vcog_hpc_prediction_simulated_data.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# A highly predictive signature (HPS) of Alzheimer's disease dementia from cognitive and structural brain features"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This notebook uses simulated data from the Alzheimer's Disease Neuroimaging Initiative (ADNI) to make high confidence predictions to classify patients with mild cognitive impairment (MCI) who will progress to Alzheimer's disease dementia within three years from those who will remain cognitively stable.\n",
    "\n",
    "This notebook is associated with the following preprint [A signature of cognitive deficits and brain atrophy that is highly predictive of progression to Alzheimer's dementia](https://doi.org/10.1101/352344)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### import packages"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/angela/anaconda3/envs/vcog_paper_py35/lib/python3.5/site-packages/sklearn/ensemble/weight_boosting.py:29: DeprecationWarning: numpy.core.umath_tests is an internal NumPy module and should not be imported. It will be removed in a future NumPy release.\n",
      "  from numpy.core.umath_tests import inner1d\n"
     ]
    }
   ],
   "source": [
    "import urllib.request\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "import scipy.io\n",
    "import matplotlib.pyplot as plt\n",
    "from matplotlib import cm\n",
    "import seaborn as sns\n",
    "from copy import deepcopy\n",
    "\n",
    "# import sklearn modules\n",
    "from sklearn.utils import shuffle\n",
    "from sklearn import preprocessing\n",
    "from sklearn.model_selection import StratifiedKFold\n",
    "from sklearn import metrics\n",
    "from sklearn.preprocessing import label_binarize\n",
    "from sklearn.metrics import roc_curve, auc\n",
    "from sklearn.metrics import average_precision_score\n",
    "from sklearn.metrics import precision_recall_curve\n",
    "\n",
    "# import proteus modules\n",
    "from Proteus.proteus.visu import sbp_visu\n",
    "from Proteus.proteus.predic import high_confidence_at\n",
    "from Proteus.proteus.predic import prediction"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### function for stats of prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "def predic_stats(y_, y_pred, lr_decision):\n",
    "    # number of AD subjects\n",
    "    n_ad = sum(y_)\n",
    "    print('Total number of TARGET subjects: ', n_ad)\n",
    "\n",
    "    # number of CN subjects\n",
    "    n_cn = len(y_) - sum(y_)\n",
    "    print('Total number of NON-TARGET subjects: ', n_cn)\n",
    "    \n",
    "    # number of subjects predicted as AD at stage 1\n",
    "    n_pos = sum(y_pred)\n",
    "    print('Stage 1 number of hits (true and false positives): ', n_pos)\n",
    "    \n",
    "    # true positives at stage 1\n",
    "    n_pos_ad = sum(y_pred[y_.astype(bool)])\n",
    "    print('Stage 1 TRUE positives: ', n_pos_ad)\n",
    "    \n",
    "    # false positives at stage 1\n",
    "    n_pos_cn = n_pos - n_pos_ad\n",
    "    print('Stage 1 FALSE positives: ', n_pos_cn)\n",
    "    \n",
    "    # number of CN subjects not identified as positive (true negatives)\n",
    "    n_neg1_cn = n_cn - n_pos_cn\n",
    "    print('Stage 1 TRUE negatives: ', n_neg1_cn)\n",
    "\n",
    "    # number of all flagged HPC-AD subjects\n",
    "    n_flag = sum(y_pred[lr_decision>0])\n",
    "    print('Total number of flagged HPC-AD subjects: ', n_flag)\n",
    "\n",
    "    # number of flagged HPC-AD subjects who are actually AD (true positives)\n",
    "    y_pred_true = y_ + y_pred\n",
    "    y_pred_true = y_pred_true==2\n",
    "    n_flag_ad = sum(y_pred_true[lr_decision>0])\n",
    "    print('Number of flagged HPC-AD subjects that are TRUE positives: ', n_flag_ad)\n",
    "\n",
    "    # number of flagged HPC-AD subjects that are actually CN (false positives)\n",
    "    n_flag_cn = n_flag - n_flag_ad\n",
    "    print('Number of flagged HPC-AD subjects that are FALSE positives: ', n_flag_cn)\n",
    "\n",
    "    # number of CN subjects that were not flagged (true negatives)\n",
    "    n_neg_cn = n_cn - n_flag_cn\n",
    "    print('Number of true negatives: ', n_neg_cn)\n",
    "    \n",
    "    print('#############################')\n",
    "    print('Stage 1 stats for TARGET vs NON-TARGET')\n",
    "    print('Precision for AD: ', n_pos_ad/(n_pos_ad + n_pos_cn))\n",
    "    prec = n_pos_ad/(n_pos_ad + n_pos_cn)\n",
    "    print('Recall (or sensitivity)  for AD: ', n_pos_ad/n_ad)\n",
    "    sens = n_pos_ad/n_ad\n",
    "    print('Specificity: ', n_neg1_cn/n_cn)\n",
    "    spec = n_neg1_cn/n_cn\n",
    "    fp = (1-spec)*664\n",
    "    tp = sens*336\n",
    "    adj_prec = tp/(tp+fp)\n",
    "    print('Adjusted precision for 33.6% baseline rate: ', adj_prec)\n",
    "    print('Accuracy: ', (n_pos_ad + n_neg1_cn)/(n_ad + n_cn))\n",
    "    acc = (n_pos_ad + n_neg1_cn)/(n_ad + n_cn)\n",
    "\n",
    "    print('#############################')\n",
    "    print('Stage 2 stats for TARGET vs NON-TARGET')\n",
    "    print('Precision for HPC-AD: ', n_flag_ad/n_flag)\n",
    "    prec_2 = n_flag_ad/n_flag\n",
    "    print('Recall (or sensitivity) for HPC-AD: ', n_flag_ad/n_ad)\n",
    "    sens_2 = n_flag_ad/n_ad\n",
    "    print('Specificity: ', n_neg_cn/n_cn)\n",
    "    spec_2 = n_neg_cn/n_cn\n",
    "    fp_2 = (1-spec_2)*664\n",
    "    tp_2 = sens_2*336\n",
    "    adj_prec_2 = tp_2/(tp_2 + fp_2)\n",
    "    print('Adjusted precision for 33.6% baseline rate: ', adj_prec_2)\n",
    "    print('Accuracy: ', (n_flag_ad + n_neg_cn)/(n_ad + n_cn))\n",
    "    acc_2 = (n_flag_ad + n_neg_cn)/(n_ad + n_cn)\n",
    "    \n",
    "    return sens, spec, prec, acc, sens_2, spec_2, prec_2, acc_2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### set the random seed"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [],
   "source": [
    "np.random.seed(1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Load simulated data"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We created simulated data derived from the ADNI dataset and hosted it on [Figshare](https://figshare.com/articles/Simulated_cognitive_and_structural_MRI_data_from_ADNI/7132757). The following URL request will retrieve these data and load them using pandas."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "('simulated_data.csv', <http.client.HTTPMessage at 0x7f3941ce5278>)"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Download the csv file from the web\n",
    "urllib.request.urlretrieve(\"https://ndownloader.figshare.com/files/13123139\",\"simulated_data.csv\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# read the data with pandas\n",
    "data = pd.read_csv('simulated_data.csv')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# create 1-hot variable for initial AD diagnosis\n",
    "for i,row in data.iterrows():\n",
    "    dx = row['DX']\n",
    "    if dx == 'Dementia':\n",
    "        data.loc[i,'AD'] = 1\n",
    "    else:\n",
    "        data.loc[i,'AD'] = 0"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train a model to classify patients with AD dementia vs cognitively normal (CN) subjects"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this section, we will train a machine learning model to classify patients with AD dementia from cognitively normal subjects in ADNI1. This section displays the code that was used to generate the results presented in *\"Prediction of AD dementia vs cognitively normal individuals\"* and *\"Identification of easy AD cases for prediction\"* in the *Results* section of the preprint.\n",
    "\n",
    "We used [Proteus](https://github.com/SIMEXP/vcog_hps_ad/tree/master/Proteus) which is built on top of [scikitlearn](http://scikit-learn.org) to generate our predictive models."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Organize the data for the training set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# grab only adni1 subjects\n",
    "train_set = data[data.dataset != 'ADNI2']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [],
   "source": [
    "# grab only Dementia & CN subjects\n",
    "train_set = train_set[train_set.DX != 'pMCI']\n",
    "train_set = train_set[train_set.DX != 'sMCI']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set instance of standard scaler\n",
    "scaler = preprocessing.StandardScaler()\n",
    "\n",
    "# define the x and y variables\n",
    "x_ = train_set.iloc[:,train_set.columns.get_loc(\"ADAS13\"):train_set.columns.get_loc(\"sub7\")+1].values\n",
    "y_ = train_set[['AD']].values.ravel()\n",
    "\n",
    "# add these variables to the x\n",
    "confounds = train_set[['gender','age_scan','mean_gm','tiv']].values\n",
    "\n",
    "# scale the x variables\n",
    "x_ = scaler.fit_transform(np.hstack((x_,confounds)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Cross-validation of the classifier in ADNI1 AD & CN"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Stage 1\n",
      "Proba:\n",
      "[1.         0.93076923 1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         0.85062241 1.         1.\n",
      " 1.         1.         1.         0.99579832 1.         1.\n",
      " 1.         1.         0.90725806 1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         0.996139   1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         0.99176955 1.         1.         1.         0.85887097\n",
      " 1.         1.         0.91935484 1.         0.9921875  1.\n",
      " 1.         1.         1.         0.996139   1.         1.\n",
      " 0.88655462 0.99190283 1.         1.         1.         0.82758621\n",
      " 1.         0.99585062 1.         1.         0.65182186 0.61354582\n",
