simulated_data_dim.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/az/anaconda/lib/python2.7/site-packages/sklearn/cross_validation.py:44: DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. Also note that the interface of the new CV iterators are different from that of this module. This module will be removed in 0.20.\n",
      "  \"This module will be removed in 0.20.\", DeprecationWarning)\n",
      "/Users/az/anaconda/lib/python2.7/site-packages/sklearn/grid_search.py:43: DeprecationWarning: This module was deprecated in version 0.18 in favor of the model_selection module into which all the refactored classes and functions are moved. This module will be removed in 0.20.\n",
      "  DeprecationWarning)\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[2.43349633 1.34400588 1.36532069 2.50839624 2.21611601 0.74632222\n",
      " 1.58068352 1.8676711  1.96042976 3.41867412]\n",
      "(2000, 10)\n",
      "xshape\n",
      "81.11497247772766\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(array([298., 161., 139., 135., 126., 136., 141., 142., 211., 511.]),\n",
       " array([3.53811817e-08, 9.97357913e-03, 1.99471229e-02, 2.99206666e-02,\n",
       "        3.98942104e-02, 4.98677541e-02, 5.98412979e-02, 6.98148417e-02,\n",
       "        7.97883854e-02, 8.97619292e-02, 9.97354729e-02]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 1,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYEAAAEICAYAAAC55kg0AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nSeqYISBJHTMEJKljhoAkdez/AbHwwjSlzWagAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x118c5b990>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAEICAYAAABRSj9aAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x10f900590>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAD8CAYAAAB5Pm/hAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nJalxBr0kNe7/AYr9VflGEgV+AAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x1207aef50>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "%matplotlib inline \n",
    "import matplotlib.pyplot as plt\n",
    "import numpy as np \n",
    "import cvxpy as cvx \n",
    "# from gurobipy import * \n",
    "from off_pol_eval_functions import * \n",
    "\n",
    "\n",
    "from scipy.optimize import minimize\n",
    "import datetime as datetime\n",
    "import pickle\n",
    "from matplotlib.backends.backend_pdf import PdfPages\n",
    "from matplotlib import collections as matcoll\n",
