Part2-NN/cur_neural.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Part2 - Neural Networks"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In this part we aim to build a manual neural network based on SGD, and to optimize it's structure using grid search and genetic algorithm method."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "import pandas as pd\n",
    "import os\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib as mpl\n",
    "%matplotlib inline\n",
    "import numpy as np\n",
    "from tqdm import tqdm\n",
    "from tqdm import trange\n",
    "from sklearn.metrics import roc_auc_score, roc_curve\n",
    "import time\n",
    "import pickle\n",
    "from sklearn.preprocessing import MinMaxScaler\n",
    "import itertools # iterations creator"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "data = pd.read_csv('..\\\\dataset_ver1.csv')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# PreProcess\n",
    "### Create dataset"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Unlike in Bayesian case, neural network input should contain as much continuos values as possible"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "data.sort_values('date',ascending=True, inplace=True)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dataset = data[[u'action_type', \n",
    "                u'minutes_remaining', u'period', u'seconds_remaining', \n",
    "                u'shot_distance', u'shot_made_flag',u'shot_zone_area', \n",
    "                u'loc_x', u'loc_y', \n",
    "                u'shot_zone_basic', u'shot_zone_range',  u'shot_type', \n",
    "                u'position', u'season', u'is_home']]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Filter outlier periods"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dataset = dataset[dataset.period <=4] ### 4 and not 5 this time"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Label "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dataset.rename(columns={'shot_made_flag' : 'label'}, inplace=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Position"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def transpos(position):\n",
    "    '''transfers position to numeric value (10-50)'''\n",
    "    dic = {'Guard':15, 'Center':50, 'Forward':40, 'Guard-Forward':22, 'Center-Forward':48,\n",
    "       'Forward-Center':42, 'Forward-Guard':28}\n",
    "    return dic[position]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dataset['numpos'] = dataset.position.apply(transpos)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Angle feature"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dataset['angle'] = dataset.apply(lambda x: np.arctan(x.loc_y*1.0/(x.loc_x+0.001)), axis = 1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Shot type feature"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "for index, row in dataset.iterrows():\n",
    "    ls_words = row['action_type'].split(' ')\n",
    "    for wd in ls_words:\n",
    "        dataset.set_value(index,wd ,1)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "dataset = dataset.fillna(0)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "new_dset = dataset[[ u'minutes_remaining', u'period', u'seconds_remaining',\n",
    "       u'shot_distance', u'label', u'loc_x', u'loc_y',\n",
    "       u'season', u'is_home', u'numpos', u'angle', u'Driving', u'No', u'Tip', u'Turnaround', u'Finger',\n",
    "       u'Pullup', u'Roll', u'Running', u'Dunk', u'Putback', u'Pull-Up',\n",
    "       u'Layup', u'Step', u'Cutting', u'Bank', u'Reverse', u'Fadeaway',\n",
    "       u'Back', u'Alley', u'Jump', u'Hook', u'Slam', u'Floating', u'Oop']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(406630, 35)"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "new_dset.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Scale data using Min-Max Scaling between 0-1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>minutes_remaining</th>\n",
       "      <th>period</th>\n",
       "      <th>seconds_remaining</th>\n",
       "      <th>shot_distance</th>\n",
       "      <th>label</th>\n",
       "      <th>loc_x</th>\n",
       "      <th>loc_y</th>\n",
       "      <th>season</th>\n",
       "      <th>is_home</th>\n",
       "      <th>numpos</th>\n",
       "      <th>...</th>\n",
       "      <th>Bank</th>\n",
       "      <th>Reverse</th>\n",
       "      <th>Fadeaway</th>\n",
       "      <th>Back</th>\n",
       "      <th>Alley</th>\n",
       "      <th>Jump</th>\n",
       "      <th>Hook</th>\n",
       "      <th>Slam</th>\n",
       "      <th>Floating</th>\n",
       "      <th>Oop</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "      <td>406630.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>0.446306</td>\n",
       "      <td>0.486850</td>\n",
       "      <td>0.487648</td>\n",
       "      <td>0.144109</td>\n",
       "      <td>0.452045</td>\n",
       "      <td>0.497957</td>\n",
       "      <td>0.145621</td>\n",
       "      <td>0.512867</td>\n",
       "      <td>0.500096</td>\n",
       "      <td>0.430878</td>\n",
       "      <td>...</td>\n",
       "      <td>0.020124</td>\n",
       "      <td>0.018073</td>\n",
       "      <td>0.021024</td>\n",
       "      <td>0.018152</td>\n",
       "      <td>0.011209</td>\n",
       "      <td>0.649664</td>\n",
       "      <td>0.038312</td>\n",
       "      <td>0.006293</td>\n",
       "      <td>0.026648</td>\n",
       "      <td>0.011209</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>0.287715</td>\n",
       "      <td>0.372832</td>\n",
       "      <td>0.295821</td>\n",
       "      <td>0.115735</td>\n",
       "      <td>0.497696</td>\n",
       "      <td>0.219723</td>\n",
       "      <td>0.097717</td>\n",
       "      <td>0.499835</td>\n",
       "      <td>0.500001</td>\n",
       "      <td>0.389321</td>\n",
       "      <td>...</td>\n",
       "      <td>0.140424</td>\n",
       "      <td>0.133215</td>\n",
       "      <td>0.143465</td>\n",
       "      <td>0.133500</td>\n",
       "      <td>0.105279</td>\n",
       "      <td>0.477076</td>\n",
       "      <td>0.191950</td>\n",
       "      <td>0.079080</td>\n",
       "      <td>0.161054</td>\n",
       "      <td>0.105279</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>0.166667</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.237288</td>\n",
       "      <td>0.023256</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.394000</td>\n",
       "      <td>0.063947</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>0.416667</td>\n",
       "      <td>0.333333</td>\n",
       "      <td>0.491525</td>\n",
       "      <td>0.151163</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.101433</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.714286</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>0.666667</td>\n",
       "      <td>0.666667</td>\n",
       "      <td>0.745763</td>\n",
       "      <td>0.267442</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.600000</td>\n",
       "      <td>0.228225</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.714286</td>\n",
       "      <td>...</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>1.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>8 rows × 35 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       minutes_remaining         period  seconds_remaining  shot_distance  \\\n",
       "count      406630.000000  406630.000000      406630.000000  406630.000000   \n",
       "mean            0.446306       0.486850           0.487648       0.144109   \n",
       "std             0.287715       0.372832           0.295821       0.115735   \n",
       "min             0.000000       0.000000           0.000000       0.000000   \n",
       "25%             0.166667       0.000000           0.237288       0.023256   \n",
       "50%             0.416667       0.333333           0.491525       0.151163   \n",
       "75%             0.666667       0.666667           0.745763       0.267442   \n",
       "max             1.000000       1.000000           1.000000       1.000000   \n",
       "\n",
       "               label          loc_x          loc_y         season  \\\n",
       "count  406630.000000  406630.000000  406630.000000  406630.000000   \n",
       "mean        0.452045       0.497957       0.145621       0.512867   \n",
       "std         0.497696       0.219723       0.097717       0.499835   \n",
       "min         0.000000       0.000000       0.000000       0.000000   \n",
       "25%         0.000000       0.394000       0.063947       0.000000   \n",
       "50%         0.000000       0.500000       0.101433       1.000000   \n",
       "75%         1.000000       0.600000       0.228225       1.000000   \n",
       "max         1.000000       1.000000       1.000000       1.000000   \n",
       "\n",
       "             is_home         numpos      ...                 Bank  \\\n",
       "count  406630.000000  406630.000000      ...        406630.000000   \n",
       "mean        0.500096       0.430878      ...             0.020124   \n",
       "std         0.500001       0.389321      ...             0.140424   \n",
       "min         0.000000       0.000000      ...             0.000000   \n",
       "25%         0.000000       0.000000      ...             0.000000   \n",
       "50%         1.000000       0.714286      ...             0.000000   \n",
       "75%         1.000000       0.714286      ...             0.000000   \n",
       "max         1.000000       1.000000      ...             1.000000   \n",
       "\n",
       "             Reverse       Fadeaway           Back          Alley  \\\n",
       "count  406630.000000  406630.000000  406630.000000  406630.000000   \n",
       "mean        0.018073       0.021024       0.018152       0.011209   \n",
       "std         0.133215       0.143465       0.133500       0.105279   \n",
       "min         0.000000       0.000000       0.000000       0.000000   \n",
       "25%         0.000000       0.000000       0.000000       0.000000   \n",
       "50%         0.000000       0.000000       0.000000       0.000000   \n",
       "75%         0.000000       0.000000       0.000000       0.000000   \n",
       "max         1.000000       1.000000       1.000000       1.000000   \n",
       "\n",
       "                Jump           Hook           Slam       Floating  \\\n",
       "count  406630.000000  406630.000000  406630.000000  406630.000000   \n",
       "mean        0.649664       0.038312       0.006293       0.026648   \n",
       "std         0.477076       0.191950       0.079080       0.161054   \n",
       "min         0.000000       0.000000       0.000000       0.000000   \n",
       "25%         0.000000       0.000000       0.000000       0.000000   \n",
       "50%         1.000000       0.000000       0.000000       0.000000   \n",
       "75%         1.000000       0.000000       0.000000       0.000000   \n",
       "max         1.000000       1.000000       1.000000       1.000000   \n",
       "\n",
       "                 Oop  \n",
       "count  406630.000000  \n",
       "mean        0.011209  \n",
       "std         0.105279  \n",
       "min         0.000000  \n",
       "25%         0.000000  \n",
       "50%         0.000000  \n",
       "75%         0.000000  \n",
       "max         1.000000  \n",
       "\n",
       "[8 rows x 35 columns]"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "'''transfrom all features valuse to scale between 0 and 1 using Min-Max'''\n",
    "from sklearn.preprocessing import MinMaxScaler\n",
    "scaler = MinMaxScaler()\n",
    "minmax_new_dset = pd.DataFrame(scaler.fit_transform(new_dset), columns=new_dset.columns)\n",
    "minmax_new_dset.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We tried to normalize the data to mean 0 and std 1, poor results due to redundant of binary variables"
   ]
  },
  {
   "cell_type": "raw",
   "metadata": {},
   "source": [
    "# N(0,1) scaler - not good for binary features and labels\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "norm = StandardScaler()\n",
    "norm_new_dset = pd.DataFrame(norm.fit_transform(new_dset), columns=new_dset.columns)\n",
    "norm_new_dset.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Save preprocessed data to csv:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "'''this saving is in order to reduce pre process time '''\n",
    "minmax_new_dset.to_csv(os.getcwd()+'\\\\minmax.csv',index=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "#### Load preprocesses csv:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "minmax_new_dset = pd.read_csv(os.getcwd()+'\\\\minmax.csv')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Divide the data to train , validation and test"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "TrainProp=0.7 #0.7\n",
    "validationsize = 0.1\n",
    "train=minmax_new_dset.sample(frac=TrainProp,random_state=200)\n",
    "test=minmax_new_dset.drop(train.index)\n",
    "validation = test.sample(frac=validationsize/(1.0-TrainProp),random_state=200)\n",
    "test=test.drop(validation.index)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((284641, 35), (40663, 35), (81326, 35))"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train.shape, validation.shape, test.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Model"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Part2  - SGD code"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "'''defining energy function (error function) logloss'''\n",
    "def logloss(Y,Z, deriv=False):\n",
    "    if deriv:\n",
    "        return -Y*1.0/Z + (1.0-Y)/(1-Z)\n",
    "    else:\n",
    "        return -Y*np.log(Z) - (1-Y)*np.log(1-Z)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "def quadloss(Y,Z, deriv=False):\n",
    "    if deriv:\n",
    "        return (Z-Y)\n",
    "    else:\n",
    "        return 0.5*((Z-Y)**2)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "'''defining activation function'''\n",
    "def sigmoid(x):\n",
    "    ''' sigmoid function'''\n",
    "    output = 1/(1+np.exp(-x))\n",
    "    return output\n",
    "\n",
    "def sigmoid_derivative(output):\n",
    "    return output*(1-output)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def nnet_structure_init(in_list, bias = True, randrange = 1.0):\n",
    "    ''' takes as input list of neurons in each layer, output will be list of synapses (matrices)'''\n",
    "    net = []\n",
    "    # in_list where in_list[0] is the input layer, in+list[-1] is the output layer\n",
    "    for l in range(len(in_list)-1):\n",
    "        syn = 2.0*randrange*np.random.random((in_list[l]+int(bias),in_list[l+1])) -randrange*1.0\n",
    "        net.append(syn)\n",
    "    return net"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def nnet_run(net, X, Y, epochs=50000, alpha =0.01, bias=True, batchsize = 200, diminish=1.0, prints=0.1, loss=quadloss, goal_error = 0.1):\n",
    "    '''takes net structure and run the X and Y (batches) on it epochs times, return update net'''\n",
    "    errors = [] # errors values via the run\n",
    "    pbar = xrange(iterations)\n",
    "    for j in pbar:\n",
    "        sample_idx = np.random.randint(low=0, high=X.shape[0], size=batchsize)\n",
    "        Xn = X[sample_idx,:]\n",
    "        Yn = Y[sample_idx,:]\n",
    "        \n",
    "        # feed forward\n",
    "        l = [Xn] # first layer value = input\n",
    "        for i in range(len(net)): #feed forward via synapses\n",
    "            if bias: # add vector of -1s to current layer (all except last layer)\n",
    "                l[-1] = np.concatenate((l[-1],-1*np.ones((l[-1].shape[0],1))), axis=1)\n",
    "            l.append(sigmoid(np.dot(l[i],net[i])))\n",
    "\n",
    "        # Cost - error function\n",
    "        err = loss(Yn,l[-1], deriv=True) # deriv of error cost - d the output and label\n",
