWzbt17gxly/pdpUj+UBi8CKdOQb168Pnn0Lx5\naOoSEZHQUhgMjsJgzlbvWc0tY2+hX5N+jOg2gqJFivpdkoRIerq3KM3s2TBrlhcU4+Kga1cvHF5z\nDRQv7neVIuGhMHgRZs+GJ57wfoGIiEhkUhgMjsLgue0/vp+B4wdSvGhxxiSM0Qb1hdTx496w0lmz\nvNuGDd5w0tM9h82aabVSKTwUBi/C7bd7G6H+z/+EqCgREQk5hcHgKAyeX3pGOo/NeIwvNn7B1Fun\n0rhyY79LkjDbtw/mzMkKh8eOZfUadu3qDTEVKagUBvPo6FGoVQvWr4fLLgthYSIiElIKg8FRGMyd\nUd+P4unZT/O3nn8jITaBYkWL+V2S5JPNm7OGlM6eDZdemtVrGB/v7XsoUlAoDObRhx/ChAkwdWoI\nixIRkZBTGAyOwmDuLdy6kMe/fJwNBzYwoOkAhjYfSvva7SliRfwuTfJJRoa3OunpXsMFC6Bp06xe\nww4doIS2qZQIpjCYR9dfDw88AP37h7AoEREJOYXB4CgMBm/zfzYz5ocxJK5K5NjJYwxpPoQhzYfQ\n7LJmfpcm+Sw1Fb75Jiscrl4N7dtn9RzGxUERfVYgEURhMA9SUry5gjt2wCWXhLgwEREJKYXB4CgM\n5p1zjpW7V5K4KpExP4zh0pKXMqTZEAY3H0yd8ppYFo0OHoSkpKwhpXv3eh0Kp8NhgwZ+VyjRTmEw\nD155BbZtg3/8I8RFiYhIyCkMBkdhMDQyXAZfp3zN6FWjmZA8gaZVmjK0+VD6N+1PpVLa4TxabduW\nNd9w1iwoWTIrGF5/PVSu7HeFEm0UBoPkHDRuDB99BO3ahaEwEREJKYXB4CgMhl5qeiozNs4gcVUi\nX/z4BZ3rdmZI8yHc0vgWShUr5Xd54hPnYM2arGA4bx40bJgVDjt1glL65yFhpjAYpG+/hTvugLVr\ntceMiEhBoDAYHIXB8DqSeoRJaycxetVovt32Lb0a92Jo86F0a9CNmCIxfpcnPkpLg8WLs8LhsmXQ\nooW3WmnZsue+lSmT8/HSpfW3qlyYwmCQ7r/f21LimWfCUJSIiIScwmBwFAbzz+6ju/l09ackrkpk\n88HNDGw6kCHNh9CuVjtMf8VHvSNHvEB48KC3pdmRIxe+ZT/vxAkvEF4oNOY2XJYsqXBZGCkMBiE1\nFWrU8P5jaoNREZGCQWEwOAqD/th4YCOjV40mcVUiJ0+dZEjzIQxtPpTYKrF+lyYF1KlT5w6ReQmX\naWlZQbFKFbjhBujbF9q00QqpBZnCYBDGj/cWjZk9O0xFiYhIyCkMBkdh0F/OOZbtWsboVaMZ88MY\nLit9GUObD+XWZrdSq1wtv8uTKJaWlhUOt2+H6dNh8mT4z3+gd2/o0we6dIHixf2uVIKhMBiEXr28\nfQXvuCNMRYmISMgpDAZHYTBynMo4xbyUeSSuSmRi8kRaVmvJ0OZDSYhNoGLJin6XJwLA+vUwZQpM\nmgTJydCzpxcMe/SAcuX8rk4uRGEwl/bsgUaNvCWBy5QJY2EiIhJSCoPBURiMTKnpqXy+4XNG/zCa\nLzd+yfX1r2dIsyHc3OhmShYr6Xd5IgDs3AnTpnk9hvPneyui9ukDt9wC1ar5XZ3kRGEwl157Db7/\n3ttSQkRECo5oCYNm1gN4DSgCjHLOjczhnHjg/4BiwF7nXJcczlEYjHCHThxi0tpJJK5K5Lsd39G7\ncW+GNh9Kl/pdtCKpRIzDh+Hf//aC4RdfQNOmXjDs0weuuMLv6uQ0hcFcatUK/vIX6No1jEWJiEjI\nRUMYNLMiwHqgK7ADWALc6pxbm+2c8sBC4Abn3HYzq+yc25fDtRQGC5CdR3YybvU4Rq8azdbDWxl0\n5SCGNB9CmxpttCKpRIzUVEhK8oLhlClQsaIXCvv2hauu0iqlflIYzIWVK+Hmm+Gnn7RakohIQRMl\nYbAd8IJzrmfg8VOAy947aGb3A9Wdc89f4FoKgwXU+v3rGbNqDImrEgG4o+Ud3NbyNuqU1xLoEjky\nMrz9FCdP9uYZHj+etQBN585QrJjfFUYXhcFcePxxb2WkP/whzEWJiEjIRUkYTABudM7dG3j8a6Ct\nc+7hbOecHh56JVAG+Jtz7uMcrqUwWMA551i8fTEfrviQcavH0bp6a4a1HEbf2L6UKlbK7/JEzrB2\nrRcKJ0+GDRvgppu8YHjjjVqnIz/kpY0Me9+YmfUws7Vmtt7Mnszh+53N7KCZfR+4PZvb5wYrPR0S\nE7WCqIiIFHgxQGugJ9ADeM7MLve3JAkHM+OaWtfw1k1vsf3R7dzT+h4SVyVS69Va3DP1HhZsWYAC\nv0SKJk3g6adh0SJYtQo6dIB33vH29u7VC0aNgr17/a5SsgvrzOTAvIe/k23eg5lNyT7vIWCec+6W\nPD4312bOhLp1oXHjvF5BREQk7LYD2ccC1gocy24bsM85dwI4YWbzgJbAj2dfbPjw4Zn34+PjiY+P\nD3G5kl9KxJRg4JUDGXjlQHYc2cEnKz/hnmn3kJaRxrCWwzSMVCJKzZpw//3e7eBB+Pxzr8fwsceg\nRYusBWgaNPC3zpMnYf9+L6Tu3Qv79p359exjR45A69bePozx8dC+PZQo4U/tSUlJJCUlXdQ1wjpM\nNJfzHjoDjzvnegX73Gzn5moYzK23wnXXwQMPXMxPJSIifomSYaJFgXV4H4buBBYDg51zydnOaQK8\ngdcreAmwCBjknFtz1rU0TLSQc86xZMcSPlj+AeNWj6NVtVYMixtGv9h+GkYqEenECfjqq6wFaKpW\n9Raf6dMH4uIubgEa57ywdqFAl/3YsWNQqRJUqQKVK+f8Nfv9kiW9eZJz5ni31auhTRsvGHbpAtdc\nA5dcErK3KygRN2cwl/MeOgMT8D7l3A484Zxbk5vnZrvGBRu7gwe9XsHNm+HSS0P0A4qISL6KhjAI\nmVtLvE7W1hIjzOw+vA9F3wmc8zhwJ3AKeNc590YO11EYjCIn0k8wdd1UPlj+Ad9s+4aE2ASGxQ2j\nY+2OWo1UItKpU/Dtt1kL0KSlZfUYXnutd86BA7kPdvv2QUzMuYNcTsfKl7+4RSUPH/b2YZwzx1tl\nNTnZC4Rduni3Nm289UryQ0ENg2WADOfccTPrCbzunGsU6jD47rvevigTJoTsxxMRkXwWLWEwVBQG\no9eOIztIXJnI+8vfJy0jjTta3sHtLW/XMFKJWM7BmjVZwXD1ai8cVqx44UCX/WvJkv7+HAcPwtdf\ne8FwzhxvIZ327bPC4VVXhW+V1UgMg+2A4c65HoHH5xzqme05m4GrgEa5fa6ZuRdeeCHzcU5zIjp1\ngt/9Dm65BRERKSDOng/x4osvKgwGQWFQnHN8t+O7zGGkcdXiGBY3jL5N+lK6eGm/ywuZXUd3sWLX\nClbuXsnuY7vpVKcTXep1oXyJ8n6XJnl09KgX7IoW9buSi3PggBcOTw8r3bwZOnbMmnPYurXXmxkK\nkRgGczPvoapzbnfgflvgU+dcvdw8N9s1ztvYbdzoJfLt27XfiYhIQaaeweAoDEp2J9JPMG3dND5Y\n8QELty4skMNIT546ydp9a1mxawUrdnu3lbtXkp6RTsuqLWlRtQVVSlVhbspcvtn2Dc0ua0b3Bt3p\n3qA77Wq1o1hR/SEo/tq/H+bOzeo53LLF67Q63XMYF5f3ABxxYRAuPO/BzB4E7gfSgJ+B3zrnFp3r\nued4jfM2di+84HXZvv56KH8yERHJbwqDwVEYlHPZeWQnn6z8hA9WfEBqemrmMNK6Fer6XVqmPcf2\nZPb2nQ5+6/evp16FerSs2tK7VfMCYM2yNX8RaE+kn2DBlgXM3DSTmZtm8uOBH7mu7nV0q9+N7g27\nE1s5tsCEYCm89uw5Mxzu3OnNlzwdDlu0yP2cxogMg/nhfI1dRgY0bOjNFWzdOp8LExGRkFIYDI7C\noFzI2cNIW1ZrybCW3mqk+TWMNO1UGuv2r8vs7Tsd/k6kn8js7Tsd/K6sciUli+VtUti+4/uYoC+v\nBgAAEIdJREFUvWl2Zjg8lXGKbg260b1Bd7o16EbVMlVD/JNFrr3H9rJs1zKW7VzGrqO7eKDNA1xR\n6Qq/yxJg1y4vHJ5ekGbvXm83hNPDSps1O3c4VBjMwbx58OCDsHLlxS1VKyIi/lMYDI7CoATj7GGk\n/Zr0Y1jcMDrV6RSyHrR9x/d5YS/bMM91+9ZRu3ztX/T21S5XO2w9d845fjzwY2YwTPopidrlantD\nSht257q61xWKrTmcc/x08KfM4Ld893KW7VzG0ZNHaVW9FXFV4yhVrBT/XPpPBl05iOc7Px9Vobgg\n2LHDC4Wnew4PHoTOnbO2smjaNCvjKAzm4K67oEkTeOKJfC5KRERCTmEwOAqDklc7j+wkcZW3GumJ\n9BMMazksqGGk6RnprN+//hfDPI+dPEaLqi3O6O1rdlkz34NXekY63+34jpkbvXC4bNcy2tRok9lr\n2Lp6a4oWieyVTNJOpbF239ozgt/yXcspXaw0cdXiaFWtFa2qt6JVtVbUq1DvjKC97/g+Xp73Mh+v\n/JhHrnmER9s/SpniZXz8aeRctm3LCoZz5ngL7ZwOhg88oDB4huPHoWZNb2naGjV8KExEREJKYTA4\nCoNysZxzLN25lA+Wf8DYH8bmOIz0wM8HzujtW7l7Jcn7kqlRtkZmb1+Lqi1oWa0ldcvXLRDz9I6k\nHmFuylxmbpzJrM2z2HV0F9fXvz5zMZr6Fev7Wt+xk8dYuXsly3YtY/mu5SzbtYzVe1ZTu3xtL/QF\ngl9ctTguK31Zrq+76T+beParZ0n6KYnnOz/PXa3u0qI7ES4lxQuHc+fC++8rDJ5h9Gj46CNvf0ER\nESn4FAaDozAooZSansq09dP4YPkHLNi6gNbVW7N+/3oOnThE86rNzxjm2eyyZoWqZ2n74e3M2jSL\nmZtmMmvTLEoXL50ZDK+vfz0VS1YM22vvO77P6+kLhL5lu5aRcjCF2CqxZwS/FlVbhOw9X7pjKb+b\n9Tu2Hd7GiK4j6NOkT4EI8dFOw0TPcuONMGwYDB6c/zWJiEjoKQwGR2FQwmXnkZ18v/N7YqvEUq9C\nPYpYLpc7LAScc6zasyozHC7YsoDYKrGZQ0o71O5A8aLF83TdlEMpXujbuSwz+B1OPZw1zDMQ/GIr\nx4a9x845x4yNM3hy1pOUKV6GP3X7Ex3rdAzra8rFURjMZvt2aN7c+1oyb4tOiYhIhFEYDI7CoEj4\npaansnDrwszFaNbtW0enOp0yF6O5ssqVv+hVS89IZ92+dZnz+04P9ywRUyJzXt/p4Od32D6VcYrE\nVYk8N+c5WlVrxR+7/pHYKrG+1SNnSs9I59jJYxw9eZRa5WspDJ72pz/B+vXw3ns+FSUiIiGnMBgc\nhUGR/Lf/+H7m/DQnczGan9N/pluDbrSq1iozAK7eu5qaZWueEfziqsVF9EqeJ9JP8PfFf2fkgpH0\nbdKX4fHDqVFWi3LklnOO1FOpHD15NFe30wHvaNr5zzt56iRlipehdLHS7Hx8p8IggHPeHhxvv+1t\n2igiIoWDwmBwFAZF/LfxwEZmbprJqt2raFK5Ca2qt6Jl1ZaUvaSs36XlyYGfD/DHr//Iv5b/iweu\nfoAnOj5BuUvK+V1Wvjicepg5m+dw4OcDOYezcwS308EupkgMZYqXOe+tdLHSFzwn+61ETInMnmcN\nEw1YuhQGDICNG7