      " 1.         1.         1.         1.         0.95149254 1.\n",
      " 0.99593496 1.         1.         1.         1.         1.\n",
      " 0.90225564 1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 0.7892562  1.         1.         1.         1.         1.\n",
      " 0.94067797 1.         1.         1.         0.79775281 1.\n",
      " 1.         1.         1.         1.         0.76171875 1.\n",
      " 1.         1.         1.         1.         0.9254902  0.98850575\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         0.996      1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 0.35849057 1.         1.         1.         1.         1.\n",
      " 0.94186047 1.         1.         1.         0.99239544 1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         0.98449612\n",
      " 1.         1.         1.         1.         1.         0.75\n",
      " 1.         1.         1.         0.99583333 1.         1.\n",
      " 1.         1.         1.         0.81568627 1.        ]\n",
      "Average hm score 0.8653061224489796\n",
      "Stage 2\n",
      "Adjusted gamma:  1.0\n",
      "Adjusted gamma:  1.0\n",
      "Classifying AD vs CN at stage 2...\n",
      "[0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.\n",
      " 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.\n",
      " 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 1. 0.\n",
      " 1. 0. 1. 1. 0. 0. 1. 0. 0. 0. 1. 0. 1. 0. 0. 0. 0. 0. 0. 1. 1. 1. 0. 1.\n",
      " 0. 0. 0. 1. 1. 1. 0. 1. 1. 1. 1. 0. 1. 1. 1. 1. 1. 0. 1. 0. 1. 1. 1. 1.\n",
      " 1. 1. 1. 1.]\n",
      "Total number of TARGET subjects:  55.0\n",
      "Total number of NON-TARGET subjects:  69.0\n",
      "Stage 1 number of hits (true and false positives):  34.0\n",
      "Stage 1 TRUE positives:  33.0\n",
      "Stage 1 FALSE positives:  1.0\n",
      "Stage 1 TRUE negatives:  68.0\n",
      "Total number of flagged HPC-AD subjects:  29.0\n",
      "Number of flagged HPC-AD subjects that are TRUE positives:  29\n",
      "Number of flagged HPC-AD subjects that are FALSE positives:  0.0\n",
      "Number of true negatives:  69.0\n",
      "#############################\n",
      "Stage 1 stats for TARGET vs NON-TARGET\n",
      "Precision for AD:  0.9705882352941176\n",
      "Recall (or sensitivity)  for AD:  0.6\n",
      "Specificity:  0.9855072463768116\n",
      "Adjusted precision for 33.6% baseline rate:  0.95444066308047\n",
      "Accuracy:  0.8145161290322581\n",
      "#############################\n",
      "Stage 2 stats for TARGET vs NON-TARGET\n",
      "Precision for HPC-AD:  1.0\n",
      "Recall (or sensitivity) for HPC-AD:  0.5272727272727272\n",
      "Specificity:  1.0\n",
      "Adjusted precision for 33.6% baseline rate:  1.0\n",
      "Accuracy:  0.7903225806451613\n",
      "------------------------------------------------------\n",
      "Stage 1\n",
      "Proba:\n",
      "[1.         1.         1.         1.         0.08520179 1.\n",
      " 1.         1.         1.         0.53333333 1.         0.97285068\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         0.7480315  1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         0.94468085\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 0.97107438 1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         0.89795918 1.         1.         0.96525097\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         0.50607287 1.         1.         1.\n",
      " 0.99616858 1.         1.         0.92887029 1.         1.\n",
      " 1.         1.         1.         1.         0.99203187 1.\n",
      " 1.         0.5766129  0.99595142 1.         1.         1.\n",
      " 0.18461538 1.         1.         1.         1.         1.\n",
      " 1.         0.00409836 1.         0.9488189  1.         1.\n",
      " 1.         0.04150943 1.         0.33603239 1.         1.\n",
      " 1.         0.21544715 1.         0.828      0.9924812  1.\n",
      " 1.         0.         1.         1.         1.         1.\n",
      " 0.17509728 1.         1.         0.70817121 0.20392157 1.\n",
      " 1.         1.         0.91735537 0.98804781 1.         1.\n",
      " 0.87250996 0.056      1.         0.93877551 1.         1.\n",
      " 1.         1.         1.         0.9755102  1.         1.\n",
      " 1.         1.         1.         1.         1.         0.9962963\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         0.68510638 1.         1.         1.         1.\n",
      " 1.         0.98496241 1.         1.         1.         0.76734694\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 0.94650206 1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         0.99183673 0.99180328]\n",
      "Average hm score 0.8414634146341463\n",
      "Stage 2\n",
      "Adjusted gamma:  1.0\n",
      "Adjusted gamma:  1.0\n",
      "Classifying AD vs CN at stage 2...\n",
      "[0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.\n",
      " 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.\n",
      " 0. 0. 1. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 1. 1. 1.\n",
      " 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.\n",
      " 1. 1. 1. 1. 1. 1. 1. 1. 0. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.\n",
      " 1. 1. 1.]\n",
      "Total number of TARGET subjects:  55.0\n",
      "Total number of NON-TARGET subjects:  68.0\n",
      "Stage 1 number of hits (true and false positives):  56.0\n",
      "Stage 1 TRUE positives:  54.0\n",
      "Stage 1 FALSE positives:  2.0\n",
      "Stage 1 TRUE negatives:  66.0\n",
      "Total number of flagged HPC-AD subjects:  46.0\n",
      "Number of flagged HPC-AD subjects that are TRUE positives:  46\n",
      "Number of flagged HPC-AD subjects that are FALSE positives:  0.0\n",
      "Number of true negatives:  68.0\n",
      "#############################\n",
      "Stage 1 stats for TARGET vs NON-TARGET\n",
      "Precision for AD:  0.9642857142857143\n",
      "Recall (or sensitivity)  for AD:  0.9818181818181818\n",
      "Specificity:  0.9705882352941176\n",
      "Adjusted precision for 33.6% baseline rate:  0.9441091127245124\n",
      "Accuracy:  0.975609756097561\n",
      "#############################\n",
      "Stage 2 stats for TARGET vs NON-TARGET\n",
      "Precision for HPC-AD:  1.0\n",
      "Recall (or sensitivity) for HPC-AD:  0.8363636363636363\n",
      "Specificity:  1.0\n",
      "Adjusted precision for 33.6% baseline rate:  1.0\n",
      "Accuracy:  0.926829268292683\n",
      "------------------------------------------------------\n",
      "Stage 1\n",
      "Proba:\n",
      "[1.         0.99583333 1.         1.         0.         1.\n",
      " 1.         1.         1.         0.08130081 1.         0.99541284\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         0.97111913 1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         0.9561753\n",
      " 1.         1.         1.         1.         0.98828125 1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 0.97222222 1.         1.         1.         0.87596899 1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         0.52083333 1.         1.         0.99591837 1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 0.88492063 1.         1.         1.         0.99606299 1.\n",
      " 0.94979079 1.         1.         1.         1.         0.63673469\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         0.0122449  1.         0.87603306 1.         1.\n",
      " 1.         0.         1.         0.04       1.         1.\n",
      " 1.         0.23170732 1.         0.63059701 1.         1.\n",
      " 0.99170124 0.         1.         1.         1.         1.\n",
      " 0.37301587 1.         1.         0.60869565 0.092      1.\n",
      " 1.         1.         1.         0.9921875  1.         1.\n",
      " 0.95983936 0.01132075 1.         0.99159664 1.         1.\n",
      " 1.         0.99586777 1.         0.99568966 1.         1.\n",
      " 1.         1.         1.         1.         0.98755187 1.\n",
      " 1.         0.76245211 1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         0.97491039 1.         1.         1.\n",
      " 0.99583333 1.         1.         1.         1.         1.\n",
      " 0.76923077 1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.        ]\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Average hm score 0.8502024291497976\n",
      "Stage 2\n",
      "Adjusted gamma:  1.0\n",
      "Adjusted gamma:  1.0\n",
      "Classifying AD vs CN at stage 2...\n",
      "[0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0.\n",
      " 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 0. 0.\n",
      " 0. 0. 0. 0. 0. 0. 0. 0. 0. 0. 1. 0. 0. 0. 0. 1. 0. 0. 0. 0. 1. 1. 1. 1.\n",
      " 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.\n",
      " 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1. 1.\n",
      " 1. 1.]\n",
      "Total number of TARGET subjects:  54.0\n",
      "Total number of NON-TARGET subjects:  68.0\n",
      "Stage 1 number of hits (true and false positives):  57.0\n",
      "Stage 1 TRUE positives:  54.0\n",
      "Stage 1 FALSE positives:  3.0\n",
      "Stage 1 TRUE negatives:  65.0\n",
      "Total number of flagged HPC-AD subjects:  50.0\n",
      "Number of flagged HPC-AD subjects that are TRUE positives:  49\n",
      "Number of flagged HPC-AD subjects that are FALSE positives:  1.0\n",
      "Number of true negatives:  67.0\n",
      "#############################\n",
      "Stage 1 stats for TARGET vs NON-TARGET\n",
      "Precision for AD:  0.9473684210526315\n",
      "Recall (or sensitivity)  for AD:  1.0\n",
      "Specificity:  0.9558823529411765\n",
      "Adjusted precision for 33.6% baseline rate:  0.9198067632850243\n",
      "Accuracy:  0.9754098360655737\n",
      "#############################\n",
      "Stage 2 stats for TARGET vs NON-TARGET\n",
      "Precision for HPC-AD:  0.98\n",
      "Recall (or sensitivity) for HPC-AD:  0.9074074074074074\n",
      "Specificity:  0.9852941176470589\n",
      "Adjusted precision for 33.6% baseline rate:  0.9689668065306801\n",
      "Accuracy:  0.9508196721311475\n",
      "------------------------------------------------------\n"
     ]
    }
   ],
   "source": [
    "s1_spec = []\n",
    "s1_sens = []\n",
    "s1_prec = []\n",
    "s1_acc = []\n",
    "\n",
    "s2_spec = []\n",
    "s2_sens = []\n",
    "s2_prec = []\n",
    "s2_acc = []\n",
    "\n",
    "skf = StratifiedKFold(n_splits=3)\n",
    "for train_index, val_index in skf.split(x_,y_):\n",
    "    X_training, X_val = x_[train_index], x_[val_index]\n",
    "    y_training, y_val = y_[train_index], y_[val_index]\n",
    "    \n",
    "    # parameters of classifier\n",
    "    hpc = high_confidence_at.TwoStagesPrediction(\n",
    "        n_iter=500,\n",
    "        shuffle_test_split=0.5,\n",
    "        min_gamma=.99,\n",
    "        thresh_ratio=0.1)\n",
    "    \n",
    "    # train the model\n",
    "    hpc.fit(X_training, X_training, y_training)\n",
    "    \n",
    "    # test in separate sample\n",
    "    _, dic_results = hpc.predict(X_val, X_val)\n",
    "    \n",
    "    print('Classifying AD vs CN at stage 2...')\n",
    "    print((dic_results['s1df'][:,0]>0).astype(float))\n",
    "    # get the predicted labels from stage 2\n",
    "    y_pred = (dic_results['s1df'][:,0]>0).astype(float)\n",
    "    lr_decision = dic_results['s2df'][:,1]\n",
    "    \n",
    "    # print statistics of prediction\n",
    "    sens, spec, prec, acc, sens_2, spec_2, prec_2, acc_2 = predic_stats(y_val, y_pred, lr_decision)\n",
    "    s1_spec.append(spec)\n",
    "    s1_sens.append(sens)\n",
    "    s1_prec.append(prec)\n",
    "    s1_acc.append(acc)\n",
    "    s2_spec.append(spec_2)\n",
    "    s2_sens.append(sens_2)\n",
    "    s2_prec.append(prec_2)\n",
    "    s2_acc.append(acc_2)\n",
    "    \n",
    "    print('------------------------------------------------------')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Mean stats of prediction across the k folds from cross-validation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These average cross-validated statistics are presented as the results of the AD vs CN classification in ADNI1. Stage 1 represents the basic support vector machine (SVM) with a linear kernel, while Stage 2 represents the two-stage classifier that makes high confidence predictions. The statistics for Stage 1 are reported under *\"Prediction of AD dementia vs cognitively normal individuals\"* in the *Results* section of the paper. The statistics for Stage 2 are reported under *\"Identification of easy AD cases for prediction\"* in the *Results* section of the paper."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Stage 1\n",
      "Mean sensitivity:  0.8606060606060607\n",
      "Mean specificity:  0.9706592782040353\n",
      "Mean precision:  0.9607474568774879\n",
      "Mean accuracy:  0.9218452403984642\n",
      "##########\n",
      "Stage 2\n",
      "Mean sensitivity:  0.7570145903479236\n",
      "Mean specificity:  0.9950980392156863\n",
      "Mean precision:  0.9933333333333333\n",
      "Mean accuracy:  0.8893238403563307\n"
     ]
    }
   ],
   "source": [
    "print('Stage 1')\n",
    "print('Mean sensitivity: ', np.mean(s1_sens))\n",
    "print('Mean specificity: ', np.mean(s1_spec))\n",
    "print('Mean precision: ', np.mean(s1_prec))\n",
    "print('Mean accuracy: ', np.mean(s1_acc))\n",
    "print('#'*10)\n",
    "print('Stage 2')\n",
    "print('Mean sensitivity: ', np.mean(s2_sens))\n",
    "print('Mean specificity: ', np.mean(s2_spec))\n",
    "print('Mean precision: ', np.mean(s2_prec))\n",
    "print('Mean accuracy: ', np.mean(s2_acc))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Train highly predictive model on whole training set of ADNI1 AD & CN"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We will train the highly predictive classifier on the entire set of AD and CN subjects in ADNI1. The resulting model is what we will use for the rest of the classifications in other samples."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Stage 1\n",
      "Proba:\n",
      "[1.         1.         1.         1.         0.28404669 1.\n",
      " 1.         1.         1.         0.47257384 1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         0.99215686 1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         0.86770428 1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         0.99186992\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         0.94323144 1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         0.94067797 1.\n",
      " 1.         1.         0.93385214 1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         0.27799228 1.         1.\n",
      " 1.         1.         0.00746269 1.         1.         1.\n",
      " 1.         1.         1.         0.         1.         0.61382114\n",
      " 1.         1.         1.         0.         1.         0.\n",
      " 1.         1.         1.         0.         1.         0.1417004\n",
      " 0.95471698 1.         0.99621212 0.         1.         1.\n",
      " 1.         1.         0.         1.         1.         0.59274194\n",
      " 0.00806452 1.         1.         1.         0.99145299 0.99173554\n",
      " 1.         1.         0.95275591 0.         1.         0.96062992\n",
      " 1.         1.         1.         1.         1.         0.99590164\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 0.99180328 1.         1.         0.54347826 1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         0.99193548 1.\n",
      " 1.         1.         1.         1.         0.97735849 1.\n",
      " 1.         1.         0.98418972 1.         1.         1.\n",
      " 1.         1.         0.83404255 1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         0.62992126 1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         0.93360996 1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         1.         1.         0.975      1.         1.\n",
      " 1.         1.         1.         1.         1.         1.\n",
      " 1.         0.91935484 1.        ]\n",
      "Average hm score 0.8970189701897019\n",
      "Stage 2\n",
      "Adjusted gamma:  1.0\n",
      "Adjusted gamma:  1.0\n"
     ]
    }
   ],
   "source": [
    "# create instance of classifier\n",
    "hpc = high_confidence_at.TwoStagesPrediction(\n",
    "    n_iter=500,\n",
    "    shuffle_test_split=0.5,\n",
    "    min_gamma=.99,\n",
    "    thresh_ratio=0.1)\n",
    "\n",
    "# train on the data\n",
    "hpc.fit(x_, x_, y_)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Print the coefficients of each feature in the highly predictive model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ 8.43851668, -5.22286411, -3.53234506, -3.13500374, -1.88113384,\n",
       "         0.        ,  0.08946603, -1.66361914,  0.10769513,  0.        ,\n",
       "         0.80499281, -1.58166452, -0.03631993, -0.22931601, -0.93913661,\n",
       "        -1.10079405]])"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "hpc.confidencemodel.clfs[1].coef_"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Plot the model weights"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following plot is presented in Figure 4 of the paper."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "# plot weights from the model\n",
    "w_df = pd.DataFrame(data=np.transpose(hpc.confidencemodel.clfs[1].coef_),columns=['Coefficient'])\n",
    "w_df['Factor'] = ['ADAS13','MEM','EXEC','BNT','CLOCK',\n",
    "                  'VBM 1','VBM 2','VBM 3','VBM 4','VBM 5','VBM 6','VBM 7',\n",
    "                  'sex','age','GMV','TIV']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/home/angela/anaconda3/envs/vcog_paper_py35/lib/python3.5/site-packages/seaborn/categorical.py:3666: UserWarning: The `factorplot` function has been renamed to `catplot`. The original name will be removed in a future release. Please update your code. Note that the default `kind` in `factorplot` (`'point'`) has changed `'strip'` in `catplot`.\n",
      "  warnings.warn(msg)\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 1152x595.44 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots()\n",
    "fig.set_size_inches(16, 8.27)\n",
    "\n",
    "sns.factorplot(ax=ax, y='Coefficient', x='Factor', data=w_df, kind='bar', palette=\"Greys_d\")\n",
    "for item in ax.get_xticklabels():\n",
    "    item.set_rotation(50)\n",
    "ax.set_ylim(-9,9)\n",
    "plt.close()\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Get the predicted labels from HPS model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "array_results, dic_results = hpc.predict(x_, x_)\n",
    "y_pred = (dic_results['s1df'][:,0]>0).astype(float)\n",
    "lr_decision = dic_results['s2df'][:,1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Generate ROC curves for classifiying AD vs CN in ADNI1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following section of code was used to generate the ROC curves in Supplementary Figure 1 for the AD vs CN classification in the ADNI1 dataset."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Set up ROC curve for base SVM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set the classifer and learn to predict AD vs CN\n",
    "base = high_confidence_at.BaseSvc()\n",
    "base.fit(x_, y_)\n",
    "y_predicted = base.predict(x_)\n",