    "        realerr= loss(Yn,l[-1], deriv=False)\n",
    "        errors.append(np.mean(np.abs(realerr)))\n",
    "        \n",
    "        if j%int(epochs*0.01)==0.0:\n",
    "            err_to_print = np.mean(errors[max(-100,-len(errors)):])\n",
    "            if err_to_print<=goal_error:\n",
    "                print \"Stopped with error (overfitting):\" ,err_to_print\n",
    "                break # stop condition\n",
    "        if j%int(epochs*prints)==0.0:\n",
    "            err_to_print = np.mean(errors[max(-100,-len(errors)):])\n",
    "            print \"Epoch:\",j,\"Error:\", str(err_to_print),time.ctime()\n",
    "#             pbar.set_description('epoch: '+str(j)+' error: '+str(round(np.mean(np.abs(err)),3)))\n",
    "        \n",
    "        #back-propagation\n",
    "        for i in range(len(net),0,-1): # run over synapses backwards\n",
    "            delta = err*sigmoid_derivative(l[i]) # delta in i synapses\n",
    "            if i<=len(net)-1 and bias:\n",
    "                delta = np.delete(delta,-1,axis=1) # remove bias row from synapses matrix\n",
    "            # print 'i',i, 'error:',error.shape,'delta:',delta.shape, 'net',net[i-1].shape\n",
    "            err = delta.dot(net[i-1].T)\n",
    "            net[i-1] -= alpha*l[i-1].T.dot(delta) # update synapse\n",
    "        # diminishing alpha\n",
    "        alpha = alpha*diminish\n",
    "    return net, errors\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def predict(X,net,bias = True):\n",
    "    'return the prediction'\n",
    "    l = [X] # first layer value = input\n",
    "    for i in range(len(net)): #feed forward via synapses\n",
    "        if bias: # add vector of -1s to current layer (all except last layer)\n",
    "            l[-1] = np.concatenate((l[-1],-1*np.ones((l[-1].shape[0],1))), axis=1)\n",
    "        l.append(sigmoid(np.dot(l[i],net[i])))\n",
    "    return l[-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# plot error\n",
    "def plot_error(alphas,errorlists,smooth=1):\n",
    "    fig  = plt.figure()\n",
    "    for a in range(len(alphas)):\n",
    "        iterations = len(errorlists[a])\n",
    "        err =[]\n",
    "        for i in range(len(errorlists[a])):\n",
    "            err.append(np.mean((errorlists[a][max(0,i-smooth):i])))\n",
    "        ax =plt.plot(range(iterations),err)\n",
    "    plt.title('Errorrate over epochs')\n",
    "    plt.ylim([0,1.1])\n",
    "    plt.xlim(-0.02*iterations,iterations)\n",
    "    leg = plt.legend(map(lambda x: 'alpha: '+str(x),alphas))\n",
    "    return fig"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "#plot ROC\n",
    "def plot_ROC(nets,alphas,X,Y, bias = True, title =\"\"):\n",
    "    AUC = []\n",
    "    fig2  = plt.figure()\n",
    "    for net in nets:\n",
    "        y_true ,y_scores = Y,  predict(X,net, bias = bias)\n",
    "        fpr, tpr, _ = roc_curve(y_true.ravel(), y_scores.ravel())\n",
    "        AUC.append(roc_auc_score(y_true, y_scores))\n",
    "        plt.xlim([-0.02,1.01])\n",
    "        plt.ylim([-0.01,1.02])\n",
    "        ax =plt.plot(fpr, tpr)\n",
    "    leg = zip(alphas,AUC)\n",
    "    plt.title(title)\n",
    "    l = plt.legend(map(lambda x: 'params: '+str(x[0])+' AUC:'+str(round(x[1],3)),leg), loc = 4)\n",
    "    return fig2, AUC"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Our data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "(284641L, 34L) (284641L, 1L)\n",
      "Epoch: 0 Error: 0.166902801808 Sat Dec 24 11:29:05 2016\n",
      "Epoch: 50000 Error: 0.109513511862 Sat Dec 24 11:30:44 2016\n",
      "Epoch: 100000 Error: 0.110343921349 Sat Dec 24 11:32:22 2016\n",
      "Epoch: 150000 Error: 0.109679935241 Sat Dec 24 11:33:59 2016\n",
      "Epoch: 200000 Error: 0.109039794289 Sat Dec 24 11:35:42 2016\n",
      "Epoch: 250000 Error: 0.108642935552 Sat Dec 24 11:37:24 2016\n",
      "Epoch: 300000 Error: 0.108892606268 Sat Dec 24 11:39:05 2016\n",
      "Epoch: 350000 Error: 0.108557350238 Sat Dec 24 11:40:47 2016\n",
      "Epoch: 400000 Error: 0.107917730631 Sat Dec 24 11:42:25 2016\n",
      "Epoch: 450000 Error: 0.107649255071 Sat Dec 24 11:44:09 2016\n"
     ]
    }
   ],
   "source": [
    "np.random.seed(777) # note the seed here\n",
    "\n",
    "trainsize = 284641\n",
    "B = True\n",
    "iterations = 500000 \n",
    "batchsize=500\n",
    "prints=0.1\n",
    "diminish=1.0\n",
    "alpha=0.001\n",
    "\n",
    "trainsize = min(trainsize,train.shape[0])\n",
    "samp =  train.sample(trainsize)\n",
    "cols = samp.columns.drop('label')\n",
    "X = np.array(samp[cols])\n",
    "Y = np.array(samp[['label']])\n",
    "\n",
    "# loss, goal  = logloss, 0.2\n",
    "loss,goal = quadloss, 0.02\n",
    "\n",
    "\n",
    "print X.shape, Y.shape\n",
    "\n",
    "inputSize = X.shape[1]\n",
    "structure = [inputSize,50,35,1]\n",
    "\n",
    "net = nnet_structure_init(structure, bias = B)\n",
    "newnet,errors = nnet_run(net, X, Y, epochs=iterations, alpha=alpha, bias=B,\n",
    "                         batchsize=batchsize, prints=prints, diminish=diminish, loss=loss, goal_error=goal)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Saving network object"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "netobj = {'params':(trainsize,B,iterations,structure,alpha,batchsize), 'net':newnet, 'errors' : errors}\n",
    "netobj['desc']= 'Tsize=%s_Bias=%s_iter=%s_struct=%s_alpha=%s_batch=%s'%netobj['params']\n",
    "netobj['name'] = ('ANN SGD '+netobj['desc']+' '+time.ctime()).replace(' ','_').replace(',',\"_\").replace(':','_')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true
   },
   "source": [
    "### save the networks to pickle file "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "pickle.dump( netobj, open( os.getcwd()+'\\\\pickles\\\\'+netobj['name']+'.pkl', \"wb\" ) )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "# unpickle\n",
    "unpick = pickle.load(open( os.getcwd()+'\\\\pickles\\\\'+netobj['name']+'.pkl', \"rb\" ) )"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Example network:"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Training error"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYMAAAEKCAYAAADw2zkCAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nS8JpSW95jUAAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x411e15f8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXUAAAEKCAYAAADticXcAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nCBcioS6EEC5EQl0IIVzI/wcTXKZ9E31SswAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x411e1588>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fe = plot_error([netobj['params'][-3:]],[netobj['errors']], smooth=200)\n",
    "froc, AUC = plot_ROC([netobj['net']],[netobj['params'][-3:]],X,Y, bias = netobj['params'][1], title='ROC curve on trainning set')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "fe.savefig(os.getcwd()+'\\\\figures\\\\'+'trainerror'+netobj['name']+'.pdf')\n",
    "froc.savefig(os.getcwd()+'\\\\figures\\\\'+'trainROC'+netobj['name']+'.pdf')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Validation"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ROC of Validation set\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXUAAAEKCAYAAADticXcAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzt3XdgVfX9//FndsgkkBDCSIAAH8IKyEYQqqBSF8662mq1\nWm0dX2t/dqsdtlVrte5VtUurde+JVRBBZpgfdlhJSEL2vOv3x73EEMkSkjvyevyVe865577vJ8nr\nfu7nnPM5YR6PBxERCQ3h/i5ARESOHYW6iEgIUaiLiIQQhbqISAhRqIuIhBCFuohICIn0dwHSPYwx\nbmAd4AY8QBxQAVxrrV3p2yYOuB04A2jwbfc68HtrbX2zfX0XuBqIBaKBxcAt1tqKbntDAc7X3qnA\n8cBJ1tobj7DNOuCH1tpP2tjPEOBua+15xpgM4AVr7awuKrtdxpjHgIettav9VYO0TT31nsMDzLXW\nTrTWHmetHQU8D9wPYIyJAD4AwoAJ1tpcYDqQCLxrjAn3bfdz4HvAmdba44BcwAm81t1vKMB5AKy1\nrx8p0DthCDDSt68Cfwa6z3y8fyMSoNRT7znCaPbP6AvxTKDUt+gCIMxae/OhbXy98xuNMauBs40x\nbwM/A3KttSW+bVzGmJt96yOttc7mL2qMOR34re+1a4Br8H5DWG+tTfRtk3Xose9bwBV4v0lUAjHA\nn621L/m2/YPvdX9mjLnCt78w3/u4zlprW75xY8xVwHV4P3yKgB9Za7cZY57yvcY4YDCwGfiWtba2\n2XOTgD3ACGvtAd+ypcBtwA7gQSAeGACs8T2/sdnzvwucZ609wxgzGngS6AVY33s8tN3PgbN87zce\nuBnvt6THgQG+tv9Bs3aKBO4BTvK9r2XA/1lra4wxO4GnfesGA89ba285Qrtcg/cbVwNQD1xtrd1s\njBkAPOB7bhTwnLX2j8aY3/ne57+MMd+x1n7Rcp/if+qp9yyLjDFrjDH7gC14e5OX+9bNAFobBvgQ\nmAWMAmqstTuar7TW1ltrnz1CoPcD/gF8x1o7Abgb+INvdctLmZs/Hg3MsdaeiDfULvftLxy4FHjc\nGHMC8B1glrV2EnAX8FLLwo0xJ+INyDnW2onAs8CrzTY5DjgZyMEbWOe3eG+Vvv1e6ttfDtDfWvsu\n8H3gaWvt8cAIYBhwmu+pzXuzh97bP4FHfW1xH5Dl22cmcCJwgm/dL4HfWGvdwJXAdmvtghb7+hWQ\nAYzzfauK8LXBIfHW2hPwDv9c5/vgbN4u4cBfgFOstdOAx/D+jsH7O3vSWjsFmAbMN8acZ639JbAf\nuFiBHrgU6j3LXF9onIa3t/jZoR63T1Qrz4vBGyZuOvc3czywzlq7DsBa+7K19rR2ngOQZ62t8f38\nPDDd9wFxKrDV96FyGpANfOb7JnEn0NsY07vFvk4B/mOtPeir4Rm8Pd9DIfeOtdbp+0BaB/Q5Qj1P\nAN/1/XwZ8JTv51uAEmPMT4CH8YZswpHekDGmDzAeb2Birf0M2OD7ebdvv5f6von8oLX9NHMq8Igv\n+ME7jLag2fpXffveDxxo+b58z3seWGqMuR/vN5YnfcdV5gC/9bXr53h77LnNnq7hlwCmUO9ZwgCs\ntWuAm/D+E2f61i0BTmj5BGNMmG/5EmAjEGWMGdZimxhjzJvGmP4tnu6kRY/cGDPOt6z53150i+dV\nH/rBNxTyAnAJ3uB73LcqAviH7/jARF8vfKq1trzFvo70Nx7Olx9gdc2WezhCYFlrlwCRxpgpwMV4\nh1AAnsPbW9+FdyhkdbPnH+mbSMv9OwGMMROBz/AdvwD+dKQ62nlfERz+odyR9/Ud4HRgK94PqJd8\n+wGY0axdZ/DlNywJcAr1Hspa+xzeILnPt+i/QI0x5l5jTCyAMaYX3h5gFfCKb6z4T8DffD1njDEx\nwL1AnLW2sMXLLANyfEMWGGMW4u2pluP9cBjl2+6cdsp9Am+gzwBe9C17D7jo0AeJMeZavAd6W3oX\n+JYxJtW33eVAibV2Wzuv2dKTeNtirbV2n2/ZfLzDJC/gDc1pfBmKh4WotbYMWIl3OAVjzHF4x/LB\n+6H5hbX2XrxDYGc324+TI3+Dehf4gTEm0jeUci3eNukQY0xfY8xuoNRa+1e8Qz651toqvL3zm33b\n9cb7gX5WO/VIgFCo9xxHmo7zOuBUY8x8a60L79hyDbDSGJMHrMAb6IfWY639I95gfdcYswpv79TD\nl//0TXwHFi8B/u7b9ka8BxIrgf8HvGOMWQa42ircWrsKcAD/PXQQ0lr7Ht4PmPeNMWuAC/GGYcvn\nfoB37Pgj3ymE3+bLce+2xvVbegbvEMTjzZb9HHjFGLMceAj4GBjexr4uxvtBtBb4Bd5vPuAd508z\nxmzA2+aVQB9jTDzeIRq3MeZzDv+g+B1QiPfg7Aa8Jz0cOsum3fdlrS3FewD7I2PMCrw98St8qy/B\nO+SVBywF/mWtfda37hXgP8aYeUd4fxIAwjT1rohI6FBPXUQkhCjURURCiEJdRCSEKNRFREJIh6YJ\nMMZMA/5orf1Gi+Vn4L2yzQE8Za19or19FRdXtXpkNiUljrKy2tZW9whqA7UBqA1AbQCHt0FaWmKH\nLvpqt6fuu1rucbxXFTZffmjuiXnAXOAqY0xa50o+XGRkRPsbhTi1gdoA1AagNoCv1wYdGX7ZxhHO\n/8U7V8ZWa22ltdaBd/rVr1yRKCIi3afd4Rdr7cstJwPyScI7294hVUDysSpMRCRUOV1uyqsaOFjV\nQHF5HeXVDdg95ZRW1FNe3UBURDhuD1xx5hhyhx5pOqLWHc3Uu5V4g/2QRLyXf7cpJSWuza8UaWmJ\nR1FSaFAbqA1AbQDB3wYul5tdBZXkbSuhorqB3UVV7D1QTUFJTZvPq8NFZv9EkhNiOt0GnQn1loP0\nm4DhvrkhavEOvdz1lWe10NaBj7S0RIqLqzpRUuhRG6gNQG0AwdkGB8pqydteyu6iarbvr6Cg9Kt5\n1ysmkuyBSST2iiYuNpLsgcn0TYqhT2IsfZJi6BUTSViYN26bt0FHw70zoe4BMMZchHeu5ieMMTfh\nnUQoDHjCWlvQif2JiAS1kvI61u88yJptJRwoq6Pw4JchHhsdwdCMRKIiIxif3ZdBaQkM7pdASmJM\nG3s8eh0KdWttPjDT9/OzzZa/CbzZNaWJiASW2noHdk85a7eVsnVv+Vd64rnZfTGZKQzNSGTEoN6E\nh3f/1PO6nZ2ISCvKqhpYt6OUz9YVsGXv4fdVj4wIY9ywvowd2oeRg3szqF88EeH+v55ToS4i4uPx\neNhRUMmarSWs33mQ/MLDx/Qz+saRm53K6KEpjMpMITLC/yHekkJdRHq02noH732xh035ZewvqaGm\n/stb7Q4flMyE4alNY+LBQKEuIj3S7qIq3l+xh2Ubi3C6vLOXpCbHMnpIH6aPTid7YDJJ8S3vtBj4\nFOoi0iN4PB52F1WzbGMRi9cVUF3nACChVxRzJw7k5CmDSegV/HfqU6iLSEjbVVjJ4rwC/rdmPy63\nt0ceHhZGanIs587JZuKIVKKjQmeeGYW6iIQch9PF0g1FLFlXwFbfWStRkeEMHZDEtJx0Zo7tT6+Y\n0Iy/0HxXItIj5Rd6x8k37DpIRXUjACMGJXPqtEzGDesbkGerHGsKdREJagWlNSzfdIDlm4oOuxjo\npOMGMW/KINJT4vxYXfdTqItI0GlwuFicV8DnGwvZvq+yafnoISnMGNOf6WPSA+JCIH9QqItIUHC5\n3WzKL2Pd9oMs3VDYdPbKyEHJTB/bn4nDU0lO6Np5VYKBQl1EApbH42HDzoMsXlfAmm0lNDrcAMTF\nRDJv8iBOnZpJn6RYP1cZWBTqIhJwnC43X2ws5Jk3NzZdqp8YF8XUUekcZ9IYM6QPUZE9c3ilPQp1\nEQkINfUO1u84yKLV+9i6p5xDd6gfn92X+ZMHkzMkhfCw7p/1MNgo1EXEb0rK6/h4zX7WbCthf7O7\nASXGRTHR9GP6qH6MykrxY4XBR6EuIt3K4/GwfX8lb3y2i7ztpQCEhUG/lF6MyuzNnAkDGdI/kX79\nkoLuzkeBQKEuIt0iv7CK5ZuK+HxjEWVVDYB33pWFs4cyc2x/YqMVR8eCWlFEutTGXQd5+ZMdbN//\n5fnkY4akMCUnnVnjMzROfowp1EXkmCupqOPTtQUs33yAIt99O0cPSeGE3AGMG9Y3ZOddCQRqWRE5\nZg6U1fLCou2s3FLctCwnK4VvTs9izNA+fqys51Coi8hRqa5zkLe9hKXrC9mwqwyAyIhwTp2WGTJz\nlAcThbqIdJrb7WHZxiKWrC9gc345bo/3rPKBqfHMmzyI48dl9IgZEQORQl1EOmzN1hIWrytgc34Z\ntQ3ee3kOSotnkunHhOGpZKYnEKYDn36lUBeRNtXUO1ix+QAfrtzH3uLqpuVjh/bhvLnZZKYn+rE6\naUmhLiJHtHxTEf/5aFvTOeUAmf0SOGfOMEYP6aPhlQClUBeRJo0OF3nbS3nxf9spKqsDICI8jFOn\nZTJzbH8y+sb7uUJpj0JdRKitd/DO8t18sGIv9Y0uAJITovnR2ePIHpjs5+qkMxTqIj1YQWkNH67c\ny9INhdQ1uIiJimDe5EFMNv0YPihZV3sGIYW6SA9TdLCW15bsYvPusqbx8vjYSBbOHsq8SYOIi9V5\n5cFMoS7SAzhdbj5bX8iyjUVsyi9rWj58YDITR6Yyb9Jg3XQiRCjURUKU0+Umb3spa7aVsMoWN51X\nnpWeyJwJAzhuZBpJ8dF+rlKONYW6SIgpq2rg07z9/G/N/qbhlV4xkcydOJCTpwwmPaWXLhAKYQp1\nkRDg9nhYaYt5f8Uetu2tACAmKoJZ4zOYPjqdkYN767zyHkKhLhLE9pfUsGjVPj7fWEhNvXd4JXtA\nEjPG9mfGmP6a4rYH0m9cJAjlF1axaPVePllbAHiHV2aNy2DuxIEMG5Dk5+rEn9oNdWNMGPAQkAvU\nA1daa3c0W38JcBPgBJ6y1j7SRbWK9Gi19U6WbSpi0aov52BJSYxh1rgMTp85RGevCNCxnvpCIMZa\nO9MYMw24x7fskLuAHKAW2GiMedZaW3HsSxXpmfYVV7No9T6WbSxqGmLJyUph1vgMpub0IyJcYS5f\n6kiozwLeAbDWLjPGTG6xfi2QAnh8jz2IyFHblF/Gu8t3k7e9FIDY6AgWTM/kpOMG0Scp1s/VSaDq\nSKgnAc173k5jTLi11u17vAFYCVQDL1lrK1vuQEQ6prrOwadr9/PZhkL2FdcAMDQjiflTvJfu6wwW\naU9HQr0SaD5hclOgG2PGAacBWUAN8C9jzLnW2hdb21lKShyRkRGtvlhamuZmVhv0vDZodLh4cdE2\nXvhwCw6nm7AwmDamPydPy2JyTjrh4T3zvPKe9ndwJJ1tg46E+hLgdOC/xpjpwLpm6yrwjqU3WGs9\nxpgDeIdiWlVWVtvqurS0RIqLqzpQUuhSG/SsNthVWMm7y/ewZmsJDQ4XUZHhnD4zi/Pnj8LV4ACg\ntLS6nb2Epp70d9Ca5m3Q0XDvSKi/DMw3xizxPb7cGHMREG+tfcIY8xiw2BjTAGwHnu5s4SI9zd4D\n1by1LJ/PNxQBkBQXxZwJA1gwPYvk+Gj6JMVSXOzwc5USjNoNdWutB7imxeItzdY/Cjx6jOsSCTku\nt3culneX7WaL76rP1ORYLpo3ggnDU3XpvhwTuvhIpIs1Olws3VDIm0vzKamoB2DkoGROmZpJ7vDU\nHjteLl1DoS7SRWrrnSzO28/by3ZTUdNIWBgcP64/c3IHMnyQ7iYkXUOhLnIMeTwe7O5y3l+xh7zt\npbjcHiLCw/jGxIGcOi2TtN69/F2ihDiFusgx0OBw8d7y3Xy4ah+VNY0ADEyNZ9rodGaPzyA5IcbP\nFUpPoVAXOQoej4dVW0r41/uW8upG4mIimZrTj1njMxgzpI8Ofkq3U6iLfA0NjS4+zdvPR6v2UXjQ\ne+3F7PEZfOvE4brHp/iVQl2kE/aX1PDO8t2s2HyA+kYXYcBxI9M4Y+YQsvrr6kfxP4W6SDvcHg95\n20t5fckudhZ4pzZKiovipEmDOPG4QaQkarxcAodCXaQVDY0uPlq9lw9W7G2612f2wCROPG6QpryV\ngKVQF2mhvLqBVz7dybJNRTQ0uoiJimBqTj/mTxlM9gCdXy6BTaEu4lNV28g7y3bzwcq9OJxuIiPC\nWDAtkwXTs0jopYOfEhwU6tLjNThcvP15Pu8s202j001yfDQL5mRxQm4GsdH6F5Hgor9Y6bFq6h28\n+PF2lm0qoq7BRa+YCM6fnc1Jxw0iOqr1Of9FAplCXXqcsqoG3l2+m/+t2U+DwwXAgumZLJimYRYJ\nfgp16THyC6t4bclO1m4rxe3xEB8bySlThzBv8mCFuYQMhbqEvL3F1by+ZBdfbD4AwIDUeOZOGMAJ\nuQM0zCIhR6EuIcnhdLN8UxEfrdrXdMFQv5RenDcnm0kmTXOySMhSqEtIcXs8fLx6H/96fwsej3fZ\n6CEpzJkwkEkmjXCFuYQ4hbqEBKfLzfodB/nn+5aDlQ1EhIcxaVQa55wwjH4pcf4uT6TbKNQlqDmc\nLhat2sd7K/ZwsNJ7Kf/00eksPGEY/XRDCumBFOoSlBodLt79Yg8frdpLRbX3phTHj+vPSZMGMaR/\nkp+rE/EfhboEnbztpTz7wRaKyuoIDwsjN7svF88fqVvFiaBQlyDhcLpZsq6A91fsoaC0ljC8N6U4\na9ZQ+iTF+rs8kYChUJeAVlvvZNHqvXy8eh+llQ2EhcGE4amcNiOL7IGaMVGkJYW6BKT6RicfrNjL\nS5/sACA8LIw5EwYwf/JgBqTG+7k6kcClUJeAsrOgkteW5vPhF7upqnU09cy/c+ookuOj/V2eSMBT\nqEtAOFBWy6uLd7F0QyEAkRHhnDJ1MCdPydTt4kQ6QaEuflVV28jrn+3io5X7cHs89EvpxffOGEtW\nWhwxmpdFpNMU6uIXbreHT/P28/KnO6msaSQ5IZqzZw9j1rgM0tOTKC6u8neJIkFJoS7dbnN+GU+/\ns5kDZXUAfHN6FqfNyKJXjP4cRY6W/ouk2+zYX8lzH25l274KAKbm9OOMmUMYmJbg58pEQodCXbqU\n2+1h9dZiPlq1j035ZQCYwb1ZMD2T8dmpfq5OJPQo1KXL5G0v4R/vWkp9E21l9I3jgm8MJ3e4wlyk\nqyjU5Zg7UFbLvz/YSt72UgAmmTROnjKYEYN6+7kykdCnUJdjJr+wijeX7mLllmI8Hsjqn8jF80Yo\nzEW6kUJdjprD6eKf723h07wCADL7JXDK1Eymj0nXbeNEulm7oW6MCQMeAnKBeuBKa+2OZuunAH/2\nPSwELrXWNnZBrRJgPB4P63Yc5OVPdpBfVEVG3zhOnZbJrHEZCnMRP+lIT30hEGOtnWmMmQbc41t2\nyGPAudbaHcaY7wFZwNZjX6oECo/Hw6otxbyxNJ/8Qu9FQtNHp/PdBaN0FaiIn3Uk1GcB7wBYa5cZ\nYyYfWmGMGQmUAjcZY8YCb1hrFegh7PONhby+ZBcFpbUAjM/uy5nHD2XYAN1tSCQQdCTUk4CKZo+d\nxphwa60bSAVmANcCO4A3jDErrLUfH/NKxa/2l9Twn4+2sW6H74yWkWksmJ6lMBcJMB0J9Uogsdnj\nQ4EO3l76NmvtFgBjzDvAZODj1naWkhJHZGTrX9HT0hJbXddTBFIbVNc5eOLVdXz4xR4Axmb35aqF\n4xg6oGtvUBFIbeAvagO1AXS+DToS6kuA04H/GmOmA+uardsBJBhjhvkOns4GnmhrZ2Vlta2uS0tL\n7PETOQVKG7g9Hhat2sdrS3ZSVesgJTGG8+dmM3V0OuFhYV1aY6C0gT+pDdQGcHgbdDTcOxLqLwPz\njTFLfI8vN8ZcBMRba58wxlwBPGuMAfjMWvt2pyuXgFJW1cDf3trEhp0HiY4M58zjh/DN6VlE6yCo\nSMBrN9SttR7gmhaLtzRb/zEw7diWJf7g9nhYklfAS5/soKKmkZysFK44LUc3dhYJIrr4SHB7PKzZ\nWsKbS3exs6CKqMhwzp49lNNmDiFc55uLBBWFeg9XdLCWR17b0HS++ZRR/Th/bjapvXv5uTIR+ToU\n6j1Ug8PFy5/s4P0Ve/B4YOzQPpx9wjCGZugURZFgplDvgYrL63jgpXXsOVBNfGwk55wwjLkTB+rS\nfpEQoFDvYRat3sezH2zF6XIzfUw63z7Z6DZyIiFE/809xJ4D1bzy6Q5Wby0hNjqCS+YbTsgdoN65\nSIhRqIe4kvI6nl+0jRW2GIChGYlcefpoMvrG+7kyEekKCvUQVVPv4PUlu3jPd3n/wLR4zjx+KJNN\nmnrnIiFMoR6C8raX8PCrG2hodJEYF8Vp07OYP2WwwlykB1CohxC328OzH2zlw1V7CQ8LY+6EAZwz\nJ5uEXlH+Lk1EuolCPUTU1jt4+JX1bNhVRmpyLFefOYbsgV07k6KIBB6FeggoOljLAy+tY19JDcMG\nJHHdueNJjo/2d1ki4gcK9SBXUFrDH/65iuo6B7PHZ/DtUwyREeH+LktE/EShHsQKSmu489+rqa5z\ncME3hnPqtEx/lyQifqZQD1Lrd5by1Fubqahp5KTjBnHK1MH+LklEAoBCPci43G5e/mQnb32eD8Dp\nM4dwzgnD/FyViAQKhXoQcTjd3POfNdg95fRNiuH7Z4xh5ODe/i5LRAKIQj1IHKys55HXNrBtbwXD\nByVz/bnjdf65iHyFQj0IbNx1kEde3UB1nYMJw1O5+swxxETrfqEi8lUK9QDmdnt4+dMdvLU0Hw+w\ncPZQzpg5RJf7i0irFOoBqrKmkYdeWc+WPeUk9IrimrPGkDOkj7/LEpEAp1APQLuLqrj/xXWUVtaT\nk5XClaePJiUxxt9liUgQUKgHEI/Hwyv/28bTb2zE5fZw8pTBXHDicMI13CIiHaRQDxBOl5t7X1jL\nxl1lxERHcO3po5k4Ms3fZYlIkFGoBwCH081fnl/D5t3lpKX04sbzxuvORCLytSjU/ayiuoE//msV\nRWV15GSl8IvvTaOxrtHfZYlIkFKo+4nH4+HDlXv59wdbAchKT+S6c8eRnBBDsUJdRL4mhbofOF1u\nnnhjI8s3HSAmOoI5uQM48/ghxEbr1yEiR0cp0s0qqhv483/WsLe4hoGp8fzfBbn0SYr1d1kiEiIU\n6t2o+fj51Jx+fOeUUcTF6lcgIseOEqWbVNU28sd/r24K9KvPHKPL/UXkmFOod4Paegd/+vdqig7W\nMnfCAC49xSjQRaRL6GaWXay23smd/17N/pIaJo5I5eL5I3WFqIh0GfXUu9CGXQd5+q1NlFY2MCqz\nN9csHKubQotIl1KodwGX281ri3fx+me7AJg7cSCXzB9BRLgCXUS6VruhbowJAx4CcoF64Epr7Y4j\nbPcoUGqt/fkxrzKIlFU18PjrG9i8u5yk+GiuOmM0ozVlroh0k4701BcCMdbamcaYacA9vmVNjDFX\nA2OB/x37EoPH5xsLeeYdS0Oji9zsvlz+zRyS4qP9XZaI9CAdGQ+YBbwDYK1dBkxuvtIYMwOYAjx6\nzKsLIss3FfHYaxtxOt1cPG8E1503XoEuIt2uI6GeBFQ0e+w0xoQDGGP6A7cCPwJ67CkdS9cX8thr\nG4kID+PG83OZN3mwznAREb/oyPBLJZDY7HG4tdbt+/l8oC/wFpAB9DLGbLbW/r21naWkxBEZ2fpN\nk9PSEltdF2hcLjd/e2MDr32yg6jIcG759mSmjc046v0GUxt0FbWB2gDUBtD5NuhIqC8BTgf+a4yZ\nDqw7tMJaez9wP4Ax5ruAaSvQAcrKaltdl5aWSHFxVQdK8r+6BiePvraBvO2lpCbH8sOzx5GVnnDU\n9QdTG3QVtYHaANQGcHgbdDTcOxLqLwPzjTFLfI8vN8ZcBMRba5/4OoUGu+o6B39+bg35RVUMG5DE\njefnktAryt9liYi0H+rWWg9wTYvFW46w3TPHqqhAVlvv4Na/LaesqoHRQ1K44bzxRLUxnCQi0p10\n8VEnNDhc3PXcGsqqGpg8qh8/OHMM4eE6ICoigUOXOHaQw+m9sUV+YRXDByZzxWk5CnQRCTjqqXfA\ngfI6HnllPbsKqxiakciPL5xATJSGXEQk8CjU21FZ08gf/rGSippGcrP78v0zxijQRSRgKdTb8cSb\nG6moaWT+5MFceNJwzYMuIgFNod6KQzMtrt9xkOiocM6bO0yBLiIBT6F+BBU1jdz7/Fryi6pITojW\naYsiEjQU6i00NLr4y3/WsPtANTlZKfzgrDEkxmliLhEJDgr1ZhxOF3c+u5rdB6qZMDyV684dpyEX\nEQkqCnWf6joHf/7PGvILq8jJSuGahWMU6CISdBTqQH5hFY+8up6isjrGDevLD88eqzF0EQlKPT7U\n8wuruOOfK3E43Uwbnc6Vp+foXqIiErR6dKg3NLq4/ekvALjwpBGcPGWwnysSETk6PbZL6vF4+Md7\nFoBJI9MU6CISEnpsqL/1eT6frS8kvU8cV5ye4+9yRESOiR4Z6qu3FvPSJzuIjAjjx9/KJTa6R49C\niUgI6XFptmZrCQ+8uA4PcP2540lN7uXvkkREjpkeFeplVQ088cZGwsPD+MFZYxg7rK+/SxIROaZ6\nzPBLbb2Tu59bTW2DkzNnDWWS6efvkkREjrkeEeoej4en395EQWktE4anctqMLH+XJCLSJUJ++MXh\ndPO3tzaxwhaTmZ7AtWePJVyX/4tIiArpnrrb4+Ff729h2cYiMtMTuOG8XCIjQvoti0gPF9I99Zc/\n2cEna/eT3ieOWy4+jl4xIf12RURCt6e+fmcpby7NJyUxhp9dqkAXkZ4hJEO9vtHJY69tBODyb44i\nSTe5EJEeIuRC3ePx8M/3tlBd52DB9EzGDtW56CLSc4RcqB+a0yUzPYEzZw71dzkiIt0qpEK9rKqB\n15bsIjY6guvPHU9MtG50ISI9S8iEutvt4eFX1+Nwujlr1lD6JMX6uyQRkW4XMqH+93c3s21vBWZw\nb+ZrbnQR6aFCItTXbivhk7UF9O8Txw/PGacrRkWkxwr6UHc4XU13MLritBwSekX5uSIREf8J+lB/\n7qNtHKyPfqzmAAAPXklEQVRsYMaYdLIHJvu7HBERvwrqUN9VWMmiVftI6BXFhSeN8Hc5IiJ+F7Sh\n7vZ4eP6jbQB868ThJOqqURGR9if0MsaEAQ8BuUA9cKW1dkez9RcBNwAOYJ219touqvUwq7eUsHl3\nOWZwb44fl9EdLykiEvA60lNfCMRYa2cCPwPuObTCGBML/AaYY62dDfQ2xpzeJZU24/F4eGPpLgAu\nmqdhFxGRQzoS6rOAdwCstcuAyc3WNQAzrbUNvseReHvzXWrt9lLyC6sYO7QPmemJXf1yIiJBoyOh\nngRUNHvsNMaEA1hrPdbaYgBjzHVAvLX2g2Nf5pc8Hg8vLPKOpZ81W3O7iIg015FJxiuB5t3hcGut\n+9AD35j7ncAI4Jz2dpaSEkdkZOtzsqSltd3zXrx2HwWltUwb05/puYPae7mg1F4b9ARqA7UBqA2g\n823QkVBfApwO/NcYMx1Y12L9Y0CdtXZhR16wrKy21XVpaYkUF1e1ut7j8fCyr5c+b9LANrcNVu21\nQU+gNlAbgNoADm+DjoZ7R0L9ZWC+MWaJ7/HlvjNe4oGVwOXAp8aYRYAHuM9a+2ona++QNdtK2LKn\nnFGZvckeoAuNRERaajfUrbUe4JoWi7d0Zh/Hyvtf7AHgvLnDu+slRUSCStBcfLRqSzGbd5czclAy\nwwYk+bscEZGAFBSh7nS5m854OfuEYX6uRkQkcAVFqK/aUkxRWR2zxmdgMlP8XY6ISMAKilD/NK8A\ngJOOC81TGEVEjpWAD/W6Bid2dzmD+yWQ1V/nrIqItCXgQ/39FXtwutxMMmn+LkVEJOAFdKjX1jt4\nd/luesVEMm+S7jsqItKegA71Fz/ZQV2Di/mTBxEX222nw4uIBK2ADvXVW4oBOGmSDpCKiHREwIZ6\ndZ2D8upG4mMjdVcjEZEOCthQ/2LzAQBO1GmMIiIdFrChvqfIOzPZyMzefq5ERCR4BGSouz0e1u88\nSFRkOCMHaTZGEZGOCshQ37GvkpKKenKz+xLVxg01RETkcAEZ6iusdzx98qh+fq5ERCS4BFyouz0e\nFucVEBMVQW52qr/LEREJKgEX6gUlNdQ2OJkwIpWYaA29iIh0RsCF+pa9FQAM1eRdIiKdFnChvtI3\nnj5isE5lFBHprIAL9V0F3vPTB/dL8HMlIiLBJ6BCvdHhorbBSUKvKCIjAqo0EZGgEFDJWVJRB8BI\nDb2IiHwtARXq+4trABiQGu/nSkREglNAhXp+QSUAmRpPFxH5WgLqzhPvLcsHoG9yrJ8rEZGu8NRT\nj/PZZ4uJjIzk+utvIidnzGHrN2xYz3333U1kZCRz5szmW9/6btO6vXv38Itf/IRnnnnuiPtubGzk\n/PPP4MILv81FF10KQGFhAbfe+nMeffSppu1eeeVFysoOcvnl36eqqooHH7yXvXv34HK5SE/vz09+\n8jPi4w/vWLZX9759e7nrrjtwOp1ER0dz2213kJSUxP33/4W8vDVERETwwx/ewLhxuRQVFfLb3/4a\ngKSkZG699XfExMR8/UZtIWB66h6Ph8qaRgAy09VTFwk1W7ZsZs2a1Tz++DPcdtvvueeeP31lm7vv\nvoPbb7+Dhx56gry8PLZu3QLAu+++xW23/YLy8vJW9//xxx8xb94pvP3264ctDwsLa/U5t932C44/\nfjYPPPAYDz/8JKNHj+Guu/7Q6brvvPP3XHXVtTzwwGOcdda57Nmzm23btrJhwzoef/wZfvnL27n3\n3rsBeP75fzNv3sk88MBjDBkylDfeeLX1RvsaAqanXlbVQHWdg1GZvYkID5jPGpGv7fmPtjXdF6Cz\nIiLCcLk8X1k+ZVQ/LjhxeKvPe/vtN/jkk4+pra2lsrKcyy67kjlzTuTjjz/kpZdewOVyERYWxh13\n3MX27dt4+OH7iY6O5swzzyY6OvqI2/zjH08THR3FgQMHOOusc1i16gu2b9/GeeddyMKF5/Loow+y\nZs1KXC43c+eeyMUXf4f333+H+vp6zjhjYVNteXlrmDp1GgDp6f1xudxUVJSTnOw9MaK2tgaHw0lG\nxgAAZs2axYoVyxkxYiRJSUk8+OBjXHDBwq++aZ833niFG264mbKyMpYuXcyMGbPabOPCwkLKykqZ\nPXtu07Lzz7+IurpaAB566K+ceOI81q/Pa7PuhoYGysoOsnjxJzz88P3k5Izmmmuup7KygpiYWBob\nG6muriYqKgqA4cNHUlx8oOk9p6f3b7POzgqY9CyragBgYJp66SJHo6Ghnvvue4h77nmA++//C263\nmz17dnPXXffx4IOPk5U1hGXLlgLgcDTywAOPcfLJC9izZ88RtykpOcAdd9zNj3/8U/7+97/x61//\njrvuuo/XXnsJgA8+eI9bb72DBx98nIQE75Xg8+efeligA9TU1Bw2rBEXF0d1dXWL9V+eJBEfH09N\njXf9jBmziIlpfVh279491NfXk509nG9+8wxefPH5NtsoLCyMkpLipg+Q5svj4rw1XHvt9YwaNbrd\nuquqKtm5cwdTp07n/vsfpbKykrfffoOIiEjCwuCSS87jppt+1DQk1K9fOi+++Dzf/vYFfP75Ur7x\njXlt1tpZAdNTL6/2Dr3w1c6JSFC64MThbfaq25KWlkhxcdXXeu6ECccBkJLSh8TERMrLy0hJSeH3\nv7+N2NhY9uzJZ+zY8QBkZmY1PS8lpfcRtxk6NJvw8HASExMYOHAQERERJCYm0djo/Z/99a9/w8MP\n/5WysoNMnz6z1bri4+Opra1telxbW0NiYuJh62tqapoe19TUkJDQsU7e66+/Qn19HTfffD1ut4cN\nG/LYt28vcXFxNDY2HLZtXV0d0dEx9O/fnwMHig5b53Q6+eijDzj55FM7XHdiYhLx8fFN7T5z5my+\n+GIZtbU19O2byr33PkRNTTXXXHMFo0eP5cEH7+OXv7ydKVOmsXTpYn73u19z5533duh9dkTA9NT3\nFns/+UYPSfFzJSLBzdpNABw8WEptbS0xMTE8+eRj3H77Hfz0p78iOvrLg3JhYd4IqKmpbmObL8ek\nPZ7De11Op5NFiz7g9tvv4K9/fYS33nqdoqLCI9Y1btwEli9fisfjobCwEI/HQ1LSlzfBiYuLJzo6\niv379+HxeFi8eDG5uRNb7OWrvT6n08mHH77HQw89yd13/5V77rmfSy+9jJdeeoGUlD7U1taxa9dO\nAFwuF198sYycnNGkpqbRu3cKixf/r2lfzz//LEuW/O+w/bdXd0xMDIMHZ5KXtwaAtWtXMWxYNklJ\nyfTqFQdAbGwvoqNjqK+vJykpqekbSd++qVRVfb0P79YETE99X4n3E1rTA4gcndLSUm644Vpqa6u5\n+eafEh+fwPjxuVx11WVERkaQmJhMSUkx/ftnND2nI9vAVw86RkZGkpSUzFVXXUZsbCzTps0gPb3/\nEcfUjRlFbu5Err76cjweDz/+8U8BWLVqBXl5a7jssiv58Y9/xu23/xK3283cuSd85SwT+OpBzyVL\nPmXUqJzDevULFpzB5ZdfzFVXXcvPf/5r/vCH3xAeHo7T6WT27DlMnDgJgF/96jf8+c9/5Lnn/oXD\n4WDgwEHccssvgS/H1EeNGt1u3bfc8ivuuedPuN1uMjIGcM011xMeHk5e3hquueZ7uN0eTj55AYMH\nZ3LjjT/hL3+5E5fLBcBNN93SmV9vu8JafvJ2teLiqiO+4O//voKdhVU8evOcHn2g9Gi+docKtcHX\nb4O3336D3bvzufrqH3ZBVd1LfweHt0FaWmLrp/E0EzDpWVHTSJ/EmB4d6CIiRytghl+q6xwMSNXQ\ni8jRWLDgdH+XIH4WEN1ij8dDo8NNbIzudCQicjQCItQbnW7cHg8xUQp1EZGj0e7wizEmDHgIyAXq\ngSuttTuarT8D+BXgAJ6y1j7R2SIaGr1HgeNiozr7VBERaaYjPfWFQIy1dibwM+CeQyuMMZG+x/OA\nucBVxpi0zhbhcLoBiIoKiC8OIiJBqyMpOgt4B8BauwyY3GxdDrDVWltprXUAi4ETOltEfaMTgOhI\nDb+IiByNjoR6ElDR7LHTGBPeyroqIJlOivDduq68qqGdLUVEpC0dOaWxEkhs9jjcWututi6p2bpE\noPW5MYGUlDgiW/TIU1MTuPHCieSOSCO1d68OlBTa0tIS298oxKkN1AagNoDOt0FHQn0JcDrwX2PM\ndGBds3WbgOHGmN5ALd6hl7va2llZWe0Rl48fkkJq7166gkxX0akNUBuA2gC+ckVph57T7jQBzc5+\nGe9bdDkwCYi31j5hjDkNuBXvpAxPWmsf+VrVi4jIUev2uV9ERKTr6BxCEZEQolAXEQkhCnURkRCi\nUBcRCSEKdRGRENLt86l3xwRhga4DbXARcAPeNlhnrb3WL4V2ofbaoNl2jwKl1tqfd3OJXa4DfwdT\ngD/7HhYCl1prG7u90C7UgTa4BLgJcOLNg5A9ZdoYMw34o7X2Gy2WdyoT/dFT7/IJwoJAW20QC/wG\nmGOtnQ30NsaE4p0PWm2DQ4wxVwNju7uwbtReGzwGXGatPQHv/EtZ3Vxfd2ivDe4CTsQ7B9WPjTGd\nnoYkGBhjfgI8DsS0WN7pTPRHqHf5BGFBoK02aABmWmsPTYQTibcHE2raagOMMTOAKcCj3V9at2m1\nDYwxI4FS4CZjzMdAH2vtVn8U2cXa/DsA1gIpwKH5Q0L1wpptwNlHWN7pTPRHqHf5BGFBoNU2sNZ6\nrLXFAMaY6/BeufuBH2rsaq22gTGmP96rlH/EkW4fHzra+l9IBWYAf8XbS5tnjJnbveV1i7baAGAD\nsBLv9CRvWGsru7O47mKtfRnvEFNLnc5Ef4T6MZ0gLEi11QYYY8KMMXcBJwHndHdx3aStNjgf6Au8\nBfwUuNgY851urq87tNUGpcA2a+0Wa60Tb2+2ZS82FLTaBsaYccBpeIedhgDpxphzu71C/+p0Jvoj\n1JcA3wRoa4IwY0w03q8ZS7u/xC7XVhuAdyw1xlq7sNkwTKhptQ2stfdba6dYa08E/gj821r7d/+U\n2aXa+jvYASQYY4b5Hs/G22sNNW21QQXeiQIbrLUe4ADeoZhQ1vKbaaczsdvnftEEYW23Ad6vml8A\nn/rWeYD7rLWvdnedXam9v4Nm230XMCF+9ktr/wtzgT/51n1mrf2/7q+ya3WgDa4Gvof3WNN24Pu+\nby4hxxiTBTxrrZ3pOwPua2WiJvQSEQkhuvhIRCSEKNRFREKIQl1EJIQo1EVEQohCXUQkhCjURURC\niEJdRCSEKNRFRELI/wfODxpAAHGo1gAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x3ff0add8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "Xv = np.array(validation[cols])\n",
    "Yv = np.array(validation[['label']])\n",
    "print 'ROC of Validation set'\n",
    "fvroc, vAUC = plot_ROC([newnet],[0.01],Xv,Yv, bias = True, title=\"ROC curve on validation set\")\n",
    "fvroc.savefig(os.getcwd()+'\\\\figures\\\\'+'valROC'+netobj['name']+'.pdf')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Test performance (best configuration)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "ROC of Test set\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXUAAAD9CAYAAABDaefJAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x3ba8bb70>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "Xv = np.array(test[cols])\n",
    "Yv = np.array(test[['label']])\n",
    "print 'ROC of Test set'\n",
    "fvroc, vAUC = plot_ROC([newnet],[0.01],Xv,Yv, bias = True)\n",
    "fvroc.savefig(os.getcwd()+'\\\\figures\\\\'+'valROC'+netobj['name']+'.pdf')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Results + Sensitivity analysis"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We ran multiple simulations of network structure on external server, and here is the comparison:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def from_folder(path):\n",
    "    dfs = [pd.read_csv(\"{}/{}\".format(path, filename)) for filename in os.listdir(path)]\n",
    "    df_ret = pd.concat(dfs)\n",
    "    return df_ret"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "res = from_folder(os.getcwd()+'\\\\results')\n",
    "res = res[res.columns[1:]]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def add_layers(row):\n",
    "    '''decdode layers structure, converts the proportions to numbers'''\n",
    "    hidden = [int(row[4])]\n",
    "    for l in row.index[5:8]:\n",
    "        if row[l]>0:\n",
    "            hidden.append(int(hidden[-1]*row[l]))\n",
    "    num = len(hidden)\n",
    "    hidden = hidden + (4-len(hidden))*[0]\n",
    "    row['hl2']=hidden[1]\n",
    "    row['hl3']=hidden[2]\n",
    "    row['hl4']=hidden[3]\n",
    "    row['nhl'] = num\n",
    "    row['hl12'] =str(int(row[4]))+' : '+str(hidden[1])\n",
    "    row['totalnodes'] = round(int(row[4])+sum(hidden[1:]),-1)\n",
    "    return row"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {
    "collapsed": true,
    "scrolled": true
   },
   "outputs": [],
   "source": [
    "results = res.apply(add_layers, axis = 1)\n",
    "res = results[[u'alpha', u'batch', u'trainsize', u'epochs', u'trainerror', u'TrainAUC', u'ValAUC',u'hl1', u'hl2',\n",
    "       u'hl3', u'hl4','nhl','hl12','totalnodes']]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\22334\\AppData\\Local\\Continuum\\Anaconda2\\lib\\site-packages\\ipykernel_launcher.py:1: SettingWithCopyWarning: \n",
      "A value is trying to be set on a copy of a slice from a DataFrame.\n",
      "Try using .loc[row_indexer,col_indexer] = value instead\n",
      "\n",
      "See the caveats in the documentation: http://pandas.pydata.org/pandas-docs/stable/indexing.html#indexing-view-versus-copy\n",
      "  \"\"\"Entry point for launching an IPython kernel.\n"
     ]
    }
   ],
   "source": [
    "res['trainsize'] = res.trainsize.apply(lambda x:min(x,284641) )  # fixing max train size"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# By ALL"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>TrainAUC</th>\n",
       "      <th>ValAUC</th>\n",
       "      <th>trainerror</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>trainsize</th>\n",
       "      <th>epochs</th>\n",
       "      <th>alpha</th>\n",
       "      <th>batch</th>\n",
       "      <th>nhl</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">2000.0</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">100000.0</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">0.010</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">100.0</th>\n",
       "      <th>1</th>\n",
       "      <td>0.937295</td>\n",
       "      <td>0.580701</td>\n",
       "      <td>0.184723</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.967596</td>\n",
       "      <td>0.588763</td>\n",
       "      <td>0.059081</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.970596</td>\n",
       "      <td>0.573794</td>\n",
       "      <td>0.040359</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"15\" valign=\"top\">10000.0</th>\n",
       "      <th rowspan=\"12\" valign=\"top\">100000.0</th>\n",
       "      <th rowspan=\"6\" valign=\"top\">0.001</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">500.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.711320</td>\n",
       "      <td>0.662602</td>\n",
       "      <td>0.423715</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.717361</td>\n",
       "      <td>0.658452</td>\n",
       "      <td>0.419068</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.711247</td>\n",
       "      <td>0.659917</td>\n",
       "      <td>0.423183</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">2000.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.790554</td>\n",
       "      <td>0.617330</td>\n",
       "      <td>0.365937</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.802894</td>\n",
       "      <td>0.616669</td>\n",
       "      <td>0.348957</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.797318</td>\n",
       "      <td>0.620328</td>\n",
       "      <td>0.351209</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"6\" valign=\"top\">0.010</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">500.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.858271</td>\n",
       "      <td>0.601985</td>\n",
       "      <td>0.279050</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.883176</td>\n",
       "      <td>0.589948</td>\n",
       "      <td>0.245486</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.887137</td>\n",
       "      <td>0.583062</td>\n",
       "      <td>0.258031</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">2000.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.894404</td>\n",
       "      <td>0.588991</td>\n",
       "      <td>0.203615</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.912621</td>\n",
       "      <td>0.584070</td>\n",
       "      <td>0.189963</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.905081</td>\n",
       "      <td>0.581458</td>\n",
       "      <td>0.209198</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">200000.0</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">0.010</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">50.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.725684</td>\n",
       "      <td>0.653059</td>\n",
       "      <td>0.414735</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.729642</td>\n",
       "      <td>0.649255</td>\n",
       "      <td>0.408992</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.718951</td>\n",
       "      <td>0.651138</td>\n",
       "      <td>0.412695</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"34\" valign=\"top\">284641.0</th>\n",
       "      <th rowspan=\"24\" valign=\"top\">50000.0</th>\n",
       "      <th rowspan=\"6\" valign=\"top\">0.010</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">50.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.678835</td>\n",
       "      <td>0.676337</td>\n",
       "      <td>0.441563</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.675915</td>\n",
       "      <td>0.673522</td>\n",
       "      <td>0.441969</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.654088</td>\n",
       "      <td>0.652069</td>\n",
       "      <td>0.449293</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">200.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.683707</td>\n",
       "      <td>0.681004</td>\n",
       "      <td>0.437802</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.682757</td>\n",
       "      <td>0.680202</td>\n",
