src/04-pretrained-vec-dot-classifier.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Pretrained Vectors - Dot Product\n",
    "\n",
    "We train traditional (non neural network) classifiers from Scikit-Learn by looking up the vectors for each given image pair. These image vectors were [generated using pre-trained networks from Keras Model Zoo](http://localhost:8888/notebooks/pretrained-nets-vectorizers.ipynb). In this notebook, we merge the vectors for the image pair by calculating their element-wise dot product normalized by the L2 norms of the two vectors."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false
   },
   "outputs": [],
   "source": [
    "from __future__ import division, print_function\n",
    "from sklearn.ensemble import RandomForestClassifier\n",
    "from sklearn.externals import joblib\n",
    "from sklearn.model_selection import train_test_split, KFold\n",
    "from sklearn.naive_bayes import GaussianNB\n",
    "from sklearn.metrics import accuracy_score, confusion_matrix, classification_report\n",
    "from sklearn.svm import LinearSVC\n",
    "from xgboost import XGBClassifier\n",
    "import itertools\n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import os\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "DATA_DIR = \"../data\"\n",
    "IMAGE_DIR = os.path.join(DATA_DIR, \"holiday-photos\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def get_holiday_triples(image_dir):\n",
    "    image_groups = {}\n",
    "    for image_name in os.listdir(image_dir):\n",
    "        base_name = image_name[0:-4]\n",
    "        group_name = base_name[0:4]\n",
    "        if image_groups.has_key(group_name):\n",
    "            image_groups[group_name].append(image_name)\n",
    "        else:\n",
    "            image_groups[group_name] = [image_name]\n",
    "    num_sims = 0\n",
    "    image_triples = []\n",
    "    group_list = sorted(list(image_groups.keys()))\n",
    "    for i, g in enumerate(group_list):\n",
    "        if num_sims % 100 == 0:\n",
    "            print(\"Generated {:d} pos + {:d} neg = {:d} total image triples\"\n",
    "                  .format(num_sims, num_sims, 2*num_sims))\n",
    "        images_in_group = image_groups[g]\n",
    "        sim_pairs_it = itertools.combinations(images_in_group, 2)\n",
    "        # for each similar pair, generate a corresponding different pair\n",
    "        for ref_image, sim_image in sim_pairs_it:\n",
    "            image_triples.append((ref_image, sim_image, 1))\n",
    "            num_sims += 1\n",
    "            while True:\n",
    "                j = np.random.randint(low=0, high=len(group_list), size=1)[0]\n",
    "                if j != i:\n",
    "                    break\n",
    "            dif_image_candidates = image_groups[group_list[j]]\n",
    "            k = np.random.randint(low=0, high=len(dif_image_candidates), size=1)[0]\n",
    "            dif_image = dif_image_candidates[k]\n",
    "            image_triples.append((ref_image, dif_image, 0))\n",
    "    print(\"Generated {:d} pos + {:d} neg = {:d} total image triples\"\n",
    "          .format(num_sims, num_sims, 2*num_sims))\n",
    "    return image_triples\n",
    "\n",
    "def load_vectors(vector_file):\n",
    "    vec_dict = {}\n",
    "    fvec = open(vector_file, \"rb\")\n",
    "    for line in fvec:\n",
    "        image_name, image_vec = line.strip().split(\"\\t\")\n",
    "        vec = np.array([float(v) for v in image_vec.split(\",\")])\n",
    "        vec_dict[image_name] = vec\n",
    "    fvec.close()\n",
    "    return vec_dict\n",
    "\n",
    "def preprocess_data(vector_file, train_size=0.7):\n",
    "    xdata, ydata = [], []\n",
    "    vec_dict = load_vectors(vector_file)\n",
    "    for image_triple in image_triples:\n",
    "        X1 = vec_dict[image_triple[0]]\n",
    "        X2 = vec_dict[image_triple[1]]\n",
    "        xdata.append(np.multiply(X1, X2) / (np.linalg.norm(X1, 2) * np.linalg.norm(X2, 2)))\n",
    "        ydata.append(image_triple[2])\n",
    "    X, y = np.array(xdata), np.array(ydata)\n",
    "    Xtrain, Xtest, ytrain, ytest = train_test_split(X, y, train_size=train_size)\n",
    "    return Xtrain, Xtest, ytrain, ytest\n",
    "\n",
    "def cross_validate(X, y, clf, k=10):\n",
    "    best_score, best_clf = 0.0, None\n",
    "    kfold = KFold(k)\n",
    "    for kid, (train, test) in enumerate(kfold.split(X, y)):\n",
    "        Xtrain, Xtest, ytrain, ytest = X[train], X[test], y[train], y[test]\n",
    "        clf.fit(Xtrain, ytrain)\n",
    "        ytest_ = clf.predict(Xtest)\n",
    "        score = accuracy_score(ytest_, ytest)\n",
    "        print(\"fold {:d}, score: {:.3f}\".format(kid, score))\n",
    "        if score > best_score:\n",
    "            best_score = score\n",
    "            best_clf = clf\n",
    "    return best_clf, best_score\n",
    "\n",
    "def test_report(clf, Xtest, ytest):\n",
    "    ytest_ = clf.predict(Xtest)\n",
    "    print(\"\\nAccuracy Score: {:.3f}\".format(accuracy_score(ytest_, ytest)))\n",
    "    print(\"\\nConfusion Matrix\")\n",
    "    print(confusion_matrix(ytest_, ytest))\n",
    "    print(\"\\nClassification Report\")\n",
    "    print(classification_report(ytest_, ytest))\n",
    "    \n",
    "def get_model_file(data_dir, vec_name, clf_name):\n",
    "    return os.path.join(data_dir, \"models\", \"{:s}-{:s}-dot.pkl\"\n",
    "                        .format(vec_name, clf_name))\n",
    "\n",
    "def save_model(model, model_file):\n",
    "    joblib.dump(model, model_file)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Generated 0 pos + 0 neg = 0 total image triples\n",
      "Generated 300 pos + 300 neg = 600 total image triples\n",
      "Generated 1100 pos + 1100 neg = 2200 total image triples\n",
      "Generated 1800 pos + 1800 neg = 3600 total image triples\n",
      "Generated 2072 pos + 2072 neg = 4144 total image triples\n"
     ]
    }
   ],
   "source": [
    "image_triples = get_holiday_triples(IMAGE_DIR)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "NUM_VECTORIZERS = 5\n",
    "NUM_CLASSIFIERS = 4\n",
    "scores = np.zeros((NUM_VECTORIZERS, NUM_CLASSIFIERS))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# VGG-16 Vectors"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2900, 4096) (1244, 4096) (2900,) (1244,)\n"
     ]
    }
   ],
   "source": [
    "VECTOR_FILE = os.path.join(DATA_DIR, \"vgg16-vectors.tsv\")\n",
    "Xtrain, Xtest, ytrain, ytest = preprocess_data(VECTOR_FILE)\n",
    "print(Xtrain.shape, Xtest.shape, ytrain.shape, ytest.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Naive Bayes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.890\n",
      "fold 1, score: 0.859\n",
      "fold 2, score: 0.859\n",
      "fold 3, score: 0.852\n",
      "fold 4, score: 0.876\n",
      "fold 5, score: 0.834\n",
      "fold 6, score: 0.828\n",
      "fold 7, score: 0.848\n",
      "fold 8, score: 0.848\n",
      "fold 9, score: 0.859\n",
      "\n",
      "Accuracy Score: 0.847\n",
      "\n",
      "Confusion Matrix\n",
      "[[542 117]\n",
      " [ 73 512]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.88      0.82      0.85       659\n",
      "          1       0.81      0.88      0.84       585\n",
      "\n",
      "avg / total       0.85      0.85      0.85      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = GaussianNB()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[0, 0] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"vgg16\", \"nb\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SVM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.852\n",
      "fold 1, score: 0.817\n",
      "fold 2, score: 0.797\n",
      "fold 3, score: 0.797\n",
      "fold 4, score: 0.772\n",
      "fold 5, score: 0.807\n",
      "fold 6, score: 0.814\n",
      "fold 7, score: 0.831\n",
      "fold 8, score: 0.786\n",
      "fold 9, score: 0.728\n",
      "\n",
      "Accuracy Score: 0.805\n",
      "\n",
      "Confusion Matrix\n",
      "[[544 172]\n",
      " [ 71 457]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.88      0.76      0.82       716\n",
      "          1       0.73      0.87      0.79       528\n",