    "y_score = base.decision_function(x_)\n",
    "\n",
    "# make sure to binarize the actual outputs\n",
    "y_true = y_.astype(int)\n",
    "y_true = label_binarize(y_true, classes=[0, 1])\n",
    "y_score = np.reshape(y_score, (y_score.shape[0],1))\n",
    "\n",
    "# Compute ROC curve\n",
    "fpr_b, tpr_b, thresholds_b = metrics.roc_curve(y_true, y_score, pos_label=1)\n",
    "# Compute the AUC\n",
    "roc_auc_b = auc(fpr_b, tpr_b)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Set up ROC curve for HPS model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get the predicted output from HPS\n",
    "y_score = np.reshape(lr_decision, (y_score.shape[0],1))\n",
    "\n",
    "# Compute the ROC curve\n",
    "fpr_h, tpr_h, thresholds_h = metrics.roc_curve(y_true, y_score, pos_label=1)\n",
    "# Compute the AUC\n",
    "roc_auc_h = auc(fpr_h, tpr_h)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Plot both ROC curves for base SVM and HPS models"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots()\n",
    "fig.set_size_inches(10, 8)\n",
    "lw = 1\n",
    "plt.plot(fpr_b, tpr_b, color='blue',\n",
    "         lw=lw, label='Base (area = %0.4f)' % roc_auc_b)\n",
    "plt.plot(fpr_h, tpr_h, color='red',\n",
    "         lw=lw, label='HPS (area = %0.4f)' % roc_auc_h)\n",
    "plt.plot([0, 1], [0, 1], color='grey', lw=lw, linestyle='--')\n",
    "plt.xlim([-0.05, 1.00])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('ADNI1 AD vs CN')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Test on ADNI1 MCI stable vs progressors"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this section, we will test our model on a new sample. This time, we will try to classify different types of patients with mild cognitive impairment (MCI). The task will be to distinguish between MCI patients who will progress to dementia within three years of follow-up from patients who will remain cognitvely stable. For now, we will do this in the MCI patients from the ADNI1 sample.\n",
    "\n",
    "The followig code was used to generate the section of the paper titled *\"High confidence prediction of progression to AD dementia\"* in the *Results* for the ADNI1 subjects."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Organize the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [],
   "source": [
    "# grab only adni1\n",
    "adni1_mci = data[data.dataset != 'ADNI2']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "# grab only MCI\n",
    "adni1_mci = adni1_mci[adni1_mci.DX != 'Dementia']\n",
    "adni1_mci = adni1_mci[adni1_mci.DX != 'CN']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set up the x variables\n",
    "x_ = adni1_mci.iloc[:,adni1_mci.columns.get_loc(\"ADAS13\"):adni1_mci.columns.get_loc(\"sub7\")+1].values\n",
    "# set up the y \n",
    "y_ = adni1_mci['conv_2_ad'].values.ravel()\n",
    "# add these variables to the x\n",
    "confounds = adni1_mci[['gender','age_scan','mean_gm','tiv']].values\n",
    "# scale the x\n",
    "x_ = scaler.transform(np.hstack((x_,confounds)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Do the prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [],
   "source": [
    "array_results, dic_results = hpc.predict(x_, x_)\n",
    "y_pred = (dic_results['s1df'][:,0]>0).astype(float)\n",
    "lr_decision = dic_results['s2df'][:,1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Get the stats of the prediction"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These stats are presented in the paper for the classification of stable MCI (non-target subjects) vs progressor MCI (target subjects) in the ADNI1 sample under the section *\"High confidence prediction of progression to AD dementia\"* in the *Results* of the paper. Stage 1 refers to the basic linear SVM model, while Stage 2 refers to the highly predictive signature (HPS) model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total number of TARGET subjects:  147\n",
      "Total number of NON-TARGET subjects:  88\n",
      "Stage 1 number of hits (true and false positives):  129.0\n",
      "Stage 1 TRUE positives:  117.0\n",
      "Stage 1 FALSE positives:  12.0\n",
      "Stage 1 TRUE negatives:  76.0\n",
      "Total number of flagged HPC-AD subjects:  75.0\n",
      "Number of flagged HPC-AD subjects that are TRUE positives:  70\n",
      "Number of flagged HPC-AD subjects that are FALSE positives:  5.0\n",
      "Number of true negatives:  83.0\n",
      "#############################\n",
      "Stage 1 stats for TARGET vs NON-TARGET\n",
      "Precision for AD:  0.9069767441860465\n",
      "Recall (or sensitivity)  for AD:  0.7959183673469388\n",
      "Specificity:  0.8636363636363636\n",
      "Adjusted precision for 33.6% baseline rate:  0.7470613844144537\n",
      "Accuracy:  0.8212765957446808\n",
      "#############################\n",
      "Stage 2 stats for TARGET vs NON-TARGET\n",
      "Precision for HPC-AD:  0.9333333333333333\n",
      "Recall (or sensitivity) for HPC-AD:  0.47619047619047616\n",
      "Specificity:  0.9431818181818182\n",
      "Adjusted precision for 33.6% baseline rate:  0.8091954022988507\n",
      "Accuracy:  0.6510638297872341\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.7959183673469388,\n",
       " 0.8636363636363636,\n",
       " 0.9069767441860465,\n",
       " 0.8212765957446808,\n",
       " 0.47619047619047616,\n",
       " 0.9431818181818182,\n",
       " 0.9333333333333333,\n",
       " 0.6510638297872341)"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predic_stats(y_, y_pred, lr_decision)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### ROC curve for classification of stable vs progressor MCI in ADNI1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following section of code was used to generate the ROC curves in Supplementary Figure 1 for the progressive MCI (pMCI) vs stable MCI (sMCI) classification in the ADNI1 dataset."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Set up for base SVM ROC curve"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set up the classifier\n",
    "base = high_confidence_at.BaseSvc()\n",
    "base.fit(x_, y_)\n",
    "y_predicted = base.predict(x_)\n",
    "y_score = base.decision_function(x_)\n",
    "\n",
    "# make sure to binarize the outputs\n",
    "y_true = y_.astype(int)\n",
    "y_true = label_binarize(y_true, classes=[0, 1])\n",
    "y_score = np.reshape(y_score, (y_score.shape[0],1))\n",
    "\n",
    "# Compute ROC curve\n",
    "fpr_b, tpr_b, thresholds_b = metrics.roc_curve(y_true, y_score, pos_label=1)\n",
    "# Compute AUC\n",
    "roc_auc_b = auc(fpr_b, tpr_b)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Set up the ROC curve for HPS model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get the output of HPS model\n",
    "y_score = np.reshape(lr_decision, (y_score.shape[0],1))\n",
    "\n",
    "# compute ROC curve\n",
    "fpr_h, tpr_h, thresholds_h = metrics.roc_curve(y_true, y_score, pos_label=1)\n",
    "# compute AUC\n",
    "roc_auc_h = auc(fpr_h, tpr_h)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Plot the two ROC curves "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots()\n",
    "fig.set_size_inches(10, 8)\n",
    "lw = 1\n",
    "plt.plot(fpr_b, tpr_b, color='blue',\n",
    "         lw=lw, label='Base (area = %0.4f)' % roc_auc_b)\n",
    "plt.plot(fpr_h, tpr_h, color='red',\n",
    "         lw=lw, label='HPS (area = %0.4f)' % roc_auc_h)\n",
    "plt.plot([0, 1], [0, 1], color='grey', lw=lw, linestyle='--')\n",
    "plt.xlim([-0.05, 1.00])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('ADNI1 pMCI vs sMCI')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Replication in ADNI2 "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this section, we will replicate our results in a new independent sample: the ADNI2 cohort."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Classify AD vs CN in ADNI2 with the model that was trained in ADNI1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "First, we will classify AD vs CN in ADNI2 with the same model that had been trained to classify AD vs CN in ADNI1. These results are represented in the paper's sections *\"Prediction of AD dementia vs cognitively normal individuals\"* for the Stage 1 (basic linear SVM) model and *\"Identification of easy AD cases for prediction\"* for the Stage 2 (highly predictive signature/HPS) model."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Organize the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "# grab only adni2\n",
    "adni2_adcn = data[data.dataset != 'ADNI1']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "# grab only AD and CN\n",
    "adni2_adcn = adni2_adcn[adni2_adcn.DX != 'pMCI']\n",
    "adni2_adcn = adni2_adcn[adni2_adcn.DX != 'sMCI']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set the x\n",
    "x_ = adni2_adcn.iloc[:,adni2_adcn.columns.get_loc(\"ADAS13\"):adni2_adcn.columns.get_loc(\"sub7\")+1].values\n",
    "# set the y\n",
    "y_ = adni2_adcn[['AD']].values.ravel()\n",
    "# add these variables to the x\n",
    "confounds = adni2_adcn[['gender','age_scan','mean_gm','tiv']].values\n",
    "# scale the x\n",
    "x_ = scaler.transform(np.hstack((x_,confounds)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Do the prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get the predicted labels\n",
    "array_results, dic_results = hpc.predict(x_, x_)\n",
    "y_pred = (dic_results['s1df'][:,0]>0).astype(float)\n",