       "      <td>0.438671</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.662264</td>\n",
       "      <td>0.659978</td>\n",
       "      <td>0.445064</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"9\" valign=\"top\">0.100</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">50.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.684642</td>\n",
       "      <td>0.682060</td>\n",
       "      <td>0.435413</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.682864</td>\n",
       "      <td>0.680464</td>\n",
       "      <td>0.437941</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.661111</td>\n",
       "      <td>0.658867</td>\n",
       "      <td>0.444284</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">200.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.689572</td>\n",
       "      <td>0.685964</td>\n",
       "      <td>0.433572</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.688925</td>\n",
       "      <td>0.685491</td>\n",
       "      <td>0.435300</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.666406</td>\n",
       "      <td>0.663468</td>\n",
       "      <td>0.440214</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">2000.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.519607</td>\n",
       "      <td>0.519422</td>\n",
       "      <td>0.487840</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.499999</td>\n",
       "      <td>0.499990</td>\n",
       "      <td>0.481783</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.499998</td>\n",
       "      <td>0.500009</td>\n",
       "      <td>0.455657</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"9\" valign=\"top\">1.000</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">50.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.661676</td>\n",
       "      <td>0.659024</td>\n",
       "      <td>0.442277</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.670741</td>\n",
       "      <td>0.669190</td>\n",
       "      <td>0.430938</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.645736</td>\n",
       "      <td>0.645785</td>\n",
       "      <td>0.436684</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">200.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.505856</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.512222</td>\n",
       "      <td>0.512245</td>\n",
       "      <td>0.454890</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.512277</td>\n",
       "      <td>0.525013</td>\n",
       "      <td>0.461141</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">2000.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.516213</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.495046</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.484051</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">100000.0</th>\n",
       "      <th>0.001</th>\n",
       "      <th>400.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.682380</td>\n",
       "      <td>0.679462</td>\n",
       "      <td>0.440003</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">0.010</th>\n",
       "      <th>400.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.690592</td>\n",
       "      <td>0.686327</td>\n",
       "      <td>0.433375</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>500.0</th>\n",
       "      <th>3</th>\n",
       "      <td>0.692772</td>\n",
       "      <td>0.687490</td>\n",
       "      <td>0.436187</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"6\" valign=\"top\">500000.0</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">0.001</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">500.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.692606</td>\n",
       "      <td>0.688206</td>\n",
       "      <td>0.432709</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.693053</td>\n",
       "      <td>0.688576</td>\n",
       "      <td>0.432228</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.692560</td>\n",
       "      <td>0.688390</td>\n",
       "      <td>0.432267</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">0.010</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">500.0</th>\n",
       "      <th>2</th>\n",
       "      <td>0.697578</td>\n",
       "      <td>0.687489</td>\n",
       "      <td>0.429204</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.698201</td>\n",
       "      <td>0.686889</td>\n",
       "      <td>0.428699</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.698381</td>\n",
       "      <td>0.687215</td>\n",
       "      <td>0.429035</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10000000.0</th>\n",
       "      <th>0.010</th>\n",
       "      <th>1000.0</th>\n",
       "      <th>4</th>\n",
       "      <td>0.716502</td>\n",
       "      <td>0.664801</td>\n",
       "      <td>0.415642</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                                       TrainAUC    ValAUC  trainerror\n",
       "trainsize epochs     alpha batch  nhl                                \n",
       "2000.0    100000.0   0.010 100.0  1    0.937295  0.580701    0.184723\n",
       "                                  2    0.967596  0.588763    0.059081\n",
       "                                  3    0.970596  0.573794    0.040359\n",
       "10000.0   100000.0   0.001 500.0  2    0.711320  0.662602    0.423715\n",
       "                                  3    0.717361  0.658452    0.419068\n",
       "                                  4    0.711247  0.659917    0.423183\n",
       "                           2000.0 2    0.790554  0.617330    0.365937\n",
       "                                  3    0.802894  0.616669    0.348957\n",
       "                                  4    0.797318  0.620328    0.351209\n",
       "                     0.010 500.0  2    0.858271  0.601985    0.279050\n",
       "                                  3    0.883176  0.589948    0.245486\n",
       "                                  4    0.887137  0.583062    0.258031\n",
       "                           2000.0 2    0.894404  0.588991    0.203615\n",
       "                                  3    0.912621  0.584070    0.189963\n",
       "                                  4    0.905081  0.581458    0.209198\n",
       "          200000.0   0.010 50.0   2    0.725684  0.653059    0.414735\n",
       "                                  3    0.729642  0.649255    0.408992\n",
       "                                  4    0.718951  0.651138    0.412695\n",
       "284641.0  50000.0    0.010 50.0   2    0.678835  0.676337    0.441563\n",
       "                                  3    0.675915  0.673522    0.441969\n",
       "                                  4    0.654088  0.652069    0.449293\n",
       "                           200.0  2    0.683707  0.681004    0.437802\n",
       "                                  3    0.682757  0.680202    0.438671\n",
       "                                  4    0.662264  0.659978    0.445064\n",
       "                     0.100 50.0   2    0.684642  0.682060    0.435413\n",
       "                                  3    0.682864  0.680464    0.437941\n",
       "                                  4    0.661111  0.658867    0.444284\n",
       "                           200.0  2    0.689572  0.685964    0.433572\n",
       "                                  3    0.688925  0.685491    0.435300\n",
       "                                  4    0.666406  0.663468    0.440214\n",
       "                           2000.0 2    0.519607  0.519422    0.487840\n",
       "                                  3    0.499999  0.499990    0.481783\n",
       "                                  4    0.499998  0.500009    0.455657\n",
       "                     1.000 50.0   2    0.661676  0.659024    0.442277\n",
       "                                  3    0.670741  0.669190    0.430938\n",
       "                                  4    0.645736  0.645785    0.436684\n",
       "                           200.0  2    0.500000  0.500000    0.505856\n",
       "                                  3    0.512222  0.512245    0.454890\n",
       "                                  4    0.512277  0.525013    0.461141\n",
       "                           2000.0 2    0.500000  0.500000    0.516213\n",
       "                                  3    0.500000  0.500000    0.495046\n",
       "                                  4    0.500000  0.500000    0.484051\n",
       "          100000.0   0.001 400.0  2    0.682380  0.679462    0.440003\n",
       "                     0.010 400.0  2    0.690592  0.686327    0.433375\n",
       "                           500.0  3    0.692772  0.687490    0.436187\n",
       "          500000.0   0.001 500.0  2    0.692606  0.688206    0.432709\n",
       "                                  3    0.693053  0.688576    0.432228\n",
       "                                  4    0.692560  0.688390    0.432267\n",
       "                     0.010 500.0  2    0.697578  0.687489    0.429204\n",
       "                                  3    0.698201  0.686889    0.428699\n",
       "                                  4    0.698381  0.687215    0.429035\n",
       "          10000000.0 0.010 1000.0 4    0.716502  0.664801    0.415642"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "res.groupby(by=['trainsize','epochs','alpha','batch','nhl'])[['TrainAUC','ValAUC','trainerror']].mean()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# By Train Size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style>\n",
       "    .dataframe thead tr:only-child th {\n",
       "        text-align: right;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: left;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>TrainAUC</th>\n",
       "      <th>ValAUC</th>\n",
       "      <th>trainerror</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>trainsize</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>2000.0</th>\n",
       "      <td>0.965562</td>\n",
       "      <td>0.581214</td>\n",
       "      <td>0.064721</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10000.0</th>\n",
       "      <td>0.813300</td>\n",
       "      <td>0.617517</td>\n",
       "      <td>0.319089</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>284641.0</th>\n",
       "      <td>0.624072</td>\n",
       "      <td>0.621859</td>\n",
       "      <td>0.450983</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           TrainAUC    ValAUC  trainerror\n",
       "trainsize                                \n",
       "2000.0     0.965562  0.581214    0.064721\n",
       "10000.0    0.813300  0.617517    0.319089\n",
       "284641.0   0.624072  0.621859    0.450983"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bysize = res.groupby(by=['trainsize'])[['TrainAUC','ValAUC','trainerror']].mean()\n",
    "bysize"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXIAAAEMCAYAAADZDD24AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n4yTIhRAix0mQCyFEjvv/eHi3508/KdEAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1ea122e8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure()\n",
    "plt.plot(list(bysize.index),bysize['TrainAUC'])\n",
    "plt.plot(list(bysize.index),bysize['ValAUC'])\n",
    "plt.plot(list(bysize.index),bysize['trainerror'])\n",
    "plt.legend(['TrainAUC','ValAUC','trainerror'])\n",
    "plt.xscale('log')\n",
    "t = plt.title('AUC and error as a function of Train size')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As train size increases, train error gets bigger, since the nnet has hard times to fit to the data.  \n",
    "On the other hand, big train size makes the validation error slightly decrease."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# By training epochs"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>TrainAUC</th>\n",
       "      <th>ValAUC</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>epochs</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>50000</th>\n",
       "      <td>0.610836</td>\n",
       "      <td>0.609725</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>100000</th>\n",
       "      <td>0.687184</td>\n",
       "      <td>0.683405</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>500000</th>\n",
       "      <td>0.695694</td>\n",
       "      <td>0.687719</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10000000</th>\n",
       "      <td>0.716502</td>\n",
       "      <td>0.664801</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          TrainAUC    ValAUC\n",
       "epochs                      \n",
       "50000     0.610836  0.609725\n",
       "100000    0.687184  0.683405\n",
       "500000    0.695694  0.687719\n",
       "10000000  0.716502  0.664801"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "byep = res[res.trainsize>10000].groupby(by=['epochs'])[['TrainAUC','ValAUC']].mean()\n",
    "byep"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([ 4.69897,  5.     ,  5.69897,  7.     ])"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.log10(list(byep.index))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYgAAAEZCAYAAACNebLAAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0xc716f28>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure()\n",
    "plt.plot(np.log10(list(byep.index)),byep['TrainAUC'])\n",
    "plt.plot(np.log10(list(byep.index)),byep['ValAUC'])\n",
    "# plt.plot(np.log10(list(byep.index)),byep['trainerror'])\n",
    "plt.ylim([0,1.0])\n",
    "# plt.xlim([42000, 550050])\n",
    "# plt.xscale('log')\n",
    "# plt.xticks(np.log10(list(byep.index)))\n",
    "plt.legend(['TrainAUC','ValAUC','trainerror'])\n",
    "plt.xlabel('alpha exponent (base 10)')\n",
    "plt.ylabel('AUC')\n",
    "t = plt.title('AUC as a function of number of epochs')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "As epochs increase, the nnet performance improves, however we dont see major improvement when epochs>10000  \n",
    "All the analysis from now on are made on the whole training set, hence we won't see overfitting."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# By alpha"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>TrainAUC</th>\n",
       "      <th>ValAUC</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>alpha</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0.001</th>\n",
       "      <td>0.691078</td>\n",
       "      <td>0.686941</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.010</th>\n",
       "      <td>0.681293</td>\n",
       "      <td>0.676275</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>0.100</th>\n",
       "      <td>0.622076</td>\n",
       "      <td>0.620140</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1.000</th>\n",
       "      <td>0.557134</td>\n",
       "      <td>0.557724</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       TrainAUC    ValAUC\n",
       "alpha                    \n",
       "0.001  0.691078  0.686941\n",
       "0.010  0.681293  0.676275\n",
       "0.100  0.622076  0.620140\n",
       "1.000  0.557134  0.557724"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "bya = res[res.trainsize>10000].groupby(by=['alpha'])[['TrainAUC','ValAUC']].mean()\n",
    "bya"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYkAAAEeCAYAAAB/vulGAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0xc54ee48>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure()\n",
    "plt.plot(list(bya.index),bya['TrainAUC'])\n",
    "plt.plot(list(bya.index),bya['ValAUC'])\n",
    "# plt.plot(list(bya.index),bya['trainerror'])\n",
    "plt.ylim([0,1.0])\n",
    "plt.xscale('log')\n",
    "#plt.yscale('log')\n",
    "plt.xticks(list(bya.index))\n",
    "plt.xlabel('Alpha (log10)')\n",
    "plt.ylabel('AUC')\n",
    "plt.legend(['TrainAUC','ValAUC','Train error'])\n",