      "\n",
      "avg / total       0.82      0.80      0.81      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = LinearSVC()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[0, 1] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"vgg16\", \"svm\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## XGBoost"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.934\n",
      "fold 1, score: 0.893\n",
      "fold 2, score: 0.948\n",
      "fold 3, score: 0.931\n",
      "fold 4, score: 0.903\n",
      "fold 5, score: 0.897\n",
      "fold 6, score: 0.903\n",
      "fold 7, score: 0.934\n",
      "fold 8, score: 0.886\n",
      "fold 9, score: 0.931\n",
      "\n",
      "Accuracy Score: 0.917\n",
      "\n",
      "Confusion Matrix\n",
      "[[572  60]\n",
      " [ 43 569]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.93      0.91      0.92       632\n",
      "          1       0.90      0.93      0.92       612\n",
      "\n",
      "avg / total       0.92      0.92      0.92      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = XGBClassifier()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[0, 2] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"vgg16\", \"xgb\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random Forest"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.893\n",
      "fold 1, score: 0.862\n",
      "fold 2, score: 0.886\n",
      "fold 3, score: 0.872\n",
      "fold 4, score: 0.879\n",
      "fold 5, score: 0.866\n",
      "fold 6, score: 0.879\n",
      "fold 7, score: 0.876\n",
      "fold 8, score: 0.817\n",
      "fold 9, score: 0.845\n",
      "\n",
      "Accuracy Score: 0.861\n",
      "\n",
      "Confusion Matrix\n",
      "[[568 126]\n",
      " [ 47 503]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.92      0.82      0.87       694\n",
      "          1       0.80      0.91      0.85       550\n",
      "\n",
      "avg / total       0.87      0.86      0.86      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = RandomForestClassifier()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[0, 3] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"vgg16\", \"rf\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "# VGG-19"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2900, 4096) (1244, 4096) (2900,) (1244,)\n"
     ]
    }
   ],
   "source": [
    "VECTOR_FILE = os.path.join(DATA_DIR, \"vgg19-vectors.tsv\")\n",
    "Xtrain, Xtest, ytrain, ytest = preprocess_data(VECTOR_FILE)\n",
    "print(Xtrain.shape, Xtest.shape, ytrain.shape, ytest.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Naive Bayes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.838\n",
      "fold 1, score: 0.862\n",
      "fold 2, score: 0.855\n",
      "fold 3, score: 0.859\n",
      "fold 4, score: 0.866\n",
      "fold 5, score: 0.876\n",
      "fold 6, score: 0.824\n",
      "fold 7, score: 0.848\n",
      "fold 8, score: 0.869\n",
      "fold 9, score: 0.866\n",
      "\n",
      "Accuracy Score: 0.841\n",
      "\n",
      "Confusion Matrix\n",
      "[[527 128]\n",
      " [ 70 519]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.88      0.80      0.84       655\n",
      "          1       0.80      0.88      0.84       589\n",
      "\n",
      "avg / total       0.84      0.84      0.84      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = GaussianNB()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[1, 0] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"vgg19\", \"nb\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SVM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.741\n",
      "fold 1, score: 0.831\n",
      "fold 2, score: 0.786\n",
      "fold 3, score: 0.783\n",
      "fold 4, score: 0.766\n",
      "fold 5, score: 0.755\n",
      "fold 6, score: 0.810\n",
      "fold 7, score: 0.766\n",
      "fold 8, score: 0.834\n",
      "fold 9, score: 0.800\n",
      "\n",
      "Accuracy Score: 0.781\n",
      "\n",
      "Confusion Matrix\n",
      "[[545 220]\n",
      " [ 52 427]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.91      0.71      0.80       765\n",
      "          1       0.66      0.89      0.76       479\n",
      "\n",
      "avg / total       0.82      0.78      0.78      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = LinearSVC()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[1, 1] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"vgg19\", \"svm\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## XGBoost"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.903\n",
      "fold 1, score: 0.934\n",
      "fold 2, score: 0.910\n",
      "fold 3, score: 0.924\n",
      "fold 4, score: 0.931\n",
      "fold 5, score: 0.893\n",
      "fold 6, score: 0.893\n",
      "fold 7, score: 0.910\n",
      "fold 8, score: 0.941\n",
      "fold 9, score: 0.931\n",
      "\n",
      "Accuracy Score: 0.913\n",
      "\n",
      "Confusion Matrix\n",
      "[[563  74]\n",
      " [ 34 573]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.94      0.88      0.91       637\n",
      "          1       0.89      0.94      0.91       607\n",
      "\n",
      "avg / total       0.92      0.91      0.91      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = XGBClassifier()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[1, 2] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"vgg19\", \"xgb\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random Forest"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.841\n",
      "fold 1, score: 0.907\n",
      "fold 2, score: 0.890\n",
      "fold 3, score: 0.855\n",
      "fold 4, score: 0.879\n",
      "fold 5, score: 0.845\n",
      "fold 6, score: 0.848\n",
      "fold 7, score: 0.900\n",
      "fold 8, score: 0.897\n",
      "fold 9, score: 0.900\n",
      "\n",
      "Accuracy Score: 0.853\n",
      "\n",
      "Confusion Matrix\n",
      "[[552 138]\n",
      " [ 45 509]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.92      0.80      0.86       690\n",
      "          1       0.79      0.92      0.85       554\n",
      "\n",
      "avg / total       0.86      0.85      0.85      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = RandomForestClassifier()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[1, 3] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"vgg19\", \"rf\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Inception V3"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2900, 2048) (1244, 2048) (2900,) (1244,)\n"
     ]
    }
   ],
   "source": [
    "VECTOR_FILE = os.path.join(DATA_DIR, \"inception-vectors.tsv\")\n",
    "Xtrain, Xtest, ytrain, ytest = preprocess_data(VECTOR_FILE)\n",
    "print(Xtrain.shape, Xtest.shape, ytrain.shape, ytest.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Naive Bayes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.917\n",
      "fold 1, score: 0.883\n",
      "fold 2, score: 0.831\n",
      "fold 3, score: 0.897\n",
      "fold 4, score: 0.900\n",
      "fold 5, score: 0.859\n",
      "fold 6, score: 0.900\n",
      "fold 7, score: 0.897\n",
      "fold 8, score: 0.893\n",
      "fold 9, score: 0.872\n",
      "\n",
      "Accuracy Score: 0.883\n",
      "\n",
      "Confusion Matrix\n",
      "[[612 121]\n",
      " [ 24 487]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.96      0.83      0.89       733\n",
      "          1       0.80      0.95      0.87       511\n",
      "\n",
      "avg / total       0.90      0.88      0.88      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = GaussianNB()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[2, 0] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"inceptionv3\", \"nb\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SVM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.817\n",
      "fold 1, score: 0.859\n",
      "fold 2, score: 0.852\n",
      "fold 3, score: 0.876\n",
      "fold 4, score: 0.845\n",
      "fold 5, score: 0.841\n",
      "fold 6, score: 0.869\n",
      "fold 7, score: 0.797\n",
      "fold 8, score: 0.828\n",
      "fold 9, score: 0.866\n",
      "\n",
      "Accuracy Score: 0.839\n",
      "\n",
      "Confusion Matrix\n",