    "from sklearn import svm\n",
    "import sys\n",
    "from sklearn.linear_model import LinearRegression\n",
    "from scipy.stats import norm\n",
    "\n",
    "np.random.seed(2)\n",
    "def multipage(filename, figs=None, dpi=200):\n",
    "    pp = PdfPages(filename)\n",
    "    if figs is None:\n",
    "        figs = [plt.figure(n) for n in plt.get_fignums()]\n",
    "    for fig in figs:\n",
    "        fig.savefig(pp, format='pdf')\n",
    "    pp.close()\n",
    "\n",
    "d = 10  # dimension of x \n",
    "n = 2000; \n",
    "mu_x = np.zeros(d); \n",
    "sigma_x = np.random.normal(size = (d,1))\n",
    "sigma_x += np.abs(np.min(sigma_x))+0.5\n",
    "sigma_x = np.multiply(sigma_x, np.eye(d))\n",
    "sigma_x /= 2# normalize covariances a little bit \n",
    "\n",
    "W = 1.5 #treatment effect\n",
    "# interact_x = 2\n",
    "white_noise_coef = 0.1\n",
    "\n",
    "\n",
    "# generate propensity model \n",
    "def real_prop(x, beta_prop): \n",
    "    T_SIG = 5\n",
    "    if len(x.shape) > 1: \n",
    "        n= x.shape[1]\n",
    "    else:\n",
    "        n= len(x)\n",
    "    return np.dot(x, beta_prop) + np.random.normal(size = (n,1))*T_SIG\n",
    "    # T is normally distributed conditional on covariates \n",
    "    \n",
    "# coefficient of treatment effect\n",
    "beta_cons = -5\n",
    "beta_x = np.random.normal(size = (d,1))\n",
    "# interaction term with treatment \n",
    "beta_x_T = np.random.normal(size = (d,1))*1.5\n",
    "\n",
    "# sparse interaction terms \n",
    "sparse_entries = np.random.choice(range(d),size =  int(round(0.7*d)),replace = False)\n",
    "beta_x_T[sparse_entries] = 0     \n",
    "sparse_entries = np.random.choice(range(d),size =  int(round(0.35*d)),replace = False)\n",
    "beta_x[sparse_entries] = 0  \n",
    "\n",
    "FREQ = 20 \n",
    "beta_x_quad_T = np.random.normal(size = (d,1))\n",
    "sparse_entries = np.random.choice(range(d),size =  int(round(0.65*d)),replace = False)\n",
    "beta_x_quad_T[sparse_entries] = 0\n",
    "\n",
    "TRUE_PROP_BETA = np.asarray(beta_x_quad_T + np.random.normal( loc= np.ones((d,1))*2, size = (d,1))).flatten()\n",
    "print TRUE_PROP_BETA\n",
    "def real_risk(T, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x):    \n",
    "    n = len(T); risk = np.zeros(n)\n",
    "    if np.isscalar(T):\n",
    "        risk = T*beta_cons + np.dot(beta_x.T, x) + np.dot(beta_x_T.T, x*T) + (T-np.dot(beta_x_quad_T.T,x))**2\n",