W3oIhIYaIwGByFQREJl5SDKTyf9DwzfpzBM9c+w31X35enuZKR7sDPB5i6birj\n14xnXso82tduT42yNShT7DyBrvgvA13pYqXDPs9TYTDg4Yfh0kth+HD/ahIRkdBTGAyOwqCIhNuK\nXSt4ctaT/HjgR/7Q9Q8MaDqgwK88uufYHiavncyE5Al8u+1butbvSv+m/bnpipsoX6K83+Wdk8Ig\ncPKkt7fgokXQoIHPhYmISEgpDAZHYVBE8svsTbP53azfUdSK8qfufyK+XrzfJQVlx5EdTEyeyITk\nCSzbuYwel/egf9P+9Ly8Z+aempFOYRCYMgX++leYN8/nokREJOQUBoOjMCgi+SnDZTDuh3E889Uz\nxFaJZUTXETSv2tzvss4p5WAKE5InMCF5Asl7k7m50c0kxCZwQ8MbKFms4G1HoDAI9OsHv/oV3H23\nz0WJiEjIKQwGR2FQRPyQmp7K29+9zR/m/4GbrriJl7q8RK1ytfwuC4AN+zdkBsCfDv5E78a9SYhN\noGuDrgV+zmPUh8H9+72hoVu2QPnIHc4rIiJ5pDAYHIVBEfHToROHGLlgJP9c+k/uaX0PT3V6igol\nKuR7HWv2rmH8mvFMSJ7AnmN76NukLwmxCXSu15mYIjH5Xk+4RH0YfPNNmD8fxozxuyIREQkHhcHg\nKAyKSCTYfng7LyS9wNR1U3m609M80OYBLom5JGyv55xjxe4VTFgzgfHJ4zl68igJsQkkxCbQoXYH\nihYpGrbX9lPUh8G2beGll6BHD78rEhGRcFAYDI7CoIhEktV7VvPU7Kf4Yc8PvNzlZQY3H0wRKxKS\nazvnWLJjSWYAdM7Rv2l/EmITaFOzTcheJ5JFdRhcs8bRtas3RDSm8PT2iohINgqDwVEYFJFINC9l\nHk/MfIK0U2mM7DaS7g275+k6GS6DhVsXMmGNNwewVLFSmQEwrlpcgd/iIlhRHQafesqRng5//rPf\n1YiISLgoDAZHYVBEIpVzjgnJE3h69tPUr1Cfkd1G0qp6qws+Lz0jnXkp85iwZgKT1k6icqnKJMQm\n0L9pf5pWaRp1ATC7qA6DNWs6vvgCmjXzuxoREQkXhcHgKAyKSKRLO5XGu9+/y0tzX6J7w+683OVl\n6laoe8Y5J0+dZM7mOYxfM54p66ZQp3wdbw5g0wQaVWrkU+WRJ6rDYOvWjqVL/a5ERETCSWEwOAqD\nIlJQHEk9wl8W/oW/L/k7w1oO47EOj7F0x1LGJ49n+vrpNK7UmP5N+9Mvth/1KtTzu9yIFNVh8LXX\nHI884nclIiISTgqDwVEYFJGCZtfRXbyY9CL/Wv4v2tdqT0JsAv1i+1GzXE2/S4t4UR0Gd+92XHaZ\n35WIiEg4KQwGR2FQRAqqDJcRFSuAhlJe2siwv8Nm1sPM1prZejN78jzntTGzNDPrl+3YT2a2wsyW\nmdni872OgqCIiBQWF9N2iogUBgqC+SOs77KZFQH+DtwIXAkMNrMm5zhvBDDjrG9lAPHOuVbOubbh\nrDWaJCUl+V1CgaL3Kzh6v4Kj90vOFoK2U/JI/x+Do/crOHq/gqP3K3+EO3K3BTY451Kcc2nAWKB3\nDuc9BIwH9px13MiH3stoo/9cwdH7FRy9X8HR+yU5uNi2U/JI/x+Do/crOHq/gqP3K3+EO2jVBLZm\ne7wtcCyTmdUA+jjn/oEX/rJzwEwzW2Jm94S1UhERkchwsW2niIhIrsT4XQDwGpB9PkT2Rq2jc26n\nmVXBC4XJzrn5+VueiIhIxDlf2ykiIpIrYV1N1MzaAcOdcz0Cj58CnHNuZLZzNp2+C1QGjgH3Ouem\nnnWtF4AjzrlXc3gdLZUmIhIlCvtqoiFuO9U+iohEkYjaWsLMigLrgK7ATmAxMNg5l3yO898Hpjnn\nJppZKaCIc+6omZUGvgRedM59GbaCRUREfHYxbWf+VSkiIoVBWIeJOudOmdl/4wW5IsAo51yymd3n\nfdu9c/ZTst2vCkwKfKoZAyQqCIqISGF3kW2niIhIrhWKTedFREREREQkOAV624bcbsorYGa1zOwr\nM1ttZqvM7GG/ayoIzKyImX1vZlMvfHZ0M7PyZvaZmSUH/p1d43dNkczMfmtmP5jZSjNLNLPiftcU\nacxslJntNrOV2Y5VNLMvzWydmc0ws/J+1hjJ1EbmntrIvFEbmXtqI4OjNvL8Qtk+FtgwmNtNeSVT\nOvCoc+5KoD3woN6vXHkEWON3EQXE68DnzrlYoCWQ4/wmydwW4CGgtXOuBd5Q+Fv9rSoivY/3Oz67\np4BZzrnGwFfA0/leVQGgNjJoaiPzRm1k7qmNzCW1kbkSsvaxwIZBcr8prwDOuV3OueWB+0fxfgnV\nPP+zopuZ1QJ+Bbzndy2RzszKAdc6594HcM6lO+cO+1xWpCsKlDazGKAUsMPneiJOYCuh/5x1uDfw\nYeD+h0CffC2q4FAbGQS1kcFTG5l7aiPzRG3keYSyfSzIYfCCm/JKzsysHhAHLPK3koj3f8ATaHGG\n3KgP7DOz9wNDht4xs5J+FxWpnHM7gL8CW4DtwEHn3Cx/qyowLnPO7QbvD3jgMp/riVRqI/NIbWSu\nqY3MPbWRQVAbmWd5ah8LchiUPDCzMsB44JHAp5+SAzO7Cdgd+KTY0IbOFxIDtAbedM61Bo7jDVeQ\nHJhZBbxP8OoCNYAyZjbE36oKLP0hKiGjNjJ31EYGTW1kENRGhkyu2seCHAa3A3WyPa4VOCbnEOhq\nHw987Jyb4nc9Ea4jcEtgY+cxQBcz+8jnmiLZNmCrc+67wOPxeA2f5KwbsMk5d8A5dwqYCHTwuaaC\nYreZVQUws2rAHp/riVRqI4OkNjIoaiODozYyOGoj8yZP7WNBDoNLgMvNrG5ghaFbAa1mdX7/AtY4\n5173u5BI55z7vXOujnOuAd6/ra+cc7f7XVekCgxL2GpmjQKHuqJFBc5nC9DOzEqYmeG9X1pMIGdn\n9zpMBYYF7t8B6I/2nKmNDJ7ayFxSGxkctZFBUxuZOyFpH8O66Xw4nWtTXp/Lilhm1hEYCqwys2V4\nXce/d8594W9lUog8DCSaWTFgE3Cnz/VELOfcYjMbDywD0gJfz95IPOqZ2WggHqhkZluAF4ARwGdm\n9hsgBRjoX4WRS21kcNRGSj5QG5lLaiMvLJTtozadFxERERERiUIFeZioiIiIiIiI5JHCoIiIiIiI\nSBRSGBQREREREYlCCoMiIiIiIiJRSGFQREREREQkCikMioiIiIiIRCGFQZFCysw6m9k0v+sQERGJ\nNGojRTwKgyKFmzYSFRERyZnaSIl6CoMiPjOzoWa2yMy+N7N/mFkRMztiZq+a2Q9mNtPMKgXOjTOz\nb8xsuZlNMLPygeMNA+ctN7PvzKx+4PJlzewzM0s2s499+yFFRETyQG2kSHgpDIr4yMyaAIOADs65\n1kAGMBQoBSx2zjUD5gEvBJ7yIfCEcy4O+CHb8UTgjcDxDsDOwPE44GGgKdDQzDqE/6cSERG5eGoj\nRcIvxu8CRKJcV6A1sMTMDCgB7MZr8D4NnPMJMMHMygHlnXPzA8c/BD41szJATefcVADn3EkA73Is\nds7tDDxeDtQDFubDzyUiInKx1EaKhJnCoIi/DPjQOffMGQfNnjvrPJft/GCkZrt/Cv2fFxGRgkNt\npEiYaZioiL9mA/3NrAqAmVU0szpAUaB/4JyhwHzn3GHggJl1DBy/DZjrnDsKbDWz3oFrFDezkvn6\nU4iIiISe2kiRMNMnICI+cs4lm9mzwJdmVgQ4Cfw3cAxoG/j0czfenAmAO4B/BhqyTcCdgeO3Ae+Y