    "lr_decision = dic_results['s2df'][:,1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Get the stats of the classifcation of AD vs CN in ADNI2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These stats are presented in the paper for the classification of AD (target subjects) vs CN (non-target subjects) in ADNI2 under the *Results* section of the preprint. See \"Prediction of AD dementia vs cognitively normal individuals\" for the Stage 1 (basic linear SVM) model and \"Identification of easy AD cases for prediction\" for the Stage 2 (highly predictive signature/HPS) model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total number of TARGET subjects:  88.0\n",
      "Total number of NON-TARGET subjects:  188.0\n",
      "Stage 1 number of hits (true and false positives):  88.0\n",
      "Stage 1 TRUE positives:  82.0\n",
      "Stage 1 FALSE positives:  6.0\n",
      "Stage 1 TRUE negatives:  182.0\n",
      "Total number of flagged HPC-AD subjects:  70.0\n",
      "Number of flagged HPC-AD subjects that are TRUE positives:  70\n",
      "Number of flagged HPC-AD subjects that are FALSE positives:  0.0\n",
      "Number of true negatives:  188.0\n",
      "#############################\n",
      "Stage 1 stats for TARGET vs NON-TARGET\n",
      "Precision for AD:  0.9318181818181818\n",
      "Recall (or sensitivity)  for AD:  0.9318181818181818\n",
      "Specificity:  0.9680851063829787\n",
      "Adjusted precision for 33.6% baseline rate:  0.9366060269407027\n",
      "Accuracy:  0.9565217391304348\n",
      "#############################\n",
      "Stage 2 stats for TARGET vs NON-TARGET\n",
      "Precision for HPC-AD:  1.0\n",
      "Recall (or sensitivity) for HPC-AD:  0.7954545454545454\n",
      "Specificity:  1.0\n",
      "Adjusted precision for 33.6% baseline rate:  1.0\n",
      "Accuracy:  0.9347826086956522\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.9318181818181818,\n",
       " 0.9680851063829787,\n",
       " 0.9318181818181818,\n",
       " 0.9565217391304348,\n",
       " 0.7954545454545454,\n",
       " 1.0,\n",
       " 1.0,\n",
       " 0.9347826086956522)"
      ]
     },
     "execution_count": 32,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predic_stats(y_, y_pred, lr_decision)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Generate ROC curves for AD vs CN in ADNI2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This section of code will generate the ROC curves presented in Supplementary Figure 1 for the classifcation of AD vs CN in the ADNI2 sample"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Set up the ROC curve for base SVM"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set up classifier\n",
    "base = high_confidence_at.BaseSvc()\n",
    "base.fit(x_, y_)\n",
    "y_predicted = base.predict(x_)\n",
    "y_score = base.decision_function(x_)\n",
    "\n",
    "# make sure to binarize the outputs\n",
    "y_true = y_.astype(int)\n",
    "y_true = label_binarize(y_true, classes=[0, 1])\n",
    "y_score = np.reshape(y_score, (y_score.shape[0],1))\n",
    "\n",
    "# Compute ROC curve\n",
    "fpr_b, tpr_b, thresholds_b = metrics.roc_curve(y_true, y_score, pos_label=1)\n",
    "# Compute AUC\n",
    "roc_auc_b = auc(fpr_b, tpr_b)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Set up the ROC curve for the HPS model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get the outputs from HPS model\n",
    "y_score = np.reshape(lr_decision, (y_score.shape[0],1))\n",
    "\n",
    "# compute ROC curve\n",
    "fpr_h, tpr_h, thresholds_h = metrics.roc_curve(y_true, y_score, pos_label=1)\n",
    "# compute AUC\n",
    "roc_auc_h = auc(fpr_h, tpr_h)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Plot the two curves"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots()\n",
    "fig.set_size_inches(10, 8)\n",
    "lw = 1\n",
    "plt.plot(fpr_b, tpr_b, color='blue',\n",
    "         lw=lw, label='Base (area = %0.4f)' % roc_auc_b)\n",
    "plt.plot(fpr_h, tpr_h, color='red',\n",
    "         lw=lw, label='HPS (area = %0.4f)' % roc_auc_h)\n",
    "plt.plot([0, 1], [0, 1], color='grey', lw=lw, linestyle='--')\n",
    "plt.xlim([-0.05, 1.00])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('ADNI2 AD vs CN')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Classification of MCI stable vs progressors in ADNI2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We will now use the model we trained in ADNI1 on AD vs CN to classify stable MCI vs progressive MCI patients in the ADNI2 sample. This section displays the code that was used to generate the results within the section titled *\"High confidence prediction of progression to AD dementia\"* for ADNI2 subjects."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Organize the data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [],
   "source": [
    "# grab adni2 subjects only\n",
    "adni2_mci = data[data.dataset != 'ADNI1']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [],
   "source": [
    "# grab the mci only\n",
    "adni2_mci = adni2_mci[adni2_mci.DX != 'Dementia']\n",
    "adni2_mci = adni2_mci[adni2_mci.DX != 'CN']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set up the x\n",
    "x_ = adni2_mci.iloc[:, adni2_mci.columns.get_loc(\"ADAS13\"):adni2_mci.columns.get_loc(\"sub7\")+1].values\n",
    "# set up the y\n",
    "y_ = adni2_mci[['conv_2_ad']].values.ravel()\n",
    "# add these variables to the x\n",
    "confounds = adni2_mci[['gender','age_scan','mean_gm','tiv']].values\n",
    "# scale the x\n",
    "x_ = scaler.transform(np.hstack((x_,confounds)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Do the prediction"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get the predicted labels\n",
    "array_results, dic_results = hpc.predict(x_, x_)\n",
    "y_pred = (dic_results['s1df'][:,0]>0).astype(float)\n",
    "lr_decision = dic_results['s2df'][:,1]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Stats for the prediction of stable vs progressor MCI in ADNI2"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These stats are presented in the paper for the classification of stable MCI (non-target subjects) vs progressor MCI (target subjects) in the ADNI2 sample under the section \"High confidence prediction of progression to AD dementia\" in the Results of the paper. Stage 1 refers to the basic linear SVM model, while Stage 2 refers to the highly predictive signature (HPS) model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Total number of TARGET subjects:  55\n",
      "Total number of NON-TARGET subjects:  180\n",
      "Stage 1 number of hits (true and false positives):  68.0\n",
      "Stage 1 TRUE positives:  42.0\n",
      "Stage 1 FALSE positives:  26.0\n",
      "Stage 1 TRUE negatives:  154.0\n",
      "Total number of flagged HPC-AD subjects:  37.0\n",
      "Number of flagged HPC-AD subjects that are TRUE positives:  28\n",
      "Number of flagged HPC-AD subjects that are FALSE positives:  9.0\n",
      "Number of true negatives:  171.0\n",
      "#############################\n",
      "Stage 1 stats for TARGET vs NON-TARGET\n",
      "Precision for AD:  0.6176470588235294\n",
      "Recall (or sensitivity)  for AD:  0.7636363636363637\n",
      "Specificity:  0.8555555555555555\n",
      "Adjusted precision for 33.6% baseline rate:  0.727906283670709\n",
      "Accuracy:  0.8340425531914893\n",
      "#############################\n",
      "Stage 2 stats for TARGET vs NON-TARGET\n",
      "Precision for HPC-AD:  0.7567567567567568\n",
      "Recall (or sensitivity) for HPC-AD:  0.509090909090909\n",
      "Specificity:  0.95\n",
      "Adjusted precision for 33.6% baseline rate:  0.8374577176428698\n",
      "Accuracy:  0.8468085106382979\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(0.7636363636363637,\n",
       " 0.8555555555555555,\n",
       " 0.6176470588235294,\n",
       " 0.8340425531914893,\n",
       " 0.509090909090909,\n",
       " 0.95,\n",
       " 0.7567567567567568,\n",
       " 0.8468085106382979)"
      ]
     },
     "execution_count": 40,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "predic_stats(y_, y_pred, lr_decision)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### These stats are presented in the paper for the classification of stable MCI (non-target subjects) vs progressor MCI (target subjects) in the ADNI2 sample"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Generate ROC curves"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The following code will generate ROC curves for the sMCI vs pMCI classification in ADNI2. These curves are displayed in Supplementary Figure 1 of the preprint."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Set up the ROC curve for base SVM "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [],
   "source": [
    "# set the classifier\n",
    "base = high_confidence_at.BaseSvc()\n",
    "base.fit(x_, y_)\n",
    "y_predicted = base.predict(x_)\n",
    "y_score = base.decision_function(x_)\n",
    "\n",
    "# make sure to binarize outputs\n",
    "y_true = y_.astype(int)\n",
    "y_true = label_binarize(y_true, classes=[0, 1])\n",
    "y_score = np.reshape(y_score, (y_score.shape[0],1))\n",
    "\n",
    "# Compute ROC curve\n",
    "fpr_b, tpr_b, thresholds_b = metrics.roc_curve(y_true, y_score, pos_label=1)\n",
    "# Compute AUC\n",
    "roc_auc_b = auc(fpr_b, tpr_b)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Set up the ROC curves for HPS model"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [],
   "source": [
    "# get the output for HPS model\n",
    "y_score = np.reshape(lr_decision, (y_score.shape[0],1))\n",
    "\n",
    "# Compute ROC curve\n",
    "fpr_h, tpr_h, thresholds_h = metrics.roc_curve(y_true, y_score, pos_label=1)\n",
    "# Compute AUC\n",