      "[[498  62]\n",
      " [138 546]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.78      0.89      0.83       560\n",
      "          1       0.90      0.80      0.85       684\n",
      "\n",
      "avg / total       0.85      0.84      0.84      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = LinearSVC()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[2, 1] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"inceptionv3\", \"svm\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## XGBoost"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.959\n",
      "fold 1, score: 0.941\n",
      "fold 2, score: 0.938\n",
      "fold 3, score: 0.962\n",
      "fold 4, score: 0.928\n",
      "fold 5, score: 0.952\n",
      "fold 6, score: 0.966\n",
      "fold 7, score: 0.959\n",
      "fold 8, score: 0.952\n",
      "fold 9, score: 0.952\n",
      "\n",
      "Accuracy Score: 0.937\n",
      "\n",
      "Confusion Matrix\n",
      "[[593  35]\n",
      " [ 43 573]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.93      0.94      0.94       628\n",
      "          1       0.94      0.93      0.94       616\n",
      "\n",
      "avg / total       0.94      0.94      0.94      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = XGBClassifier()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[2, 2] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"inceptionv3\", \"xgb\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random Forest"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.934\n",
      "fold 1, score: 0.893\n",
      "fold 2, score: 0.914\n",
      "fold 3, score: 0.907\n",
      "fold 4, score: 0.883\n",
      "fold 5, score: 0.890\n",
      "fold 6, score: 0.931\n",
      "fold 7, score: 0.917\n",
      "fold 8, score: 0.917\n",
      "fold 9, score: 0.900\n",
      "\n",
      "Accuracy Score: 0.891\n",
      "\n",
      "Confusion Matrix\n",
      "[[605 105]\n",
      " [ 31 503]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.95      0.85      0.90       710\n",
      "          1       0.83      0.94      0.88       534\n",
      "\n",
      "avg / total       0.90      0.89      0.89      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = RandomForestClassifier()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[2, 3] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"inceptionv3\", \"rf\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# ResNet 50"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2900, 2048) (1244, 2048) (2900,) (1244,)\n"
     ]
    }
   ],
   "source": [
    "VECTOR_FILE = os.path.join(DATA_DIR, \"resnet-vectors.tsv\")\n",
    "Xtrain, Xtest, ytrain, ytest = preprocess_data(VECTOR_FILE)\n",
    "print(Xtrain.shape, Xtest.shape, ytrain.shape, ytest.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Naive Bayes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.824\n",
      "fold 1, score: 0.831\n",
      "fold 2, score: 0.828\n",
      "fold 3, score: 0.845\n",
      "fold 4, score: 0.852\n",
      "fold 5, score: 0.848\n",
      "fold 6, score: 0.883\n",
      "fold 7, score: 0.859\n",
      "fold 8, score: 0.872\n",
      "fold 9, score: 0.845\n",
      "\n",
      "Accuracy Score: 0.863\n",
      "\n",
      "Confusion Matrix\n",
      "[[571 116]\n",
      " [ 54 503]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.91      0.83      0.87       687\n",
      "          1       0.81      0.90      0.86       557\n",
      "\n",
      "avg / total       0.87      0.86      0.86      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = GaussianNB()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[3, 0] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"resnet50\", \"nb\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SVM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.869\n",
      "fold 1, score: 0.824\n",
      "fold 2, score: 0.834\n",
      "fold 3, score: 0.869\n",
      "fold 4, score: 0.838\n",
      "fold 5, score: 0.852\n",
      "fold 6, score: 0.838\n",
      "fold 7, score: 0.793\n",
      "fold 8, score: 0.831\n",
      "fold 9, score: 0.824\n",
      "\n",
      "Accuracy Score: 0.853\n",
      "\n",
      "Confusion Matrix\n",
      "[[587 145]\n",
      " [ 38 474]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.94      0.80      0.87       732\n",
      "          1       0.77      0.93      0.84       512\n",
      "\n",
      "avg / total       0.87      0.85      0.85      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = LinearSVC()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[3, 1] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"resnet50\", \"svm\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## XGBoost"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.883\n",
      "fold 1, score: 0.907\n",
      "fold 2, score: 0.886\n",
      "fold 3, score: 0.910\n",
      "fold 4, score: 0.890\n",
      "fold 5, score: 0.852\n",
      "fold 6, score: 0.903\n",
      "fold 7, score: 0.869\n",
      "fold 8, score: 0.848\n",
      "fold 9, score: 0.862\n",
      "\n",
      "Accuracy Score: 0.879\n",
      "\n",
      "Confusion Matrix\n",
      "[[558  83]\n",
      " [ 67 536]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.89      0.87      0.88       641\n",
      "          1       0.87      0.89      0.88       603\n",
      "\n",
      "avg / total       0.88      0.88      0.88      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = XGBClassifier()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[3, 2] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"resnet50\", \"xgb\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random Forest"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.838\n",
      "fold 1, score: 0.879\n",
      "fold 2, score: 0.824\n",
      "fold 3, score: 0.845\n",
      "fold 4, score: 0.876\n",
      "fold 5, score: 0.841\n",
      "fold 6, score: 0.876\n",
      "fold 7, score: 0.834\n",
      "fold 8, score: 0.776\n",
      "fold 9, score: 0.848\n",
      "\n",
      "Accuracy Score: 0.830\n",
      "\n",
      "Confusion Matrix\n",
      "[[546 132]\n",
      " [ 79 487]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.87      0.81      0.84       678\n",
      "          1       0.79      0.86      0.82       566\n",
      "\n",
      "avg / total       0.83      0.83      0.83      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = RandomForestClassifier()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[3, 3] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"resnet50\", \"rf\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Xception"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(2900, 2048) (1244, 2048) (2900,) (1244,)\n"
     ]
    }
   ],
   "source": [
    "VECTOR_FILE = os.path.join(DATA_DIR, \"xception-vectors.tsv\")\n",
    "Xtrain, Xtest, ytrain, ytest = preprocess_data(VECTOR_FILE)\n",
    "print(Xtrain.shape, Xtest.shape, ytrain.shape, ytest.shape)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Naive Bayes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.914\n",
      "fold 1, score: 0.917\n",
      "fold 2, score: 0.917\n",
      "fold 3, score: 0.931\n",
      "fold 4, score: 0.924\n",
      "fold 5, score: 0.910\n",
      "fold 6, score: 0.903\n",
      "fold 7, score: 0.907\n",
      "fold 8, score: 0.924\n",
      "fold 9, score: 0.907\n",
      "\n",
      "Accuracy Score: 0.911\n",
      "\n",
      "Confusion Matrix\n",
      "[[590  71]\n",
      " [ 40 543]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.94      0.89      0.91       661\n",
      "          1       0.88      0.93      0.91       583\n",
      "\n",
      "avg / total       0.91      0.91      0.91      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = GaussianNB()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[4, 0] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"xception\", \"nb\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SVM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.876\n",
      "fold 1, score: 0.897\n",
      "fold 2, score: 0.914\n",
      "fold 3, score: 0.890\n",
      "fold 4, score: 0.886\n",
      "fold 5, score: 0.886\n",