    "    else: \n",
    "        for i in range(len(T)): \n",
    "            risk[i] = T[i]*beta_cons + np.dot(beta_x.T, x[i,:]) + np.dot(beta_x_T.T, x[i,:]*T[i]) + (T[i]-np.dot(beta_x_quad_T.T,x[i,:]))**2#+ np.dot(beta_x_quad_T.T, (x[i,:]**2)*T[i]) + np.dot(beta_x_high_freq.T, np.sin(x[i,0:HIGH_FREQ_N]*FREQ)*T[i])\n",
    "    return risk\n",
    "\n",
    "T_SIG = 4\n",
    "def generate_data(mu_x, sigma_x_mat, n, beta_cons, beta_x, beta_x_T): \n",
    "#     x = np.random.normal(mu_x, sigma_x, size = n)\n",
    "    # generate n datapoints from the same multivariate normal distribution\n",
    "    x = np.random.multivariate_normal(mean = mu_x, cov= sigma_x_mat, size = n ) \n",
    "    print x.shape \n",
    "    print \"xshape\"\n",
    "    T = np.random.normal(0, T_SIG, n) + np.dot(x, TRUE_PROP_BETA) + 2*x[:,1] + 4*x[:,2] - 2*x[:,3]\n",
    "    true_resid = T - np.dot(x, TRUE_PROP_BETA)\n",
    "    true_Q = norm.pdf( T - np.dot(x, TRUE_PROP_BETA), loc = 0, scale = T_SIG )\n",
    "    y_sigma = 0.5\n",
    "    white_noise_coef = 5\n",
    "    \n",
    "    clf = LinearRegression(); clf.fit(x, T)\n",
    "    y_hat = clf.predict(x)\n",
    "    Y = np.zeros(n)\n",
    "    for i in range(n): \n",
    "        Y[i] = T[i]*beta_cons + np.dot(beta_x.T, x[i,:]) + T[i]*np.dot(beta_x_T.T, x[i,:]) + (T[i] - np.dot(beta_x_quad_T.T,x[i,:]))**2 #+ np.dot(beta_x_quad_T.T, (x[i,:]**2)*T[i]) + np.dot(beta_x_high_freq.T, np.sin(x[i,0:HIGH_FREQ_N]*FREQ)*T[i])\n",
    "    Y += np.random.multivariate_normal(mean = np.zeros(n), cov=white_noise_coef * np.eye(n))\n",
    "    # get pdf from residuals \n",
    "    resid = Y - y_hat\n",
    "    # get norm pdf \n",
    "    Q = norm.pdf(resid, loc = np.mean(resid), scale=np.std(resid))\n",
    "    T = T.flatten()\n",
    "    return [x, T, Y, true_Q, clf]\n",
    "\n",
    "[x_full, T_full, Y_full, true_Q_full, clf] = generate_data(mu_x, sigma_x, n, beta_cons, beta_x, beta_x_T)\n",
    "\n",
    "#compute real risk \n",
    "print np.mean( real_risk(T_full, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_full))\n",
    "\n",
    "plt.hist(Y_full); plt.title('Y')\n",
    "plt.figure()\n",
    "plt.title('T')\n",
    "plt.hist(T_full)\n",
    "Q = true_Q_full\n",
    "plt.figure()\n",
    "plt.hist(true_Q_full)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## In this setting the optimal treatment is similar to the observed treatment policy. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([198., 131., 119., 113., 116., 129., 132., 131., 197., 483.]),\n",