    "roc_auc_h = auc(fpr_h, tpr_h)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Plot the figure"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAmkAAAHwCAYAAAD98PjEAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDMuMC4wLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvqOYd8AAAIABJREFUeJzs3Xl8VXWe7vvnm5BAmAmTQEBkcCBgmHEAAYXrgKIiqGgpyhDb03bfW9Xdx67uunX71L3nVnWdvnWq6pRVJYMTjYWATAoWUqKWFiIiIiIgMxJACYQACWTc3/tHAhUDSTaQnbWHz/v12i/3sPZezw4ID7/fWr9l7i4AAABEl6SgAwAAAOB8lDQAAIAoREkDAACIQpQ0AACAKERJAwAAiEKUNAAAgChESQMAAIhClDQAdTKz98zsuJk1rvb8S2ZWYmanKm9bzOynZtaqyjZPmJmb2T9Ve2+OmY2qvP9vZvaflfc7mNkfzOyQmZ0ws7+Y2bDLyL6vMmO7as9vqszVvcpzQ81spZnlm1mema03sycrXxtlZjmXmuNymFn3yqwbqz3frvK77av2/CNmtsHMCszssJm9ZWbDK18797MGEN0oaQBqVVliRkhySeMvsMnP3b2FpPaSnpR0g6S/mFmzKtvkSXrWzFqGscvmkj6RNEhSuqSXJa0ws+aX+h0k7ZU0+ewDM+snKa3qBmZ2o6Q1kt6X1EtSW0lPS7rzMvZb35qZWd8qjx9RxXc7x8x+IOmXkv5fSR0ldZP0W0n3NlRIAPWDkgagLo9LWifpJUlTatrI3Yvc/RNVFLm2qihsZ22T9JGk79e1M3ff4+6/cPfD7l7u7jMlpUq65kLbV44MLTKz1ypH8zaaWVa1zeZWfo+zpkh6pdo2/0PSy+7+7+5+1Ct86u4P1pXZzP7ZzBZVe+5XZvbryvtPmNmeynx7zezRGj5naOUI2Ekz+9bMfnGB71H11+Dxqt+jcgTzJ5L+1t0Xu3uhu5e6+xvu/k8CEFMoaQDq8rikeZW3282sY20bu/spSatVMfpW1f8p6ftmln4xOzez/qooabtq2exeSQtVMfL2qqSlZpZS5fV1klqa2XVmlizpIUnnpvzMrKmkGyV9p2hdhD9IuuvsSGHlPh6U9GrliOKvJd1ZOeJ4k6RNNXzOryT9yt1bSuopaUG11/9T0sNmlmxm10lqIenjKq/fKKmJpCWX+D0ARBFKGoAaVR7HdKWkBe7+qaTdqphiq8shVRSmc9x9k6S3JT17EftvqYrRo//m7idq2fRTd1/k7qWSfqGKonJDtW3OjqaNlbRd0sEqr7VRxZ+Hh8PNVpW775e0UdJ9lU/dKum0u6+rfByS1NfM0ipHCL+s4aNKJfUys3buXlDl/WflSPpK0hhdeDSwraSj7l52Kd8DQHShpAGozRRJb7v70crHr6qWKc8quqjiOLTqfizpaTO7oq4PMLM0SW9IWufuP61j8wNn77h7SBVlpnO1beaqomA+ofPLzXFVFKlOdeWqxav663Fvj1Q+lrsXqmLk7m8kHTazFWZ2bQ2fMU3S1ZK2m9knZnb3BbZ5pfI7TFaV0cBKxyS1M7NGl/E9AEQJShqAC6osSQ9KGmlm35jZN6o4pizrAsd8VX1fc1WM9HxQ/TV33y5psaR/qWPfjSUtVcVo11NhxO1a5b1JkjJUMZpXdd/7VXGQ/V2VGaq+dloVx8w9EMa+arJQ0igzy5B0vypLWuXnr3L3saoogdslzbrQB7j7TnefLKmDpH+XtKjaCRiS9LqkcZL2VH6nqj6SVKS/jugBiGGUNAA1uU9SuaQ+kvpX3q5TRfl6vPrGZtbYzAapolwdl/RiDZ/731RxUkHrC71YeSzZIklnJD1eOTJWl0FmNqFyBOn/kFSsiuPQqpsm6dbK0a3q/qukJ8zsn8ysbWWWLDObH8b+5e65kt5Txffe6+7bKj+jo5mNryxbxZIKVPFzPY+Zfc/M2ld+5/zKp7+zbWX2WyVNv0CGE6oYrXzOzO4zs6ZmlmJmd5rZz8P5HgCiByUNQE2mSHrR3b9292/O3iT9RtKjVabU/quZnVLF9OYrkj6VdFMNRUjuvlcVU4/VR4jOuknS3ZL+N0n5lWt9FZhZ9RMRqlqmiinF45IekzSh8vi06vve7e4basi1VhXl51ZJe8wsT9JMSStr2W91r6piFPHVKs8lSfoHVYzs5UkaKem/1PD+OyR9aWYFqjiJ4GF3L7pA1g3uvruG7/ELST+Q9CNJuaqYCn5GFeUZQAwxdw86AwBcMjP7N0m93P17QWcBgPrESBoAAEAUoqQBAABEIaY7AQAAohAjaQAAAFGIkgYAABCFYm5V6nbt2nn37t2DjgEAAFCnTz/99Ki7t7+U98ZcSevevbs2bLjgMkcAAABRxcyqXxkkbEx3AgAARCFKGgAAQBSipAEAAEQhShoAAEAUoqQBAABEIUoaAABAFKKkAQAARCFKGgAAQBSipAEAAEQhShoAAEAUoqQBAABEIUoaAABAFKKkAQAARCFKGgAAQBSipAEAAEShiJU0M3vBzI6Y2ZYaXjcz+7WZ7TKzzWY2MFJZAAAAYk0kR9JeknRHLa/fKal35S1b0u8imAUAACCmRKykufufJeXVssm9kl7xCusktTazTpHKAwAAEEuCPCati6QDVR7nVD4HAAAQ8zZu3HhZ729UTzkuhV3gOb/ghmbZqpgSVbdu3SKZCQCA+pGeLh0/HnQKBOBU8+ZKLSlRj6ZNL+tzgixpOZK6VnmcIenQhTZ095mSZkrS4MGDL1jkAACIKsePS85fWRfDLPZ/ZPv379frr7+uO++8U9ddd13Fl7pEQU53Lpf0eOVZnjdIOuHuhwPMAwAAcMnWrVunhQsXavz48RUF7TJFbCTNzP4gaZSkdmaWI+n/kpQiSe7+e0krJd0laZek05KejFQWAACASAmFQkpKSlJaWpqmT5+u1q1b18vnRqykufvkOl53SX8bqf0DAABEWm5urhYuXKgJEyYoKyurXj87yGPSAAAAYtbWrVu1YsUK3Xbbbbriiivq/fMpaQAAABcpFApp8+bNevTRR9W5c+eI7IOSBgA1YAUFXKxjSle6Kn7T5KmN2l76iX0JqU2boBPUraCgQO+8847uuOMOPfzwwxHdFyUNAGrACgq4aPbX3zTpqmHxT8Ssr7/+WosWLdLAgQOVkpIS8f1R0gAAAOpQUFCgRYsW6Z577lHv3r0bZJ+UNAAAgBqUlJRo586dyszM1DPPPKPU1NQG23eQi9kCAABErWPHjmn27NnatWuX3L1BC5rESBoAAMB5Dh48qFdffVWjR4/WoEGDZJdxeadLRUkDAACoFAqFVFBQoA4dOuh73/ueOnXqFFgWpjsBAAAkFRYWau7cuXrvvfeUkpISaEGTGEkDUA/idT2xWFizCUD9yMnJ0cKFC5WVlaVRo0YFHUcSJQ1APWA9MQCxyiv/8AqFQrrrrrt0zTXXBJzoryhpAAAgIZWWlurNN99Ux44dddNNNwUd5zwckwYAABJOXl6eZs+eLUkaMmRIwGkujJE0AACQcD799FMNHjxYgwcPDmR5jXBQ0gAAQEIIhUJ6//33de2112rs2LFBx6kT050AACDunT59WvPmzdOBAwfUsmXLoOOEhZE0IEFEcpkMlqpA1GuodWL4nyEqubvmz5+vbt266dZbb1VSUmyMUVHSgATBMhlIaPwPkJDcXdu3b9fVV1+tRx55RE2aNAk60kWhpAEAgLhTWlqqlStX6uDBg8rIyFCLFi2CjnTRYmO8DwAAIEwlJSV64YUXVFZWpunTp8dkQZMYSQMAAHHk1KlTatGihW6//XZdeeWVUbu8RjgYSQMAADEvFArp3Xff1YsvvqiysjJ17949pguaxEgaAACIcadPn9bixYtVVlamqVOnqlGj+Kg38fEtAABAQnJ3ubu6du2qESNGxMzyGuGgpAFxrOrSUCzfFMUaag2vRMb/AHFp48aN2rFjhx5++GGNHDky6Dj1jpIGxDGWhooR/EIBF+Xs8ho5OTl66KGHgo4TMZQ0AAAQU/bu3avS0lJNnz5djRs3DjpOxFDSAABATNi1a5dOnTqlAQMGqHfv3jF/9mZd4ufoOgAAEJfcXe+//76WL1+u9PR0SYr7giYxkgYAAKLcn//8Z+3Zs0czZsyI2asHXApKGgAAiErffPON0tLSNGzYMA0fPlzJyclBR2pQlDSgAQS1wkKNqw6w5EN0YXkI4DybNm3S6tWrdf/996tXr15BxwkEJQ1oAFG3wkLUBQKAv3r77be1Y8cOPfHEE2rfvn3QcQJDSQMAAFGhsLBQTZs21bXXXquRI0fG9fIa4eDsTgAAELg9e/bod7/7nb755ht169Yt4QuaxEgaAAAIkLvrww8/1Pr16/XAAw+oU6dOQUeKGpQ0AAAQCK88NrakpEQzZsxQy5YtA04UXZjuBAAADe7bb7/VnDlzVFRUpNtuu42CdgGMpAEAgAa1efNmrVq1SrfffrvS0tKCjhO1KGnABdT3MmIsgwUAFQoLC7Vu3To9/vjj6tixY9BxoholDbgAlhEDgPp18uRJbdy4USNHjtSMGTMS4tqbl4tj0gAAQETt3btXs2bNUqNGFWNDFLTwMJIGAAAiJicnR4sXL9b999+vHj16BB0nplDSAABAvSsqKtLRo0fVpUsXPfXUU2revHnQkWIO050AAKBeHTlyRLNmzdK2bdtkZhS0S8RIGgAAqDc7d+7U0qVLNXbsWPXv3z/oODGNkoaoVt9LYYQrLpfMqPrDjMsvCCBI5eXlKi8vV9u2bfXYY4/piiuuCDpSzKOkIaqxFEY94ocJIEJOnTqlhQsX6pprrtHNN98cdJy4QUkDAACXbN++fVq8eLEGDx6sm266Keg4cYWSBgAALllubq7uvfde9ezZM+gocYeSBgAALkpxcbHeeOMNDRw4UEOGDAk6TtxiCQ4AABC23NxczZ49W40bN1a3bt2CjhPXGEkDAABhW7VqlW666SYNGDAg6Chxj5IGAABqVV5errVr12rIkCF69NFHufZmA2G6E1EhPV0yO//Gcl6XqeoPlh8mgEtQUFCguXPn6uuvv5a7U9AaECNpiAos4RUh/GABXIbS0lLNmTNHWVlZGjlyJAWtgVHSAADAd7i7Dhw4oG7duumJJ55Qq1atgo6UkJjuBAAA55SUlGjx4sV66623VFZWRkELECUNAABIkk6ePKnZs2erUaNGmjp1qho1YsItSPz0AQCAiouL1bRpU40cOVJ9+vTh+LMowEgaAAAJLBQKafXq1Zo/f74aNWqkzMxMClqUYCQNiHbp6RVnaV4Klt0AUIuCggK9/vrrSkpK0qRJk4KOg2ooaUC0YxkNABGSn5+vrl27atSoUUpKYnIt2lDSAABIIO6uTz75RIWFhRo9erQyMjKCjoQaUJsBAEgQJSUlWrJkiTZu3KisrKyg46AOjKQBAJAg1q1bJzPTtGnTlJKSEnQc1IGSBgBAnPvqq6/UokULDR8+XGbG2ZsxgulOAADiVCgU0jvvvKOVK1fK3ZWUlERBiyGMpAEAEKeWL1+ukydPKjs7W82aNQs6Di4SJQ0RcbFLe7Gcl2r+ofHDAXCRvvnmG7Vv316jRo1Sy5YtWV4jRvGrhog4u7RXuLe8vKATR4Gafmj8cACEyd21YcMGzZ07V7m5uWrdujUFLYYxkgYAQBxwdy1btkyHDx/W1KlT1bZt26Aj4TJR0gAAiHGlpaVKSUlRjx49dNdddyk1NTXoSKgHER0DNbM7zOwrM9tlZv98gde7mdm7ZvaZmW02s7simQcAgHizY8cO/eY3v9Hp06d1/fXXU9DiSMRG0swsWdJzksZKypH0iZktd/etVTb7kaQF7v47M+sjaaWk7pHKBABAvAiFQnr//fe1adMmTZw4UU2bNg06EupZJKc7h0ra5e57JMnM5ku6V1LVkuaSWlbebyXpUATzAAAQN0pKSpSfn68ZM2aoefPmQcdBBERyurOLpANVHudUPlfVv0n6npnlqGIU7e8u9EFmlm1mG8xsQ25ubiSy4hKlp0tm598ua9WImj403m8stQEgDIcOHdKiRYuUmpqq+++/n4IWxyJZ0i60pLFXezxZ0kvuniHpLklzzey8TO4+090Hu/vg9u3bRyAqLlVEVo242PU74uXGUhsA6rBx40bNmzdPffr0YWmNBBDJ6c4cSV2rPM7Q+dOZ0yTdIUnu/pGZNZHUTtKRCOYCACDmHDx4UOvWrdOTTz6pdu3aBR0HDSCSNfwTSb3N7CozS5X0sKTl1bb5WtJtkmRm10lqIon5TAAAKh0/flxbtmxRly5d9NRTT1HQEkjESpq7l0l6RtIqSdtUcRbnl2b2EzMbX7nZP0iaYWafS/qDpCfcvfqUKAAACWnnzp2aM2eOCgsLJUnJyckBJ0JDiuhitu6+UhUnBFR97sdV7m+VdHMkMwAAEIu2bNmit99+W5MmTdKVV14ZdBwEgCsOAAAQRc6cOaPS0lL17NlTM2bMUIsWLYKOhIBwaggAAFHi8OHDmjlzprZt26a0tDQKWoJjJA3nSU+vWAUjHCztBQD1Y/PmzVq1apXuuusuZWZmBh0HUYCShvOcXaYMABB55eXlSkpKUtOmTfXEE0+I9UBxFiUNAICA5Ofna+HChbrlllt0zTXXBB0HUYaSBgBAAHbv3q0lS5bopptu0tVXXx10HEQhShoAAAHYtm2bJk6cqO7duwcdBVGKszsBAGggRUVFWrJkifLz83X33XdT0FArShoAAA3gm2++0cyZM9WkSROW1kBYmO4EACDCysrKtHDhQo0ePVr9+vULOg5iBCNpCSw9XTI7/3ZRa5/V9CGXc2PxNQBxoqysTJ9++qmSk5P19NNPU9BwURhJS2D1sh4ai6oBwAWdPHlSCxYsUIsWLXT99dcrJSUl6EiIMZQ0AADqWX5+vubMmaNhw4bp5ptvlpkFHQkxiJIGAEA9cXfl5eUpPT1djz76qK644oqgIyGGcUwaAAD1oKioSK+99preeustSaKg4bIxkgYAwGXKzc3V/Pnz1aNHD91+++1Mb6JeUNIAALgMoVBIjRo10siRI3X99dcHHQdxhJKWANLTK07CrI6VLgDg0pWXl+vtt99WKBTSuHHj1IY/VFHPKGkJgFUyAKB+nTp1SgsXLlRaWpruv//+oOMgTlHSAAC4SF999ZV69eqlESNGcPwZIoaSBgBAGNxdH330kdLT0zV48OCg4yABsAQHAAB1KC4u1sKFC/Xll1+ytAYaDCNpAADUYcWKFWratKkmTJigRo34qxMNg99pAADUYPv27bryyit19913KzU1Neg4SDBMdwIAUE15eblWrVqlVatWqbCwkIKGQDCShgo1LaZWF9YFAhBn3F3z5s1TcnKysrOzlZaWFnQkJChKGiqwmBoAKD8/X61bt9att96qLl26sLwGAsV0JwAg4bm71q1bp9mzZ6uwsFAZGRkUNASOkTQAQEIrKSnR8uXLlZeXp2nTpqlZs2ZBRwIkUdIAAAksFAopKSlJHTt21L333quUlJSgIwHnMN0JAEhIW7du1Zw5c5SUlKQRI0ZQ0BB1GEkDACSUUCikd955R19++aUefPBBJSUxXoHoxO/MGJeeLpnVfqtxlYyqb2YpDQAJIjc3V0ePHlV2drY6d+4cdBygRoykxbjLWjmDZTcAJJADBw5oz549GjlypCZPnhx0HKBOjKQBAOKau2v9+vWaP3++OnXqFHQcIGyMpAEA4trmzZu1ceNGTZs2Tenp6UHHAcJGSQMAxKVjx46ppKREffv2VZ8+fTh7EzGH6U4AQNzZvn27XnjhBeXm5io5OZmChpjESBoAIK589NFHWrdunSZPnqyMjIyg4wCXjJIGAIgLp0+fVmpqqnr27Knrr7+eyzsh5jHdmWhYGw1AHMrJydHzzz+v3bt3q0OHDhQ0xAVG0hINa6MBiDMbNmzQu+++q3vuuUfXXHNN0HGAekNJAwDEJHeXmamsrExTp05V27Ztg44E1CumOwEAMScvL0+zZ8/W0aNHdcMNN1DQEJcYSQMAxJQdO3Zo2bJlGjlyJOUMcY2SBgCIGeXl5Vq7dq0efvhhde3aNeg4QEQx3QkAiHqnT5/WqlWr5O6aMmUKBQ0JgZIWI6qunFH1xioaAOLdoUOHNHPmTCUlJSkpKUlmFnQkoEEw3RkjWDkDQCI6ceKE5s2bp3HjxqlPnz5BxwEaFCUNABB1SktL9fXXX6tnz57627/9WzVt2jToSECDo6QBAKLK8ePHtWDBAnXo0EE9evSgoCFhUdIAAFEjJydH8+fP1/DhwzVs2DCOP0NCo6QBAALn7jpz5ozatWunhx56iLM3AXF2JwAgYGfOnNGrr76qNWvWqEmTJhQ0oBIjaQCAwBw+fFgLFizQtddeqzFjxgQdB4gqlLQok55esdxGdRe9Hlq9fRAARIa769SpUxozZowyMzODjgNEHUpalKm39dBYWA1AlCorK9Nbb72lzp07a9CgQUHHAaIWJQ0A0GDy8/O1YMECtWnTRn379g06DhDVKGkAgAbzwQcfqF+/frrhhhtYXgOoAyUNABBR7q6//OUv6tOnj+6++27KGRAmluAAAETMmTNnNH/+fO3YsUMpKSkUNOAiMJIGAIgId9d//ud/KiMjQw8++KCSk5ODjgTEFEpaFKi6WsZlrZBRbx8EAJdnz549uuqqqzR58mQ1b9486DhATGK6MwqcXS3DXcrLi4YPAoBLU1ZWphUrVmjlypUqLCykoAGXIaySZmapZtYr0mEAALGruLhYL730kgoKCjRjxgwKGnCZ6ixpZjZO0heSVlc+7m9mSyIdDAAQO86cOaPU1FQNHz5cDz74oBo3bhx0JCDmhTOS9hNJwyTlS5K7b5LEqBoAQO6uDz/8UHPmzJG769prr+UMTqCehHPiQKm751f7n47rDQFAgisqKtLSpUtVWFioxx9/XElJHOYM1KdwSto2M3tQUpKZXSXpf5e0LrKxAADRrqioSG3bttWkSZNYXgOIgHD+2fOMpEGSQpIWSypSRVEDACSgL774QkuXLlXr1q01duxYChoQIeGUtNvd/Vl3H1B5+2dJd0Y6WDxKT5fMzr+xpBmAWFBeXq6VK1fq3Xff1Y033hh0HCDuhVPSfnSB5/61voMkgqrLmFW9saQZgFiwbds2nThxQtnZ2erYsWPQcYC4V+MxaWZ2u6Q7JHUxs19UeamlKqY+AQAJYO/evSouLlZmZqYyMzM5exNoILWNpB2RtEUVx6B9WeX2tpjuBIC45+76y1/+otdff12pqakyMwoa0IBqHElz988kfWZm89y96FI+3MzukPQrScmSZrv7zy6wzYOS/k0Vy3p87u6PXMq+AAD167333tOuXbs0Y8YMtWrVKug4QMIJZwmOLmb23yX1kdTk7JPufnVtbzKzZEnPSRorKUfSJ2a23N23Vtmmt6QfSrrZ3Y+bWYdL+A4AgHqUm5urZs2aaejQoRoxYoQaNQrnrwoA9S2cEwdekvSiJFPFNOcCSfPDeN9QSbvcfY+7l1S+595q28yQ9Jy7H5ckdz8SZm4AQARs2bJFL730kg4dOqRmzZpR0IAAhVPSmrr7Kkly993u/iNJo8N4XxdJB6o8zql8rqqrJV1tZn8xs3WV06PnMbNsM9tgZhtyc3PD2HX0qLrsRkSW2oj4DgAkitWrV2vNmjV67LHH1KsXV/8DghbOP5GKreJI0d1m9jeSDkoKZ1ryQkeXVr+cVCNJvSWNkpQh6QMz6+vu+d95k/tMSTMlafDgwTF1Saqzy27E7g4AxLvi4mI1btxY3bt31/Dhw5WWlhZ0JAAKbyTt+5KaS/p7STerYopyahjvy5HUtcrjDEmHLrDNMncvdfe9kr5SRWkDADSA/fv367nnnlNubq569+5NQQOiSJ0jae7+ceXdU5IekyQzywjjsz+R1Lvyep8HJT0sqfqZm0slTZb0kpm1U8X0557wogMALpW7a926dfrLX/6i++67T+3btw86EoBqah1JM7MhZnZfZYGSmWWa2SsK4wLr7l6miut+rpK0TdICd//SzH5iZuMrN1sl6ZiZbZX0rqR/cvdjl/F9AABhCIVCOnbsmKZPn87xZ0CUMq/heCYz+6mkByR9LukqSUtUcWH1f5f0O3c/3VAhqxo8eLBv2LAhiF1fErMIHzIW8R0AiCdHjx7VH//4R02aNEmNGzcOOg4Q98zsU3cffCnvrW26815JWe5+xszSVXE8WZa7f3UpOwIABGvr1q1asWKFbrvtNgoaEANqK2lF7n5Gktw9z8y2U9AAIDadOnVK7777rh599FF17tw56DgAwlBbSethZosr75uk7lUey90nRDQZAOCyFRQUaOvWrRo6dKiefvppJSWFc1I/gGhQW0l7oNrj30QyCACgfn399ddatGiRBgwYIHenoAExprYLrL/TkEEAAPVn//79WrhwocaPH6+rr671UssAohQXZQOAOFJSUqITJ04oIyNDM2bMUKtWrYKOBOASMfYNAHHi2LFjmj17tjZt2qTk5GQKGhDjwh5JM7PG7l4cyTAAgEuzc+dOLV26VKNHj9agQYOCjgOgHtQ5kmZmQ83sC0k7Kx9nmdn/ingyAECdQqGQysrK1KJFCz3yyCMaPHiwzCzoWADqQTjTnb+WdLekY5Lk7p9LGh3JULiA9PSKqwtUv7VpE3QyAAEpLCzU3LlztWHDBl1xxRXq0qVL0JEA1KNwpjuT3H1/tX+ZlUcoD2py/DiXfwJwTk5OjhYuXKjrr79eQ4cODToOgAgIp6QdMLOhktzMkiX9naQdkY0FAKjNnj17dOedd+raa68NOgqACAmnpD2tiinPbpK+lfSnyucAAA2otLRUK1eu1KBBg3TLLbcEHQdAhIVT0src/eGIJwEA1CgvL08LFixQx44d1bFjx6DjAGgA4ZS0T8zsK0mvSVrs7qcinAkAUIW7a+nSpRo0aBBnbwIJpM6zO929p6T/R9IgSV+Y2VIzY2QNACIsFArp448/VllZmaZMmaIhQ4ZQ0IAEEtYVB9x9rbv/vaSBkk5KmhfRVACQ4E6fPq158+YTzazvAAAgAElEQVRp+/btKi0tVXJyctCRADSwcBazbW5mj5rZG5LWS8qVdFPEkyUq1kMDEl5xcbFmzZqlTp066bHHHlPTpk2DjgQgAOEck7ZF0huSfu7uH0Q4D1gPDUhY7q4jR46oY8eOeuSRR9S+ffugIwEIUDglrYe7hyKeBAAS2NnlNQ4fPqwZM2ZQ0ADUXNLM7P9z93+Q9LqZnTe04+4TIpoMABLEyZMn9Yc//EHt2rXT1KlTOf4MgKTaR9Jeq/zvbxoiCAAkorKyMjVu3FhDhw5V//79OXsTwDk1njjg7usr717n7u9UvUm6rmHiAUB8cne99957WrBggRo3bqwBAwZQ0AB8RzhLcEy9wHPT6jsIACSKM2fO6NVXX9W+ffs0fvz4oOMAiFK1HZP2kKSHJV1lZourvNRCUn6kgyWU9PSKszolltoAEsChQ4fUrl07jRkzhuPPANSotmPS1ks6JilD0nNVnj8l6bNIhko4LLsBJITPPvtMxcXFuuGGG9SzZ8+g4wCIcjWWNHffK2mvpD81XBwAiD9lZWVauXKlDhw4oAcffDDoOABiRG3Tne+7+0gzOy6p6jCPSXJ3T494OgCIA3/+859VXFys6dOnq3HjxkHHARAjapvuHF3533YNEQQA4s3u3bvVunVr3XLLLUpOTubsTQAXpbYlOM5eZaCrpGR3L5d0o6SnJDVrgGwAEJPcXe+//76WLl2q06dPq1GjRhQ0ABctnCU4lkpyM+sp6RVVrJH2akRTAUAMW7ZsmXbv3q3s7Gx17do16DgAYlQ41+4MuXupmU2Q9Et3/7WZcXYnAFRz7Ngxpaen66abblLbtm1ZXgPAZQlnJK3MzCZJekzSm5XPpUQuUoxKT5fMzru5zn/uvBtrowExb9OmTXrhhReUm5urDh06UNAAXLZwRtKmSvovkn7u7nvM7CpJf4hsrBhUw1pnZiyBBsSzUCiklStXat++fZoyZYo6dOgQdCQAcaLOkubuW8zs7yX1MrNrJe1y9/8e+WgAEN1CoZDMTO3bt9fYsWNZXgNAvapzutPMRkjaJWmOpBck7TCzmyMdDACi2Z49e/Tcc8+puLhYw4YNo6ABqHfhTHf+T0l3uftWSTKz6yTNlTQ4ksEAIBq5uz788EOtX79eEyZMUJMmTYKOBCBOhVPSUs8WNEly921mlhrBTAAQtQoLC5WTk6MZM2aoZcuWQccBEMfCObtzo5k9b2bDK2+/ExdYB5Bgvv32W7355ptq1qyZJk+eTEEDEHHhlLS/kbRb0n+V9KykPaq46gAAJITNmzfrlVdeUbdu3bhyAIAGU+t0p5n1k9RT0hJ3/3nDRAKA6LF//369//77evzxx9WxY8eg4wBIIDWOpJnZv6jiklCPSlptZlMbLBUABOzkyZPatWuXunXrpqeeeoqCBqDB1Tbd+aik6919kqQhkp5umEgAEKy9e/dq1qxZys3NlZkpNZVzpQA0vNqmO4vdvVCS3D3XzMI5fg0AYtoXX3yhVatWacKECerRo0fQcQAksNpKWg8zW1x53yT1rPJY7j4hoskAoAEVFxervLxc3bt314wZM9SqVaugIwFIcLWVtAeqPf5NJIMAQFCOHDmiBQsWaOjQoRo6dGjQcQBAUi0lzd3facggABCEL7/8UitXrtTYsWPVv3//oOMAwDnhXHEAAOJOKBRSUlKS3F2PPfaYrrjiiqAjAcB3UNIAJJxTp05p4cKFGjlypPr27Rt0HAC4oLDP2DSzxpEMAgANYf/+/Zo1a5Z69uzJ2ZsAolqdJc3MhprZF5J2Vj7OMrP/FfFkAFDP3F3r16/X+PHjNXLkSC7xBCCqhTOS9mtJd0s6Jknu/rmk0ZEMBQD1qbi4WCtWrNDp06c1adIk9erVK+hIAFCncEpakrvvr/ZceSTCAEB9y83N1ezZsxUKhdS4MUdtAIgd4Zw4cMDMhkpyM0uW9HeSdkQ2FgBcvpKSEs2bN08jR47UgAEDgo4DABclnJL2tCqmPLtJ+lbSn8R1PAFEsVAopO3bt6tPnz56+umnGUEDEJPqLGnufkTSww2QJW6kp0vHj1fcb9Mm2CxAoikoKNCiRYuUkpKiq6++moIGIGbVWdLMbJYkr/68u2dHJFEcOH5c8vN+YgAiLS8vTy+99JIGDhyoW265RUlJYa8yBABRJ5zpzj9Vud9E0v2SDkQmDgBcPHfXqVOn1Lp1a02cOFHdunULOhIAXLZwpjtfq/rYzOZKWh2xRABwEUpKSvTGG2+ouLhYjzzyCAUNQNy4lLmAqyRdWd9BAOBiHTt2TLNnz1ajRo00adKkoOMAQL0K55i04/rrMWlJkvIk/XMkQwFAXdxdZWVlGjZsmAYOHMjVAwDEnVpLmlX8qZcl6WDlUyF3DokHEJxQKKR33nlHkjR27Fh17Ngx4EQAEBm1TndWFrIl7l5eeaOgVZeeLpmx1gbQAAoKCjR37lx98803uvnmm4OOAwARFc7ZnevNbKC7b4x4mljEehtAg9m0aZO6du2qUaNGsbwGgLhXY0kzs0buXiZpuKQZZrZbUqEkU8Ug28AGyggggbm7PvnkE3Xo0EE333wzx54BSBi1jaStlzRQ0n0NlAUAvqOkpERvvvmmjhw5ogcffJCCBiCh1FbSTJLcfXcDZQGA71i6dKlSUlI0bdo0paSkBB0HABpUbSWtvZn9oKYX3f0XEcgDANq9e7e6deume+65R02aNGEEDUBCqu3I22RJzSW1qOEGAPXq7PIay5cv14kTJ5SWlkZBA5CwahtJO+zuP2mwJAASWigU0quvvqpQKKTs7Gw1a9Ys6EgAEKg6j0nDBaSnVyy9IZ1bH+0CTwEIU2FhoZo1a6Ybb7xRV111FctrAIBqn+68rcFSxJqza6O5S3l5NT0FoA7urg0bNuj3v/+9zpw5o549e1LQAKBSjSNp7k7VABAxpaWlWrFihQ4fPqwnnnhCaWlpQUcCgKgS0X+ymtkdZvaVme0ysxovym5mE83MzWxwJPMAiA7urlAopObNm2vatGlq27Zt0JEAIOpErKSZWbKk5yTdKamPpMlm1ucC27WQ9PeSPo5UFgDRY+fOnZo7d65SU1M1ZswYpaamBh0JAKJSJEfShkra5e573L1E0nxJ915gu/9b0s8lFUUwC4CAhUIhvfvuu3rjjTc0evRoltYAgDpEsqR1kXSgyuOcyufOMbMBkrq6+5sRzAEgChw6dEgHDhxQdna2unbtGnQcAIh6tS3Bcbku9M9kP/eiWZKk/ynpiTo/yCxbUrYkdevWrZ7i1Y+zS2+w7AZwYYcOHdLBgwc1ZMgQPfbYY4ygAUCYIjmSliOp6j+XMyQdqvK4haS+kt4zs32SbpC0/EInD7j7THcf7O6D27dvH8HIF+/s0hssuwGcb+PGjZo3b965hWkpaAAQvkiOpH0iqbeZXSXpoKSHJT1y9kV3PyGp3dnHZvaepH909w0RzASggXz22Wf66KOP9OSTT6pdu3Z1vwEA8B0RK2nuXmZmz0hapYrrgL7g7l+a2U8kbXD35ZHaN4Dg5Ofnq7y8XJmZmcrMzOTsTQC4RJEcSZO7r5S0stpzP65h21GRzAIg8nbt2qWlS5dq7NixysrKCjoOAMS0iJY0AInjo48+0kcffaRJkybpyiuvDDoOAMQ8ShqAy1JUVKTU1FRlZGRoxowZatGiRdCRACAucCVjAJfs8OHDev7557Vnzx517dqVggYA9YiRNACXZNOmTVq9erXuvPNO9erVK+g4ABB3KGkALpq7Ky8vT1OmTFGHDh2CjgMAcYnpTgBhO3HihF5++WWdOHFCt956KwUNACKIkgYgLHv27NGsWbPUu3dvtWrVKug4ABD3mO4EUKfS0lKtXr1aDzzwgK666qqg4wBAQmAkDUCNioqK9P777ys5OVnZ2dkUNABoQJQ0ABf07bffaubMmTp9+rTcnYujA0ADY7oTwHny8vL0yiuv6I477lC/fv2CjgMACYmSBuCc8vJyHT58WF26dFF2djYnCABAgJjuBCBJOnnypF566SWtX79eZkZBA4CAMZIGQAcPHtT8+fM1bNgw3XzzzUHHAQCIkgYkNHdXaWmpWrZsqQkTJnD2JgBEEUoakKCKi4u1bNkytWzZUnfccQcXRweAKMMxaUACOnLkiGbNmqVmzZppzJgxQccBAFwAI2lAAjp8+LBGjBihrKysoKMAAGpASQMSRHl5ud5++2117dqVcgYAMYDpTiABnDp1Si+//LLy8/PVs2fPoOMAAMLASBqQAN5++2316tVLI0aM4PJOABAjKGlAnHJ3ffLJJ+rTp4/uv/9+JSUxcA4AsYSSBsSh4uJiLV++XMePH9c111xDQQOAGMSf3ECccXe9/PLLatKkiaZOncrlnQAgRjGSBsSRgwcPqkuXLpo4caLS09ODjgMAuAyMpAFxoLy8XKtWrdKiRYt0+vRpChoAxAFG0oAYV1RUpPnz5yslJUXZ2dlKS0sLOhIAoB5Q0oAYVlpaqtTUVPXv319ZWVksrwEAcYTpTiAGubvWrVunOXPmyMzUv39/ChoAxBlG0oAYU1JSouXLl+vYsWN66KGHKGcAEKcoaUCMOXHihNLS0jR16lSlpKQEHQcAECFMdwIxYtu2bVq1apXat2+vcePGUdAAIM5R0oAoFwqFtHr1aq1atUp9+/YNOg4AoIEw3XmJzh4G1KZNsDkQ/zZt2qRvv/1W2dnZatq0adBxAAANhJJ2idyDToB4d+DAAZWXl6t///7q378/198EgATDn/pAlHF3rV+/XvPnz1dpaamSkpIoaACQgBhJA6LMmjVrtHPnTk2bNo3LOwFAAqOkAVEiLy9PzZs316BBg3TLLbdw9iYAJDjmUIAosH37ds2ZM0c5OTlq3bo1BQ0AwEgaELQ1a9bo888/1+TJk5WRkRF0HABAlGAkLVzp6RXrbpgpT6y7gctXVlYmSWrfvr2ys7MpaACA76Ckhev48Yp1N9zVVnlBp0GMO3jwoJ577jnl5eWpX79+atasWdCRAABRhulOoAG5uz799FO9++67uueeezh7EwBQI0oa0IDKy8u1b98+TZ06VW3btg06DgAgijHdCTSAvLw8LVq0SO6uiRMnUtAAAHWipAERtmPHDs2ZM0fdunVTo0YMXgMAwsPfGEAE5efna+XKlXr44YfVtWvXoOMAAGIIJQ2IgNOnT2v37t3q16+fnnnmGUbQAAAXjelOoJ4dOnRIM2fO1Lfffit3p6ABAC4Jf3sA9Wjv3r1atGiRxo0bpz59+gQdBwAQwyhpQD0oKytTQUGBMjIyWF4DAFAvmO4ELtPx48f1wgsv6JNPPlFKSgoFDQBQLxhJAy7D7t27tWTJEg0fPlzDhg0LOg4AII5Q0oBL4O7nTgqYNGmSrrzyyqAjAQDiDCUNuEhnzpzR4sWLdc0112jw4MFBxwEAxCmOSQMuwuHDhzVz5ky1a9dOAwYMCDoOACCOMZIGXITNmzdrzJgxyszMDDoKACDOUdKAOpSVlentt9/W4MGDdfvttwcdBwCQIChpQC3y8/O1cOFCtWrVSq1atQo6DgAggVDSgBq4u1577TX169dPN954o8ws6EgAgARCSQOqcXdt3rxZ/fr105NPPqnU1NSgIwEAEhAlDaiiqKhIS5Ys0ZkzZ3T11VcrLS0t6EgAgATFEhxApTNnzmjmzJlq3bq1pkyZQkEDAASKkTRAFdffbNOmjSZMmKCMjIyg4wAAwEgaEltZWZlWrFih1157TaFQiIIGAIgajKQhYZ08eVILFixQixYt9MQTTygpiX+zAACiByUNCSkUCikpKUn9+vXT0KFDWV4DABB1GDpAQnF3ffjhh1q8eLGaN2+uYcOGUdAAAFGJkTQkjKKiIi1btkynTp3Sgw8+GHQcAABqRUlDwtizZ4+aN2+uBx54QI0a8VsfABDd+JsKce+LL75QeXm5+vfvrz59+gQdBwCAsHBMGuJWeXm53nrrLb377rvq1KlT0HEAALgojKQhbr3zzjvKz89Xdna2mjRpEnQcAAAuCiUNcWffvn1q27atRo4cqdTUVM7eBADEpIhOd5rZHWb2lZntMrN/vsDrPzCzrWa22czeMbMrI5kH8c3dtXbtWi1atEj5+flq3LgxBQ0AELMiNpJmZsmSnpM0VlKOpE/MbLm7b62y2WeSBrv7aTN7WtLPJT0UqUyIb8uWLVNubq6mT5+u1q1bBx0HAIDLEsnpzqGSdrn7Hkkys/mS7pV0rqS5+7tVtl8n6XsRzIM4dfLkSbVo0UKDBw/WFVdcwfIaAIC4EMnpzi6SDlR5nFP5XE2mSXorgnkQh7Zs2aLnn39ex44dU0ZGBgUNABA3Ivk32oUOBvILbmj2PUmDJY2s4fVsSdmS1K1bt/rKV6f0dOn48Yr7Luns4U1t2jRYBNQgFApp9erV2r59u773ve+pXbt2QUcCAKBeRbKk5UjqWuVxhqRD1TcyszGS/lXSSHcvvtAHuftMSTMlafDgwRcsepFw/LjkZ/dmVe4jUF75C9G4cWNlZ2crLS0t4EQAANS/SE53fiKpt5ldZWapkh6WtLzqBmY2QNLzksa7+5EIZkGc2L9/v2bOnKmysjKNGjWKggYAiFsRG0lz9zIze0bSKknJkl5w9y/N7CeSNrj7ckn/Q1JzSQsrl0r42t3HRyoTYpe76+OPP9aHH36o++67T6mpqUFHAgAgoiJ6lLW7r5S0stpzP65yf0wk94/4cfLkSW3fvl3Tpk1TGw4KBAAkAK7diah29OhRrVmzRi1bttSUKVMoaACAhEFJQ9TaunWrXnzxRbVu3VpmxtUDAAAJhUWlanFM6ZJVrsHBCE6D2r17t1avXq1HH31UnTt3DjoOAAANjpJWi3QdZ92NBlZQUKD8/Hz16NGD5TUAAAmN6U5EjQMHDmjWrFnav3+/zIyCBgBIaIykISp88cUXWrVqlcaPH6+rr7466DgAAASOkoZAlZSUSJI6d+6sqVOnKj09PeBEAABEB6Y7EZhjx45pzpw52rx5s9q2bUtBAwCgCkbSEIjt27frjTfe0OjRozVo0KCg4wAAEHUoaWhQ7i4z06lTpzR58mRlZGQEHQkAgKhESUODKSws1OLFizVq1CgNGTIk6DgAAEQ1jklDg8jJydHMmTPVuXNndenSJeg4AABEPUbSEHHurjVr1ujOO+/UtddeG3QcAABiAiNpiJjS0lL96U9/UklJiR577DEKGgAAF4GRNEREXl6eFixYoA4dOnBxdAAALgElDfWuqKhIL774okaMGKEhQ4ZQ0AAAuASUNNSbUCikffv2qUePHnrqqafUvHnzoCMBABCzKGmoF6dPn9brr78ud1f37t0paAAAXCZKGi7bsWPHNHfuXGVmZuq2225TUhLnowAAcLkoabhk7q6ioiK1bNlSd999t3r16hV0JAAA4gZDHrgkpaWlWr58uZYtW6aUlBQKGgAA9YyShot2/PhxvfDCCyorK9OECROCjgMAQFxiuhMX7eTJk8rKytKwYcNYXgMAgAihpCEs7q73339fSUlJuuWWW3TllVcGHQkAgLhGSUOdzpw5o8WLF6u0tFQTJ04MOg4AAAmBkoY6rV27Vu3atdOYMWOUnJwcdBwAABICJQ01+uyzz9SpUyfdeuutHHsGAEAD4+xOnKesrExvvPGG1q5dq+TkZAoaAAABYCQN51m0aJGSk5M1ffp0NW7cOOg4AAAkJEoazjlw4IA6d+6su+66Sy1atGAEDQCAADHdiXPLayxYsED5+flq2bIlBQ0AgIAxkpbgysvL9dprr6moqEjZ2dlq0aJF0JEAAIAoaQmtuLhYjRs3VlZWlq699lqW1wAAIIow3ZmgPv/8c/32t79VcXGxMjMzKWgAAEQZRtISTFlZmf74xz9q3759evTRRzl7EwCAKEVJSzDFxcWSpBkzZlDQAACIYkx3Jog9e/Zo4cKFatq0qe6++24KGgAAUY6SFufcXR988IGWLFmiwYMHs7QGAAAxgunOOLdv3z7t2LFDM2bMUMuWLYOOAwAAwkRJi1PffvutcnNz1bdvXz355JNKSmLQFACAWEJJi0ObN2/WqlWrdMcdd0gSBQ0ALkJpaalycnJUVFQUdBTEkCZNmigjI0MpKSn19pmUtDjz6aefau3atXr88cfVsWPHoOMAQMzJyclRixYt1L17d47jRVjcXceOHVNOTo6uuuqqevtcSlqcOHnypCSpT58+yszMVJMmTQJOBACxqaioiIKGi2Jmatu2rXJzc+v1c5kHiwN79+7VrFmztHfvXqWlpVHQAOAyUdBwsSLxe4aRtBj38ccf64MPPtCECRPUo0ePoOMAAIB6wkhajCotLZW7q127dpoxYwYFDQDiSHJysvr376+srCwNHDhQa9eujej+fvnLX+qVV16J6D4uR3FxsR566CH16tVLw4YN0759+y643a9+9Sv17dtXmZmZ+uUvf3nu+by8PI0dO1a9e/fW2LFjdfz4cUnSiRMndM899ygrK0uZmZl68cUXJUn79+/XoEGD1L9/f2VmZur3v//9uc8aM2bMufdHGiUtBh05ckTPP/+89u3bp549e