      "fold 6, score: 0.879\n",
      "fold 7, score: 0.893\n",
      "fold 8, score: 0.897\n",
      "fold 9, score: 0.903\n",
      "\n",
      "Accuracy Score: 0.890\n",
      "\n",
      "Confusion Matrix\n",
      "[[600 107]\n",
      " [ 30 507]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.95      0.85      0.90       707\n",
      "          1       0.83      0.94      0.88       537\n",
      "\n",
      "avg / total       0.90      0.89      0.89      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = LinearSVC()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[4, 1] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"xception\", \"svm\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## XGBoost"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.897\n",
      "fold 1, score: 0.883\n",
      "fold 2, score: 0.903\n",
      "fold 3, score: 0.886\n",
      "fold 4, score: 0.879\n",
      "fold 5, score: 0.859\n",
      "fold 6, score: 0.900\n",
      "fold 7, score: 0.879\n",
      "fold 8, score: 0.862\n",
      "fold 9, score: 0.866\n",
      "\n",
      "Accuracy Score: 0.883\n",
      "\n",
      "Confusion Matrix\n",
      "[[580  95]\n",
      " [ 50 519]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.92      0.86      0.89       675\n",
      "          1       0.85      0.91      0.88       569\n",
      "\n",
      "avg / total       0.89      0.88      0.88      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = XGBClassifier()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[4, 2] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"xception\", \"xgb\"))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Random Forest"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "fold 0, score: 0.831\n",
      "fold 1, score: 0.838\n",
      "fold 2, score: 0.862\n",
      "fold 3, score: 0.834\n",
      "fold 4, score: 0.807\n",
      "fold 5, score: 0.821\n",
      "fold 6, score: 0.834\n",
      "fold 7, score: 0.834\n",
      "fold 8, score: 0.831\n",
      "fold 9, score: 0.831\n",
      "\n",
      "Accuracy Score: 0.830\n",
      "\n",
      "Confusion Matrix\n",
      "[[560 142]\n",
      " [ 70 472]]\n",
      "\n",
      "Classification Report\n",
      "             precision    recall  f1-score   support\n",
      "\n",
      "          0       0.89      0.80      0.84       702\n",
      "          1       0.77      0.87      0.82       542\n",
      "\n",
      "avg / total       0.84      0.83      0.83      1244\n",
      "\n"
     ]
    }
   ],
   "source": [
    "clf = RandomForestClassifier()\n",
    "best_clf, best_score = cross_validate(Xtrain, ytrain, clf)\n",
    "scores[4, 3] = best_score\n",
    "test_report(best_clf, Xtest, ytest)\n",
    "save_model(best_clf, get_model_file(DATA_DIR, \"xception\", \"rf\"))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 0.88965517,  0.85172414,  0.94827586,  0.89310345],\n",
       "       [ 0.87586207,  0.83448276,  0.94137931,  0.90689655],\n",
       "       [ 0.91724138,  0.87586207,  0.96551724,  0.93448276],\n",
       "       [ 0.88275862,  0.86896552,  0.91034483,  0.87931034],\n",
       "       [ 0.93103448,  0.9137931 ,  0.90344828,  0.86206897]])"
      ]
     },
     "execution_count": 31,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "scores"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": false
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.text.Text at 0x7ff2b6d5b6d0>"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAhUAAAGjCAYAAACbn+eLAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAAPYQAAD2EBqD+naQAAIABJREFUeJzs3Xl8VNX9//HXJwKGEFbZLZvIWhUELIhCFURAWSMiIQhi\nf6KiiBQrVQoKWvwWQYQqBVoVKAZBZVPr0lgsqCiVxWqNqAgEkS3KIpCw5fz+uJNxZrJPBpKQ9/Px\nmEcy555z7rlzM5nPnHvOueacQ0RERKSwooq6ASIiInJuUFAhIiIiEaGgQkRERCJCQYWIiIhEhIIK\nERERiQgFFSIiIhIRCipEREQkIhRUiIiISEQoqBAREZGIUFAhZ4yZNTCzDDMbWtRtKUnM7Ne+161z\nMWjLo2aWUdTtOFvMbL6ZbSvqdoTDzHqY2SYzSzOz02ZWqajbJKWPggqRfDKzFmb2iJnVPwu7Ky7r\n5zvCbIuZ3W1mwyLcnjMt7OPNDzO70vc3FNEPfDOrBiwBjgEjgVuBo5HcR2GYWR3fcV9W1G2RM0tB\nhUj+tQQeARqeyZ045/4NlHfOrTmT+zkLRgIlLag40zoCE4EqEa73CiAW+INz7gXnXKJz7nSE91EY\ndfHeO62LuiFyZimoEMk/4yz1IDjnTpyN/ZwrzCymqNuQT3aG6q3l+3noDNVfWGfkuEvQeS81FFRI\nWMysrpk9b2Z7zCzdzD43s+H5LNvMzF4xsx9813//Y2a9Q/IM840ruMrMZpnZPjM7YGZzzKyMmVU2\ns4Vm9qPv8ads9mNmdr+vbWm+ts4xsyoh+bab2Srfvj725d1qZrcGtgdY6nv6nq9tp3Mb92Bmr5rZ\nhpC013xlewWk/cqX1t33PMuYCjO72Fffbl/7dprZYjOrGFL/EDP7xMyO+V7fxWb2i9zOR0DZq33n\nIs3MvjazETnkO8/MJpjZN75zv83M/mhm5QLybAN+CVzjO5YMM/tXLvvOHH/zW9852+47hvfM7Jch\neeeb2U9mdpGZ/cPMDgOLArbfHPAa7Dezv5tZ3Wz22S/gb+O/ZtYvmzzZjm+xHMYL+f62l/r+Xo+Z\n2Zdm9rhv2yPAVF/W7QF/Q7leTsvreMxsNTDf9/QTX73P51DXTb7tnbLZdqdvW8uQ48n1verLV9nM\nZvj+FtJ9f58LzKyamf0aWI8XkM8POO6hAeXzPGe5nXcza5Kf94eceWWKugFS8phZTeBj4DQwC0gF\negLPmVlF59ysXMr+Engf+A54Au+670BghZnFOedWhhT5M7Abr8u4A3AHcBCvG3kH8BBwA/CAmX3m\nnFsUUHYeMBR4HpgJNAJGAa3N7KqA7mEHNAFeBp7D+wd9O/CCmX3inEsG1viOdRTwOPClr2xyLi/V\nWqCPmcU654740jr6XrdOwOu+tM6+tA8Cyvp7RMysLPAOUNbXhj3AhUAvvG70n3z5xgOTgZeAvwI1\ngPuAf5vZ5c65wzk11MwuAd4G9uG91mWBR33PQz2H97ouBaYB7fHOQ3PgJl+e0cAzvrY9jvdNdW9O\n+w8wDK8b/xkg2lfPu2Z2qXNuf8BrU8bX3rXAWLyxBJjZbXjn+2Pg93jf4O8HOga+BmZ2PfAK8Lkv\n3wXAC3h/l6Hy1Ttl3niBtcBxYC7e32djvPP0B2AZ0BQY5DuuH3xF92ep7Oc683M8jwNb8N4bfwC2\nA1tzqPIN4Ajee25tyLaBwOfOuS98+87Xe9XMKvjyNcP729gEVAf6AL8AvsD7m5rse10y9/thAY4R\ncjjvvvfH2+Tx/pCzxDmnhx4FegB/w/tHUyUkPRH4ETjf97wBkAEMDciThPdPp0xI2feBLwOeD/OV\nfSMk3wd4H8DPBKRFASnAvwLSrvaVvyWkfDdf+qCAtG2+OjsGpFUH0oCpAWk3+fJ1zufr1Na3r+6+\n55f4nr8EfBiQbwXwScDzXwfuB2jlK9c/l33VB04C40LSWwIngN/n0dbleB8aFwakNfPVeTog7TJf\nW+aElJ/qa/OvA9I+Czwneew/82/lCFA7IP0KX/q0gLQXfPt6PKSOMngfKJuBcgHpN/jqeCQgbZPv\nbzg2IK2rL9+3OZ2LbNob+Lf9b7yA98JcjnOsr776+XhNCnI8w3z1tslHvS/iBeoWkFYLOAU8HMZ7\ndZJv333y8V4YWohjzOm85/n+0OPsPXT5Q8IRB7wGnGdmF2Q+8L5NVwbaZFfIzKoC1+L1CFTOpmwT\nM6sTUMThfYMJ9LHvpz/dOZcBfAJcFJBvAN4/+HdD9rMJ74Pr2pB6v3DOfRhQZyret7+LCF/mvjK7\nzjsBO4GFQFszi/alX03Wb42BMq+T9zCz8jnkuQmvN+DlkOPdB3xN1uP1M7Mo4HpguXNuV2a6c24L\n3jfAQDfgnZcZIenTffu/MZfjyI/lzrk9AW34D945vyGbvHNCnrcDagKzXcCYFOfcP/B6lm4EMLPa\neB9E893PPUg4597F+1ZdYGZWHe/8Phf4GhZSvo4nDEt89V4TkHYz3vlbCgV+r8YBnzrnVoXRlnCO\nMfS85+f9IWeJggopEDOrgdelOAKv2zbwkflBXzOH4hfj/eN6LJuyj+ZQNiXkeeY/kJ3ZpFcNeN7E\n1859IfvZB1TIx34ADoTUmS0zq2RmtQIeVcEf7KzD+7DB93MtXm/LeUAH3/XrauQSVDjntuN9aP8/\nINXM3jKzkRY8LfFivPfzN9kcb/NsjjdQDaC8r2yoLSHPM7+hB+V1zu3FC+Ia5LKf/MiuDV+RdcbN\nKedc6KWKBngBz1fZ1PFlQNsyf+bnePMrM/j8X5jls5Pf4ymot4DDwC0BaQOBzc65zNekIO/VxniX\nkcJR0GPMct7z+f6Qs0RjKqSgMgPRRcCCHPL8N4+y08j6DThT6D/6nKbFZZceOMI8Cu8a/uCQ9Eyh\n17Fz2k9+Rq3PJHjq5HtAF9/v7wMPm9n5eEHFY865Q2b2ue/5Prx/qrn1VOCc+52ZzQf64vUqzAIe\nMrP2zrnv8Y43A+jh+xnqSDZphVHU62gcP0v7yek4zztL+48459wJM1sB9DezkUAd4Cq88QyZwnmv\nng3Znvcc3h+/N7MOvveHnCUKKqSg9uMNfDrPOZfjaP4cfOv7eTKMsgW1Fe8a+YfOuUh9AOX0AfMn\n4O8Bzw8E/L4WKAfE483Vzwwe1uBdFtkLfOV+HoSY886d+x/eN+EpZtYBb6DbXXiD4LbiBUDbA75t\n5td+vPEjTbLZ1jzk+Q68D5wmBHyr9w3ereLb7m9yAdtBDm1oijf4MC878F6DZniBXaBmAW3L/Jnd\nvpqFPD/gqzN0XYmGIc8z/7YvyaONBXlN8ns84ViCN9i2K94sHfh5dhMU7L26lfCPO2LHmMf7Q84S\nXf6QAvF16b8K3GQhU/3Af205p7L78f5x3Om7rp3vsmFYihc0Z/mHYt6UyMph1HmUbD5gnHNfOuf+\nFfDYFLD5Y7wBcOOAH503kwS84KIDXmCRay+FmVU0s9Bvxv/D65E43/d8me/5IznUUS2n+n3n9G2g\nnwVMPzWzFnjf+gL9A+81uD8kfSzeB8cbAWlHKfgiT/0seLrkr/Bml/wjH2U/wev5ucs3IyCzjp5A\nC3yzbXxjNjYDwwKnHJpZN7yBrYF24BuoGZI+koAPSt8YnDXA7WZWL5c2Zq5ymZ/XJV/HE6YkvIBp\nEN6lj/XOOf8HeAHfq68Crcysby77y+m4C32M+Xx/yFmingoJx+/xBnl9bGZ/xRvcVg1vhHcXvJkT\nObkH70P0M1/Zb/FGnl+JNw3s8oC8YS+Y45xbY2Zz8bpAW+MNLjuJ9613AN5Uy2UFrHYz3gfMOPPW\nujgOvOv7QMmpHWnmrVXRAQgcyLYGb2xHDNkHFYHH3gV4xsxexrv2XAbvW+YpvH/oOOe+NbM/4H1L\na4Q3o+QnvGv9/fCm8j2Vy7E9gnfp5H0zm403Pe9evGvl/qWVnXP/NbMFwAjf2JF/433oDwWWOW81\n0Ewb8D4sxuN1le9zzq3OpQ348r1vZn/h5yml+4En8yiHc+6UmY3DG9uzxswWA7XxzvW3wNMB2R/C\n+8D6wLw1HS4ION7YgDoP+173+8wMvG/lvfDGoYS6D+9cbjSzeXizihoBNzjnMv+uN+Cd2ylm9hLe\n3+Qq51xaIY8HCvB+8dW9DC+oiMELCkPl9736JN576mUze8F3jBcAvYE7nXOf4b1uB/H+Ho7gBRkf\nO+e2F/AYs5Pn+0POoqKefqJHyXzgBQ6z8Lql04FdeB/ctwfkaYD3IRw6jawh3vSwXb6yKcBKAqaE\nkcMUObwPv9NAtZD0F4BD2bTzN3gL7xzB+6e2GZgC1ArI8y2wMpuyq/GChsC02/FmU5wgn9NL8S6P\nnAbGhqR/hfePr2FIeuiU0oZ46058hffPeD/eN81rstlXP7wP+sO+x//wxnxcnI92Xu17rdJ8x3hH\n5usdki8Kbz2Eb3znbzvegL6yIflq4gVSB33Hk+P0Un4eAPpbvF6Q7XhrT6wGLsnPuQ7YPgDvG/Ax\n32u1AKiTw2v1uS/fZ3jX418AtobkuwCv5+snvDVZnsX7Fp3d33YLvPUvfvCdqy8ImBbpy/Mw3t/8\nSfIxvTQ/x0MBppQGlOnqK3MSqJtDnoZk/17tF5Kviu/vLMX397MDb82KqgF5evle5+Ohr10+jzGn\n93hD8vn+0OPMP8x3UkREioyZNcD7Zv+Acy63HhURKcaKxZgKM+tk3jLJu8xbwrVPPspcY2YbzFsS\n9isreXdDFBEROacUi6AC79ryZkIGP+XEzBriXQ99F28Rm5nA33wDrURERKQIFIuBms65t/AWZMF8\no6HycDfeUroP+p5vMbOrgTHAP89MK0XkDHMU/foXIlIIxSKoCEMHvIE4gd4m69LBIlICOG86Y4ld\nUEpEPMXl8kdB1SbrHQ/3ApV8KxeKiIjIWVZSeyoKzHcjnO78PAVSRERE8icab/ru2865H3LKVFKD\nij14i7AEqgUcdjkvydwd75a/IiIiEp4EIDGnjSU1qFgH9AxJu96XnpPtAIsWLaJFixZnqFn5N2bM\nGGbM0BCQc4XO57lF5/PcovNZeMnJyQwZMgTyuA9PsQgqzKwCP99qF+AiM2uFd6+EnWb2BN6Kb5lr\nUcwB7jGzP+Et79oVb0W2G3LZTTpAixYtaNOmzZk4jAKpXLlysWiHRIbO57lF5/PcovMZUbkOHygu\nAzXbAZvw1ox3wHRgIzDJt7024L9Jj3NuO3AjcB3e+hZjgN8450JnhIiIiMhZUix6Kpx3E6IcAxzn\n3PBs0tbg3cBKREREioHi0lMhIiIiJZyCiiISHx9f1E2QCNL5PLfofJ5bdD7PnlJzl1IzawNs2LBh\ngwbsiIiIFMDGjRtp27YtQFvn3Mac8qmnQkRERCKiWAzUFBEpTlJSUkhNTS1UHdWrV6d+/foRapFI\nyaCgQkQkQEpKCs2aNyM9rXCr+UeXj2bLl1sUWEipoqBCRCRAamqqF1DEAdXDrQTSl6WTmpqqoEJK\nFQUVIiLZqQ7ULepGiJQsGqgpIiIiEaGgQkRERCJClz9EpFiIxIwL0KwLkaKkoEJEilxKSgotmjXj\nWHrhZlwAxERHk7xFsy5EioKCChEpcqmpqRxLT2cR0KIQ9SQDQ9LTWbt2LS1ahFdTcnJyIVogUrop\nqBCRYqMFUJhF9HcDEMWQIUMi0h4RKRgFFSJyzjgIQAYUqs/jH8CECLVIigON1zl7FFSUcnqzybmp\nMH0euvxxLonUCqmgVVLzQ0FFKZaSkkKzFi1IP3as0HVFx8SwJTlZbzYRKVYiskIqaJXUfFJQUYql\npqZ6AcWiRRDmoDYAkpNJHzJEbzYRKb60QupZoaBCvICiTWGGx4mIiCioKDK6tbKIiJxrFFQUAd1a\nWc4kBawiUlQUVBQB3VpZzpRIDb7VwNvIiMRCWgrwpCRRUFGUNHBIIiwig2818LbwjkBUFBFZhCsm\nJprkZPVISsmgoELkXKTBt0UrHTIy4OGHoUGD8KvZsQOmTCndPZKFvZynZdfPLgUVIiJnSIMG0LRp\n4esp7AdjSb2EkpKSQrNmLUhPL/xaOnJ2KKgQESmmfvwRiCr8vUxK6hiZ1NRUX0ChZddLCgUVIiLF\n1JEjeNdRSv0YGS27XlIoqBARKe5K4BiZSExt1niIkqfUBRWl9dqkiMjZkpKSQotmzTiWXvibeEnJ\nUuqCisJem4yJjiZ5i6Z3iYjkJDU1lWPp6YUaCQEaDVESlbqg4jHghjDLJgND0ovX9K7C9Lyoa1FE\nzqTCjIQAjYYoiUpdUNGIwv2RFxsRXFxHRETyR5fQc1fqgopIKOwfVUR6CCKwuM7HH8Pzzxe+KaL7\nbYic8yL0Re5cXyFVQUUB7Aag8HPGI6kwi+ukpES2LaWVbhAnUgpE4ItcaVghVUFFARwEIIPCLcQC\nGn50bonkDeLWrl1Li3DXI0DjZETOtEitknquUlARFg0/kmwU5gZxGiMjIucABRUixUGEbkClcTIi\nxV8kehSL6xgsBRUixUhhu1Y1Tkak+IrUvVyg+N7PRUGFiIjIWRCRe7lAsb6fi4IKERGRs6kE3ssl\nvxRUSImltSFERIoXBRVSIkXqhkW6l4uISOQoqJASKRI3LCqO93IRESnJFFRIiVbYFUNERCRyooq6\nASIiInJuUE+FlHrF4gZxIiLnAAUVUmoVxxvEiYiUZAoqpNTSDeJERCJLQYWIbhAnIhIRGqgpIiIi\nEaGgQkRERCKi2AQVZnaPmW0zszQz+8jMrsgjf4KZbTazo2b2vZk9Z2bVzlZ7RUREJFixCCrM7BZg\nOvAIcDnwKfC2mVXPIf9VwALgr0BLYADwK2DeWWmwiIiIZFEsggpgDDDXObfQOfclcBdwDLg9h/wd\ngG3OuWedczuccx8Cc/ECCxERESkCRR5UmFlZoC3wbmaac84BScCVORRbB9Qzs56+OmoBNwNvnNnW\nioiISE6KPKgAqgPnAXtD0vcCtbMr4OuZGAIsMbMTeOsYHQDuPYPtFBERkVwUh6CiwMysJTATeBRv\ngYHuQCO8SyAiIiJSBIrD4lepwGmgVkh6LWBPDmV+D3zgnHvK9/xzMxsJrDWz8c650F4Pv+nAkpC0\neN9DRESktFu8eDGLFy8OSjt06FC+yhZ5UOGcO2lmG4CuwCoAMzPf81k5FIsBToSkZQAOsNz2NxZI\nKEyDRUREzmHx8fHExwd/1d64cSNt27bNs2xxufzxFHCHmQ01s+bAHLzAYT6AmT1hZgsC8r8G3GRm\nd5lZI98U05nAx865nHo3RERE5Awq8p4KAOfcUt+aFJPxLntsBro75/b7stQG6gXkX2BmscA9wDS8\ne0O9i3dZRERERIpAsQgqAJxzs4HZOWwbnk3as8CzZ7pdIiIikj/F5fKHiIiIlHAKKkRERCQiFFSI\niIhIRCioEBERkYhQUCEiIiIRoaBCREREIkJBhYiIiESEggoRERGJCAUVIiIiEhEKKkRERCQiFFSI\niIhIRCioEBERkYhQUCEiIiIRoaBCREREIkJBhYiIiESEggoRERGJCAUVIiIiEhEKKkRERCQiFFSI\niIhIRCioEBERkYhQUCEiIiIRoaBCREREIkJBhYiIiESEggoRERGJCAUVIiIiEhEKKkRERCQiFFSI\niIhIRCioEBERkYhQUCEiIiIRoaBCREREIkJBhYiIiESEggoRERGJCAUVIiIiEhEKKkRERCQiFFSI\niIhIRCioEBERkYhQUCEiIiIRoaBCREREIkJBhYiIiESEggoRERGJCAUVIiIiEhEKKkRERCQiFFSI\niIhIRCioEBERkYhQUCEiIiIRoaBCREREIkJBhYiIiESEggoRERGJCAUVIiIiEhEKKkRERCQiFFSI\niIhIRBSboMLM7jGzbWaWZmYfmdkVeeQvZ2Z/NLPtZpZuZt+a2W1nqbkiIiISokxRNwDAzG4BpgMj\ngPXAGOBtM2vqnEvNodjLQA1gOLAVqEMxCpJERERKm2IRVOAFEXOdcwsBzOwu4EbgdmBqaGYz6wF0\nAi5yzh30JaecpbaKiIhINor8m72ZlQXaAu9mpjnnHJAEXJlDsd7AJ8A4M/vOzLaY2ZNmFn3GGywi\nIiLZKg49FdWB84C9Iel7gWY5lLkIr6ciHejnq+MvQDXgN2emmSIiIpKb4hBUhCMKyAAGO+eOAJjZ\nb4GXzWykc+54kbZORESkFCoOQUUqcBqoFZJeC9iTQ5ndwK7MgMInGTDgF3gDN7M1HVgSkhbve4iI\niJR2ixcvZvHixUFphw4dylfZsIIKM7vWObc6nLKhnHMnzWwD0BVY5avffM9n5VDsA2CAmcU45475\n0prh9V58l9v+xgIJkWi4iIjIOSg+Pp74+OCv2hs3bqRt27Z5lg13oOZbZrbVzP5gZvXCrCPQU8Ad\nZjbUzJoDc4AYYD6AmT1hZgsC8icCPwAvmFkLM+uMN0vkOV36EBERKRrhBhUXAs8AA4BvzextMxto\nZuXCqcw5txR4AJgMbAIuA7o75/b7stQG6gXkPwp0A6oA/wH+DqwERod3OCIiIlJYYV3+8C1INQOY\nYWZt8Bagmg3MNrNEvB6DTwtY52xfHdltG55N2ldA94K2XURERM6MQq9T4ZzbCDyB13MRi7dg1QYz\nW2tmvyxs/SIiIlIyhB1UmFlZMxtgZv8AduD1GtyLN2vjYl/ayxFppYiIiBR74c7++DPeLEzDG8/w\noHPu84AsR83sAeD7wjdRRERESoJw16loCYwCluUy2yIVuDbM+kVERKSECXegZtd85DkF/Duc+kVE\nRKTkCWtMhZk9ZGZZZmSY2e1mNq7wzRIREZGSJtyBmncCX2ST/j/grvCbIyIiIiVVuEFFbWBfNun7\ngTrhN0dERERKqnCDip3AVdmkX4VmfIiIiJRK4c7++CvwtJmVBf7lS+uKd/+N6ZFomIiIiJQs4QYV\nTwIX4C2rnXm/j3TgT865JyLRMBERESlZwp1S6oBxZvYY0AJIA77WHUJFRERKr3B7KgBwzh3Bu0uo\niIiIlHJhBxVm1g4YCNTn50sgADjn4grZLhERESlhwl38ahDwId6lj/5AWeCXQBfgUMRaJyIiIiVG\nuFNKHwbGOOd6AyeA0UBzYCmQEqG2iYiISAkSblDRGHjD9/sJoIJv8OYMYEQkGiYiIiIlS7hBxQGg\nou/3XcAlvt+rADGFbZSIiIiUPOEO1FwDdAM+A14GZppZF1/auxFqm4iIiJQg4QYV9wLRvt//CJwE\nOgKvAo9HoF0iIiJSwhQ4qDCzMkAv4G0A51wG8H8RbpeIiIiUMAUeU+GcOwXM4eeeChEREZGwB2qu\nB1pHsiEiIiJSsoU7pmI28JSZ1QM2AEcDNzrn/lvYhomIiEjJEm5Q8ZLv56yANAeY7+d5hWmUiIiI\nlDzhBhWNItoKERERKfHCvfX5jkg3REREREq2sIIKMxua23bn3MLwmiMiIiIlVbiXP2aGPC+Ltzz3\nCeAYoKBCRESklAn38kfV0DQzawL8BXiysI0SERGRkifcdSqycM59DfyerL0YIiIiUgpELKjwOQXU\njXCdIiIiUgKEO1CzT2gSUAfvRmMfFLZRIiIiUvKEO1BzRchzB+wH/gWMLVSLREREpEQKd6BmpC+b\niIiISAkXbk+FiIiEae9eOHQo73y7d/t+SU4Of2e+ssmFqaPAu/Tts5D1bPu5xsLXklq4tnDA+5GS\nEn4VETmfAeUjeU6rV69O/fr1C11PuGMqXgU+cs49GZL+IHCFc+7mQrdMROQctHcv3HZbFOnpGfkv\nNGRIofc7JAJ1FHifxammZYWvAuCPf4xAJRE6F5E8pzExMSQnJxc6sAi3p6IzMDGb9DfRmAoRkRwd\nOgTp6RksWrSIFi1aFHVzREhOTmbIkCGkpqYWWVARizd9NNRJoFL4zRERKR1atGhBmzZtiroZIhEV\n7oDLz4BbskkfBHwRfnNERESkpAq3p+IxYJmZNcabRgrQFYgHNJ5CRESkFAp3SulrZtYPeBgYAKQB\n/wWuc879O4LtExERkRIi7Cmlzrk3gDci2BYREREpwcIaU2FmV5hZ+2zS25tZu8I3S0REREqacHsq\nngWeyCb9QmAckCXgEBGRvKWkpJCaWtiVmgon3IWQ+vbtS1JSEvv27aNChQrZ5klISOCVV15hz549\nVK1alRMnTjB37lyWLl3K//73P44ePUr16tVp164d8fHxDBw4kKio4O+/P/30E7NmzWLFihV8/fXX\npKenU7t2bTp06MDQoUO54YYb8mzr0aNHmTp1KuvXr2f9+vUcOHCA+fPnM3To0GzzO+eYM2cO8+bN\nY8uWLcTExNCqVSuefvppLr300gK/VueqcIOKlsDmbNI3+baJiEgBpaSk0KJZM46lpxdpO2Kio0ne\nsqXAgUVCQgKvv/46y5cvz3ZhprS0NFatWsUNN9xA1apVSU1NpUePHmzatInu3bszYcIEqlWrxp49\ne0hKSiIhIYGtW7cyfvx4fx3ffPMN3bt3Z+fOnfTv359hw4YRGxvLzp07+cc//kHv3r1ZuHAhCQkJ\nubY1NTWVxx57jAYNGtC6dWvee++9XPMPHz6cxYsXM3ToUEaNGsXRo0fZtGkT+/btK9BrdK4LN6g4\nDtQmcBVVTx2yX79CRETykJqayrH0dBYBRbUsVjIwJD09rIWQ+vTpQ2xsLImJidkGFStWrODo0aP+\nD/whQ4bw6aefsmzZMvr27RuUd9y4cWzcuJEtW7b4006fPk3//v3Zv38/a9asoUOHDkFlJkyYQFJS\nEqdPn86zrXXr1mXPnj3UrFmTDRs2cMUVV+SYd+nSpSxcuJAVK1bQp0/oTbolULhBxTvAE2bW1zl3\nCMDMqgBTgH9GqnEiIqVRC6AkLosVHR1NXFwciYmJpKamUr169aDtiYmJVKpUid69e7Nu3Treeecd\nRo4cmSWgyNSmTZugBcIyL5FMnTo1S0CR6brrrstXW8uWLUvNmjXzlXfGjBm0b9+ePn364JwjLS2N\nmJiYfJUtbcJd/OoBoB6ww8xWm9lqvF6L2miZbhGRUishIYGTJ0+ydOnSoPQDBw7wzjvvEBcXx/nn\nn89rr72e6fnuAAAgAElEQVSGmeV5mSLQ66+/XuAyhfXTTz+xfv16rrjiCsaPH0/lypWJjY2lcePG\nvPzyy2etHSVFuOtU7DKzy4AEoBXeOhUvAIudcycj2D4RESlBunTpQp06dUhMTGTkyJH+9KVLl3Lq\n1Cl/QJB5WeOSSy4JKn/8+HGOHDnif16mTBkqV64MwJdffkmVKlWoU6dOUJljx46Rlpbmf16uXDkq\nVqwYkePZunUrzjkWL15M2bJlmTZtGpUqVWLmzJkMGjSIypUrc/3110dkX+eCcHsqcM4dBd4HXgPW\nAAeBnmamC04iIqVUVFQUgwYNYt26daQE3Cc8MTGRWrVq0aVLFwAOHz4MQGxsbFD5OXPmUKNGDf+j\nU6dO/m2HDx/Okh9g/PjxQWUi2ZORGeD8+OOPrFq1ihEjRjBo0CCSkpK44IILePzxxyO2r3NBuOtU\nXGRmnwKf4y2AtQJYHvAIp857zGybmaWZ2UdmlvOomeByV5nZSTPbGM5+RUQkshISEnDOkZiYCMCu\nXbt4//33iY+Px8wA/D0Jgb0SAAMGDCApKYmkpCQuu+yyoG0VK1bMkh/gnnvu8ZepVauWPz0jI4O9\ne/cGPU6eLFhnevny5QFo1KgR7dr9vAxThQoV6N27N+vXrycjowC3sT/HhdtTMRNvDEVN4BhwCfBr\n4BPgmoJWZma3ANOBR4DLgU+Bt82seh7lKgMLgKSC7lNERM6MNm3a0Lx5cxYvXgzgDy4GDx7sz9O8\neXMAPv/886CyF154IV26dKFLly5UrVo1aFvz5s05ePAgu3fvDkq/+OKL/WWio6P96Tt37qROnTrU\nrVvX/3PdunUFOpa6desCBAUrmWrWrMnJkyc5evRogeo8l4UbVFwJTHTOpQIZwGnn3PvAQ8CsMOob\nA8x1zi10zn0J3IUXrNyeR7k5wIvAR2HsU0REzpCEhAQ+//xzPvvsMxYvXkyTJk1o27atf3uvXr1w\nzvHiiy/mu86ClqlduzZJSUn885//9P9s1apVgY6jTp061K5dm127dmXZtmvXLqKjoyM2fuNcEG5Q\ncR7wk+/3VKCu7/cdQLOCVGRmZYG2wLuZac45h9f7cGUu5YYDjYBJBdmfiIiceZmXQCZOnMjmzZuz\nrFvRsWNHunXrxrx581i1alW2dXgfBT8bOHAgLVu25LHHHuPjjz/Os8z555/v78HIfGQO+iyIW265\nhZ07d/Luu/6PKVJTU1m1ahVdu3YtcH3nsnDXqfgcb9bHNuBj4EEzOwGMAL4tYF3V8YKUvSHpe8kh\nQDGzJnhrYlztnMvIvEYnInIuSD4H9t2wYUM6duzIypUrMbOgSx+ZFi1aRM+ePenfvz89evTguuuu\no2rVqv4VNdeuXRu05HaZMmVYvnw5PXr04OqrryYuLo5OnTpRoUIFdu3axapVq9i5cye9e/fOVxuf\nffZZDh486O+FyCwPcN999/l7IB566CGWLl3KTTfdxJgxY6hUqRJz587l1KlTTJkypbAv1Tkl3KDi\ncSBzYfeJwOvAWuAH4JYItCtHZhaFd8njEefc1szkM7lPEZGzoXr16sRERzOkGCzTHbpwVTgSEhJY\nt24d7du356KLLsqyvUaNGnz44YfMnTuXJUuWMHnyZI4