\n2UuBawzI6eXC95OIiIiEltpIkfAz5/RvXyTSmNkR51xZv+sQERGJNGojRUJHw0RFIpM+pREREcmZ\n2kiREFHPoIiIiIiISBRSz6CIiIiIiEgUUhgUERERERGJQgqDIiIiIiIiUUhhUEREREREJAopDIqI\niIiIiEQhhUEREREREZEo9P8BG7R14RY2saAAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff25e4c0810>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# summarize history for accuracy\n",
    "plt.figure(figsize=(15, 5))\n",
    "plt.subplot(1, 2, 1)\n",
    "plt.plot(hist.history['acc']); plt.plot(hist.history['val_acc']);\n",
    "plt.title('model accuracy'); plt.ylabel('accuracy');\n",
    "plt.xlabel('epoch'); plt.legend(['train', 'valid'], loc='upper left');\n",
    "\n",
    "# summarize history for loss\n",
    "plt.subplot(1, 2, 2)\n",
    "plt.plot(hist.history['loss']); plt.plot(hist.history['val_loss']);\n",
    "plt.title('model loss'); plt.ylabel('loss');\n",
    "plt.xlabel('epoch'); plt.legend(['train', 'valid'], loc='upper left');\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "model.save_weights('../pickles/vgg16_aug_lessdrop.pkl')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test generator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Predictions for Augmented - 0\n",
      "Found 1000 images belonging to 1 classes.\n",
      "Predictions for Augmented - 1\n",
      "Found 1000 images belonging to 1 classes.\n",
      "Predictions for Augmented - 2\n",
      "Found 1000 images belonging to 1 classes.\n",
      "Predictions for Augmented - 3\n",
      "Found 1000 images belonging to 1 classes.\n",
      "Predictions for Augmented - 4\n",
      "Found 1000 images belonging to 1 classes.\n"
     ]
    }
   ],
   "source": [
    "test_aug = 5\n",
    "test_datagen = ImageDataGenerator(\n",
    "        shear_range=0.2,\n",
    "        zoom_range=0.1,\n",
    "        rotation_range=10.,\n",
    "        width_shift_range=0.2,\n",
    "        height_shift_range=0.2,\n",
    "        horizontal_flip=True)\n",
    "\n",