6pVq1ZBRwIA1KO0tDRt2rRJn3/+uX7605/qhz/8YcT2VVZWphdeeEGPPPLIRb2nIc2ZM0dt2rTRrl279P3vf1/PPvvsedts2bJFs2bN0vr16/X555/rzTff1M6dOyVJP/vZz3Tbbbdp586duu222/Szn/1MkvTcc8+pT58++vzzz/Xee+/pH/7hH1RSUqJOnTpp7dq12rRpkz7++GP97Gc/06FDhyRJjz32mH772982yPempMWYLVu26OWXX9bw4cPr9QwSAEB0OnnypNq0aSNJKigo0G233aaBAweqX79+WrZsmSSpsLBQ48aNU1ZWlvr27avXXntNUsUZ/yNHjtSgQYN0++236/Dhw+d9/po1azRw4EA1alRxBNSsWbM0ZMgQZWVl6YEHHtDp06clSU888YR+8IMfaPTo0Xr22WdVWFioqVOnasiQIRowYMC5LPv27dOIESM0cODAehsFXLZsmaZMmSJJmjhxot555x25+3e22bZtm2644QY1bdpUjRo10siRI7VkyZLz3j9lyhQtXbpUUsVxZKdOnZK7q6CgQOnp6WrUqJFSU1PPXT6xuLhYoVDo3H7Gjx+vP/zhD5f9ncLi7jF1GzRokDcYqeH2FYZQKORvvvmmHz58OOgoABC3tm7d+p3HUv3f6pKUlORZWVl+zTXXeMuWLX3Dhg3u7l5aWuonTpxwd/fc3Fzv2bOnh0IhX7RokU+fPv3c+/Pz872kpMRvvPFGP3LkiLu7z58/35988snz9vXjH//Yf/3rX597fPTo0XP3//Vf//Xca1OmTPFx48Z5WVmZu7v/8Ic/9Llz57q7+/Hjx713795eUFDghYWFfubMGXd337Fjh9f09/bw4cM9KyvrvNvq1avP2zYzM9MPHDhw7nGPHj08Nzf3O9ts3brVe/fu7UePHvXCwkK/4YYb/JlnnnF391atWn1n29atW7u7+8mTJ33UqFF+xRVXeLNmzfzNN988t83XX3/t/fr187S0NP/Nb37znff36tXrOz+nqhmqk7TBL7HzcOJADDh16pRWrFihcePGady4cUHHAYCEUm3ApkGcne6UpI8++kiPP/64tmzZInfXv/zLv+jPf/6zkpKSdPDgQX377bfq16+f/vEf/1HPPvus7r77bo0YMUJbtmzRli1bNHbsWElSeXm5OnXqdN6+Dh8+rOuuu+7c4y1btuhHP/qR8vPzVVBQoNtvv/3ca5MmTVJycrIk6e2339by5cv1H//xH5Iqli75+uuv1blzZz3zzDPatGmTkpOTtWPHjgt+xw8++CDsn4df4Beh+tmU1113nZ599lmNHTtWzZs3V1ZW1rnRwZqsWrVK/fv315o1a7R7926NHTtWI0aMUMuWLdW1a1dt3rxZhw4d0n333aeJEyeeW3+0Q4cOOnTokNq2bRv2d7gUlLQot3//fr3++usaNGiQmjdvHnQcAEADu/HGG3X06FHl5uZq5cqVys3N1aeffqqUlBT9/+3de1TU5dbA8e/GMIjSJIxUVCivoTYaEdVKMRPtVPqapkdL0yQr66SdcC07nW5vubpYy26eyk4JZhdvZWhqmmGxPF4pMNO8rNLS8Ia+ZiYFzH7/+I1zuIwyIjOA7M9as9b87nt4hNk+z/P77djYWAoLC2nXrh05OTksWrSIhx9+mJSUFAYMGEB8fDyrVq066fnDw8PLVFcYOXIk8+fP57LLLiM9PZ0VK1Z4t0VERHjfqyrz5s2jffv2Zc73xBNPEB0dTV5eHm63+4SPhbr22ms5cuRIhfUvvPAC119/fZl1MTEx/Pzzz8TExFBcXMzhw4eJjIyscOzo0aMZPXo0AP/4xz+IiYkBIDo6mvz8fJo1a0Z+fj4XXnghANOnT2fixImICG3atCEuLo7vv/+exMRE7zmbN29OfHw82dnZDBo0CHAS0vDwcJ+fqzrZnLRa7I8//iAzM5N+/frRo0cPe26PMcbUQ99//z0lJSVccMEFHD58mAsvvJDQ0FCysrLYuXMnAL/88gvnnHMOt99+O2lpaXz99de0b9+e/fv3e5O0oqIivvvuuwrn79ixI9u3b/cuHzlyhGbNmlFUVMR77713wrj69OnDq6++6u3l+uabbwDnjslmzZoREhLCu+++S0lJic/js7Ozyc3NrfAqn6CBMw8sIyMDgLlz53Ldddf5/E7ct28fAD/99BMfffQRQ4cOrXB8RkYG/fv3B6BVq1YsX74ccGpeb9myhYsvvphdu3Zx7NgxAA4dOsTKlSu9yaiqsmfPHmJjY0/4s6ku1pNWC/3xxx/k5uaSmJjI2LFjvV3Lxhhj6odjx47hcrkAJynIyMigQYMG3Hbbbdx8880kJCTgckt7i50AABMWSURBVLno0KEDAN9++y0TJkwgJCSE0NBQXn/9dRo2bMjcuXN54IEHOHz4MMXFxYwfP574+Pgy17rhhhsYPny4d/mpp57iyiuvpHXr1nTu3NlnbxfAo48+yvjx4+nSpQuqSmxsLAsXLmTs2LEMHDiQOXPm0LNnzzK9b1U1evRohg8fTps2bYiMjOTDDz8EnOQ0NTWVRYsWATBw4EAKCgoIDQ1l6tSp3hsuJk6cyODBg3n77bdp1aoVc+bM8X6GkSNH0rlzZ1SV5557jqioKJYtW8ZDDz2EiKCqpKWl0blzZ8C5GSMpKanSodTqIL7GeWuzhIQEXb9+fXAuJhL0yQgHDhxg1qxZtGzZkhtvvNESNGOMCbLNmzeXmaNVHwwYMIDnn3+etm3b1nQotd64cePo168fvXr1qrDN178dEclR1YSqXMuGO2uR/fv3M336dK666ir69etnCZoxxpigePbZZ30+nsNU1KlTJ58JWiDYcGct4Ha7OXDgAE2bNmXUqFFERUXVdEjGGGPqkfbt21e4AcD4dtdddwXtWtaTVsN+++03ZsyYQXZ2NiJiCZoxxhhjAOtJq1G7d+9m9uzZdO3ale7du9d0OMYYY4ypRSxJqwGqSklJCeHh4dx00002UdMYY4wxFViSFmR//vknCxYsoHHjxlx//fU+H8ZnjDHGGGNz0oKooKCAt99+mwYNGtCjR4+aDscYY0wtVb7CTHp6Ovfffz/gPNG/RYsWuFwuOnXqRGZmJgBbtmwhOTkZl8tFx44dGTNmjM9z5+fnc9NNNwX2A5ymjIwM2rZtS9u2bb0PoS0vNzeXpKQkXC4XCQkJrF271rttxYoVuFwu4uPjvd+3W7ZsweVyeV+NGjXipZdeApznpXXp0gWXy0VKSgq//PILAAsXLuTxxx8P8Kc9iaoW/aypV10usL569Wpdt26dut3uaj2vMcaY6uOrSHawRURElFmePn263nfffaqq+vjjj+vkyZNV1Yn1ggsu0JKSEk1JSdH58+d7j9mwYYPPc6elpZXZrzLHC6oHS0FBgcbFxWlBQYEePHhQ4+Li9ODBgxX26927ty5atEhVVT/99FPt0aOHqjrF3jt27Kg7d+5UVdW9e/dWOLa4uFijo6N1x44dqqreovWqqi+//LLefffdqqrqdrvV5XLp0aNH/Yq9ugusW09agLndbj7//HO2bdvGlVdeSUJCgpV3MsYYUy06duzIWWedxYEDB8jPz/fWqgS8T8gvb968efTt2xeAHTt2cO2119KtWze6devGf/7zH8DpierZsyfDhg3znmfmzJkkJibicrm4++67veWe7r33XhISEoiPj6+WXqfPPvuM3r17ExkZSZMmTejduzdLliypsJ+I8OuvvwJOKarmzZsD8P7773PLLbfQqlUrAG+dztKWL1/OJZdcQuvWrQFo1KiRd9vRo0e939MiQnJyMgsXLjztz1UVNictgI4ePcq8efMQEa6++uqaDscYY0xVBOI/1pVUsyldFgrg4MGD9OvXr8J+a9asISQkhKZNm/Lggw9y3XXXcfXVV5OSksKoUaM4//zzy+z/448/0qRJE84++2zASWCWLVtGWFgY27ZtY+jQoRyv6rN27Vo2btxIXFwcmzdvZtasWaxcuZLQ0FDGjh3Le++9x4gRI5g0aRKRkZGUlJTQq1cvNmzYQJcuXcpcd/LkyT7rgHbv3p1XXnmlzLrdu3fTsmVL73JMTAy7d++ucOxLL71Enz59SEtLw+12exPMrVu3UlRURHJyMkeOHGHcuHGMGDGizLEffviht67ncY888ggzZsygcePGZGVledcnJCSQnZ3N4MGDK8QQaJakBdCCBQuIiYkhOTmZkBDrtDTGmDqpBsonhoeHk5ub611OT0+ndEnEKVOmMHPmTM477zxmzZqFiDBq1Cj69OnDkiVL+OSTT3jzzTfJy8vzJmTgzEdr2rSpd7moqIj777+f3NxcGjRowNatW73bEhMTiYuLA5yep5ycHK644grASSKP91DNnj2badOmUVxcTH5+Pps2baqQpE2YMIEJEyb49dnVx8/b1wjU66+/zpQpUxg4cCCzZ89m9OjRfP755xQXF5OTk8Py5cs5duwYV111FUlJSbRr1w5wbuDLzMzkmWeeKXO+SZMmMWnSJJ555hlee+01nnzyScBJZI/PUQs2S9KqmaqSl5dHhw4dGDRoUFAKsBpjjKlfHnzwQdLS0iqsb968OXfeeSd33nknnTp1YuPGjVx++eXe7eHh4RQWFnqXp0yZQnR0NHl5ebjdbsLCwrzbShdGV1XuuOOOConNjz/+yAsvvMC6deto0qQJI0eOLHP+406lJy0mJoYVK1Z4l3ft2kVycnKFYzMyMnj55ZcBuPXWW0lNTfUeHxUVRUREBBEREXTv3p28vDxvkrZ48WK6detGdHR0hXMCDBs2jBtvvNGbpBUWFhIeHu5z30Cz7p1qVFRUxPz581m1ahWFhYWWoBljjAmaJUuWUFRUBMCePXsoKCigRYsWZfZp164dO3bs8C4fPnyYZs2aERISwrvvvuudZ1Zer169mDt3Lvv27QOc4dedO3fy66+/EhERQePGjdm7dy+LFy/2efyECRPIzc2t8CqfoAH06dOHpUuXcujQIQ4dOsTSpUvp06dPhf2aN2/Ol19+CcAXX3zhfeZo//79yc7Opri4mN9//501a9aUKXr+wQcfVBjq3LZtm/d9ZmYmHTp08C5v3bqVTp06+fxcgWZZRDVxu91Mnz6dpk2bkpqaSmhoaE2HZIwxph5ZunQp48aN8/aGTZ48mYsuuqjMPhEREVxyySVs376dNm3aMHbsWAYOHMicOXPo2bNnmd6z0i699FKefvppUlJScLvdhIaGMnXqVJKSkujatSvx8fFcfPHFXHPNNaf9OSIjI3n00Ue9Q6uPPfaY95miqamp3HPPPSQkJPDWW28xbtw4iouLCQsLY9q0aYBzM0Xfvn3p0qULISEhpKamepOs33//nWXLlvHmm2+WuebEiRPZsmULISEhtG7dmjfeeMO7LSsrq0IPYrCIr7Hf2iwhIUFLj8sHlIhfcxEOHDhAVFQUe/bsITo62u7eNMaYOmzz5s1lel7ONB9//DE5OTk8/fTTNR1Krbd3716GDRvG8uXL/drf178dEclR1YSqXN+GO0+D2+3miy++YObMmRQWFnLRRRdZgmaMMaZWGzBgALGxsTUdRp3w008/8eKLL9bY9W24s4qOHTvG3Llzcbvd3HXXXWUmWxpjjDG12fFJ9ubkjg+51hRL0k7mBEOdJSUlhIaG0rZtWxITE+3xGsYYc4ZRVRsZMackENPHLLs4BapKTk4O6enpNGjQgKSkJEvQjDHmDBMWFkZBQUFAvnTNmUlVKSgoqPZRNetJ81NRURGLFi1i9+7dDBkyxP6HZYwxZ6iYmBh27drF/v37azoUU4eEhYWVKctVHQKapIlIX+BloAHwb1V9ttz2s4EZwOVAATBEVXcEMqaq2rdvH263m9TUVBo2bFjT4RhjjAmQ0NBQ75P2jalJARurE5EGwFTgBuBSYKiIXFput9HAIVVtA0wBngtUPFW1bds2vvrqK1q0aMGAAQMsQTPGGGNMUARyQlUisF1Vf1DVP4EPgf7l9ukPZHjezwV6SS0ZR3S73WRlZbFgwQK7VdkYY4wxQRfI4c4WwM+llncBV55oH1UtFpHDwAXAgQDG5Zf169ezc+dOxowZw7nnnlvT4RhjjDGmnglkkuarR6z8rTL+7IOIjAHGeBZ/E5Etpxmbv6JGjRpV4wmjqVQUtSCxN5Wydqo7rK3qBmunuqF9VQ8MZJK2C2hZajkG+OUE++wSkbOAxsDB8idS1WnAtADFeUIisr6qpRxM8Fg71Q3WTnWHtVXdYO1UN4hIlWtZBnJO2jqgrYjEiUhD4K9AZrl9MoE7PO8HAV+oPZjGGGOMMSZwPWmeOWb3A5/hPILjHVX9TkT+F1ivqpnA28C7IrIdpwftr4GKxxhjjDGmLgnoc9JUdRGwqNy6x0q9LwRuDWQMpynoQ6ymSqyd6gZrp7rD2qpusHaqG6rcTmKji8YYY4wxtY8VnjTGGGOMqYXqfZImIn1FZIuIbBeRiT62ny0iszzb14hIbPCjNOBXW/1dRDaJyAYRWS4irWsizvqusnYqtd8gEVERsbvTaoA/7SQigz2/U9+JyPvBjtE4/Pjb10pEskTkG8/fv7/URJz1mYi8IyL7RGTjCbaLiLziacMNItLNn/PW6yTtTCldVR/42VbfAAmq2gWngsXzwY3S+NlOiMh5wAPAmuBGaMC/dhKRtsDDwDWqGg+MD3qgxt/fqX8Cs1W1K84NeP8KbpQGSAf6nmT7DUBbz2sM8Lo/J63XSRp1vHRVPVNpW6lqlqr+7llcjfNsPhNc/vxOATyFk0QXBjM44+VPO90FTFXVQwCqui/IMRqHP22lQCPP+8ZUfCapCTBV/Qofz3ktpT8wQx2rgfNFpFll563vSZqv0lUtTrSPqhYDx0tXmeDyp61KGw0sDmhExpdK20lEugItVXVhMAMzZfjz+9QOaCciK0VktYicrJfABI4/bfUEcLuI7MJ5osLfghOaOQWn+h0GBPgRHHVAtZWuMgHndzuIyO1AAtAjoBEZX07aTiISgjNtYGSwAjI++fP7dBbO0EwyTq90toh0UtX/C3Bspix/2mookK6qL4rIVTjPH+2kqu7Ah2f8VKVcor73pJ1K6SpOVrrKBJw/bYWIXA88AvRT1T+CFJv5r8ra6TygE7BCRHYASUCm3TwQdP7+7ftEVYtU9UdgC07SZoLLn7YaDcwGUNVVQBhOXU9Te/j1HVZefU/SrHRV3VFpW3mG0d7ESdBs/kzNOGk7qephVY1S1VhVjcWZO9hPVatc285UiT9/++YDPQFEJApn+POHoEZpwL+2+gnoBSAiHXGStP1BjdJUJhMY4bnLMwk4rKr5lR1Ur4c7rXRV3eFnW00GzgXmeO7t+ElV+9VY0PWQn+1kapif7fQZkCIim4ASYIKqFtRc1PWTn231EPCWiDyIM4Q20joTgktEPsCZGhDlmRv4OBAKoKpv4MwV/AuwHfgdGOXXea0djTHGGGNqn/o+3GmMMcYYUytZkmaMMcYYUwtZkmaMMcYYUwtZkmaMMcYYUwtZkmaMMcYYUwtZkmaMqVYiUiIiuaVesSfZN1ZENlbDNVeIyBYRyfOUMWpfhXPcIyIjPO9HikjzUtv+7atQ/GnGuU5EXH4cM15Ezjndaxtj6h5L0owx1e2YqrpKvXYE6bq3qeplQAbOM/NOiaq+oaozPIsjgealtqWq6qZqifK/cf4L/+IcD1iSZkw9ZEmaMSbgPD1m2SLyted1tY994kVkraf3bYOItPWsv73U+jdFpEEll/sKaOM5tpeIfCMi34rIOyJytmf9syKyyXOdFzzrnhCRNBEZhFP79T3PNcM9PWAJInKviDxfKuaRIvJqFeNcRakCyyLyuoisF5HvRORJz7oHcJLFLBHJ8qxLEZFVnp/jHBE5t5LrGGPqKEvSjDHVLbzUUOfHnnX7gN6q2g0YArzi47h7gJdV1YWTJO3ylLgZAlzjWV8C3FbJ9W8GvhWRMCAdGKKqnXEqrNwrIpHAACBeVbsAT5c+WFXnAutxerxcqnqs1Oa5wC2llocAs6oYZ1+c0kvHPaKqCUAXoIeIdFHVV3Dq+/VU1Z6e8kz/BK73/CzXA3+v5DrGmDqqXpeFMsYExDFPolJaKPCaZw5WCU4dyPJWAY+ISAzwkapuE5FewOXAOk+pr3CchM+X90TkGLAD+BvQHvhRVbd6tmcA9wGvAYXAv0XkU2Chvx9MVfeLyA+e2nvbPNdY6TnvqcQZgVPip1up9YNFZAzO3+VmwKXAhnLHJnnWr/RcpyHOz80YcwayJM0YEwwPAnuBy3B68AvL76Cq74vIGuBG4DMRSQUEyFDVh/24xm2lC7WLyAW+dvLUQkzEKUj9V+B+4LpT+CyzgMHA98DHqqriZEx+xwnkAc8CU4FbRCQOSAOuUNVDIpKOUyS7PAGWqerQU4jXGFNH2XCnMSYYGgP5quoGhuP0IpUhIhcDP3iG+DJxhv2WA4NE5ELPPpEi0trPa34PxIpIG8/ycOBLzxyuxqq6CGdSvq87LI8A553gvB8B/wMMxUnYONU4VbUIZ9gyyTNU2gg4ChwWkWjghhPEshq45vhnEpFzRMRXr6Qx5gxgSZoxJhj+BdwhIqtxhjqP+thnCLBRRHKBDsAMzx2V/wSWisgGYBnOUGClVLUQGAXMEZFvATfwBk7Cs9Bzvi9xevnKSwfeOH7jQLnzHgI2Aa1Vda1n3SnH6Znr9iKQpqp5wDfAd8A7OEOox00DFotIlqrux7nz9APPdVbj/KyMMWcgUdWajsEYY4wxxpRjPWnGGGOMMbWQJWnGGGOMMbWQJWnGGGOMMbWQJWnGGGOMMbWQJWnGGGOMMbWQJWnGGGOMMbWQJWnGGGOMMbWQJWnGGGOMMbXQ/wPBYQk9fCKZMQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 720x576 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, ax = plt.subplots()\n",
    "fig.set_size_inches(10, 8)\n",
    "lw = 1\n",
    "plt.plot(fpr_b, tpr_b, color='blue',\n",
    "         lw=lw, label='Base (area = %0.4f)' % roc_auc_b)\n",
    "plt.plot(fpr_h, tpr_h, color='red',\n",
    "         lw=lw, label='HPS (area = %0.4f)' % roc_auc_h)\n",
    "plt.plot([0, 1], [0, 1], color='grey', lw=lw, linestyle='--')\n",
    "plt.xlim([-0.05, 1.00])\n",
    "plt.ylim([0.0, 1.05])\n",
    "plt.xlabel('False Positive Rate')\n",
    "plt.ylabel('True Positive Rate')\n",
    "plt.title('ADNI2 pMCI vs sMCI')\n",
    "plt.legend(loc=\"lower right\")\n",
    "plt.show()"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python [default]",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.5.6"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}