dO+a/90diYiIDBw4MKtOkSRM2b97MrFmz\nWL58OW+99RYnTpygVq1atG/fnkmTJtGzZ898tW/atGn+GSpmxvLly1m+3JtrcOutt/qDipo1a/LB\nBx/wwAMP8PTTT3Py5Ek6duxIYmJilimxpV2461S8HfD7N0BzM6sGHHCh/VV5SwVOA6GjYGoBe7LJ\nXxFoB7Q2s2d9aVGA+XpLrnfOvZfTzqYDS0LS4n0PEZGiVL9+fZK3bCmxNxQLdffdd3P33Xfnmqdc\nuXKMGjWKUaNG5bveihUrMn78+KB7goRj27bQO03krGHDhrzyyiuF2l9JsXjxYv8g20yHDh3KV9lw\neyqycM79GGa5k2a2AegKrAIvOvA9z27Q52G82SaB7gGuBW4Ctue2v7F4K3aJiBRH9evXj8gHuki4\n4uPjiY8P/qq9cePGoIG2OYlYUFFITwHzfcHFerzZIDHAfAAzewKo65wb5usJ+SKwsJntA9Kdc0V5\nKVJERKRUKxZBhXNuqW9Nisl4lz02A92dc/t9WWrj3WtEREREiqliEVQAOOdmA7Nz2DY8j7KT0NRS\nERGRIhX2vT9EREREAimoEBERkYhQUCEiIiIRoaBCREREIkJBhYiIiESEggoRERGJCAUVIiIiEhEK\nKkRERCQiis3iVyIiAikpKSX2hmJ9+/YlKSmJffv2UaFChWzzJCQk8Morr7Bnzx6qVq3KiRMnmDt3\nLkuXLuV///sfR48e9d+lND4+noEDBxIVFfz996effmLWrFmsWLGCr7/+mvT0dGrXrk2HDh0YOnRo\n0O3Sc3L06FGmTp3K+vXrWb9+PQcOHGD+/PkMHTo02/zPPPMMs2fP5ttvv6V69erccsstPPbYY8TE\nxBT4dTqXKagQESkmUlJSaNasBenpx4q0HdHRMWzZklzgwCIhIYHXX3+d5cuXM2TIkCzb09LSWLVq\nFTfccANVq1YlNTWVHj16sGnTJrp3786ECROoVq0ae/bsISkpiYSEBLZu3Rp0N9JvvvmG7t27s3Pn\nTvr378+wYcOIjY1l586d/OMf/6B3794sXLiQhITcbx2ZmprKY489RoMGDWjdujXvvfdejnnHjRvH\nk08+ycCBA7n//vv54osv+POf/8wXX3zBm2++WaDX6FynoEJEpJhITU31BRSLgBZF1Ipk0tOHkJqa\nWuCgok+fPsTGxpKYmJhtULFixQqOHj3q/8AfMmQIn376KcuWLaNv375BeceNG8fGjRvZsmWLP+30\n6dP079+f/fv3s2bNGjp06BBUZsKECSQlJXH69Ok821q3bl327NlDzZo12bBhA1dccUW2+fbs2cOM\nGTMYNmwYL7zwgj+9SZMm3HfffbzxxhvceOONee6vtFBQISJS7LQA2hR1IwosOjqauLg4EhMTSU1N\npXr16kHbExMTqVSpEr1792bdunW88847jBw5MktAkalNmza0afPz65B5iWTq1KlZAopM1113Xb7a\nWrZsWWrWrJlnvnXr1nH69GluueWWoPRBgwYxatQoXnrpJQUVATRQU0REIiYhIYGTJ0+ydOnSoPQD\nBw7wzjvvEBcXx/nnn89rr72GmeV5mSLQ66+/XuAyhXX8+HEAypcvH5SeOZZiw4YNZ60tJYGCChER\niZguXbpQp04dEhMTg9KXLl3KqVOn/AFB5mWNSy65JCjf8ePH+eGHH/yPQ4cO+bd9+eWXVKlShTp1\n6gSVOXbsWFCZn376KWLH06xZM5xzfPDBB0Hpa9asAWDXrl0R29e5QEGFiIhETFRUFIMGDWLdunWk\npKT40xMTE6lVqxZdunQB4PDhwwDExsYGlZ8zZw41atTwPzp16uTfdvjw4Sz5AcaPHx9UJpI9GZdf\nfjnt27fnT3/6E/Pnz2fHjh28+eab3HXXXZQtW5a0tLSI7etcoKBCREQiKiEhAeecv7di165dvP/+\n+8THx2NmAFSsWBGAI0eOBJUdMGAASUlJJCUlcdlllwVtq1ixYpb8APfcc4+/TK1atfzpGRkZ7N27\nN+hx8uTJAh/PsmXLaNWqFb/5zW9o1KgRffv25ZZbbuHyyy/PNsgpzRRUiIhIRLVp04bmzZuzePFi\nAH9wMXjwYH+e5s2bA/D5558Hlb3wwgvp0qULXbp0oWrVqkHbmjdvzsGDB9m9e3dQ+sUXX+wvEx0d\n7U/fuXMnderUoW7duv6f69atK/Dx1KlThzVr1vDVV1+xdu1avvvuO/7v//6PnTt30rRp0wLXdy5T\nUCEiIhGXkJDA559/zmeffcbixYtp0qQJbdu29W/v1asXzjlefPHFfNdZ0DK1a9cmKSmJf/7zn/6f\nrVq1KvCxZGrcuDFXXXUVNWvW5IsvvmD37t1069Yt7PrORQoqREQk4jIvgUycOJHNmzdnWbeiY8eO\ndOvWjXnz5rFq1aps63DOBT0fOHAgLVu25LHHHuPjjz/Os8z555/v78HIfFSuXLmQR+bt48EHH6RC\nhQrceeedha7vXKJ1KkREJOIaNmxIx44dWblyJWYWdOkj06JFi+jZsyf9+/enR48eXHfddVStWtW/\noubatWuDltwuU6YMy5cvp0ePHlx99dXExcXRqVMnKlSowK5du1i1ahU7d+6kd+/e+Wrjs88+y8GD\nB/0zODLLA9x3333+cR/3338/6enptG7dmpMnT/Liiy/yySefsHDhQn7xi18U9qU6pyioEBEpdpLP\niX0nJCSwbt062rdvz0UXXZRle40aNfjwww+ZO3cuS5YsYfLkyRw7dsx/74/ExEQGDhwYVKZJkyZs\n3ryZWbNmsXz5ct566y1OnDhBrVq1aN++PZMmTaJnz575at+0adP8M1TMjOXLl7N8+XIAbr31Vn9Q\ncfnllzNz5kwSExOJioriV7/6Ff/617/o3LlzYV6ec5KCChGRYqJ69epER8eQnp51ieuzKTo6Jstq\nmOG4++67ufvuu3PNU65cOUaNGsWoUaPyXW/FihUZP3580D1BwrFt27Z85Rs2bBjDhg0r1L5KCwUV\nIiLFRP369dmyJbnE3qVUREGFiEgxUr9+fX2gS4ml2R8iIiISEQoqREREJCIUVIiIiEhEKKgQERGR\niFBQISIiIhGhoEJEREQiQkGFiIiIRISCChEREYkIBRUiIiISEQoqREREsnHbbbfRqFGjom5GiaJl\nukVEipGUlJQSfe+PBQsWMHz4cD755BPatGkT4ZZF3u7du5k3bx79+/fnsssuC9pmZkRFnf3v3i1b\ntuT48eNs3bo1xzzt27dn165d7Ny5k7S0NO69917Wr1/Pd999R0ZGBo0bN+Y3v/kNd911F2XKnL2P\negUVIiLFREpKCs2aNyM9Lb1I2xFdPpotX24JO7Awswi36Mz5/vvvmTRpEo0aNcoSVPztb38jIyPj\nrLdpyJAhTJgwgY8++ogOHTpk2f7tt9/yn//8h7Fjx2JmHDt2jC1bttCrVy8aNmxIVFQU77//PqNH\nj+aTTz5h/vz5Z63tCipERIqJ1NRUL6CIAwp/5/EwGwHpy9JJTU0tFTc2c87luO28887jvPPOO4ut\n8QwePJgJEyaQmJiYbVDx4osvYmYMHjwY8HqWPvjgg6A8I0aMIDY2lnnz5jF9+nQuuOCCs9J2jakQ\nESluqgN1i+gR4WDmtttuo2LFinz//ff069ePihUrUrNmTX73u99l+UB3zjFz5kwuu+wyypcvT82a\nNenZsycbN24Myrdo0SLatWtHTEwMF1xwAfHx8Xz33XdBea655houu+wyNm7cyFVXXUVMTAwXXXQR\nc+fO9ef597//za9+9SvMjNtuu42oqCjOO+88Fi5c6G976JiKY8eOMXbsWOrXr090dDTNmzdn+vTp\nWY47KiqK++67j5UrV3LppZcSHR3NJZdcwttvv53na9awYUM6duzIyy+/nG1PyUsvvUSzZs24/PLL\nc62nQYMGOOc4dOhQnvuMFAUVIiJyxpgZGRkZdO/enRo1ajB9+nSuueYannrqKebNmxeU9/bbb2fM\nmDE0aNCAqVOn8tBDD1G+fHk++ugjf54//vGPDBs2jGbNmjFjxgzGjBnDu+++y69//WsOHz4ctN8f\nf/yRG2+8kXbt2vHkk09Sr1497r77bv/lgBYtWjB58mScc9x5550sWrSIv//973Tu3NlfR+ilnN69\nezNz5kxuuOEGZsyYQfPmzfnd737H2LFjsxz72rVrueeee4iPj+fJJ5/k+PHjDBgwgAMHDuT5uiUk\nJLBv3z6SkpKC0jdv3kxycjJDhgzJUubEiRP88MMP7Ny5k1dffZUZM2bQuHHjszrYVJc/RETkjEpP\nTyc+Pp6HH34Y8Lrm27Zty3PPPcedd94JwOrVq1mwYAH3338/Tz31lL/smDFj/L+npKTw6KOPMmXK\nFMaNG+dPj4uLo3Xr1syePZvf//73/vTdu3fz1FNPMXr0aP9+27dvz0MPPcStt97q7wmZOHEiV155\npf9yQk5WrlzJ6tWrmTJlin8/d999NwMHDmTmzJnce++9QR/gX375JcnJyTRs2BDwek9atWrF4sWL\nGTlyZK77GjhwIKNHjyYxMZHrr7/en5556SM+Pj5LmZdffplbb73V/7x9+/Y8//zzZ3WMi3oqRETk\njMsMHjJ16tSJb7/91v/81VdfJSoqiokTJ+ZYx6uvvopzjptvvpkffvjB/6hZsyZNmjRh9erVQfnL\nlCnDiBEj/M/Lli3LnXfeyb59+9iwYUOBj+HNN9+kTJkyjBo1Kih97NixZGRk8Oabbwald+vWzR9Q\nAFx66aVUqlQp6LhzUq1aNXr06MGKFSs4fvy4P33JkiV06NAh296Hbt26kZSUxCuvvMKdd95J2bJl\nOXLkSAGPsnAUVIiIyBkVHR2dZaBg1apVgy4DfPvtt9StW5cqVarkWM8333xDRkYGF198MTVq1PA/\natasyZdffsm+ffuC8tetW5fy5csHpTVt2hTnHNu3by/wcezYsYO6detSoUKFoPQWLVr4tweqV69e\nljoCj/vkyZPs3bs36BE4hiIhIYHDhw/z2muvAd4YkO+++y7bSx8ANWvWpEuXLsTFxfGXv/yFbt26\n0a1bt7M6RVmXP0RE5IyK1AyKjIwMoqKieOutt7JdPyI2NjYi+4mUnI47c4Dqhx9+yLXXXouZ4ZzD\nzNi2bZt/1k2fPn2oWLEiiYmJDBgwgMTERMqWLcvAgQPztf8BAwbwyCOP8NprrzF8+PDIHFQeFFSI\niEiRa9y4Me+88w4HDx7MsbeicePGOOdo2LAhF198cZ51fv/996SlpQX1VmzZsgUz81+WKMh4gwYN\nGvDuu+9y9OjRoN6K5ORk//aCaNWqVZaBmLVr1/b/Hh0dTVxcHC+99BL79+/n1Vdf5frrr8/39NC0\ntDQAzf4QEZHS5aabbiIjI4NJkyblmCcuLo6oqKgc8/z4449Bz0+dOsWcOXP8z0+ePMncuXOpUaMG\nbdu2BfAHBwcPHsyzjTfccAOnTp3imWeeCUqfMWMGUVFR9OzZM886AlWpUoUuXboEPcqVKxeUJyEh\ngePHjzNixAgOHDhAQkJClnp++OGHbOv/61//ipnRrl27ArWrMNRTISIiEZXbglI5ueaaa7j11luZ\nNWsWX331FT169CAjI4O1a9fSpUsXRo4cyUUXXcTjjz/Oww8/zLZt2/zrXnz77besWLGCO++8k9/+\n9rf+OuvWrcvUqVPZvn07TZs25aWXXuK///0vf/3rX/2XJho3bkyVKlWYM2cOsbGxVKhQgQ4dOmTb\n69C7d2+uvfZaxo8fz7Zt22jVqhVvv/02r732GmPGjDkjUze7du1KnTp1WLlyJbGxsfTt2zdLnvnz\n5/P888/Tt29fGjVqxOHDh3nrrbf417/+RVxcHFdffXXE25UTBRUiIsVNUd76IwL7Dr2kkNMlhtD0\n+fPn06pVK5577jkefPBBKleuTLt27ejYsaM/z7hx4/xrVEyePBnwBkT26NGDPn36BNVXtWpVFixY\nwL333svf/vY3atWqxbPPPsvtt9/uz1OmTBkWLlzIQw89xN13382pU6d44YUXGDp0aJY2mhmvvfYa\nEydOZMmSJcyfP5+GDRsybdq0oKmvmXmzO+6c0nNiZgwaNIinn36afv36ZRl4CtC5c2fWr19PYmIi\ne/fupWzZsjRv3pyZM2fmOXU10hRUiIgUE9WrVye6fDTpy4r+3h/Vq4e3tOawYcMYNmyY//kLL7zA\nCy+8kCXfI488wiOPPBKUZmb89re/DeptyE6/fv3o169fvtpz+eWXZ1nCOlSvXr3o1atXlvTs2h0T\nE8O0adOYNm1arnWePn062/T8TCcNNX369GxX7cx0xRVXsGTJkgLXeyYoqBARKSbq16/Pli+3lOi7\nlErppqBCRKQYqV+/vj7QpcTS7A8RETknlaRbsJ8rik1QYWb3mNk2M0szs4/M7Ipc8vY3s3fMbJ+Z\nHTKzD83s+pzyi4hI6bJ69Wo+/fTTom5GqVMsggozuwWYDjwCXA58CrxtZjmNFOoMvAP0BNoAq4HX\nzKzVWWiuiIiIZKNYBBXAGGCuc26hc+5L4C7gGHB7dpmdc2Occ9Occxucc1udc+OBr4HeZ6/JIiIi\nEqjIgwozKwu0Bd7NTHPeyilJwJX5rMOAisCPeeUVERGRM6PIgwqgOnAesDckfS9QO2v2bP0OqAAs\njWC7REREpABK/JRSMxsMTAD6OOeKdnK3iIhIKVYcgopU4DRQKyS9FrAnt4JmNgiYBwxwzq3Oz86m\nA6HrjsX7HiIiIqXd4sWLWbx4cVBafu90WuRBhXPupJltALoCq8A/RqIrMCuncmYWD/wNuMU591Z+\n9zcWyHqPNxEREQGIj48nPj74q/bGjRv9d3bNTZEHFT5PAfN9wcV6vNkgMcB8ADN7AqjrnBvmez7Y\nt+0+4D9mltnLkeacO3x2my4iIiJQTIIK59xS35oUk/Eue2wGujvn9vuy1AbqBRS5A29w57O+R6YF\n5DANVURERM6sYhFUADjnZgOzc9g2POT5tWelUSIiZ1lKSkqJvqHYggULGD7853/Z5513HrVq1aJb\nt2788Y9/pG7dupFqZpBrrrmGNWvW0Lt3b1auXBm0bceOHTRq1Ihp06bleQfUUGlpaUydOpVrr72W\nzp0756vMpEmTmDRpUpb06Ohojh07liV91apVTJo0iS+++IKaNWsyfPhwJkyYwHnnnVegthYHxSao\nEBEp7VJSUmjRohnHjhXtrc9jYqJJTt4SdmBhZjz22GM0bNiQ9PR0PvroI1544QU++OADPv/8c8qV\nKxfhFnv7NDNef/11Nm3axOWXXx6Reo8dO8akSZMws3wHFZntmTNnDhUqVPCnZRckvPnmm/Tv358u\nXbrwzDPP8Nlnn/H444+zf/9+nn322Sz5izsFFSIixURqairHjqXz8MPQoEHRtGHHDpgyJZ3U1NRC\n3S21R48etGnTBoDbb7+dCy64gKlTp7Jq1SoGDBgQqeYGqV+/Pj/99BOTJk1ixYoVEanTW4sxPDfd\ndBPVqlXLNc8DDzxA69atefvtt4mK8paOqlixIk888QSjR4+madOmYe+/KBSHxa9ERCRAgwbQtGnR\nPM5UMNOpUyecc2zdujUo/c0336Rz587ExsZSqVIlevXqxRdffBGUZ+/evQwfPpx69eoRHR1N3bp1\n6devHykpKUH5KlasyJgxY1i1ahWbN2/Os02HDh3i/vvvp379+kRHR9OkSROmTp3qDyR27NhBzZo1\nMTMeffRRoqKiiIqKYvLkyfk65oyMDH766acctycnJ5OcnMyIESP8AQXAyJEjycjI4JVXXsnXfooT\nBRUiInLGbdu2DYCqVav60/7+97/Tq1cvKlasyNSpU5k4cSLJycl06tQpKGCIi4tj5cqV/OY3v+Ev\nf/kLo0eP5siRI1mCCoDRo0dTtWpVHn300Vzbk5aWRufOnUlMTOS2227jz3/+M1dffTUPPfQQY8eO\nBaBGjRrMmTMH5xxxcXEsWrSIRYsWERcXl+fxOue46KKLqFy5MhUrVuTWW29l3759QXk2bdqEmWWZ\nqlmnTh1+8YtfsGnTpjz3U9zo8oeIiETcoUOH+OGHH/xjKiZPnkz58uXp1asXAEePHmX06NGMGDGC\nv/zlL/5yw4YNo2nTpkyZMoU5c+Zw6NAh1q1bl2WQ5bhx47Ldb2xsLPfffz+PPvoomzdvpnXr1tnm\nmz59Otu2bWPz5s1cdNFFANxxxx3UqVOHadOmMXbsWC688EJuuukm7rrrLi677DIGDx6cr2OvWrUq\no0aN4sorr+T8889n7dq1PPPMM/znP//hk08+ITY2FoDdu3cDXhARqk6dOnz//ff52l9xop4KERGJ\nKOccXbt2pUaNGtSrV4+bb76Z2NhYVq1a5Z/98c9//pNDhw4xaNAgfvjhB//DzGjfvj2rV3uLJJcv\nX55y5crx3nvvcfDgwXztf/To0VSpUiXbGRiZXnnlFTp16kTlypWD9t+1a1dOnTrFmjVrwj7+++67\nj5kzZzJo0CD69+/PU089xYIFC/jqq6+YPfvnSY5paWkAnH/++VnqiI6O9m8vSdRTISIiEWVmzJ49\nmyZNmnDo0CGef/551qxZEzTr4+uvv8Y5x7XXZl0hwMyoVKkSAOXKleNPf/oTDzzwALVq1aJDhw70\n6tWLoUOHUqtW6N0dPJUqVfL3Vnz66adUqVIlS56vv/6azz77jBo1amS7/9BLFaHS09OzLF2dU3vA\nW6Vy7NixJCUl8eCDDwJewARw/PjxbOvP3F6SKKgQEZGIu+KKK/yzP/r27cvVV1/N4MGD2bJlCzEx\nMWRkZGBmLFq0KNsP4zJlfv54Gj16NH369GHFihW8/fbbTJw4kSeeeILVq1fTqlWrbPc/evRoZsyY\nwaRJk5gxY0aW7RkZGXTr1o1x48ZlO8Mjr1kXS5YsCVqPw8w4ffp0rmXq1avHjz/+6H+eedlj9+7d\nXHjhhUF5d+/eTfv27XOtrzhSUCEiImdUVFQUTzzxBNdeey3PPPMMDz74II0bN8Y5R40aNejSpUue\ndTRq1IgxY8YwZswYtm7dSqtWrZg+fToLFy7MNn9mb8WkSZMYOnRolu2NGzfmyJEj2faUBPJuRZVV\njx49SEpKyrPdgbZv3+4PtABat26Nc45PPvmEdu3a+dN3797Nd999x1133VWg+osDjakQEZEz7te/\n/jW/+tWvePrppzlx4gTdu3enUqVKTJkyhVOnTmXJn7mqaFpaWpbLA40aNaJixYrZXjYIdP/991O5\ncmUmT56cJTgYOHAg69at45133slS7tChQ/5eh5iYGIAs4zlq1apFly5dgh6hbQ80e/Zs9u/fT8+e\nPe2c/vgAACAASURBVP1pLVu2pHnz5sybNy+ot2T27NlERUVx00035Xp8xZF6KkREipkdO0r2vnNa\nMOp3v/sdN998M/Pnz/fP+hg6dCht2rRh0KBB1KhRg5SUFN544w2uvvpqZs2axVdffUXXrl0ZOHAg\nLVu2pEyZMixbtox9+/ZluZNmqEqVKjF69Gj/ipihbVm1ahW9evXitttuo23bthw9epT//ve/LFu2\njO3bt1OtWjWio6Np2bIlS5YsoUmTJlSrVo1LLrmEX/7ylznut0GDBtxyyy1ceumlREdHs3btWpYs\nWUKbNm0YMWJEUN4nn3ySvn370q1bNwYNGsRnn33Gs88+yx133EGzZs3y+YoXHwoqRESKierVqxMT\nE82UKUW/THf16tXDLp/TJYO4uDgaN27MtGnTuOOOO4iPj+fCCy/k//7v/5g2bRrHjx/nwgsvpFOn\nTv7xCvXq1WPw4MG8++67LFq0iDJlytC8eXNefvll+vXrl+d+77//fmbOnJllUGX58uVZs2YNU6ZM\n4eWXX+bvf/87lSpVomnTpkyePJnKlSv78z733HOMGjWK3/72t5w4cYJHHnkk16BiyJAhfPjhhyxb\ntoz09HQaNGjA73//ex5++GGio6OD8t54440sW7aMSZMmcd9991GjRg3+8Ic/MGHChNxf5GJKQYWI\nSDFRv359kpO3lOgbig0bNoxhw4Zlu83M+Prrr4PSOnfunOs9NapVq8asWbPy3G/mFNRQlStXDhoc\nGSgmJobHH3+cxx9/PNe627dvz/r16/NsQ6a5c+fmOy9Anz596NOnT4HKFFcKKkREipH69f9/e/ce\nb+Wc/n/89S4ddCC/UlsoEWqUQ2iklFHKqRjCoBwmKqJBM5SacWiQnInqO2UcMhiMUxPGYYYZRkZO\nDXsSQg1FSTR0vn5/XJ+1W3tXVHvtvfbhej4ePdj3ute977X32ve67s/n+lxXi1L13AghnyJRM4QQ\nQgg5EUFFCCGEEHIigooQQggh5EQEFSGEEELIiQgqQgghhJATEVSEEEIIISciqAghhBBCTkRQEUII\nIYSciKAihBBCCDkRQUUIIYQQciKCihBCCCHLc889R40aNXj55ZfzfSqVTvT+CCGECuSTTz6p1A3F\n+vfvz0MPPcTMmTNp3bp1scfGjBnDJZdcwtSpUzniiCMAWL58OePGjeOPf/wjs2bNYvny5bRs2ZKe\nPXsydOjQdY6RS7fddhtbbbUV/fv3X+exDXVaDd8vgooQQqggPvnkE3Zv25Zl336b1/OoW68eswoL\nNyuwuOGGG5g2bRqDBw/m2WefLdo+Z84cRo8ezfHHH18UUCxcuJBDDz2Ut99+m969e9OvXz/q16/P\nrFmzuO+++7jjjjtYunRpzl5XSePGjWPHHXdcJ6jo3r073333HbVr1y6z711VRVARQggVxMKFCz2g\nmDIF2rbNz0kUFrKsXz8WLly4WUHFtttuyzXXXMPAgQO55557ij6wzznnHGrXrs1NN91UtG+/fv14\n5513ePTRR+ndu3ex44wePZpRo0aV7rWUQgQUmydyKkIIoaJp2xY6dMjPvxwEM2eeeSadO3fml7/8\nJYsXL+b+++/n6aef5sorr2S77bYD4OWXX+Yvf/kLgwYNWiegAP9QHzt2bLFthYWFHHfccTRu3Jh6\n9erRsWNHpk2bVmyfSZMmFeVDnHXWWTRu3JhGjRpxxhlnsGTJkqL9dtxxR9577z2effZZatSoQY0a\nNejZsydA0baSORX3338/HTp0YMstt6Rp06acdtppzJ8/v9g+/fr1Y5tttmHevHn06dOHhg0b0rRp\nU4YPH775P9BKJIKKEEIIOTdx4kSWLFnC4MGDufDCC+nYsSPnnHNO0eOPP/44kujXr99GHW/mzJl0\n6tSJ999/nxEjRnDdddex5ZZb0qdPH6ZOnVq0XyYX4uyzz+aDDz7giiuuoH///tx999307du3aL9x\n48ZRUFBAu3btuPfee5kyZQojRowoOkbJnIpJkyZx8sknU7duXcaOHcuZZ57Jgw8+yEEHHVRsikYS\nq1atomfPnmy33XZcf/31dO3alWuvvZbJkydv+g+ykonpjxBCCDn3ox/9iGHDhnH11VezxRZb8OST\nTxZ7vLCwEID27dtv1PHOO+88WrduzfTp06lZsybgUyoHHHAAw4cP56ijjiq2f/369YtGHAC23357\nRo4cyVNPPcVhhx3G0UcfzfDhwykoKOCkk0763u+9YsUKRowYwT777MMLL7xArVq1ADjggAM45phj\nuPnmmxk5cmTR/t9++y2nn346F110EQADBw5k7733ZvLkyQwYMGCjXm9lFSMVIYQQykSTJk0AaN68\nOXvssUexx77++mskUa9evR88zsKFC3nxxRc54YQT+Oqrr1i0aBGLFi1i4cKF9OrVi8LCQr744oui\n/SUxaNCgooACYMiQIUhaZ7pkY7z66qssWrSIIUOGFAUUAH369KF169b8+c9/Xuc5AwcOLPZ1ly5d\n+PDDDzf5e1c2EVSEEELIublz53LppZfSvn175s6du05+xFZbbYWZ8e1GrHSZPXs2ACNGjGDbbbct\n+te0aVNGjx4NwOeff17sOSWXojZs2JBmzZrx0UcfbfJr+fjjj5HEbrvtts5jbdq04eOPPy62rUGD\nBjRq1KjYtm222YbFixdv8veubGL6I4QQQs6de+65SOLJJ5/kggsu4Morr+Tkk09mp512AvzDeOrU\nqcycOZMf//jH33usNWvWAHDxxRfTo0eP9e7TqlWrnJ5/aWSmZ0oys3I+k/IXIxUhhBBy6pFHHuGJ\nJ57gt7/9Lc2bN+emm26iVq1aDBkypGif3r17Y2ZMmTLlB4+3yy67AL4i5JBDDlnvv5LTKJnRjYxv\nvvmGBQsWFAU1sPEFrlq2bImZMWvWrHUemzVrFi1bttyo41QHEVSEEELImaVLlzJ06FD23Xdfzj33\nXAC22247Ro8ezVNPPcXDDz8MeI5Bjx49mDhxYrHVGxnLly8vSnQsKCigS5cujB8/fp1pDmCdCqRm\nxsSJE1m9enXRtnHjxmFmHH744UXb6tevz1dfffWDr6ljx440btyY8ePHs2rVqqLtTzzxBLNnz14n\nSbQ6i+mPEEIIOTNy5Ejmz5/PY489VmwkYMiQIdx1112cf/75HHbYYdSvX58pU6bQq1cvjjnmGHr3\n7k337t2pV68e7733Hvfffz+LFi0qysUYP348Xbt2pV27dpx11lm0atWKBQsW8NJLL/H555/z2muv\nFTuP7777jh49etC3b1/effddJkyYwMEHH1wsqNh3332ZPHkyV111FbvssgsFBQV069YNKD5VUbt2\nbcaMGcPAgQPp2rUrJ510Ep9++im33HILrVu3ZujQoWX5I61UIqgIIYSKJi23rGzf+/XXX2f8+PEM\nGTKEDh06FHusRo0aTJgwgU6dOjFq1ChuvPFGmjZtyiuvvMK4ceN44IEHGDlyJCtWrKBFixYcffTR\nnH/++UXP32OPPXjttde47LLL+P3vf8/ixYtp2rQp++yzD7/+9a+LfS9J3H777dx111385je/YfXq\n1Zx66qnFqnkCXHbZZcybN49rrrmGpUuX0r1796KgouTUyIABA2jQoAFjx47l4osvpkGDBhx//PGM\nGTOGBg0arPP916c69BOJoCKEECqIJk2aULdePZZtZEGoslK3Xr2i5aCbokOHDqxYsWKDj++3336s\nXLmy2LY6deowbNgwhg0b9oPHb9WqFXfddddGnUuDBg2YOHEiEydO3OA+BQUF65166d69e7Gpk4wT\nTzyRE0888Xu/7z333LPe7aNHjy5aqVKVRVARQggVRIsWLZhVWFipu5SG6i2CihBCqEBatGgRH+il\nVB2WblZUsfojhBBClVIdchcqqggqQgghVBkDBgxg9erV7Lnnnvk+lWopgooQQggh5EQEFSGEEELI\niQgqQgghhJATEVSEEEIIISdiSWkIIeRBYT6rZoaQJZfvxQgqQgihHG29NdStW4N+ea6aGUK2eptZ\nRbWkCCpCCKEcNWsGd965hiVLfnjf6dPhjjuAKVOgbdvN+4aFhdCvH1OmTKHt5h5jk79lIf369WMK\nUJrvOA3wrh6lOVI6yrFAaT4zZwN/hZEjYXNrk+Xk9wll8jvNVRXVCCpCCKGcNWvm/37IJ5+k/2nb\nFko06NpUbdu2XafJV1lrC5TmO64dlC/NkdJRmgDNS3EyqXJ6ixaw226bd4hc/j79MOX/O/0hFSZR\nU9IQSXMkfSfpFUn7/8D+B0uaIWmZpPcknVZe5xpCCCGEdVWIoELSicD1wKXAPsBbwNOS1jtYJWkn\nYCrwHLAXcDMwSdKh5XG+IYQQQlhXhQgqgAuAiWZ2t5n9BxgMfAv8fAP7nw18aGYXmdksM7sNeCgd\nJ4QQQgh5kPegQlItYF981AEA8xZzzwKdNvC0A9Lj2Z7+nv1DCCGEUMbyHlTg6TM1gQUlti8ACjbw\nnIIN7L+VpDq5Pb0QQgghbIzqtPqjLsBLpTjA2udOIzsvebOPNJuijOJNlrKIp0/PyijeRDNnpv+Z\nNs2XKG2uOXOA8i3mk/lepflNVLXfJ+Tod1pJf5+Qq99p/D5Lq8r9PiGuuWu/V93v208+05A/afrj\nW+A4M3s8a/udwNZm9tP1POcFYIaZXZi17XTgRjPbZgPf52Tg3tyefQghhFCtnGJmf9jQg3kfqTCz\nlZJmAN2BxwEkKX19ywae9k/g8BLbeqbtG/I0cArwEbCsFKccQgghVDd1gZ3wz9INyvtIBYCkE4A7\n8VUfr+KrOPoCbczsC0lXA83N7LS0/07ATOB24A48ALkJOMLMSiZwhhBCCKEc5H2kAsDM/phqUlwB\nNAPeBHqZ2RdplwJgx6z9P5J0JHAjMBSYBwyIgCKEEELInwoxUhFCCCGEyq8iLCkNIYQQQhUQQUUI\nIYQQciKCihBCKAeSaub7HEIoaxFUVBNxQcsPSV0kXSdp+3yfS8gPSa0k1TWz1fk+l1DxSKqR/qt8\nn0suRFBRxSnJXNAk9ZLUOd/nVR1I2hdf0/1z4IQ8n07IA0mnAX8COkgqkPSwpOb5Pq+QP5ngQdJZ\nktqa2Zr0UMM8nlbORFBRxVki6SBJ7wA3AAdK2jrf51ZVZd1x/BdYg9cX7pKCjFANZO4+gUfwisET\ngY/Ttm/yclKhQkjX45r4e+JUSadLWgKcn+dTy4kKUacilK1U02McMAW4GVhlZkvSCEasKc4RSS3N\n7ON00djCzOZLegyoBzQATgRm5PcsQ1lKwYSZ2ZoUXG4F7IDX37nazC7L5/mF/ErvjxpmtkrSaOAi\nYCUw3Mxuz+/Z5UaMVFQh35M3sRcwz8x+bWYLfVfVBRql51WJubx8kdRC0j+AlySdBJAuGpnhzD8B\nbwCdJR2Wr/MMZUtSTTNbk4LKDsDRwGfAGcAzQBtJbdK+ce2tRiRtAZCmOjI3cocAy4G/Af+X9qv0\n1+J4Y1cRJfImOkpql9UGvhbQQNLlkn4PjAfeB6ZKahKjFaW2K34nWhu4RtI5kpqZ2TfAKuBQYCx+\nR3K8pAb5O9VQVsxstaRt0+jUNOAkYEczex6YDOwO9En7rqkKHyBh45jZKgBJvwDuTLkUBwEHA0cB\nx0qqURWuxRFUVBHp7miv1JztHmAqcK+kA4Fb8YTBE4Gv8P4q5+Hlzy/P0ylXepkPBTN7Dvgj3mP5\nZWAffKoJ4CGgCfA1PmKxB3B8uZ9sKHOStsPfBzWATsDZpIbZZvYo8BZwqKROaVul/wAJGyfd6M3C\nr7uvAk3SFOmbeN7NcFIrisoebEZQUUmVnOqQ1AIPHl4Bfgz0ApYATwDLzWy4mbUxswvM7AYzewRP\nIJxfzqdeZaRArnOa5rgfqIknZl6OXzTG4QHEUmA18Ac8efNISbvk6bRDKUmqsYHpi9bAj4ATzWwO\n/n6oKWmH9PgteIb/AEnNJP1K0rnlc9ahvJR8b6SvLwaeN7PWZnarmf09M3qBB5/tgZ9KqpWuKzus\n71iVQSRqVlJZUx1NUp7E/kCBmXVN28/EO73+FU8S/F/a3hxvYTsMvwCOLv+zrxok7Qr8HdjbzN6W\n9CLwM2Bv4DBgEHAaUB8YYWYfSnoY/5kfAnyQnzMPmysNUa9J/98C2BKYnbYtwP+2Lpe0VXqsI9BI\n0rlm9pCk+4CT8aaJtYD++XgdIfdSTs3qrCWiGdviLcNfTfv9BKgDbAe8bGazJI0BRgH1JO0JHCxp\nDzNbVH6vIDeioVgllIbHtsGnOR40szslXQC0Ap4HrgeWAZeY2WPpOTWAA/APuV74HfNZZvZuHl5C\npZF+brXMbPl6tgt4G5hkZjfKC1zdgN+Nnm5mn0u6GM+1GGNmK9NzD0nz7KESklQfT6w7BA/W/w1M\nMLOnJJ0K/Aqf6piRHjsV2N/M2kiqja8G2c3MnsrLCwhlStLP8Ru2ucDjeBftCUBX/FowCw8oGgCr\nzax1et6NQAdgIXCemX1a/mdfehFUVAKZCDj7LiltnwvcamZj06qDe4Ev8A+2m8xsefrwOwofin0C\nOAb43MxeTMeosZ7IOgCSBgH7AQ+nD4xewHwzeys93giYBLwLjDazlZJ+BvwCeNLMrihxvPhZV3Jp\n1dQk/M7zF8DOwE/xQH0/M5uThrBXZv3dDgJOAY4ws6X5OveQWyWX5EtqiefUNAcexIPOL4GRZvbP\nFGz8Dw8y5uFLzd8BjjKzaZJqAfXMbEk5v5SciumPSiBdmGrhb8IlKcFnFfAosG/a5z5Jv8Kj41uz\n7qz3BwbiEfNqM3soc9zMRa88X