    "t = plt.title('AUC as a function of learning rate')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Small alpha, around 0.001  (best was found at 0.01), indicates good AUC for both train and validation sets.  \n",
    "Big alpha don't give good results. prevents convergence"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# By Batch size"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>TrainAUC</th>\n",
       "      <th>ValAUC</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>batch</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>50</th>\n",
       "      <td>0.670427</td>\n",
       "      <td>0.668375</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>200</th>\n",
       "      <td>0.623503</td>\n",
       "      <td>0.622609</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>400</th>\n",
       "      <td>0.686486</td>\n",
       "      <td>0.682895</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>500</th>\n",
       "      <td>0.695629</td>\n",
       "      <td>0.687714</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1000</th>\n",
       "      <td>0.716502</td>\n",
       "      <td>0.664801</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2000</th>\n",
       "      <td>0.502450</td>\n",
       "      <td>0.502426</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       TrainAUC    ValAUC\n",
       "batch                    \n",
       "50     0.670427  0.668375\n",
       "200    0.623503  0.622609\n",
       "400    0.686486  0.682895\n",
       "500    0.695629  0.687714\n",
       "1000   0.716502  0.664801\n",
       "2000   0.502450  0.502426"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "byb = res[res.trainsize>10000].groupby(by=['batch'])[['TrainAUC','ValAUC']].mean()\n",
    "byb"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYwAAAEZCAYAAACEkhK6AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nJBIFhoiIRKLAEBGRSP4/+3QudUJxmokAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0xb243400>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure()\n",
    "plt.plot(list(byb.index),byb['TrainAUC'])\n",
    "plt.plot(list(byb.index),byb['ValAUC'])\n",
    "# plt.plot(list(byb.index),byb['trainerror'])\n",
    "plt.ylim([0,1.0])\n",
    "#plt.xscale('log')\n",
    "#plt.yscale('log')\n",
    "plt.xticks(list(byb.index))\n",
    "plt.xlabel('Batch size')\n",
    "plt.ylabel('AUC')\n",
    "plt.legend(['TrainAUC','ValAUC','Train error'])\n",
    "t = plt.title('AUC as a function of batch size')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can see a maximum AUC on 400-500 batch size. Bigger apparently says too much noise, prevents convergence (?)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# By Network stucture"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {
    "scrolled": false
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>TrainAUC</th>\n",
       "      <th>ValAUC</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>nhl</th>\n",
       "      <th>hl1</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"7\" valign=\"top\">2</th>\n",
       "      <th>20</th>\n",
       "      <td>0.652937</td>\n",
       "      <td>0.650692</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25</th>\n",
       "      <td>0.529411</td>\n",
       "      <td>0.529133</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>30</th>\n",
       "      <td>0.659618</td>\n",
       "      <td>0.656995</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>32</th>\n",
       "      <td>0.694724</td>\n",
       "      <td>0.687858</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>40</th>\n",
       "      <td>0.650604</td>\n",
       "      <td>0.648015</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50</th>\n",
       "      <td>0.597906</td>\n",
       "      <td>0.593890</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>100</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"7\" valign=\"top\">3</th>\n",
       "      <th>20</th>\n",
       "      <td>0.655243</td>\n",
       "      <td>0.653327</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25</th>\n",
       "      <td>0.499998</td>\n",
       "      <td>0.499985</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>30</th>\n",
       "      <td>0.650816</td>\n",
       "      <td>0.648774</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>32</th>\n",
       "      <td>0.695258</td>\n",
       "      <td>0.688119</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>40</th>\n",
       "      <td>0.650652</td>\n",
       "      <td>0.648455</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50</th>\n",
       "      <td>0.601965</td>\n",
       "      <td>0.597389</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>100</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"7\" valign=\"top\">4</th>\n",
       "      <th>20</th>\n",
       "      <td>0.598798</td>\n",
       "      <td>0.597592</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500015</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>30</th>\n",
       "      <td>0.653609</td>\n",
       "      <td>0.651932</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>32</th>\n",
       "      <td>0.695020</td>\n",
       "      <td>0.688097</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>40</th>\n",
       "      <td>0.648534</td>\n",
       "      <td>0.653066</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50</th>\n",
       "      <td>0.607267</td>\n",
       "      <td>0.599196</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>100</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         TrainAUC    ValAUC\n",
       "nhl hl1                    \n",
       "2   20   0.652937  0.650692\n",
       "    25   0.529411  0.529133\n",
       "    30   0.659618  0.656995\n",
       "    32   0.694724  0.687858\n",
       "    40   0.650604  0.648015\n",
       "    50   0.597906  0.593890\n",
       "    100  0.500000  0.500000\n",
       "3   20   0.655243  0.653327\n",
       "    25   0.499998  0.499985\n",
       "    30   0.650816  0.648774\n",
       "    32   0.695258  0.688119\n",
       "    40   0.650652  0.648455\n",
       "    50   0.601965  0.597389\n",
       "    100  0.500000  0.500000\n",
       "4   20   0.598798  0.597592\n",
       "    25   0.500000  0.500015\n",
       "    30   0.653609  0.651932\n",
       "    32   0.695020  0.688097\n",
       "    40   0.648534  0.653066\n",
       "    50   0.607267  0.599196\n",
       "    100  0.500000  0.500000"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "res[res.trainsize>10000].groupby(by=['nhl','hl1'])[['TrainAUC','ValAUC']].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 74,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>TrainAUC</th>\n",
       "      <th>ValAUC</th>\n",
       "      <th>trainerror</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>hl1</th>\n",
       "      <th>hl2</th>\n",
       "      <th>hl3</th>\n",
       "      <th>hl4</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"11\" valign=\"top\">20</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">4</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.652195</td>\n",
       "      <td>0.649863</td>\n",
       "      <td>0.453868</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <th>0</th>\n",
       "      <td>0.665983</td>\n",
       "      <td>0.663885</td>\n",
       "      <td>0.440015</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <th>0</th>\n",
       "      <td>0.574531</td>\n",
       "      <td>0.573650</td>\n",
       "      <td>0.462162</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"4\" valign=\"top\">10</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.653151</td>\n",
       "      <td>0.650859</td>\n",
       "      <td>0.452418</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <th>0</th>\n",
       "      <td>0.648865</td>\n",
       "      <td>0.646818</td>\n",
       "      <td>0.437195</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">7</th>\n",
       "      <th>0</th>\n",
       "      <td>0.668435</td>\n",
       "      <td>0.666682</td>\n",
       "      <td>0.439296</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.648524</td>\n",
       "      <td>0.646619</td>\n",
       "      <td>0.439498</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"4\" valign=\"top\">16</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.653465</td>\n",
       "      <td>0.651353</td>\n",
       "      <td>0.454695</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <th>0</th>\n",
       "      <td>0.649218</td>\n",
       "      <td>0.647546</td>\n",
       "      <td>0.439183</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">11</th>\n",
       "      <th>0</th>\n",
       "      <td>0.649896</td>\n",
       "      <td>0.647734</td>\n",
       "      <td>0.437225</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.647870</td>\n",
       "      <td>0.646158</td>\n",
       "      <td>0.439441</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"11\" valign=\"top\">25</th>\n",
       "      <th rowspan=\"3\" valign=\"top\">5</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.588232</td>\n",
       "      <td>0.587399</td>\n",
       "      <td>0.469568</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500002</td>\n",
       "      <td>0.500041</td>\n",
       "      <td>0.455623</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500007</td>\n",
       "      <td>0.452017</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"4\" valign=\"top\">12</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.548130</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.451875</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">8</th>\n",
       "      <th>0</th>\n",
       "      <td>0.499986</td>\n",
       "      <td>0.499855</td>\n",
       "      <td>0.471298</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500046</td>\n",
       "      <td>0.458558</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"4\" valign=\"top\">20</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.453025</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500465</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">14</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.502230</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.451145</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"8\" valign=\"top\">30</th>\n",
       "      <th rowspan=\"4\" valign=\"top\">6</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.653465</td>\n",
       "      <td>0.651267</td>\n",
       "      <td>0.435531</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <th>0</th>\n",
       "      <td>0.647296</td>\n",
       "      <td>0.645819</td>\n",
       "      <td>0.439021</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">4</th>\n",
       "      <th>0</th>\n",
       "      <td>0.649614</td>\n",
       "      <td>0.647572</td>\n",
       "      <td>0.440972</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.663836</td>\n",
       "      <td>0.662532</td>\n",
       "      <td>0.441051</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"4\" valign=\"top\">15</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.661770</td>\n",
       "      <td>0.658937</td>\n",
       "      <td>0.437858</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <th>0</th>\n",
       "      <td>0.652380</td>\n",
       "      <td>0.649943</td>\n",
       "      <td>0.438358</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">10</th>\n",
       "      <th>0</th>\n",
       "      <td>0.650750</td>\n",
       "      <td>0.648647</td>\n",
       "      <td>0.451609</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>0.647858</td>\n",
       "      <td>0.645935</td>\n",
       "      <td>0.438829</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <th>...</th>\n",
       "      <th>...</th>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>40</th>\n",
       "      <th>32</th>\n",
       "      <th>22</th>\n",
       "      <th>6</th>\n",
       "      <td>0.649420</td>\n",
       "      <td>0.647606</td>\n",
       "      <td>0.454468</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"17\" valign=\"top\">50</th>\n",
       "      <th rowspan=\"4\" valign=\"top\">10</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.450230</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.502145</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">7</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.502835</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>0.499992</td>\n",
       "      <td>0.499995</td>\n",
       "      <td>0.462300</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"6\" valign=\"top\">25</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.597799</td>\n",
       "      <td>0.594001</td>\n",
       "      <td>0.464253</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.501390</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">12</th>\n",
       "      <th>0</th>\n",
       "      <td>0.696872</td>\n",
       "      <td>0.687238</td>\n",
       "      <td>0.429241</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>0.695561</td>\n",
       "      <td>0.687695</td>\n",
       "      <td>0.430623</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">17</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500390</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500110</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"3\" valign=\"top\">35</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.696829</td>\n",
       "      <td>0.687513</td>\n",
       "      <td>0.430103</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">17</th>\n",
       "      <th>0</th>\n",
       "      <td>0.696927</td>\n",
       "      <td>0.686738</td>\n",
       "      <td>0.430175</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>0.697992</td>\n",
       "      <td>0.686933</td>\n",
       "      <td>0.429380</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"4\" valign=\"top\">40</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.501435</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>12</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.450690</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">28</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.549350</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.501530</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"12\" valign=\"top\">100</th>\n",
       "      <th rowspan=\"4\" valign=\"top\">20</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.499295</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.498540</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">14</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500655</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.499420</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"4\" valign=\"top\">50</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.549655</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>15</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.450830</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">35</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.548615</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>10</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.453260</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"4\" valign=\"top\">80</th>\n",
       "      <th>0</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.548845</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>24</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.452325</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th rowspan=\"2\" valign=\"top\">56</th>\n",
       "      <th>0</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500330</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.500000</td>\n",
       "      <td>0.450200</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>83 rows × 3 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                 TrainAUC    ValAUC  trainerror\n",