    "for aug in range(test_aug):\n",
    "    print('Predictions for Augmented -', aug)\n",
    "    random_seed = np.random.random_integers(0, 100000)\n",
    "\n",
    "    test_generator = test_datagen.flow_from_directory(\n",
    "            '../test/',\n",
    "            target_size=size,\n",
    "            batch_size=bath_size,\n",
    "            shuffle = False,\n",
    "            seed = random_seed,\n",
    "            classes = None,\n",
    "            class_mode = None)\n",
    "\n",
    "    test_image_list = test_generator.filenames\n",
    "    if aug == 0:\n",
    "        predictions = model.predict_generator(test_generator, nb_test_samples)\n",
    "    else:\n",
    "        predictions += model.predict_generator(test_generator, nb_test_samples)\n",
    "\n",
    "predictions /= test_aug"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Begin to write submission file ..\n",
      "0 / 1000\n",
      "100 / 1000\n",
      "200 / 1000\n",
      "300 / 1000\n",
      "400 / 1000\n",
      "500 / 1000\n",
      "600 / 1000\n",
      "700 / 1000\n",
      "800 / 1000\n",
      "900 / 1000\n"
     ]
    }
   ],
   "source": [
    "# clip predictions\n",
    "c = 0\n",
    "preds = np.clip(predictions, c, 1-c)\n",
    "\n",
    "print('Begin to write submission file ..')\n",
    "f_submit = open(os.path.join('/home/trooper/Documents/Nature/pooled', \\\n",
    "                             'submit_' + str(pool_counter) +'.csv'), 'w')\n",
    "f_submit.write('image,ALB,BET,DOL,LAG,NoF,OTHER,SHARK,YFT\\n')\n",
    "for i, image_name in enumerate(test_generator.filenames):\n",
    "    pred = ['%.6f' % p for p in preds[i, :]]\n",
    "    if i%100 == 0:\n",
    "        print(i, '/', 1000)\n",
    "    f_submit.write('%s,%s\\n' % (os.path.basename(image_name), ','.join(pred)))\n",
    "\n",
    "f_submit.close()"
   ]
  }
 ],
 "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": 0
}