0tlIOlQ/E50KfCXtK0bPm3RKP3/AjP7StJy/MNkZXr6I8BBQDdJ\n+5nZa5lgIgKKyiMzl72e31kLPEg/wcxmADMkTcc/SMYDh6WAojmeT9EbH9a+JgKKqmMDNwgn4Nff\ng8xsRUrIfQb4DXC4md1R4hin4dOnMwDSNaRSBxQQiZoV0noSfXbAA4jnUw5FJsGnJlA7swYaH3bt\nCjwt6VJJ44Hn8Df6lJLZ5hFQrCsFDDfg1e72x+8ynjKzF/DlgJ/hdyOnpKc8CrST1BggBXMPA02B\n3mlbBBOViNbWm1ijtbVGih4GFuOJtxn/xYvK7Z+WcjfGcyWmAr8ELjCzm8vj3EPZKhlsSrpNUt/0\ncGfgmRRQjMFX3D1GVqVMSZ0k9Zb0Bzyx/gkzW1CuL6KMRVBRgcjVXM+H0Kf4xakuMFlSZknic/g6\n59VQtLTxGOB1YDf87qm7mZ1tZt9W9qVK5eQIPHCYYGbLzGxZ1tLR+fhQ95vAtekudCUwGx8GJ+33\nPHC2mV1a7mcfSi0zMihpAvBXeb+OAenh1XiZ7f0kbZnZH592XIL/jS7Gc5suM7OdzeyB8n8VoSxk\nJ+lKOhvoCXwmrz3THOgo6SO8Nk1fMzslJWLumq4juwHn4LkVe5vZdXl5IWUopj8qiKz5udVpbu58\nfN5thpn9DShMwcSZwB2SvsWH598GuuFV2Uh31C9IqpOZAvmeodywrj2BL83sKwB5Bcw2KdP/c+A+\nMztP0tfAdXhg1wnP5M/OAP9Hfk4/lJZ8ue+DeJAwCV/BcbukpmZ2taTn8CmQt4BMsuU2eKXEz9Pf\n2b/Sv1CFpGvp+cA1+NRGHzMrTI89gS8nv8TMxmQ9pw0wGC+A9gDwrJn9t7zPvbxEUFHO1pNsKUvS\n15cDF+BFlNoDv5Q0ErgnrdS4UNIqfKpjJV5Y6dvsY0HRMHzkTWy6G4GnJDXDk/GW4wWMmuFLQ88A\ndjezkZKWAofjd6d74cvD4mddSUjaG89T+jCNQmV0wDPzj0oZ+BNSUnRfSa8CvwXuwwONB/Fs/Qvx\nQGRByQS+UDll5a4VSVNi/8ZX9RSYWWHWNXYcfu1uLmkvfJVHW/zm4zvgazNbhk+XVVmx+qOcpKEv\nZQ2f7QbMyVpmWAPPIB8OXJGmMpD0N/xD61IzezrreH3wqHgvYFha0hgXsxxI0xpd8GHsF/GkzA8k\ndcU/TC4xs7tS8mx7PMi4L39nHDZFGgmciFc+XYQHFselxEsk3QrsaWbdMh8s8mWkfwWmp5GqFvhS\n0QPxyrTjzex3+Xg9oXQy180NXT/lK+sWAnPN7D9pquNC4NdASzP7VGtX/BwDjATa4DeGXfDCeENS\nQFHlRVBRDrJHC9IH03V4oPA1Ppx+W3psN2AHM3teUhfgWmAXfBj2b8DFZvZl1nH3BH4PvABcVDKq\nDrkl6WC87PapaQlYBHGVjKSb8Dnth4BL8RyJPwP/NLOfp31OxVf/bG2+LLt2Sr67HDjWzNpnHW9L\nM/uu3F9IKLV0o1ej5OhiVpDxE7wW0Ap8VLgAOAsvt98CX+n1bzM7RcWLom0HtMNrUUw3s1nl9qIq\ngEjULAcp8auZvNHQM/g8/Hn4qoz+kvZNu85JAUVvfC73FTNrCtyBJ2D2yhwzvYnfxnMqWqW7qUjE\nLCOS6gFH4kmwr0H0bqhM5KW1zwCGAqeZ2cnAx2b2IV6/pXbW7i/i+UzXAZjZirS9GfBxSuLM5ClF\nQFEJpRs9S9fmPSR9IK8xkym/3wC4Ai+tvzu+qu5ufHnoUel9czPQW9KBaVqkVnr+Z2b2jJndXd0C\nCoigolxI+n/4ioJd8aHyESmh8nb8QrUIitYpg5fXfh3PmwBvAlYHOCMl/WSbA9RJw2/xIZdDklpJ\nOkVevOgtPH9ilJl9nudTC5so3UVOB57Fy2STRh8643kyNSVdLam9mX2E1yQZIl+a3U3ScXhy5qNm\ntjKSniu3FEzUlTQFv9a+iI9YZeyFjxJPS7/vBfgKvM/wHh1b451on8HzsLKv39VaJGqWsTSi8KW8\n5n9bPEM8Y398CqSLvFX2dElN8MZE/8DrUKzCh9IewFcffApFCUN9gcvwkq7xhs69dviI0mrgZjMb\nl+fzCRspBfKD8b+fr81sAvAf4HfA79IUxyHAcfhUyGy8SNyxkk5IOTPb4r1cTgG2Bq4ys0nl/2pC\nrkk6GZ/i+he+tLOwxC4r8Z4d76f965ovL78Tr2NTx8zmS7qX1A3azF4uv1dQcUVORRnITsrMSvRq\niK9lvwhvLHMDXh9+Gh4V7wj8LM3Vj8EvZh/gc3ef4muev8gcPw3RdQB2tqwqmSG3JLXFG0ZFvkol\nIekKPBh8Eu/EuxIYm/4em+GrNwbgc+IjzOy99LyuwF14ntMlaVsd/Gbg3/EeqDokPYAnWffPJOim\n7W3x1V4rgDeAv5vZ2ZmcCUnt8AqYB5rZjHRdb2iVtE9HWYigIsdKJGWWXD46Eh9W/Rqfn7sqRbvb\n43kT25hZx5RpfjjQA3jXzG7JOkb0jwhhPeS9WG7Hg/QL8RLrdUpm3UvqgY9Y3G5m12bdhTbFlwFe\nZ2ZXaj1LCkPlprX9WPbBy6r/zcyGy8uqT8GXEvfHRyh+jjcCOxRfLr5M0iigO97HJfJp1iOCijIi\n78PRDpgPPJYZGpM0Cx9qPT77TSnpevxNvMf6ol5FvYkQ1itr5O5Y4CrgXDN7NuvxPfFg48kULDTC\n58cHAW0stZeW92K4EF/+F8XLqogNXTslXQX8BG/61RnPqRhlZnPS41sCNwHH4o2/vsZv9M43s/8r\np9OvdCJRM8ck7SzpdXzp0Tt4IZ2rJQ1OuwwHDgN+kskgT0OsuwH3lwwoMis6IqAIYS25WpJ+gzdw\nA0/A/CwTUKQVHyPx1ToNgZGSmptXS30IX+FxmaSGkp7GixdNiICiaskaOe4i6UdZD92CT411AH5u\nXlJ7TtbzvjOzQcDZwEt4UvxuEVB8vxipKIXMHVKJbRfiXep+mr7uhOdNvAMcnPIr/oa/mY/FC/Dc\nBqzB39gzCCH8oPQB8W9gRzP7r6RXgE/M7AR5k70aeFO39/Aqhs8AH6TH6+LFqyakwz0MDDCzr8v9\nhYScys5pS1/vjOfPNMGXDl8PTDazLyQNxEesbrN1u4jGVPNmiJGKUkhDrjUl9ZfUKm3eB+8mWkPe\nkOgveFfLk7PmZwfh83Jv4Be6aWa2VwQUIWySnfCAYZm8jsi7QCdJW5vZKjNbYWYPm9lM86Jxl+Cl\ntrukPIsX8LbkHczs+AgoKr8UCFhKqqyTNvfAe7S0w+v/HIOv9CF9PQ/oI2mPzDEgeiVtrggqNlIa\nbl3fz+sI/G7nf+nrVniRpHn46o6eZjbIzD6RtKekbVNBlHH46MX2ZnZx+h41y/yFhFAJSWouaayk\no7P+Tmrhy/5WmNm3eJBem9RqWqkYUWYKEe/jMg+vDYOZzTKzq8zszXJ8KaEMlAwEUr7E3yVdzdr6\nIovxwHIG0EtSp7T//+EdRn+afYyweSKo+B6SOkm6XFKDTPSb9VjmZ/dXPKDIzOv+Hm+He4OZdTWz\nf6b92+AdRtsDmNlQMzvazBam0Q5F3kQIG7Q7XjzuD8A0eRn7jni+RObv5l48sDg3jUashKIRxV2A\nXwB/xwtghSokK5jYUd6S/DB8lPhkfPXGnLSf4b04VgGnpW1/xpPnj8mMVoTNF0HF9+sG/NfMloIv\nWZN0maTWWQFGA/zCtnv6+k58OdKBkg6V1FhSNzzYaI3Xnii6e9LaVtmR3BLCBpjZX1OeUi98RdV4\nYAjQCFiW9vkSny//D/C0pD/KK2JOBmbitQfONbMl+XgNoeyk6eahwIf4Td3xZjYKr0dSiJfXBsDM\n/o7fDO6hVJobb8442MzeKd8zr3oiqFiPrKG0MZlM3zQ/tzs+lPa4vLUt5i2T6+LNZjKZxj/Dg41H\ngEeBqXjDoiPM7OO0n2XtH0LYCGb2DzM7DTgaD973x/8eB6XHn8H/Rm/C/yb3AbYBDjWzE9MQeKjE\nUhJuMekm7x08eNzBzD5ID72It6Q/RF6SPeMB/PPvSHmLg/fM7LUyPvVqIVZ//ABJO+GFcv6SCuVs\ngSdX1gDuMrM7JA3Do9xds563FbAz3qnujRR8RL2JEEohqyZFHfzD4m08eDgOnyu/z8wmZ+3fHCsw\nuQAACBtJREFU0My+yc/ZhlzZwEq7E4HFeEvyQnnRwPPx7rM7m9m8tF974EpgdWZVXtp+IDDDzJaX\n1+uoDmKkIkmJmOskSpo3F1oGHCRp/7SC40zgeWCCvP/GUuA9eZGdzPO+NrM3zexJ86qZkTcRQill\njfAtxxOh55rZmcDBwCvAzZL+Ja9SSwQUlZekH6fAIZMXk5ky7ippLjAGrzUxXd7LYzmeV/MW3gaB\n9NyZ+Ijx/pLOytr+cgQUuVctG4ppbR33zF1PZvRgtbz7XAs8l+LL9JTr8DdpH0lvpKG1y1NkPAjY\nCr9b+t96vh0Q0xwh5JK8KuYqvGskZvYW8Jakh4D5mZHBUKn9Fu8e+25aFmxp6fBofJTqYnxqawS+\nNHi5mT0saRxwq6SuZvZiOtbzeIHBd8v/ZVQv1XKkIpNkWTKvQd6I6H28L8erkvqkx1/AK6p1w+tL\nZAzHe3YU4A3BmpfTSwih2kp3rEuB+qSuv5m72DQ6GAFFJZY1Ynw1Xgn12KxlwXvhq4AyLck/B34F\nzMVbkm+D16R4iuKjFR+Z2XAze6m8Xkd1VS2DCklNJM1KQ2aZNfAP4028zsCXID0JXCTvJwBwKz6y\nc0x642Jma8zsvvS8XVNWcQihDKWbgTZApgFY0Q1CqNxSknzmZu95vKNzJ7yAFfgUR8mW5KuAyfjK\noC3NbAE+DdI2rbwL5ajKBxUbKCi1Ek/quip9/RX+Jj3NzKYC9fCs8j2BU1PBqtn4ao698IpsRczs\nXTP7IJM3UUYvJYSw1hLgWtIS7VA1pBu1NZKOlHQrUAfYA596bgi8iS8ZHpWesiL99218GnqH9PVz\neJ+OF8rv7ANUg6DCvM1tHUnbZW3LXJAaSBplXo3vFjN7N02BvA78C5+7a48XUAHvdFgLbwa21fq+\nV9wxhVD2zGyumY2KXKWqJ9WbuA9YiBet+hQ4AV8WvAa4EThDUk+8gir4jd50vBcMZrbUzP5b3uce\nqsGSUklH4lX4PgAuwOtFrEhJlr/Alx81NbMlkg4FxgIjzWyapAb4m3oWcJ6ZvSGpA1BoWW3LQwgh\nbJpMgnzJ5aKSpgLzzGxw+ro+XqzqQ+ACM/tM3lepLx5EfIVPfQwzs9vL/YWEYqr8SAW+IuN/eDOZ\nicBdkpqY2f/wKpcf4LXfwcv+bolHvABd8V4BDYDOAGb2upl9p/X3AQkhhPAD0gq8zChTnaztBXj/\npBnp65rpWj0evx5n+nMMBgbjZdc/BnaPgKJiqA4jFcKbdxXgb9SDge2Bu83smlSm9Q/4NEd7fJnS\nVPzNOgp4CHjavAlYCCGEzZRd/E/ev2UU8C3eev5P6YZtOvChmZ0kaYuUiImkT/Gqmb82s1fy9BLC\nD6jyd9tpWO0m4P/hIw6H4ctAR0m6Gy+x/SAwzszux9cz98WXlf7TzG7JBBQxOhFCCJsvTXfUT4UC\nf4cnXdYGRuL1JsCnoPtKOjAroGgHfIMnbXaPhPiKq8qPVGRIGoEn8/zKzF6U1BXoD5wOPI23MD/G\nzB6X1ApYZGZfp+euUyI2hBDCppH3TJoEfAc8bGY3p0JmQ/BCgoeb2TuS/oSvtJuA15wYik9FvwU8\nEQm6FVd1Cioa4S1vvwDOyIqAz8X7BnTD3+THZ1faBNZEQBFCCBtPa5syrimxvTVwGd4QrrOZvZ22\nt8c7zC4zsz6SauHT1p3xuhRzgONiRUfFV22G883sK3zaY1fWLhHFzMbhXQ17mtnxaVtRpc0IKEII\nYeOlvIlMvYmWkvZP5bUxs/fx6eZvSEmXyb/x3La9JB1rZivx0YmfAIeY2QERUFQO1WakAiBFv3cC\nWwMDzexTlegamp0YFEIIYf0yPZQ28Fg9fMVGDzwR8yPgTjO7V1JjPH+iJ9DHvGkjknbECxJ2M7MW\nZf8KQlmoNiMVACn6vR3YCfhx2ra6xD4RUIQQwgZk9VlZk74uKPH4j4A/42XUj8RHgl8DfitpBzNb\nBEzD60tkkjMxs7n4KMYkSVtEMmblVK2CiuRlPDp+JN8nEkIIlUkancg0cest6SXgT5LulnRw2m0N\n3g30aDN7E2+02BdoiVfDBPgH8ATQWVKPrG8x1cyuMLNVMfVcOVWr6Y+SYlVHCCFsGkk7AVPw5Z03\n4NMbA/Aigz8xs6WSmgNf4gUHe+AFBhcDY/AVHi+k6sQ3A++Z2YDyfh2hbGyR7xPIpwgoQghh46UO\nzY/jdX+2T32TSM2+RgAHAU+mfLV+wM7AsWY2PXUMrYPnTXQ2s9clnWNmM/PyYkKZqI7THyGEEDaD\nmS0GHgBmk/LSkh2AmsAnWdsOA2qZWabtwY74tMdeknql40VAUcVU65GKEEIIm+w2fETiZ5K2wKdA\n2uIltLtKWmJm84BX8eqXAwHDi1vdAvQ1sy/yc+qhrFXrnIoQQgibTtIJwNX46MMofDVHPbyAVUPg\nGuANvGLxcXg7hCvM7Hf5ON9QfiKoCCGEsElSzZ978WqXx5nZl2l7Y3ylx+XAXKAPsEVaLhqqgcip\nCCGEsElSzZ/r8MTLwVBUv+JLM5sInIXnTyyPgKJ6iZGKEEIImywFETcCewIXmNlbkmqb2Yo8n1rI\noxipCCGEsMnSkvzrgFrAsLQtAopqLlZ/hBBC2CxmNk/So8DKKCYYIKY/QgghlEIEEyFbBBUhhBBC\nyInIqQghhBBCTkRQEUIIIYSciKAihBBCCDkRQUUIIYQQciKCihBCCCHkRAQVIYQQQsiJCCpCCCGE\nkBMRVIQQQgghJyKoCCGEEEJORFARQgghhJz4/xyPDBB4aHcmAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7ff2b9480c90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "width=0.15\n",
    "plt.bar(np.arange(NUM_CLASSIFIERS), scores[0], width, color=\"r\", label=\"VGG-16\")\n",
    "plt.bar(np.arange(NUM_CLASSIFIERS)+width, scores[1], width, color=\"b\", label=\"VGG-19\")\n",
    "plt.bar(np.arange(NUM_CLASSIFIERS)+2*width, scores[2], width, color=\"g\", label=\"Inception-V3\")\n",
    "plt.bar(np.arange(NUM_CLASSIFIERS)+3*width, scores[3], width, color=\"y\", label=\"ResNet-50\")\n",
    "plt.bar(np.arange(NUM_CLASSIFIERS)+4*width, scores[4], width, color=\"cyan\", label=\"XCeption\")\n",
    "plt.legend(loc=4)\n",
    "plt.ylabel(\"accuracy\")\n",
    "plt.xticks(np.arange(NUM_CLASSIFIERS)+2*width, [\"NaiveBayes\", \"SvM\", \"XGBoost\", \"RandomForest\"],\n",
    "          rotation=30)\n",
    "plt.title(\"element-wise dot product of vectors\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [conda root]",
   "language": "python",
   "name": "conda-root-py"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.12"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}