       "hl1 hl2 hl3 hl4                                \n",
       "20  4   0   0    0.652195  0.649863    0.453868\n",
       "        1   0    0.665983  0.663885    0.440015\n",
       "        2   0    0.574531  0.573650    0.462162\n",
       "    10  0   0    0.653151  0.650859    0.452418\n",
       "        3   0    0.648865  0.646818    0.437195\n",
       "        7   0    0.668435  0.666682    0.439296\n",
       "            2    0.648524  0.646619    0.439498\n",
       "    16  0   0    0.653465  0.651353    0.454695\n",
       "        4   0    0.649218  0.647546    0.439183\n",
       "        11  0    0.649896  0.647734    0.437225\n",
       "            3    0.647870  0.646158    0.439441\n",
       "25  5   0   0    0.588232  0.587399    0.469568\n",
       "        1   0    0.500002  0.500041    0.455623\n",
       "        3   0    0.500000  0.500007    0.452017\n",
       "    12  0   0    0.500000  0.500000    0.548130\n",
       "        3   0    0.500000  0.500000    0.451875\n",
       "        8   0    0.499986  0.499855    0.471298\n",
       "            2    0.500000  0.500046    0.458558\n",
       "    20  0   0    0.500000  0.500000    0.453025\n",
       "        6   0    0.500000  0.500000    0.500465\n",
       "        14  0    0.500000  0.500000    0.502230\n",
       "            4    0.500000  0.500000    0.451145\n",
       "30  6   0   0    0.653465  0.651267    0.435531\n",
       "        1   0    0.647296  0.645819    0.439021\n",
       "        4   0    0.649614  0.647572    0.440972\n",
       "            1    0.663836  0.662532    0.441051\n",
       "    15  0   0    0.661770  0.658937    0.437858\n",
       "        4   0    0.652380  0.649943    0.438358\n",
       "        10  0    0.650750  0.648647    0.451609\n",
       "            3    0.647858  0.645935    0.438829\n",
       "...                   ...       ...         ...\n",
       "40  32  22  6    0.649420  0.647606    0.454468\n",
       "50  10  0   0    0.500000  0.500000    0.450230\n",
       "        3   0    0.500000  0.500000    0.502145\n",
       "        7   0    0.500000  0.500000    0.502835\n",
       "            2    0.499992  0.499995    0.462300\n",
       "    25  0   0    0.597799  0.594001    0.464253\n",
       "        7   0    0.500000  0.500000    0.501390\n",
       "        12  0    0.696872  0.687238    0.429241\n",
       "            6    0.695561  0.687695    0.430623\n",
       "        17  0    0.500000  0.500000    0.500390\n",
       "            5    0.500000  0.500000    0.500110\n",
       "    35  0   0    0.696829  0.687513    0.430103\n",
       "        17  0    0.696927  0.686738    0.430175\n",
       "            8    0.697992  0.686933    0.429380\n",
       "    40  0   0    0.500000  0.500000    0.501435\n",
       "        12  0    0.500000  0.500000    0.450690\n",
       "        28  0    0.500000  0.500000    0.549350\n",
       "            8    0.500000  0.500000    0.501530\n",
       "100 20  0   0    0.500000  0.500000    0.499295\n",
       "        6   0    0.500000  0.500000    0.498540\n",
       "        14  0    0.500000  0.500000    0.500655\n",
       "            4    0.500000  0.500000    0.499420\n",
       "    50  0   0    0.500000  0.500000    0.549655\n",
       "        15  0    0.500000  0.500000    0.450830\n",
       "        35  0    0.500000  0.500000    0.548615\n",
       "            10   0.500000  0.500000    0.453260\n",
       "    80  0   0    0.500000  0.500000    0.548845\n",
       "        24  0    0.500000  0.500000    0.452325\n",
       "        56  0    0.500000  0.500000    0.500330\n",
       "            16   0.500000  0.500000    0.450200\n",
       "\n",
       "[83 rows x 3 columns]"
      ]
     },
     "execution_count": 74,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "res[res.trainsize>10000].groupby(by=['hl1','hl2','hl3','hl4'])[['TrainAUC','ValAUC','trainerror']].mean()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 76,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABI4AAAJsCAYAAACbN+gqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\niiMAAAAASv0/wd37z6WeqikAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x5a993748>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(20,10))\n",
    "sns.set_style(\"whitegrid\")\n",
    "ax = sns.barplot(x=\"hl12\", y=\"ValAUC\", hue = \"nhl\", data=res)#, errwidth=0.0)\n",
    "ax.legend(loc = 2,fontsize = 'xx-large')\n",
    "plt.ylim([0.0,1.0])\n",
    "tit = plt.title('Validation AUC as a function of Number of Hidden Layers and 1st two layer size', fontsize = 'xx-large')\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Best validation AUC was achieved using 2 layers, with proportion of ~50:35, 32:22, 32:16"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 78,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAABI4AAAJsCAYAAACbN+gqAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzs3X98zfX///H72WaM/fRrfkVatiPMjzG/FaNUvEUMCZV4\nI4kUSc23H36k/IgpPyoU+fXOr4r6hPQLkUJK8T4JRX6EMcbYXt8/9t7J2V7bzuzH2Tm7XS8XF5fX\nj/M6j9frvM7r7NzP8/l8WQzDMAQAAAAAAABk4OXqAgAAAAAAAFA0ERwBAAAAAADAFMERAAAAAAAA\nTBEcAQAAAAAAwBTBEQAAAAAAAEwRHAEAAAAAAMAUwRGAfDdgwADVr19fiYmJWa4ze/ZsWa1W/fe/\n/3V6u71791b37t3t07fffrueeOKJbB+zbds2Wa1WffHFF04/jyT9+OOP6tWrl306JSVFVqtVU6ZM\nydV28ssDDzwgq9WqZcuWmS5funSprFarDh06ZLr8q6++ktVq1ddff51p2ebNmzVkyBC1adNG9evX\n11133aUXX3xRx44dy9d9KEwvv/yymjZtqoYNG2r+/PmF/vxffvmlBg8ebJ/+7bffZLVatXz58kKv\nJSfOHKvY2Fjddttt+u6770yXx8XFKTIysiDLdPDaa6/JarUqOTm50J4zN/bu3asuXbqoXr16atWq\nla5evZppnezek2bLnT2HnD02LVu21KhRo5zco7wr7OfzRM68thmvPc46deqUnnjiCe3duzdXjyvK\n17aMivo5OGfOHLVq1UqRkZF66aWXXFZHUT9OAFzDx9UFAPA8PXv21NatW7Vhwwb16NHDdJ01a9ao\ncePGuvXWW53ersVicZieO3euSpcunevHOWPNmjXat2+ffdrb21srVqxQxYoVc72tvPrtt9/0/fff\nq3bt2lqyZIlDoJXOYrHkuJ9my8eOHasPP/xQnTt3VlxcnAIDA3Xw4EG9/fbbWr9+vRYuXCir1Zpv\n+1IYtm/frsWLF+uBBx5Qp06dVLVq1UKv4d1333UI3qpVq6YVK1aoWrVqhV5Ldpw9VhaLRYZh6Jln\nntG6desyve+cOf/yU2E/X25NmTJFp0+f1uzZsxUQEKASJUqYrpeb96yz55Czx6awj19Rfr3chTOv\nbcZrj7N27typTz/9VAMGDLjR8oq8onwOHj16VDNmzNCdd96p/v37u+RvjXRF+TgBcB1aHAHId+3a\ntVO5cuW0Zs0a0+Xbt2/X0aNH1bNnzzw9j9VqVfXq1XNczzCMXG/b7DGRkZGqVKlSrreVVytWrFBo\naKhGjBihgwcPaseOHfmy3Xnz5mnNmjWaMWOGJk2apPbt2ys6Olp9+vTRsmXL5O3trdGjR+fLcxWm\nM2fOyGKxqEuXLmrUqJFCQ0NdXZJ8fX0VGRmpsmXLuroUB7k5VgEBATp27JgmTJhQiBW6pzNnzqh2\n7dpq06aNGjZsmC/bLKrnEDyDYRgEBi505swZSVLHjh0VFRWlm266ycUVAYAjgiMA+c7Hx0fdunXT\n999/r6NHj2Za/sEHHyg4OFh33XWXfd53332nwYMHq1mzZqpbt65at26tuLi4bLu7tWnTxqGr2tWr\nVzVt2jTdcccdatCggQYMGKDff/890+Nyeq6nn35a77//vq5du6batWtrzpw5pl3Vrl69qrfeekud\nOnVS/fr1dccdd2jy5Mm6dOmSfZ0ZM2bo9ttv17fffqsePXooMjJSrVu31tSpU5WamprjsUxOTtba\ntWsVExOj1q1bq2LFilqyZEmOj8vJ1atX9c4776ht27Zq3759puUVK1bUs88+q44dO+rixYtZbufs\n2bN66aWX1L59e9WrV09RUVHq37+/du/ebV/HMAxNmzZNd955p33/n332Wf3999/Z1ujMtjN68skn\n9eSTT0pKa/lWv359SWldrTK21Dp9+rSsVqu9e1Z6l4tPPvlEo0ePVnR0tBo2bKjBgwfryJEjDo/d\nt2+fBg0apOjoaEVHR2vAgAH6+eef7c/79ddf67ffflPt2rW1fv160+4ciYmJevXVV3XXXXcpMjJS\nd911l/1cu35/+vbtq/Xr16tz586qV6+e2rdvr4ULF2Z77JzZflbHKiu33HKL+vTpow8++ECbN2/O\ndt2sutR07NhRAwcOtE+3atVKs2bN0rRp09SqVSs1aNBAjzzyiI4dO6aNGzeqc+fOatCggf16ktGX\nX36pzp07KzIyUp07d9batWszrbN27Vp17dpVkZGRatGiheLi4pSQkGBfnt4lbMWKFerQoYMaNmyo\nlStXZrlv27dvV79+/RQdHa3GjRtr8ODB2r9/v6R/zqFDhw7p66+/Vu3atfOtq6TZOXTixAmNHDlS\nTZs2VXR0tOLi4pSUlJTpsZs2bdL9999v74q6bt060+dYuHCh7rnnHtWrV0+33367pkyZoitXrtiX\np3eJ3b9/v/r166cGDRqoefPmGj9+vC5fvpwv+7lx40b169dPjRs3Vt26dRUTE6OpU6fq2rVrkqRX\nXnlFdevW1dmzZx0e980338hqtdq7UyYmJurll1/W7bffrnr16unee+/N9Lqmv78mTpyoJk2aqFWr\nVjp79qz27dun/v37q2nTpmrQoIG6d++ujz/+OM+1S2nn/Ouvv66ZM2fqjjvuUGRkpLp165apy6Kz\nr+31zK49UtrnyLx58+yfVW3bttUrr7xi396yZcvsXZNiY2Pt79HU1FS99dZbuu+++9SgQQPVr19f\nXbp00erVq3M8FtebOnWq2rVrp23btql79+6KjIxUmzZtNH36dIcfacy6SO3Zs0dWq9W+L+nv1+3b\nt2vgwIFq2LChWrRooenTp+vKlSuaPHmymjdvrujoaD311FOZ/oZISUnRq6++qubNm6thw4YaOnRo\npi7eeTl3zOR0/KdOnaqePXvKYrHoySefVO3atbP8fHT2/HHm80XKv2vDjX7OA3AfdFUDUCB69Oih\n+fPna+3atRo2bJh9fmJioj777DP17t1bvr6+kqRffvlFDz30kNq1a6dXX31VJUqU0LZt2zR37lyV\nLFlS48aNM32OjL+Ojh49Wps2bdKwYcNUt25dff7555owYYLDes481/Dhw5WcnKxNmzbp/fffz7KV\n0ZAhQ7Rz504NHjxYDRo00P79+zVr1iz98MMPWrJkiXx80i6x586d05gxYzRw4EDVqlVLn3zyiebP\nn69KlSqpT58+2R7Hzz77TOfOndP9998vLy8vde3aVW+//bZOnDiRp5Y0u3fv1rlz5xQTE5PlOvfe\ne2+22zAMQwMGDFBCQoJGjBihypUr6/Dhw5o1a5aGDx+uTZs2qUSJEoqPj9e7776r0aNHKzw8XEeO\nHNErr7yiP//8U4sWLcrTtjMaOXKkatWqpZkzZ2rixIkKDw+XlLum9+PHj9c999yj119/XcePH9ek\nSZP09NNP27+w//LLL3rggQcUERGhl19+WaVKldKbb76p/v37a/Xq1Zo4caLGjRunU6dOafr06apR\no0amP54vX76s3r176+TJk3r88ccVFham7du3Kz4+Xvv379frr79uX/eXX37RjBkz9Nhjjyk0NFTv\nv/++Jk+erLCwMLVu3dp0H5zZfsZjVatWrRyPzdNPP62vv/5azz//vBo0aJBl65esutSYzVu0aJGi\noqL0yiuv6NixY3rxxRf1yCOPSJKGDx+u0qVLa+LEiRoxYoQ+//xzeXt7S0o7R8aOHathw4YpLCxM\nq1ev1pgxY5SamqquXbtKSmtVN23aNMXGxuqpp57SsWPH9Prrr2vv3r1asWKF/RokpYW848aNU4kS\nJbIM0dasWaNnnnlG7du315QpU3T58mXNnTtXvXv31uLFixUeHq4VK1ZoxIgR9vC1cuXK2R7T1NTU\nTF/m0udn59KlS+rdu7dSU1P17LPPKigoSO+9956+/fZbh/XSr4nt27fXiBEj9Ndff2ny5Mk6f/68\nw3rjx4/XBx98oIcfflgtWrTQgQMH9Prrr+vAgQN66623JP3z+g0dOlQPPPCAhg0bpm+//VazZ89W\nYGBgnsdF2bhxo4YNG6bevXtryJAhSk1N1fr16zV//nyFhobqwQcfVI8ePbRgwQKtXbtWDz30kP2x\n//nPf1SzZk01btxYycnJ6tu3r44dO6bhw4erZs2a2rJli55//nmdOXNG//73v+2P++GHH+Tl5aVZ\ns2bp77//lsVi0YABA1S3bl299tpr8vHx0fLly/XUU0+pYsWKatKkyQ3Xnm7JkiWqXbu2xo8fL0ma\nNm2aHn/8cX3xxRcKDAx0+rXNyOzaI0mDBw/W999/r8GDB6t+/fr66aefNHv2bO3evVuLFy/WnXfe\nqbNnz9qvBemt5CZPnqzly5dr5MiRqlOnjhISEvTOO+/o2WefVZ06dezXV2ecOXNGY8eO1b///W+F\nhYXp448/1ty5c1W5cmV7qJ/Vddps/siRI/XQQw/p0Ucf1apVqzR37lxt3LhRVqtVU6dO1a+//qop\nU6aoYsWKDi1nP/vsM1mtVr388su6cuWKpk2bpgcffFAffvihypYtm6dzJyQkxLT+nI7/gw8+qBo1\naui5557TyJEj1bx5cwUHB2d5LHM6f5z9fMnPa8ONfM4DcDMGABSQhx9+2IiJiXGYt3TpUsNqtRq/\n/fabfd7q1auNRx55xEhJSXFYt2fPnkanTp3s07179zbuv/9++3SbNm2M4cOHG4ZhGAcOHDAiIiKM\nuXPnOmxj7NixhtVqNbZs2ZKr53rxxReNOnXq2KevXbtmREREGK+88ophGIaxZcsWIyIiwli2bJnD\ndj799FMjIiLCWLlypWEYhjF9+nTDarUan332mcN6d9xxh/HQQw9lOmYZ9e/f3+jatat9+siRI4bV\najVmzJjhsJ7Zcb3el19+aURERBhfffWVYRiGsXbtWsNqtRo7duzIsYasHDt2zOjfv7+xbds2h/lL\nliwxrFarsX//fsMwDKNv377Gv/71L4d1Nm7caMTHx+d522Y+/vhjw2q1Gnv27LHPi42NNXr27Omw\n3qlTp4yIiAhj3rx5hmEYhs1mMyIiIozHH3/cYb2pU6caVqvVOHXqlGEYhvHYY48ZzZo1My5dumRf\n59y5c0ZMTIzxwQcfGIZhGAMGDDDuvvtu+/L0baefL4sWLTKsVquxdetWh+eaN2+ew+sycuRIw2q1\nGr/++qt9nUuXLhl16tQx4uLisjwGzm7f7FiZuf747d2716hTp44xdOhQ+/K4uDgjMjLSPv3aa68Z\nVqvVuHLlisN2OnbsaDz66KP26ZYtWxotW7Y0kpOT7fOGDh1qWK1W44cffrDPW7x4sWG1Wo1Dhw45\nbH/t2rUO2+/Ro4fRtm1bwzDSXpPIyEhjzJgxDuv88ssvhtVqNd577z3DMP55b2R8T2V07do1o3nz\n5sYDDzzgMP/ChQtGixYtjL59+2a5n2bSnze7f1ar1f6ezXgOLVy40LBarcaPP/7oUGP79u0djn2X\nLl0crm2GYRjbtm0zIiIijCeffNIwDMM4ePCgERERYcycOdNhvc8//9yIiIgwNm/ebBhG2nUmIiLC\nfuzS9ezZ0+jYsWO2+9uyZUv782Vl7ty5puu0aNHCeOyxx+zTffr0MTp37myfPnfunFGvXj3j7bff\nNgzjn+vE9u3bHbYzefJkIzIy0jhz5oxhGP+8v44cOWJfZ8eOHUZERITx6aef2uddvXrVmDx5srFr\n1648155+zl+7ds0+76uvvjKsVqvx8ccfG4bh/GtrJuO1Z+PGjQ6fSek+/vhjIyIiwli9erV9OuO1\nYMSIEcZbb73l8Lj//ve/RkREhLFo0SLDMDKfl2bS36/p51G61q1bGwMGDHA4NhmP4e7du42IiAj7\nsUl/30ycONG+TkJCghEREeGw34ZhGPfff7/Ddb9ly5ZGdHS0ceHCBfu8gwcPGrVr1zZef/11wzDy\ndu6Ycfb4Z9zPrDhz/jh7/c/Pa8ONfM4DcC90VQNQYGJjY/Xnn3863Ilp1apVatKkiWrWrGmfd999\n9+ntt9/WtWvX9N///lebN2/W3Llzdfz4cafvmrRjxw5ZLBZ16NDBYX6nTp0cmsLnx3NJaXdrs1gs\n6ty5s8P8O++8U35+ftq+fbvD/EaNGjlMh4aGZtsFTJKOHDmib7/9VnfffbcuXLigCxcuKCgoSI0a\nNdLKlSsduj84Oxhu+nrprTbMWjo4q3Llylq4cKGaNWum48ePa/v27Vq+fLk2btwoSfbj2aJFC/36\n66/q2rWr4uPjtXv3brVr106PPfZYnrddEDK+VuktztJfr507d6ply5by8/OzrxMUFKSNGzeqW7du\nTj3H9u3bVa5cOTVv3txh/n333SfDMBzOnzJlyjj8su/n56fAwMBsz5/cbD+36tWrp8GDB2vTpk1a\ntWrVDW8nXZ06dRxaj5UvX15eXl6qV6+efV76L/nX/xJusVh09913O2zrzjvv1PHjx/XHH3/ou+++\nU3Jystq3b6+UlBT7v7CwMFWvXl1ffvmlw2Nr166dbZ0HDhzQmTNnMr3n/f39FRMTo++//97hPems\n559/Xh988EGmf88991y2j9u5c6dCQ0NVt25d+zxvb2+HLsCJiYn65ZdfMnVHbdasmcqVK2ef/uab\nb2SxWBQTE+NwrFq0aKGSJUs6HCuLxWL6HsnpeuaMQYMGaerUqbp8+bJ++eUX/d///Z9mzpypy5cv\nO7znY2NjdfDgQfsNDNatWyfDMHTfffdJkrZu3aqAgAA1btzYYX/atWunK1euOJz/fn5+DuPJ1K5d\nW8HBwRozZozGjBmjtWvX6vz58xozZkym/b6R2qW091D6NVhKu94ZhuFwjcnptXXW9u3bTT+r7rnn\nHpUsWTLba8H06dM1YMAAnTt3Trt379a6deu0aNEiWSyWG7oG38jnYFauHzssMDBQvr6+DtcMSQoO\nDs7UeqZ169by9/e3T996662qWbOmdu7cKSntvXCj546ZvBz/rOR0/jhz/c/va8ONfM4DcC90VQNQ\nYNq3b6+yZcva76B28OBB7d27V9OnT3dYLzk5WS+//LLWrl2r5ORkVa5cWXXr1pWfn5/TA1ufO3dO\nkjJ1nalQoUK+P5ckJSQkyM/Pz/SubhUqVNCFCxcc5l0fMkiSl5dXjl8yV6xYYR83YOrUqfb56eHP\nJ598ok6dOjlsP6s/5tPnp9dbtWpVGYahP/74I8vnv3Tpkq5evaqgoKAs11m7dq29S5e/v7/Cw8MV\nEBDgsM7gwYMVHBysVatW6c0331R8fLwqVKiggQMHql+/fnnadkEwe62ktK5DqampSkhIUPny5fP0\nHAkJCZnOTSktNLFYLA5fdkqVKpVpPS8vr2zP19xs/0YMGTJEW7Zs0cSJE9W0adM8bev6L3HpfHx8\nHL4YmTG7W1n6F57z58/r7NmzMgxDjz/+eKZjZbFYHF5ni8WS4x0a068xZse1YsWKSklJ0aVLlxQY\nGJjtdjLWUaNGDdWpUyfTsvTBcrOrx6yr4PX1pY+5ktN6586dk2EYpsGnxWLRyZMnHeaZvUecGbMt\nJ2fPntX48eO1adMmSWl3kqtfv758fX0dXsOOHTtqwoQJWrVqlerWratVq1apbdu29v08e/asLly4\nYHpcLRaLTpw4YZ/O+Lr7+/tr+fLlmjNnjrZs2aJ169bJYrGoVatWiouLy/Kuds7Wbvac6df09GPo\nzGvrrISEBAUEBKhkyZKZlpUrVy7ba8GPP/6ol156SXv37lXJkiV1yy23qHbt2jIM44ZuOmF23tzI\ndiwWi+l1I+P2zZgdw7Jly9rHPTt37twNnztm8nL8s5LT+ePM9T+/rg3px+NGP+cBuA+CIwAFxsfH\nR127dtWyZcsUFxenVatWqVy5cplaBb344otav369Jk+erDZt2qhMmTKSpP79++uvv/5y6rnS//g5\nffq0Q9CR8ctXfjyXlPZLZlJSki5dupTpj7hTp05l+8u0M65du6Y1a9aoZcuWGjx4cKbljz/+uBYv\nXmwPjipUqCDDMHTy5ElFRERkWj9939L/IIyMjFRwcLC2bNmi7t27m9bw/vvva+rUqXrvvffUuHHj\nTMu3b9+uZ555Rr1799agQYPsLXM++OADffHFFw7r9urVS7169dKFCxe0fft2LVq0SJMmTVKdOnUU\nFRWVp207w2KxZGpddSO/dHt5eal06dKmA37u3LlTZcuWVVhYWI7bCQoK0uHDhzPNP3XqlAzDyPOd\ns3LaflZjcTjL29tbU6ZMUbdu3fTMM884tCCUMn+RSZcfrVKu35aR4U5Qp06dkpT2hSw9wJk0aZLp\n+E1mgVx20sccSX+O6504cULe3t65Co3yqmzZsqav8fXXvPTX+fTp06brpZ+rgYGBslgsevvtt02D\n4sIIbKW0Ma0OHz6s+fPnKyoqyv5lu23btg7r+fr6qkuXLvroo4/Uq1cv7d+/3z7Qe3q9VapU0axZ\ns0yDiZzujlmjRg1NmjRJUtoYY+ktU59//nktWLAgT7U7w5nX1llBQUG6cOGCkpOTHcb0ktLOi6yu\nBefPn9ejjz6qWrVqad26dbr11lvl5eWlo0eP5ktLw6zkx3U6O9cPjJ/u9OnT9nMir+dORjd6/PPC\nmc+Xgrg25PZzHoB7oasagAIVGxurS5cuacuWLdqwYYO6detmHzQ63c6dO9WwYUPdfffd9iDn77//\n1r59+5z+FbtFixYyDEMfffSRw/zPPvvM4Yuls8+V3sokK02bNpVhGJnuQPLpp58qKSlJ0dHRTtWd\nlU2bNun06dPq3bu3mjRpkulfp06dtGfPHvvdnBo2bKhSpUpleeefDRs2qFq1avZfy728vPTII49o\n06ZN9u5f1zt+/LgWLlyoGjVqmIZGkrRr1y5J0rBhwxz+mE4PdtKPZ//+/fXUU09JSvsjs0OHDnr6\n6adlGIaOHTuWp207q0yZMg6/FEtp3RtvRHR0tLZu3epwh6Pz589r0KBB9rsN5dRapmnTpvr777+1\nbds2h/lr1qyRxWLJcgBeZ+W0/byen1LaXdZGjRqlnTt3asOGDQ7L0lsDHD9+3D7v8OHDmVqu5EVK\nSoq++uorh3nr16/XzTffrNDQUEVFRcnHx0d//vmn6tSpY/9Xo0YNTZs2TVu3bs3V84WHhyskJCTT\nez4xMVGbNm3K8n1SUFq0aKHTp0/bu9hIaQOGX3/HO39/f9WvX1+ffPKJw3tm7969Dq9F+vXs5MmT\nDscqKChIr776qr1LWEH7/vvvFRMTY+8GI0m//vqrjh8/nuk937NnT505c0YTJkxQ5cqVHQaKb9as\nmU6cOKGAgACH/Tl27JhmzJiRbQDz+eefq1mzZrLZbJIkq9WqoUOHqkmTJller3Jbe06ceW2zkvHa\nk/7aZrzj4Pr165WcnGy/FmT8zDtw4IASEhLUp08fhYeH25enX4NvpKVQTvz9/TP9gJPeDT2/bNu2\nTVevXrVP//TTT/r999/VokULSXk7d8w4e/zzkzOfL/l9bbiRz3kA7oUWRwAKVPXq1RUdHa3p06fr\n5MmTio2NzbROgwYNtGHDBr333nuyWq36/fff9dZbbykpKSnH2w+nq1Gjhh544AHNnz9fhmEoOjpa\n27Zty/TLqLPPFRQUpJSUFG3YsEGRkZGZfmW844471Lx5c02aNEl///23GjVqpP3792v27NmqW7eu\nvSXQjVqxYoWCgoJ0++23my7v3r27Fi9erMWLF2vChAkqVaqURowYoVdeeUVXrlxRp06dFBwcrJMn\nT2r16tXas2ePZs+e7bCNRx55RD/88INGjBihrl27qm3btvLz89O+ffu0aNEiXbt2LdNjrpd+56mX\nXnpJsbGxunz5slavXm0PotKPZ7NmzTRr1ixVrlxZLVu2VGJioubPn6+goCC1atUqT9vOSsYvNe3a\ntdPWrVs1fvx43Xvvvdq/f7/eeecd0+4DORk2bJj69Omjhx9+WI8++qi8vLw0f/58lSlTRn379pWU\n9ivtiRMn7Ldkzyg2NtZ+961hw4bp1ltv1bfffqu33npLMTExeQ4hcrP9vHwB7Nu3rzZv3qxt27Y5\nHMu2bdtq2rRpiouL0+DBg3Xu3DnFx8fnqSVVxjq9vb31/PPPa8SIESpfvryWLl2qgwcPKj4+XlJa\nq6MBAwZozpw5unjxolq1aqULFy5o3rx5stlsGj58eJbbNuPt7a2nnnpKzz33nIYOHarY2FglJSVp\n/vz5SkxMdGjxcqP7lJvl3bp10/vvv68nnnhCI0aMUKVKlbRs2TIdPXrUYb2nn37afvepfv366ezZ\ns3r99dcdWlzddttt6tSpk1566SUdO3ZMjRo10smTJzV79mydO3cu09gxN+rQoUOmd1iqUqWKOnTo\noHr16tmvudWqVdPPP/+sefPmycvLK9N7PiwsTI0aNdKOHTsyjaMSGxur5cuX66GHHtLgwYNVo0YN\n/fzzz5o5c6ZuueWWbFsFNmzYUD4+PhoxYoSGDBmi8uXL6/vvv9e2bdsc7qiVUW5qz4mzr62ZjNee\nmJgYRUdHa8KECTp9+rQaNGign3/+WbNnz1ZkZKTuueceSWmfeYZhaNOmTSpVqpRuvfVWlS5dWvPn\nz1fJkiXl5+enr776Su+++64sFosuXbqUq31yRrt27bRgwQLNnDlTzZo1086dO+1h/PXycs06d+6c\nhgwZooceekinT5/W9OnTVb16dfXu3VtS3s4dM84e//zk7PU/P68NN/I5D8C9EBwBKHCxsbEaNWqU\nWrRoYTqQ5Lhx42SxWDRnzhxdunRJVapU0T333KOQkBBNmjRJhw4dsneFuf6Xx4y3/I6Li1OlSpW0\nYsUKLViwQHXr1tXLL7/scIvoZ5991qnn6tKlizZu3KjRo0erZ8+eGjt2bKbnmzt3rmbPnq1Vq1Zp\nzpw5qlixonr37q3HHnvMoUl6bm4xLEnHjh3T9u3b1b17d9Nbzktpv4LfdtttWr9+vcaMGaPAwEA9\n9NBDuummm/T+++8rLi5O58+fV0hIiBo2bKilS5c6DLQqpXUlfOONN/TBBx9ozZo1+vzzz5WYmKgq\nVaro7ru1aw2XAAAgAElEQVTv1qOPPqrQ0FDT55ekVq1a6fnnn9eiRYs0ePBglS1bVpGRkVq6dKn6\n9OmjHTt2qHnz5hoyZIh8fX21evVqLV68WL6+vmrcuLEmTJiQZTN9Z7edlYzHtlevXvYQbc2aNapX\nr55mz56tgQMHZvs4M3Xr1tXixYs1Y8YMjR492r4/EydOtB+vPn36aM+ePRoyZIiefvpptWrVymHb\nfn5+ev/99zV9+nTNmzdPCQkJqlatmp544gkNGDAg1zVlVNDbv97kyZPVuXNnh/G1atWqpSlTpmju\n3LkaMmSIbrrpJg0dOlQbN2506gunWU0Z5wUEBCguLk5TpkzR8ePHdeutt+rNN99UmzZt7OuMHDlS\n1apV05IlS7RkyRKVLl1a9erV0/jx4xUZGZnt85m5//77FRQUpPnz52vEiBEqWbKkmjRpookTJ8pq\ntea4D87sZ3bLr5/29fXVe++9pylTpmj69OlKTk5WTEyMBg0apFmzZtnXa9y4sRYsWKDp06dr5MiR\nCgkJ0dChQ7V06VKHbU+ZMkVvv/22/XoWGBioJk2a6PHHH89xAGBn93f//v32VpLXa968uTp06KDX\nXntNEydO1OTJk5WSkqJq1app0KBB+uuvv7Rs2TJdvHjR3lJUSvtSvmfPnkzjr5QuXVpLly7VjBkz\nFB8fr7Nnz6pChQqKjY3VsGHDHFrXZKw7ODhYCxYs0IwZMzRx4kSdP39eVatW1ciRI/Xoo49muW9T\np07VhAkTnK49u+Pn7GtrJuO1p1+/fpo/f77i4+O1cuVKvfHGGwoNDdWDDz6ooUOH2j9joqKi1LZt\nWy1cuFDffPON/vOf/+jNN9/U1KlTNWrUKPn5+SksLEyzZ89WfHy8Q2uo/DjXpbRQ/uLFi1q6dKkW\nLFigpk2bau7cueratWuO28r4+ZzVuj169FBqaqqefPJJpaam6vbbb9czzzxjbyWZl3Mnq/125vg7\nu73snieds9f//Lw23MjnPAD3YjEKoq0pAAAAUIAefPBB+fv7a86cOa4uBQAAj1bgYxzt2bPH3nT/\neps3b1b37t3Vq1cvrVy5sqDLAAAAgJu7dOmS4uPjNXz4cH3//femNw8AAAD5q0C7qr311ltau3Zt\npqa5165d0+TJk7Vq1SqVLFlSvXv3VkxMTJ7vIgMAAADPVapUKa1cuVLJyckaN26cGjRo4OqSAADw\neAXa4qhGjRqmA6vabDbVqFFD/v7+KlGihKKiohz6SgMAAAAZeXl56YsvvtC2bdvUp08fV5cDAECx\nUKDBUYcOHUxvSZyYmKiAgAD7dJkyZXThwoWCLAUAAAAAAAC55JK7qvn7+ysxMdE+ffHiRQUGBub4\nuF27dhVkWQAAAAAAAMVSVFSU6fxCCY4y3rgtLCxMhw8f1vnz51WqVCnt3Lkz0+2Bs5LVjgAAAAAA\nACD3smuoUyjBkcVikSR99NFHSkpKUo8ePTR27Fg98sgjMgxDPXr0UMWKFQujFAAAAAAAADjJYmRs\nDlSE7dq1ixZHAAAAAAAA+Si7vKVAB8cGAAAAAACA+yI4AgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAA\nAIApgiMAAAAAAACYIjgCAAAAAACAKYIjAAAAAAAAmCI4AgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAA\nAIApH1cXUFhSUlJks9lcXYYkKSwsTN7e3q4uAwAAAAAAIFvFJjiy2Wz697i3VSaogkvruJhwSnMn\nDFB4eLhL6wAAAAAAAMhJsQmOJKlMUAUFlq3s6jLyVWJiorp166YHH3xQ/fr1c3U5AAAAAADAgzDG\nkRs7e/asBg0apKNHj7q6FAAAAAAA4IEIjtzUhg0b1LVrVx06dMjVpQAAAAAAAA9FcOSG/vzzTz31\n1FNq06aN3nnnHRmG4eqSAAAAAACABypWYxx5ipCQEH366aeqVq2a/vzzT1eXAwAAAAAAPBTBkRsq\nXbq0Spcu7eoyAAAAAACAh6OrGgAAAAAAAEwRHAEAAAAAAMAUwREAAAAAAABMERwBAAAAAADAVLEa\nHPtiwilXl1AkagAAAAAAAHBGsQmOwsLCNHfCAFeXISmtlvxksVhksVjydZsAAAAAAADFJjjy9vZW\neHi4q8vId1WrVtX+/ftdXQYAAAAAAPBAjHEEAAAAAAAAUwRHAAAAAAAAMEVwBAAAAAAAAFMERwAA\nAAAAADBFcAQAAAAAAABTBEcAAAAAAAAwRXAEAAAAAAAAUwRHAAAAAAAAMEVwBAAAAAAAAFM+ri6g\nsKSkpMhms7m6DElSWFiYvL29XV0GAAAAAABAtopNcGSz2fTY3NHyLx/o0joST5/X7H9PUXh4uEvr\nAAAAAAAAyEmxCY4kyb98oIIqhbi6jHyxevVqvffeezp8+LDKlCmjhg0batSoUapevbqrSwMAAAAA\nAB6CMY7c0MKFCzV27Fg1atRIs2bN0jPPPKODBw/q/vvv159//unq8gAAAAAAgIcoVi2OPMWbb76p\ne++9V88995x9XqNGjdSuXTstXrxYY8aMcWF1AAAAAADAUxAcuZkLFy7oX//6l9q2beswv1KlSvL3\n99fJkyddVBkAAAAAAPA0BEduJiAgQOPGjcs0f+vWrTp//rysVqsLqgIAAAAAAJ6IMY48wIkTJ/Tc\nc8+pQoUK6tmzp6vLAQAAAAAAHoIWR27u8OHDevTRR3X+/Hm98847CgwMdHVJAAAAAADAQ9DiyI1t\n3bpVsbGxSkpK0oIFCxQZGenqkgAAAAAAgAchOHJTK1eu1KBBg1SxYkUtX75c9erVc3VJAAAAAADA\nwxAcuaE1a9YoLi5OjRs31tKlS1W1alVXlwQAAAAAADxQsRrjKPH0eVeXkOca/v77b73wwgsKCQnR\n4MGDZbPZHJYHBQXp5ptvztNzAAAAAAAASMUoOAoLC9Psf09xdRmS0mq5UVu2bNHly5d1+fJlPfzw\nw5mWx8TEKD4+Pi/lAQAAAAAASJIshmEYri7CWbt27VJUVJSrywAAAAAAAPAY2eUtjHEEAAAAAAAA\nUwRHAAAAAAAAMEVwBAAAAAAAAFMERwAAAAAAADBFcAQAAAAAAABTBEcAAAAAAAAwRXAEAAAAAAAA\nUwRHAAAAAAAAMEVwBAAAAAAAAFM+ri6gsKSkpMhms7m6DElSWFiYvL29XV0GAAAAAABAtopNcGSz\n2bR86DBV8vd3aR1/JSaq5xvxCg8Pz9N2Vq5cqXfffVdHjx5VuXLl1LVrVw0aNEi+vr75VCkAAAAA\nACjuik1wJEmV/P1VNTDI1WXk2cKFCzV58mQNHDhQLVu21MGDBzVt2jQdOXJEU6ZMcXV5AAAAAADA\nQxSr4MgTXLt2TXPmzFHnzp01atQoSVKzZs107tw5vfnmmxozZozKlSvn4ioBAAAAAIAnIDhyM97e\n3lqyZInKlCnjMN/Hx0eGYejq1asuqgwAAAAAAHgagiM3Y7FYFBYWZp++cOGCvvrqK73zzjuKiYlR\npUqVXFgdAAAAAADwJARHbuzHH39Ujx49ZLFYVLVqVY0ePdrVJQEAAAAAAA/i5eoCcOMqVaqk9957\nT9OnT5efn5+6d+8um83m6rIAAAAAAICHIDhyYxUqVFCTJk3UsWNHLViwQJcvX9aiRYtcXRYAAAAA\nAPAQBEdu5ty5c1q7dq3++OMPh/nly5dXaGiojh8/7qLKAAAAAACApyE4cjOpqakaO3as3nnnHYf5\nNptNf/75p2677TYXVQYAAAAAADwNg2O7mbJly6pfv35699135e/vrxYtWujIkSN64403VLVqVT38\n8MOuLhEAAAAAAHiIYhUc/ZWY6OoS8qWGMWPGqHr16lq2bJk9QGrXrp1GjBih4ODgfKgSAAAAAABA\nshiGYbi6CGft2rVLUVFRN/TYlJSUInPHsbCwMHl7e7u6DAAAAAAAgGzzlmLT4sjb21vh4eGuLgMA\nAAAAAMBtMDg2AAAAAAAATBEcAQAAAAAAwBTBEQAAAAAAAEwRHAEAAAAAAMAUwREAAAAAAABMERwB\nAAAAAADAFMERAAAAAAAATBEcAQAAAAAAwBTBEQAAAAAAAEz5uLqAwpKSkiKbzebqMiRJYWFh8vb2\ndnUZAAAAAAAA2So2wZHNZtOk8YsUElTRpXWcTTipsS/0V3h4eL5sLz4+XvHx8dq8ebOqVKmSL9sE\nAAAAAACQilFwJEkhQRVVvpznhCt79+7VnDlzZLFYXF0KAAAAAADwQIxx5KaSkpL09NNPq2JF17ag\nAgAAAAAAnovgyE1NnDhRAQEBevDBB11dCgAAAAAA8FAER25o06ZN+uijjzRlyhQG2QYAAAAAAAWG\n4MjNnD59Ws8//7xGjhypW265xdXlAAAAAAAAD0Zw5GbGjRun8PBw9evXz9WlAAAAAAAAD1es7qrm\n7pYuXarvvvtOa9asUUpKiiQpNTVVkpSSkqLU1FR5eZEFAgAAAACA/EFw5EbWr1+vS5cuqUOHDpmW\ndejQQdHR0Xr33XddUBkAAAAAAPBEBEdu5KWXXtLFixcd5n300UdauHCh5syZoxo1arioMgAAAAAA\n4IkIjtzIzTffnGned999J0mqVauWqlSpUsgVAQAAAAAAT1asgqOzCSddXUKRqAEAAAAAAMAZFsMw\nDFcX4axdu3YpKirqhh6bkpIim82WzxXdmLCwMHl7e7u6DAAAAAAAgGzzlmLT4sjb21vh4eGuLgMA\nAAAAAMBtcO92AAAAAAAAmCI4AgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAAAIApgiMAAAAAAACYIjgC\nAAAAAACAqQINjgzD0Pjx49WrVy/169dPR48edVi+bt06devWTT169NDSpUsLshQAAAAAAADkkk9B\nbnzjxo1KTk7WsmXLtGfPHk2aNElvvPGGffmUKVO0YcMGlSpVSvfee686deqkgICAgiwJAAAAAAAA\nTirQ4GjXrl1q3bq1JKl+/frat2+fw3Kr1aqEhARZLBZJsv8PAAAAAAAA1yvQ4CgxMdGhBZGPj49S\nU1Pl5ZXWQ65WrVq6//77Vbp0aXXo0EH+/v4FWQ4AAAAAAAByoUCDI39/f128eNE+fX1o9Ouvv2rL\nli3avHmzSpcuraeeekqffvqp7rrrrmy3uWvXroIsGQAAAAAAAP9ToMFRo0aN9Pnnn6tjx47avXu3\nwsPD7csCAgLk5+cnX19fWSwWlS1bVufPn89xm1FRUQVZMgAAAAAAQLGSXSOdAg2OOnTooG+++Ua9\nevWSJE2aNEkfffSRkpKS1KNHD8XGxuqBBx6Qr6+vqlevrq5duxZkOQAAAAAAAMgFi2EYhquLcNau\nXbtocQQAAAAAAJCPsstbvAq5FgAAAAAAALgJgiMAAAAAAACYIjgCAAAAAACAKYIjAAAAAAAAmCI4\nAgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAAAIApgiMAAAAAAACYIjgCAAAAAACAKYIjAAAAAAAAmCI4\nAgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAAAIApgiMAAAAAAACYIjgCAAAAAACAKYIjAAAAAAAAmCI4\nAgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAAAIApgiMAAAAAAACYIjgCAAAAAACAKYIjAAAAAAAAmCI4\nAgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAAAIApgiMAAAAAAACYIjgCAAAAAACAKYIjAAAAAAAAmCI4\nAgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAAAIApgiMAAAAAAACYIjgCAAAAAACAKYIjAAAAAAAAmCI4\nAgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAAAIApgiMAAAAAAACYIjgCAAAAAACAKYIjAAAAAAAAmCI4\nAgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAAAIApgiMAAAAAAACYIjgCAAAAAACAKYIjAAAAAAAAmCI4\nAgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAAAIApgiMAAAAAAACYIjgCAAAAAACAKYIjAAAAAAAAmCI4\nAgAAAAAAgCmCIwAAAAAAAJgiOAIAAAAAAIApgiMAAAAAAACYIjgCAAAAAACAKYIjAAAAJ8ycOVMx\nMTGaOXOmq0sBAAAoNARHAIBc8/Qv0J6+f7lRHI9FSkqKDhw44PBv7969WrdunSRp3bp12rt3r8Py\nlJQUF1cNAABQMCyGYRiuLsJZu3btUlRUlKvLAIBiLSkpSZ07d5ZhGPLy8tK6devk5+fn6rLyjafv\nX24U12Nx4MABTRq/SCFBFe3zrqUk66ffPrFP17mlo3y8fSVJZxNOauwL/RUeHl7otQIAAOSH7PIW\nn0KuBQDg5pKTk5X+m0NqaqqSk5M9Kkzw9P3LjeJ8LEKCKqp8uSr26WspV6Xf0qcsKl+uqny8S7ik\nNgAAgMJEVzUVz2b4AADAeT7eJVQttI4kqVrobYRGAACg2Cj2wVFSUpJ9zIIPP/xQSUlJLq4ofxGK\nAQCQPyJqtlZMs8GKqNna1aUAAAAUmmIfHJk1w/cUnh6KAQAAAACAglXsgyNP5smhGAAAAAAAKHgE\nR4AJuvgBcBbXCwAAAHgygiMgA7r4AXAW1wsAAAB4OoIjIAO6+AFwFtcLAAAAeDqCI7g1uogAAAAA\nAFBwCI7gtugicuMI3AAAAAAAziA4gtvKbRcRwpI0BG4AAAAAAGcRHKFYICz5B2OyoKgj5AUAAACK\nDoIjFAuEJYB7IOQFAAAAihaCIwBAkUHICwAAABQtBEdAEUaXHQDuhGsWAACA5yE4gh1/8BctdNkp\nXnj/wd1xzQIAAPBMBEeQxB/8RVFx6LJDWJKG9x88QXG4ZgEAABRHBEeQVLz/4E9JSdGBAwfs/2w2\nm8Nym83msDwlJcVFlXqWggxLikog5Wwdxfn9BwAAAKBo83F1AYCr2Ww2TRq/SCFBFSVJ11Icv7Qv\nfHOTfLx9JUlnE05q7Av9FR4eXuh1ehqzsMTPzy/P280YSA0cODBftuuudaBomTlzptauXasuXbpo\n+PDhri4HAAAAyFGxCo5SUlIytSZJTEx0mLbZbPL395ckhYWFydvbu9Dqg+uEBFVU+XJVJElXr152\nWFYupJJKlCjlirJwAwoqkHLXOlB0ECYCAADAHRWr4Mhms+mxuaPlXz7QPi812bHb0QsfTpWXr7cS\nT5/X7H9PoWUJgCKHVivuiTARAAAA7qjYjXHkXz5QQZVC7P8CQ4MdlgeGBiuoUohDuAS4g6Iyrg8K\nFgNpAwAAAChMxarFEeCp6AJTfNBqBQUpY5fu7LpzS3TpBgAAKA4IjgAPQJgAID/YbDYtHzpMlf4X\nDl0xUh2Wb33hRZW0pDVW/isxUT3fiKdLNwAAgIcjOAKKgNwO3C7xSz+AglHJ319VA4MkSUkpKVLi\nBfuyyv6B8uO6AwAAUKwQHMEj5aa7xaFDhwq1NjO5GbhdUrEcvJ0BoQEAAACg8BEcwSPlprvFvpMn\n5B0ZW+g1ZpQ+cHu6lCvXlKi/7NOBocHyLlk837KM4QQAAAAArlE8v4V6KAY1deRsd4sTiRd0xiUV\nwlmeNIZTbrslevr7FAAAAEDRRnDkQTJ2d6KrEyQCxaImY2s4KesWcZ4y+HBW3QzdMUTj/QQAAIDi\nhuDIw1zf3YmuTpDSvshOGr9IIUEVJUnXUpIdli98c5N8vH0lSWcTTmrsC/3dPqgo6q5vDSd59gDE\n2XUzzHhuSlmfn0Xl3OSuYwAAAChuSBHgNtxtwOuiJCSoosqXqyJJunr1ssOyciGVVKJEKVeUhWIg\np26G15+bknucn9x1DAAAAMUJwRHcRm664p04eEwPKzDTNgAAAAAAgPMIjuBWnO2Kl3j6vBjxGoUl\nN2P10BoOAAAAgDshOCqmjNRUhy+wDPDqXnLz+hWFoMIdB0HOjYyt4aSsW8TRGg4AAACAOyE4KqYu\nnknUjlen6ggDvLql3Lx++06ekHdkbKHXeL3cBCvuese/61vDSVm3iKM1HAAAAAB3QnBUjDHAq3tz\n9vU7kXihSOQUzgYr+EeqYWRqMeZJLbU8XcbXj5adAAAAcEd8SwMysHhd/8XNkmEaKDynLiZq+8Iv\nFRL0i31eUb9dPf6R8fXL6rWTxOsHAACAIovgCMjAx7uEqoXW0R8nflK10Nvk413CJXVYvCzXTWSY\nhqTcj50kuV+rDne8XT3+cf3rx2sHAAAAd0RwBJiIqNlaETVbu7QGrxLe8g8vq8QDZ+Rfq6y8SrhP\n2JEXGQf+lrK/Q9myhV8qJKiifRmtOgAAAAAg/xAcAUVYSHQVhURXyXlFD5Jx4G8p68G/9508oZDI\nWFrkAC7AGE4AAADFA8ERgCLn+oG/pawH/y4qA3/jHxm7D2YXJmRsWeYq7lhzUcAYTgAAAMUDwRGK\nBW/LP+MDWTJMA8g/NptNy4cOU6X/BS1ZtRaTpH0nT8g7MrbQa8zIZrPpsbmj5V8+UJKUmpzisPyF\nD6fKy/d/dyk8eEwPK7DQayyqGMMJAADA8xEcoVjw9fJSgzL+2n0xUfXL+MvXy8vVJQF2nhZsXt9i\nLKvWYpKKVIsx//KBCqoUIklKuXJNifrLviwwNFjeJdM+LhNPn1eRKRoAAAAoBARHKDZigssqJris\nq8sAMiHYBAAAAFBUFWhwZBiG/t//+3/69ddf5evrqwkTJuimm26yL9+7d69eeeUVSVL58uX16quv\nytfXtyBLAoAiiWATAAAAQFFUoD9rb9y4UcnJyVq2bJlGjRqlSZMmOSyPi4vT5MmTtWTJErVu3VrH\njh0ryHIAAAAAAACQCwXa4mjXrl1q3bq1JKl+/frat2+ffdmhQ4cUHBysBQsW6ODBg7rjjjt08803\nF2Q5gMfytDFyAAAAAABFQ4G2OEpMTFRAQIB92sfHR6mpaXfYOXv2rHbv3q2+fftqwYIF2rp1q779\n9tuCLAfwWOlj5EhijBwAAAAAQL4p0BZH/v7+unjxon06NTVVXv/7QhscHKzq1aurZs2akqTWrVtr\n3759atq0abbb3LVr1w3Xc/jw4Vytv2/fPl24cCHnFYuI3O5fbhSFY1GQ+5cbBXEs8mPf8muMnKK6\nf/nFk/evsN6n11/XJWnPnj0qU6aMJPe8DhW31y877vj6FYQbOQ7utH8AAAC5UaDBUaNGjfT555+r\nY8eO2r17t8LDw+3LbrrpJl26dElHjx7VTTfdpF27dql79+45bjMqKuqG6wkICJB+XeP0+nXr1nWo\nuajL7f7lRlE4FgW5f7lREMeiMPfN4uV9/VSGafffv5wU1v5ZvK7rLmjJMF1ACut9mpCQ4DBdv359\nBQUFSUo7Fl/kYdvZnZ8FtX9F5fwsKtfZvLx+2SkK++esgIAAff1/R3P1GHfaPwAAgIyya6RToMFR\nhw4d9M0336hXr16SpEmTJumjjz5SUlKSevTooQkTJujJJ5+UJDVs2FC33357QZYDQJKPdwlVC62j\nP078pGqht8nHu4SrS/JIXiW85R9eVokHzsi/Vll5lfDO+UFFVEpKimw2m306MTHRYbnNZpO/f1pX\nyUOHDuXpuTg/AQAAgKKlQIMji8WiF154wWFeetc0SWratKlWrlxZkCUAMBFRs7UiarZ2dRlOc9fB\nv0Oiqygkuoqry8gzm82mx+aOln/5QElSanKKw/IXPpwqL9+0YOzEwWN6WIF5ej53Oz8BAAAAT8YI\nupDkvl/MUTww+Lfr+ZcPVFClEAVVClFgaLDDssDQYPuyMiH+Lqqw6MupmyjgrJkzZyomJkYzZ850\ndSkAAKAY4NsXJPHFHEVfTHBZjapaPV8GAAdcIb0bniS64eGGJSUlad26dZKkDz/8UElJSS6uCPgH\noSYAeKYC7armDlwxgG1RlV935SosvHYojtyt1QqtGR3RDQ95lZycLMMwJKXdrTY5OVl+fn4urgrI\nHGoOHDiQcxMAPESxb1aSPoCtJLcfwLa44bVzbwR/N8bdWq14emtGgjEASGMWagIAPEOxb3Ekec4A\ntsURr5378qS7jhU2d2u14m6tGXMjPRjbfTHRI4MxAAAAgOAIgMsQ/METeHIwRosqAAAAEBwBAOBG\nUlJSZLPZnF4/LCxM3t431qKPFlUAAAAgOAIAwI3YbDb9e9zbKhNUIcd1Lyac0twJAxQeHn7Dz+fJ\nLaoAADdm5syZWrt2rbp06aLhw4e7uhwABYzgCAAAN1MmqIICy1Z2dRkAgGKIO+gBxQ/BEQAAAAAg\nE7Pu0YmJiQ530Nu/f7/8/dPuoJqX7tEAii6CIwAexeJ1/R8rlgzTAAAAcJbNZtPyocNU6X/BkCRd\nMVId1tn6wosqafHSX4mJ6vlGfJ66RwMomgiOAHgUH+8SqhZaR3+c+EnVQm+Tj3cJV5cEAADgtir5\n+6tqYJB9OiklRUq8YJ+u7B8oP1oZAR6N4AiAx4mo2VoRNVu7ugwAAAAAcHvcVxcAAAAAAACmCI48\nmMXLct1EhmkAAAAAAIAcEBx5MK8S3vIPLytJ8q9VVl4l6HsMAAAA4MZ5W/75MdqSYRqAZ2KMIw8X\nEl1FIdFVXF0GAAAAAA/g6+WlBmX8tftiouqX8ZevF20RAE/nVHC0Y8cObd68Wb///ru8vLxUo0YN\nxcTEqHHjxgVdHwAAAACgCIkJLquY4LKuLgNAIck2Ht6/f7/69u2rJUuWqGrVqurRo4d69uypatWq\n6d1331WfPn30008/FVatAAAAAAAAKETZtjhat26dZs6cqZCQkEzL+vTpo7///lvz5s1TnTp1CqxA\nAAAAAAAAuEa2wdGYMWOyfXC5cuU0duzYfC0IAAC4H4vX9TdgsGSYBgAAgLvKcYyjmTNnqkmTJv+/\nvXuPt6qu88f/OhePIgcUNEsTrwHmdYJkrL44WFqajo+8DaSilnnXMbMSDBAShJq0UuyhNo1+RRzU\nEW+NU3nBLPNKaaKpP4+3vJRXlAPE5Zz1+8OvR9EVnsO4OWdvns+/WJe9eL/XZ+2193nttdfOpz71\nqfcGDIIAACAASURBVCRvhkmbbLJJTj755IoXBwBUh8aGtbLph7fLs399KJt+eNs0NqzV3SUBFdLW\n1paWlpYV5rW2tq4w3dLSkubm5o7prbfeOg0NAmWAarTS4OjHP/5xHnnkkYwcObJj3nHHHZdp06Zl\n+vTpOfHEEyteIABQHQZvOTyDtxze3WUAFdbS0pITLvx2mjfs2zGvfWnbCutMuuHs1De9GRS1vvxG\nzj/m+xk0aNBqrROAD8ZKg6Obb745V199dZqamjrmbbHFFjn77LMzcuRIwREAAKyBmjfsm/U+8vZ9\nUNuWLE9r/tIx3ffD66dh7U79gDMAPdxKf1WtoaFhhdDoLb17905joxcCAAAAgFq20uCoV69eeeaZ\nZ94z/+mnn059/UofCgAAAECVW2n6c8wxx+SrX/1qrr322jzxxBNpaWnJddddl6OOOipf+9rXVleN\nAACwgnPPPTef+9zncu6553Z3KQBQ01b6fbMRI0akvr4+F154YSZNmpT6+vrssMMOGT9+fIYPd/NL\nAABWv8WLF+f6669Pktxwww056qij0qtXr26uCgBq0/veqGjXXXfNrrvuujpqAQCA97V06dIURZEk\naW9vz9KlSwVHAFAhKw2Oxo4du8J0fX19+vbtm09/+tOuOAIAAACocSsNjoYNG7bCdFEUeeWVV3LO\nOefkqaeeyujRoytaHAAAAADdZ6XB0X777Vc6f9SoUTn44IMFRwDUnLr6undMvGsaAADWMCv9VbW/\np0+fPmloaPigawGAble/VkOaB/VPkjQP7J/6tbzeAQCw5nrfm2OXefXVVztuSAgAtabfsE3Sb9gm\n3V0GAAB0u5UGR9OnT3/PvAULFuS2227LMcccU7GiAADWdG1tbWlpaVlhXmtr6wrTLS0taW5uTpJs\nvfXWrggHAD5wXbriqK6uLgMGDMj06dMzcODAStUEALDGa2lpyQkXfjvNG/btmNe+tG2FdSbdcHbq\nmxrS+vIbOf+Y72fQoEGru0wAoMatNDg68cQTS+cvWrQoV1xxRUaOHFmRogAASJo37Jv1PtKvY7pt\nyfK05i8d030/vH4a1l6lOw8AAHRKl26O/cgjj+SMM87I//k//ydXXnllpWoCAAAAoAd434+olixZ\nkv/+7//Of/7nf+axxx5LfX19LrzwwgwbNmx11AcAAABAN1npFUeTJ0/O5z73udx0000ZPXp07rjj\njvTr109oBAAAALAGWOkVR7/4xS+y44475vOf/3x22223NDc3p66ubnXVBgAAAEA3WukVR7/+9a9z\n0EEH5ZZbbsmIESNywgknZPHixVm6dOnqqg8AAACAbrLS4KihoSG77bZbzjvvvNx0003Zeeed86EP\nfSjDhw/P9773vdVVIwAAAADdoNO/39q/f/8cccQROeKII/Lwww9n9uzZlawLAACSJG1tbWlpaemY\nbm1tXWF5S0tLmpubO6a33nrrNDQ0rLb6SOrq33E7i7p3TQNQ1VYaHE2fPv3vLlt//fU/8GIAAODd\nWlpacsXxJ+Yj/y8cWlK0r7D8d5O+m7Xr3ryQ/i+trRn5k+kZNGjQaq9zTVa/VkOaB/VP62Ovpnlg\n/9SvJbgDqBWdvuIIAAC6y0eam/PRvuslSRa3tSWtCzqWbdzcN71cYdTt+g3bJP2GbdLdZQDwAVtp\ncHTiiSeWzi+KIs8++2xFCgIAAACgZ+jUFUeXXXZZzjnnnCxevLhj3qabbpqbbrqpYoUBAAAA0L1W\n+qtqb/mP//iPXHfddfniF7+Ym266KVOmTMmOO+5Y6doAAAAA6EadCo422GCDDBgwIIMHD85jjz2W\n/fffP08++WSlawMAAACgG3UqOOrVq1fuuuuuDB48OHPmzMlLL72UN954o9K1AQAAANCNOnWPo3Hj\nxuXqq6/Oaaedlv/6r//KXnvt9XdvnE3taS+KLl1htvXWW6fBL5sAwGpTtLd7rf5/av19S1tbW1pa\nWjq9frX1V22qcTy6UnOtf8ukGsevK7rSX1tbW5J0ur+esC9qffw6q6v7Ien6vlhpcDRnzpz80z/9\nUwYNGpSxY8cmSc4777wuFUT1e2lha+665Pb0W++R9133tddfzNhJh2fQoEGroTKgO9TV171j4l3T\nQEX9veffwldbc8+/nZ1nmpvfdxt/aW3NyJ9Mr+rX6oa6t/dD3buma/19S0tLS6ae8X/Tb72N3nfd\nauyv2rS0tOSK40/MR6roudfS0pITLvx2mjfs+77r/vX/ez5fyfuvV62qcfy6oiv9zXvxr3lj8xFV\ndW6p9fHrrK68LiSrNn4rDY4uvvjiTJw4Mfvuu28OPPDAbL755p3eMLWl33obZcMNNunuMoAeoH6t\nhjQP6p/Wx15N88D+qV+r9j65gZ5qZc+/jzQ356N91+vG6lafpvr6/EPv5ty/sDU79W5OU/2Kd1+o\n9fcttd5ftanG517zhn2z3kf6ve96rS+/kby6GgrqRtU4fl3R2f7+2rogDVV4bqn18eusSr8urDQ4\nuvTSS/PCCy/k+uuvz3HHHZd+/frlgAMOyF577ZVevXpVrCgAerZ+wzZJv2HV9cYCaoXn35s+t37/\nfG79/t1dBgDUvPe9OfbGG2+cY445JjfeeGPGjBmThx9+OPvtt1/Gjx+/OuoDAAAAoJt06ubYbxk4\ncGB22mmnPP/88/n9739fqZoAAAAA6AHeNzhqa2vLb37zm9xwww255557MmLEiBx11FH5xCc+sTrq\nAwAAAKCbrDQ4mjBhQm666aYMHDgw+++/f6ZMmZJ11llnddUGAAAAQDdaaXC0wQYb5Morr8yAAQNW\nVz0AAAAA9BArvTn2smXLsv766//d5fPnz8+//du/feBFAQAAAND9VnrF0Re/+MUcf/zx2WijjfLJ\nT34yH/nIR9LQ0JDnn38+d911V1588cWcfvrpq6tWAAAAAFajlQZH2267bWbMmJG77rort956a267\n7bbU1dVls802y8iRI/OpT31qddUJAAAAwGr2vr+qliS77LJLdtlll0rXAgAAAEAP0qng6De/+U1+\n9KMf5fXXX09RFB3zb7nllooVBgAAAED36lRwNHny5IwZMyYDBw5MXV1dpWsCAAAAoAfoVHDUr1+/\n7LbbbpWuBQAAAIAepFPB0dChQzN16tQMHz48a6+9dsf8nXfeuWKFAQAAANC9OhUc/fGPf0ySPPzw\nwx3z6urqcumll1amKgAAAAC6XaeCoxkzZlS6DgAAAAB6mE4FR/fdd19+9rOfZdGiRSmKIu3t7Xn+\n+edz6623Vro+AAAAALpJfWdWGjduXHbfffe0tbXlkEMOyeabb57dd9+90rUBAAAA0I06FRyts846\nOeCAAzJs2LD07ds3kydPzr333lvp2gAAAADoRp0KjtZee+3Mnz8/W265ZR544IHU1dVl0aJFla4N\nAAAAgG7UqeDoiCOOyCmnnJLddtst1157bfbee+9sv/32la4NAAAAgG7UqZtj77XXXtlzzz1TV1eX\n2bNn56mnnso222xT6doAAAAA6EaduuLo9ddfz/jx43PYYYdlyZIlmTFjRhYsWFDp2gAAAADoRp0K\njsaPH58ddtgh8+fPT+/evbPRRhvlW9/6VqVrAwAAAKAbdSo4evbZZzNy5MjU19enqakpp5xySv7y\nl79UujYAAAAAulGngqOGhoYsWLAgdXV1SZKnnnoq9fWdeigAAAAAVapTN8c+6aSTMnr06Lzwwgs5\n/vjjc//99+ess86qdG0AAAAAdKNOXTa0/fbbZ/fdd8+mm26aF154IXvssUfmzZtX6doAAAAA6Ead\nuuLoqKOOyuDBg7PbbrtVuh4AAAAAeohOBUdJfDUNAAAAYA3TqeBo9913z1VXXZVddtklDQ0NHfM3\n2WSTihUGAAAAQPfqVHC0YMGCXHTRRenXr1/HvLq6utxyyy0VKwwAAACA7tWp4OhXv/pV7rzzzqyz\nzjqVrgcAAACAHqJTv6o2YMCAvP7665WuBQAAAIAepFNXHNXV1WXvvffOwIEDs9Zaa3XMv/TSSytW\nGAAAAADdq1PB0bHHHlvpOgAAAADoYToVHA0bNqzSdQAAAADQw3TqHkcAAAAArHkERwAAAACUEhwB\nAAAAUEpwBAAAAEApwREAAAAApQRHAAAAAJQSHAEAAABQSnAEAAAAQCnBEQAAAAClBEcAAAAAlBIc\nAQAAAFBKcAQAAABAKcERAAAAAKUqGhwVRZEzzjgjo0aNymGHHZY///nPpetNmDAh55xzTiVLAQAA\nAKCLKhoc3XzzzVm6dGlmzZqVU089NVOnTn3POrNmzcpjjz1WyTIAAAAAWAUVDY7mzp2b4cOHJ0l2\n2mmnzJs3b4Xlf/jDH/Lggw9m1KhRlSwDAAAAgFVQ0eCotbU1ffr06ZhubGxMe3t7kuSll17K9OnT\nM2HChBRFUckyAAAAAFgFjZXceHNzcxYuXNgx3d7envr6N7OqX/ziF5k/f36OOuqovPTSS1myZEm2\n2mqrfOlLX1rpNufOnbvK9Tz99NNdWn/evHlZsGDBKv9/q1tX+6uUSu23Wu6vp/SW6G9V9JT+PPdW\nTbX1V8nXskrui2p6TV2V/VBt55ZqGo/EvninnvB+tqc8R3qCnjAeXeX59LZqHL+uqPWxrvXx66zV\ncU6uaHA0ZMiQzJkzJ3vuuWfuv//+DBo0qGPZ6NGjM3r06CTJNddckyeffPJ9Q6MkGTp06CrX06dP\nn+TRazu9/vbbb79CzT1dV/urlErtt1rur6f0luhvVfSU/jz3Vk219denT5/kxic+8O2+te1fd3rL\nXVNNr6l9+vTJb39V/oMef0+1nVuqaTwSx+Y7dfX47CnHZrXt587q6rHZE/aDc8vbqnH8uqLWz521\nPn6d9UG9b1nZRToVDY722GOP3HHHHR33MJo6dWp+/vOfZ/HixTnooIMq+V8DAAAA8L9U0eCorq4u\nkyZNWmHelltu+Z719ttvv0qWAQAAAMAqqOjNsQEAAACoXoIjAAAAAEoJjgAAAAAoJTgCAAAAoJTg\nCAAAAIBSgiMAAAAASgmOAAAAACglOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAAAKCU\n4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACglOAIAAACg\nlOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAAAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgCAAAA\noJTgCAAAAIBSgiMAAAAASgmOAAAAACglOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAA\nAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACglOAIA\nAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAAAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgC\nAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACglOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4\nAgAAAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACgl\nOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAAAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAo\nJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACglOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAA\nKCU4AgAAAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAA\nACglOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAAAKCU4AgAAACAUoIjAAAAAEoJjgAA\nAAAoJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACjVWMmNF0WRiRMn5tFHH01TU1OmTJmSAQMG\ndCz/+c9/nksvvTSNjY0ZNGhQJk6cWMlyAAAAAOiCil5xdPPNN2fp0qWZNWtWTj311EydOrVj2ZIl\nS3Luuefmsssuy+WXX54FCxZkzpw5lSwHAAAAgC6oaHA0d+7cDB8+PEmy0047Zd68eR3LmpqaMmvW\nrDQ1NSVJli9fnrXXXruS5QAAAADQBRUNjlpbW9OnT5+O6cbGxrS3tydJ6urq0r9//yTJjBkzsnjx\n4nz605+uZDkAAAAAdEFF73HU3NychQsXdky3t7envv7trKooinz/+9/P008/nenTp3dqm3Pnzl3l\nep5++ukurT9v3rwsWLBglf+/1a2r/VVKpfZbLffXU3pL9Lcqekp/nnurptr6q+RrWSX3RTW9pq7K\nfqi2c0s1jUdiX7xTT3g/21OeIz1BTxiPrvJ8els1jl9X1PpY1/r4ddbqOCdXNDgaMmRI5syZkz33\n3DP3339/Bg0atMLy8ePHZ5111slPfvKTTm9z6NChq1xPnz59kkev7fT622+//Xtq7sm62l+lVGq/\n1XJ/PaW3RH+roqf057m3aqqtvz59+iQ3PvGBb/etbf+601vummp6Te3Tp09++6s/d+kx1XZuqabx\nSByb79TV47OnHJvVtp87q6vHZk/YD84tb6vG8euKWj931vr4ddYH9b5lZRfpVDQ42mOPPXLHHXdk\n1KhRSZKpU6fm5z//eRYvXpztttsus2fPztChQzN69OjU1dXlsMMOy+67717JkgAAAADopIoGR3V1\ndZk0adIK87bccsuOfz/88MOV/O8BAAAA+F+o6M2xAQAAAKhegiMAAAAASgmOAAAAACglOAIAAACg\nlOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAAAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgCAAAA\noJTgCAAAAIBSgiMAAAAASgmOAAAAACglOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAA\nAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACglOAIA\nAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAAAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgC\nAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACglOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4\nAgAAAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACgl\nOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAAAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAo\nJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACglOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAA\nKCU4AgAAAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAA\nACglOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAAAKCU4AgAAACAUoIjAAAAAEoJjgAA\nAAAoJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACglOAIAAACglOAIAAAAgFKCIwAAAABKCY4A\nAAAAKCU4AgAAAKCU4AgAAACAUoIjAAAAAEoJjgAAAAAoJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmO\nAAAAACglOAIAAACglOAIAAAAgFKCIwAAAABKCY4AAAAAKCU4AgAAAKBURYOjoihyxhlnZNSoUTns\nsMPy5z//eYXlt956aw488MCMGjUqV111VSVLAQAAAKCLKhoc3XzzzVm6dGlmzZqVU089NVOnTu1Y\ntnz58kybNi2XXHJJZsyYkSuuuCKvvvpqJcsBAAAAoAsqGhzNnTs3w4cPT5LstNNOmTdvXseylpaW\nbL755mlubs5aa62VoUOH5t57761kOQAAAAB0QWMlN97a2po+ffq8/Z81Nqa9vT319fXvWda7d+8s\nWLCgkuW8WdPLb3yg6/U0na174Wut+Utr53LDlxYtyhuvv9ipdV/r5Hqrqpb768oxV8v9Vaq3pLb7\n89z736m2/ha+/tIHut47/aW1tVPr9aTxq4Su1Fxt55bOjnFP49h8W2fr7gnHZlfXrUadPTZ70nPP\nueVt1Th+XVHr585aH7/OqvT7lrqiKIouP6qTpk2bln/4h3/InnvumSQZMWJEbrvttiTJo48+mrPP\nPjsXXXRRkmTq1KkZOnRoPv/5z//d7c2dO7dSpQIAAACssYYOHVo6v6JXHA0ZMiRz5szJnnvumfvv\nvz+DBg3qWLb11lvn6aefzhtvvJF11lkn9957b4488siVbu/vNQEAAADAB6+iVxwVRZGJEyfm0Ucf\nTfLmVUUPPfRQFi9enIMOOii33XZbpk+fnqIocuCBB+bLX/5ypUoBAAAAoIsqGhwBAAAAUL0q+qtq\nAAAAAFQvwREAAAAApQRHAAAAAJSq6K+q9WTLly/P6aefnueeey7Lli3Lsccem4997GMZM2ZM6uvr\nM3DgwJxxxhkrPOadN/tuamrKlClTMmDAgG7qYOXa29szbty4PPnkk6mvr8+kSZPS1NRUM/0lySuv\nvJIDDjggF198cRoaGmqqt/333z/Nzc1Jkk033TTHHntsTfV30UUX5dZbb82yZcty8MEHZ+edd66Z\n/q655prMnj07dXV1WbJkSR555JHMnDkzZ511VtX3t3z58px22ml57rnn0tjYmDPPPLOmnntLly7N\n2LFj8+yzz6a5ubmjl1ro74EHHsgPfvCDzJgxI88880xpT1deeWWuuOKKrLXWWjn22GMzYsSIFbbx\n9x7XE7yzv7dMnTo1W221VUaOHJmkdvr705/+lMmTJ6ehoSFNTU35/ve/n/79+1dtf+/s7fHHH8+E\nCROSJJtvvnmmTJmS+vr6qu0tKT82b7jhhsycOTOzZs1KUlvH5jHHHJMtttgiSfLlL385e+21V830\n941vfCMvv/xyiqLIc889l0984hM5++yza6a/P/3pT5k4cWIaGxuzxRZbZMqUKUmq9/h8Z28PPfRQ\nJk6cmLXXXjvbbLNNxo0bl6Q6e+vK37C10t9nP/vZJLVx7izrb+ONN66Oc2exhrr66quLs846qyiK\nonj99deLESNGFMcee2xx7733FkVRFBMmTChuuummFR7zq1/9qhgzZkxRFEVx//33F8cdd9zqLboL\nbrrppuL0008viqIo7r777uK4446rqf6WLVtWnHDCCcUXvvCF4oknnqip3pYsWVLst99+K8yrpf7u\nvvvu4thjjy2KoigWLlxYnHfeeTXV3ztNmjSpuPLKK2umv5tvvrn4+te/XhRFUdxxxx3FSSedVDO9\nFUVRXHbZZcX48eOLoiiKJ598svjqV79aE/399Kc/LfbZZ59i5MiRRVGUn09eeumlYp999imWLVtW\nLFiwoNhnn32KpUuXrrCd99sX3eXd/b3yyivF1772tWKPPfYoZs2aVRRFUVP9HXroocUjjzxSFEVR\nzJo1q5g2bVrV9vfu3o4//vjivvvuK4qiKMaMGVNzx2ZRFMVDDz1UHH744R3zaqm/K6+8srj44otX\nWKeW+nvL66+/XnzpS18qXn755Zrq74QTTihuv/32oiiK4tRTTy3mzJlTtf29u7f999+/uP/++4ui\nKIof/vCHxfXXX1+1vXX2b9ha6G/+/PnFiBEjiqKonXNnWX9XXXVVVZw719ivqu211145+eSTkyRt\nbW1paGjIww8/nE9+8pNJkl133TV33nnnCo+ZO3duhg8fniTZaaedMm/evNVbdBfsvvvuOfPMM5Mk\nzz//fNZbb72a6u973/tevvzlL2ejjTZKURQ11dsjjzySRYsW5cgjj8wRRxyRBx54oKb6++1vf5tB\ngwbl+OOPz3HHHZcRI0bUVH9vefDBB/P444/noIMOykMPPVQT/W2xxRZpa2tLURRZsGBBGhsba2rs\nHn/88ey6665J3uz1iSeeqIn+Nt9885x//vkd0+8+Hn/3u9/lj3/8Y4YOHZrGxsY0Nzdniy22yKOP\nPrrCdt7vOO4u7+5v0aJFOemkk7Lvvvt2zKul/n74wx9m8ODBSd785LKpqalq+3t3b9OnT8/QoUOz\ndOnSvPTSS+nTp0/V9pa8t7/XXnstP/rRj/Kd73ynY14t9ffQQw/ltttuy6GHHppx48Zl4cKFNdXf\nW84999wceuih2WCDDWqqv49//ON57bXXUhRFFi5cmMbGxqrt7929/fWvf81OO+2UJBkyZEjuu+++\nqu2tM3/DVvPr+jv7a29vT2NjY+bPn18z586y/h566KHMmTOnx58719jgqFevXll33XXT2tqak08+\nOaecckqKouhY3rt37yxYsGCFx7S2tqZPnz4d042NjWlvb19tNXdVfX19xowZk8mTJ2efffapmf5m\nz56dDTbYIJ/5zGc6enpnndXcW5Kss846OfLII/Ozn/0sEydOzDe/+c2aGbvkzTfO8+bNy7nnntvR\nXy2N31suuuiinHTSSe+ZX8399e7dO88++2z23HPPTJgwIaNHj66pY/PjH/94brvttiTJ/fffn7/+\n9a81cWzuscceaWho6Jh+95i1trZm4cKFK/Sx7rrrvqfXdyrbF93l3f1tuumm2XHHHVdY593jVM39\nbbjhhkmS3//+97n88stzxBFHVG1/7+6trq4uzz//fP75n/858+fPzzbbbFO1vSUr9vfWLQTGjBmT\nXr16daxTK/0lb4bn3/72t3PZZZdlwIABmT59ek31lySvvvpq7r777uy///5Jamv83vp62t57751X\nX301w4YNq9r+3t3bgAEDct999yVJ5syZk7/97W9V21tn/oat5tf1d/d38skn5zvf+U7NnDvfzorV\nogAACcRJREFU3d/Xv/717LjjjjnttNN6/LlzjQ2OkuSFF17I4Ycfnv322y9777136uvf3h0LFy5M\n3759V1i/ubk5Cxcu7Jhub29f4TE90bRp0/LLX/4y48aNy5IlSzrmV3N/s2fPzh133JHRo0fn0Ucf\nzWmnnZbXXnutY3k195a8+cL91iflW2yxRdZff/288sorHcurvb/1118/w4cPT2NjY7bccsusvfba\naW1t7Vhe7f0lyYIFC/LUU09l5513TpKaObdccsklGT58eH75y1/m+uuvz2mnnZZly5Z1LK/m3pLk\ngAMOSO/evXPIIYfklltuyXbbbbfCG89q7+8tZcdjc3Pz+z4P3+847slqrb8bb7wxkyZNykUXXZR+\n/frVVH+bbLJJfvnLX2bkyJGZOnVq+vTpUxO9PfTQQ3nmmWcyceLEnHrqqXn88cczderUmhq73Xff\nPdtuu23Hvx955JGaGb+3/OIXv8g+++yTurq6JLV1bpkyZUouv/zy3Hjjjdl3330zbdq0mhm/s846\nKxdccEG+8pWvZIMNNki/fv2qurfO/A1bzcfmO/vbbLPNau7c+e7xq5ZzZ89+d1tBL7/8co488sh8\n61vfyn777ZfkzU+b77333iTJ7bffnqFDh67wmCFDhuTXv/51kjc/jR40aNDqLboLrrvuulx00UVJ\nkrXXXjv19fXZfvvtc8899ySp7v4uu+yyzJgxIzNmzMg222yT73//+xk+fHjNjN3VV1+dadOmJXnz\n0trW1tZ85jOfqYmxS5KhQ4fmN7/5TZI3+1u8eHF22WWXmukvSe69997ssssuHdO1cm5Zb731Om7a\n3qdPnyxfvjzbbrttzYzdgw8+mE996lOZOXNmvvCFL2SzzTbLxz/+8Zrp7y3bbrvte47HHXbYIXPn\nzs3SpUuzYMGCPPHEExk4cOAKj3u/47i7vfMT13fbcccdq76/t1x33XWZOXNmZsyYkY9+9KNJaqe/\n4447Lk8//XSSNz89ra+vr5ljc4cddsgNN9yQSy+9NOecc04+9rGPZezYsTUzdkly5JFH5sEHH0yS\n3Hnnndluu+1qZvzecuedd3Z8pTmpnede8uYHe2+9xn/4wx/OG2+8URPjlyS//vWvc/bZZ+fiiy/O\n/Pnz8+lPf7pqe+vs37C10t+OO+5YU+fOsvGrlnPnGvurahdeeGHeeOON/OQnP8n555+furq6fOc7\n38nkyZOzbNmybL311tlzzz2TJKeddlpOOeWU7LHHHrnjjjsyatSoJG/+YktP9fnPfz5jx47NoYce\nmuXLl2fcuHHZaqutMm7cuJro791OO+20jB8/viZ6O/DAAzN27NgcfPDBqa+vz7Rp07L++uvXzNiN\nGDEi9913Xw488MCOX6T66Ec/WjP9JcmTTz65wi9r1crxefjhh+f000/PIYcckuXLl+eb3/xmtttu\nu5oZu8033zw//vGPc8EFF6Rv376ZMmVKFi5cWBNj905lx2NdXV1Gjx6dgw8+OEVR5Bvf+EaamprS\n0tKSmTNnZsKECX/3OO4p3roCoMyGG25Y9f0lb17RdtZZZ2WTTTbJCSeckLq6ugwbNiwnnnhiTfR3\n9NFHZ8yYMWlqakqvXr0yefLkmhi7NeHYTJKJEyfmzDPPzFprrZUPfehD+e53v5vevXtXfX/vHL+n\nnnpqhdf3Whq/M888M1//+tfT2NiYpqamnHnmmTXT3+abb57DDz88vXr1yj/+4z92hH/V2Ftn/4at\n1tf1sv7+/d//PU1NTSusV63HZll/Y8eOzVlnndXjz511xco+ogMAAABgjbXGflUNAAAAgJUTHAEA\nAABQSnAEAAAAQCnBEQAAAAClBEcAAAAAlBIcAQAAAFBKcAQArHFaW1tzwgknrHSdsWPH5oUXXljp\nOqNHj8699977v67nmmuuydixY//X2wEA+KAJjgCANc78+fPzyCOPrHSdu+++O0VRrKaKAAB6psbu\nLgAAYHWbMmVKXnzxxZx00knZbbfdcvHFF6euri7bbbddxo8fn8suuywvvvhijj766MycOTO/+93v\ncskll2TJkiX529/+lsmTJ+eTn/xkx/buueeeXHjhhVlnnXXS0tKSwYMH5+yzz05jY2OuvvrqXHLJ\nJR3bnzBhQnr16pVrr702F1xwQfr06ZONN944vXv3TpL88Y9/zLRp0/K3v/0t/fr1y3e/+9189KMf\nzcUXX5xrr702DQ0N2WGHHTJp0qTu2n0AwBrEFUcAwBpn3Lhx2WijjfKv//qvueCCCzJz5sxcf/31\n6dWrV84///wcffTR2WijjfLTn/40ffv2zZVXXpkLL7ww1157bY466qj87Gc/e882//CHP+SMM87I\n//zP/+T555/Pb3/72zz22GO58MILV9j+9OnT8+KLL+YHP/hBLr/88lxxxRVZuHBhkmTZsmUZP358\nzjnnnMyePTtf+cpXMm7cuLS1teWiiy7K7Nmzc/XVV6e+vj4vvvji6t5tAMAayBVHAMAaqSiK3HPP\nPfnsZz+bvn37Jkn+5V/+JaeffvoK69TV1eW8887LnDlz8uSTT+aee+5JQ0PDe7Y3aNCgbLTRRkmS\nrbfeOvPnz89zzz1Xuv0dd9wxQ4YMSf/+/ZMk++67b+6666489dRTeeaZZ3Lcccd1fE1u0aJFaWho\nyJAhQ3LAAQfkc5/7XA455JCO/wsAoJIERwDAGqsoivfcx6itrW2F6UWLFuXAAw/Ml770pey8884Z\nPHhwZs6c+Z5tNTU1dfy7rq4uSdLe3l66/fr6+rS3t3fMa2xs7Fi22Wab5Zprrumo76WXXkqSnH/+\n+XnggQdy++2358gjj8zZZ5+9wtflAAAqwVfVAIA1TmNjY9rb27Pzzjtnzpw5eeONN5IkV155ZXbZ\nZZeOddra2vLUU0+loaEhxx57bHbZZZfcfvvtK4Q+KzNs2LDS7Q8ZMiQPPPBAXnzxxbS3t+fGG29M\nkmy11VZ5/fXXc9999yVJrrrqqnzzm9/Mq6++mr322iuDBg3KSSedlM985jN59NFHP+jdAgDwHq44\nAgDWOBtssEE23njjnHXWWTn66KNzyCGHpK2tLdttt13HTadHjBiRo446Kj/96U+zzTbb5Atf+ELW\nXXfd7Lzzznn++eeTvH1l0d8zePDg0u2vu+66GT9+fI444oisu+66+djHPpbkzauWfvSjH2XKlClZ\nunRpmpub873vfS/9+/fPqFGjcsABB6RXr17ZZJNNst9++1V2JwEAJKkr/M4sAAAAACV8VQ0AAACA\nUoIjAAAAAEoJjgAAAAAoJTgCAAAAoJTgCAAAAIBSgiMAAAAASgmOAAAAACglOAIAAACg1P8PqPc8\n9YMojS0AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x48b135f8>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig = plt.figure(figsize=(20,10))\n",
    "sns.set_style(\"whitegrid\")\n",
    "ax = sns.barplot(x=\"totalnodes\", y=\"ValAUC\", hue = \"nhl\", data=res)#, errwidth=1.0)\n",
    "ax.legend(loc = 2,fontsize = 'xx-large')\n",
    "plt.ylim([0.0,1.0])\n",
    "tit = plt.title('Validation AUC as a function of Number of Hidden Layers and total number of nodes', fontsize = 'xx-large')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "C:\\Users\\Shuki\\Anaconda2\\lib\\site-packages\\ipykernel\\__main__.py:1: FutureWarning: sort(columns=....) is deprecated, use sort_values(by=.....)\n",
      "  if __name__ == '__main__':\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>alpha</th>\n",
       "      <th>batch</th>\n",
       "      <th>trainsize</th>\n",
       "      <th>epochs</th>\n",
       "      <th>trainerror</th>\n",
       "      <th>TrainAUC</th>\n",
       "      <th>ValAUC</th>\n",
       "      <th>hl1</th>\n",
       "      <th>hl2</th>\n",
       "      <th>hl3</th>\n",
       "      <th>hl4</th>\n",
       "      <th>nhl</th>\n",
       "      <th>hl12</th>\n",
       "      <th>totalnodes</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>25</th>\n",
       "      <td>0.001</td>\n",
       "      <td>500</td>\n",
       "      <td>284641</td>\n",
       "      <td>500000</td>\n",
       "      <td>0.432074</td>\n",
       "      <td>0.693237</td>\n",
       "      <td>0.688970</td>\n",
       "      <td>50</td>\n",
       "      <td>35</td>\n",
       "      <td>17</td>\n",
       "      <td>8</td>\n",
       "      <td>4</td>\n",
       "      <td>50 : 35</td>\n",
       "      <td>110</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>26</th>\n",
       "      <td>0.001</td>\n",
       "      <td>500</td>\n",
       "      <td>284641</td>\n",
       "      <td>500000</td>\n",
       "      <td>0.433033</td>\n",
       "      <td>0.692874</td>\n",
       "      <td>0.688918</td>\n",
       "      <td>50</td>\n",
       "      <td>35</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>2</td>\n",
       "      <td>50 : 35</td>\n",
       "      <td>90</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>16</th>\n",
       "      <td>0.010</td>\n",
       "      <td>500</td>\n",
       "      <td>284641</td>\n",
       "      <td>500000</td>\n",
       "      <td>0.429991</td>\n",
       "      <td>0.696398</td>\n",
       "      <td>0.688882</td>\n",
       "      <td>32</td>\n",
       "      <td>16</td>\n",
       "      <td>8</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>32 : 16</td>\n",
       "      <td>60</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>27</th>\n",
       "      <td>0.001</td>\n",
       "      <td>500</td>\n",
       "      <td>284641</td>\n",
       "      <td>500000</td>\n",
       "      <td>0.431971</td>\n",
       "      <td>0.693212</td>\n",
       "      <td>0.688867</td>\n",
       "      <td>50</td>\n",
       "      <td>35</td>\n",
       "      <td>17</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>50 : 35</td>\n",
       "      <td>100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>32</th>\n",
       "      <td>0.001</td>\n",
       "      <td>500</td>\n",
       "      <td>284641</td>\n",
       "      <td>500000</td>\n",
       "      <td>0.432293</td>\n",
       "      <td>0.693443</td>\n",
       "      <td>0.688830</td>\n",
       "      <td>50</td>\n",
       "      <td>35</td>\n",
       "      <td>17</td>\n",
       "      <td>0</td>\n",
       "      <td>3</td>\n",
       "      <td>50 : 35</td>\n",
       "      <td>100</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "    alpha  batch  trainsize  epochs  trainerror  TrainAUC    ValAUC  hl1  hl2  \\\n",
       "25  0.001    500     284641  500000    0.432074  0.693237  0.688970   50   35   \n",
       "26  0.001    500     284641  500000    0.433033  0.692874  0.688918   50   35   \n",
       "16  0.010    500     284641  500000    0.429991  0.696398  0.688882   32   16   \n",
       "27  0.001    500     284641  500000    0.431971  0.693212  0.688867   50   35   \n",
       "32  0.001    500     284641  500000    0.432293  0.693443  0.688830   50   35   \n",
       "\n",
       "    hl3  hl4  nhl     hl12  totalnodes  \n",
       "25   17    8    4  50 : 35         110  \n",
       "26    0    0    2  50 : 35          90  \n",
       "16    8    0    3  32 : 16          60  \n",
       "27   17    0    3  50 : 35         100  \n",
       "32   17    0    3  50 : 35         100  "
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "best = res.sort(columns='ValAUC',ascending=False).head(5)\n",
    "best"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Bonus1 - [Genetic Notebook](Genetic.ipynb) \n",
    "implements genetic optimization algorithm in order to maximize AIC criteria"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Bonus2 - [Other Models Notebook](Other_models.ipynb) \n",
    "implements Random Forest, SVM and PCA in order to create benchmarks for prediction and find best features"
   ]
  }
 ],
 "metadata": {
  "anaconda-cloud": {},
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 1
}