       " array([2.65065187e-06, 9.97593288e-03, 1.99492151e-02, 2.99224973e-02,\n",
       "        3.98957796e-02, 4.98690618e-02, 5.98423440e-02, 6.98156262e-02,\n",
       "        7.97889085e-02, 8.97621907e-02, 9.97354729e-02]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAEICAYAAABRSj9aAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nktScQS9JzRn0ktScQS9JzRn0ktTc/wOZNYJHtZ5XCAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x120a4a710>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAEICAYAAABRSj9aAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAEWhJREFUeJzt3VGMHeV5xvH/E0NJlKQKlK3l2G4XJOfCRIqRVm6kpFUS\nmuCAKkOrIqMq8gWtU8mliZobk1yEXlhypQDqRaEyBWFFSVyrSYQVaCNjoaJIEWRBDmATihOMsGXs\nTdIoIEVubd5e7Fg5ddY+Z/fs+hx/+v+ko/PNN/OdeT0Mz87OmZlNVSFJatc7Rl2AJGlpGfSS1DiD\nXpIaZ9BLUuMMeklqnEEvSY0z6CWpcQa9dI4kb/W83k7yq57pvxh1fdJ8xRumpPNLcgT4y6p6YtS1\nSAvlEb0kNc6gl6TGGfSS1DiDXpIaZ9BLUuMMeklqnEEvSY3zOnpJapxH9JLUOINekhpn0EtS4wx6\nSWrcZaMuAODqq6+uycnJUZchSZeUZ5999qdVNdFvub5Bn+SdwFPAFd3y/1ZVX05yN/BXwEy36Ber\n6vFuzF3AHcAZ4G+r6rsXWsfk5CTT09P9SpEk9Ujy2iDLDXJEfwr4RFW9leRy4HtJ/r2bd19VfeWc\nFa8FNgHXAe8Hnkjygao6M3j5kqTF0vccfc16q5u8vHtd6OL7jcDuqjpVVa8Ch4H1Q1cqSVqQgb6M\nTbIsyQHgJLCvqp7uZt2Z5PkkDye5sutbCbzeM/xo13fuZ25JMp1kemZm5tzZkqRFMlDQV9WZqloH\nrALWJ/kg8ABwLbAOOA7cM58VV9XOqpqqqqmJib7fJUiSFmhel1dW1S+AJ4ENVXWi+wHwNvAgvz49\ncwxY3TNsVdcnSRqBvkGfZCLJ+7r2u4BPAj9KsqJnsVuBF7v2XmBTkiuSXAOsAZ5Z3LIlSYMa5Kqb\nFcCuJMuY/cGwp6q+k+SrSdYx+8XsEeCzAFV1MMke4BBwGtjqFTeSNDpj8fTKqamp8jp6SZqfJM9W\n1VS/5XwEgiQ1biwegSD1M7ntsZGt+8iOm0e2bmkxeEQvSY0z6CWpcQa9JDXOoJekxhn0ktQ4g16S\nGmfQS1LjDHpJapw3TGleRnnjkqSF8Yhekhpn0EtS4wx6SWqcQS9JjTPoJalxBr0kNc6gl6TGGfSS\n1DiDXpIa552xlyDvTpU0H32P6JO8M8kzSX6Y5GCSv+/6r0qyL8kr3fuVPWPuSnI4yctJblzKf4Ak\n6cIGOXVzCvhEVX0IWAdsSPJhYBuwv6rWAPu7aZKsBTYB1wEbgPuTLFuK4iVJ/fUN+pr1Vjd5efcq\nYCOwq+vfBdzStTcCu6vqVFW9ChwG1i9q1ZKkgQ30ZWySZUkOACeBfVX1NLC8qo53i7wBLO/aK4HX\ne4Yf7fokSSMwUNBX1ZmqWgesAtYn+eA584vZo/yBJdmSZDrJ9MzMzHyGSpLmYV6XV1bVL4AnmT33\nfiLJCoDu/WS32DFgdc+wVV3fuZ+1s6qmqmpqYmJiIbVLkgYwyFU3E0ne17XfBXwS+BGwF9jcLbYZ\neLRr7wU2JbkiyTXAGuCZxS5ckjSYQa6jXwHs6q6ceQewp6q+k+T7wJ4kdwCvAbcBVNXBJHuAQ8Bp\nYGtVnVma8iVJ/fQN+qp6Hrh+jv6fATecZ8x2YPvQ1UmShuYjECSpcQa9JDXOoJekxhn0ktQ4g16S\nGmfQS1LjDHpJapxBL0mNM+glqXEGvSQ1zqCXpMYZ9JLUOINekhpn0EtS4wx6SWqcQS9JjTPoJalx\nBr0kNc6gl6TGGfSS1DiDXpIaZ9BLUuP6Bn2S1UmeTHIoycEkn+v6705yLMmB7nVTz5i7khxO8nKS\nG5fyHyBJurDLBljmNPCFqnouyXuBZ5Ps6+bdV1Vf6V04yVpgE3Ad8H7giSQfqKozi1m4dLFMbnts\nJOs9suPmkaxX7el7RF9Vx6vqua79JvASsPICQzYCu6vqVFW9ChwG1i9GsZKk+ZvXOfokk8D1wNNd\n151Jnk/ycJIru76VwOs9w44yxw+GJFuSTCeZnpmZmXfhkqTBDBz0Sd4DfBP4fFX9EngAuBZYBxwH\n7pnPiqtqZ1VNVdXUxMTEfIZKkuZhoKBPcjmzIf+1qvoWQFWdqKozVfU28CC/Pj1zDFjdM3xV1ydJ\nGoFBrroJ8BDwUlXd29O/omexW4EXu/ZeYFOSK5JcA6wBnlm8kiVJ8zHIVTcfAT4DvJDkQNf3ReD2\nJOuAAo4AnwWoqoNJ9gCHmL1iZ6tX3EjS6PQN+qr6HpA5Zj1+gTHbge1D1CVJWiTeGStJjTPoJalx\nBr0kNc6gl6TGGfSS1DiDXpIaZ9BLUuMMeklqnEEvSY0z6CWpcQa9JDXOoJekxhn0ktQ4g16SGmfQ\nS1LjDHpJapxBL0mNM+glqXEGvSQ1bpA/Di5pBCa3PTaydR/ZcfPI1q3F5xG9JDWub9AnWZ3kySSH\nkhxM8rmu/6ok+5K80r1f2TPmriSHk7yc5Mal/AdIki5skCP608AXqmot8GFga5K1wDZgf1WtAfZ3\n03TzNgHXARuA+5MsW4riJUn99Q36qjpeVc917TeBl4CVwEZgV7fYLuCWrr0R2F1Vp6rqVeAwsH6x\nC5ckDWZe5+iTTALXA08Dy6vqeDfrDWB5114JvN4z7GjXJ0kagYGDPsl7gG8Cn6+qX/bOq6oCaj4r\nTrIlyXSS6ZmZmfkMlSTNw0BBn+RyZkP+a1X1ra77RJIV3fwVwMmu/xiwumf4qq7v/6mqnVU1VVVT\nExMTC61fktTHIFfdBHgIeKmq7u2ZtRfY3LU3A4/29G9KckWSa4A1wDOLV7IkaT4GuWHqI8BngBeS\nHOj6vgjsAPYkuQN4DbgNoKoOJtkDHGL2ip2tVXVm0SuXJA2kb9BX1feAnGf2DecZsx3YPkRdkqRF\n4p2xktQ4g16SGmfQS1LjDHpJapxBL0mNM+glqXEGvSQ1zqCXpMYZ9JLUOINekhpn0EtS4wx6SWqc\nQS9JjTPoJalxBr0kNc6gl6TGGfSS1DiDXpIaZ9BLUuMMeklqnEEvSY0z6CWpcX2DPsnDSU4mebGn\n7+4kx5Ic6F439cy7K8nhJC8nuXGpCpckDWaQI/pHgA1z9N9XVeu61+MASdYCm4DrujH3J1m2WMVK\nkubvsn4LVNVTSSYH/LyNwO6qOgW8muQwsB74/oIrHGOT2x4bdQmS1Ncw5+jvTPJ8d2rnyq5vJfB6\nzzJHu77fkGRLkukk0zMzM0OUIUm6kIUG/QPAtcA64Dhwz3w/oKp2VtVUVU1NTEwssAxJUj8LCvqq\nOlFVZ6rqbeBBZk/PABwDVvcsuqrrkySNyIKCPsmKnslbgbNX5OwFNiW5Isk1wBrgmeFKlCQNo++X\nsUm+AXwMuDrJUeDLwMeSrAMKOAJ8FqCqDibZAxwCTgNbq+rM0pQuSRrEIFfd3D5H90MXWH47sH2Y\noiRJi8c7YyWpcQa9JDXOoJekxhn0ktQ4g16SGmfQS1LjDHpJapxBL0mNM+glqXEGvSQ1zqCXpMYZ\n9JLUOINekhpn0EtS4wx6SWqcQS9JjTPoJalxBr0kNc6gl6TGGfSS1DiDXpIaZ9BLUuP6Bn2Sh5Oc\nTPJiT99VSfYleaV7v7Jn3l1JDid5OcmNS1W4JGkwgxzRPwJsOKdvG7C/qtYA+7tpkqwFNgHXdWPu\nT7Js0aqVJM1b36CvqqeAn5/TvRHY1bV3Abf09O+uqlNV9SpwGFi/SLVKkhZgoefol1fV8a79BrC8\na68EXu9Z7mjX9xuSbEkynWR6ZmZmgWVIkvoZ+svYqiqgFjBuZ1VNVdXUxMTEsGVIks5joUF/IskK\ngO79ZNd/DFjds9yqrk+SNCKXLXDcXmAzsKN7f7Sn/+tJ7gXeD6wBnhm2SEkX1+S2x0ay3iM7bh7J\nelvXN+iTfAP4GHB1kqPAl5kN+D1J7gBeA24DqKqDSfYAh4DTwNaqOrNEtUuSBtA36Kvq9vPMuuE8\ny28Htg9TlCRp8XhnrCQ1zqCXpMYZ9JLUOINekhpn0EtS4wx6SWqcQS9JjTPoJalxBr0kNc6gl6TG\nGfSS1DiDXpIaZ9BLUuMMeklqnEEvSY0z6CWpcQa9JDXOoJekxhn0ktQ4g16SGmfQS1LjLhtmcJIj\nwJvAGeB0VU0luQr4V2ASOALcVlX/PVyZkqSFWowj+o9X1bqqmuqmtwH7q2oNsL+bliSNyFKcutkI\n7Orau4BblmAdkqQBDRv0BTyR5NkkW7q+5VV1vGu/ASyfa2CSLUmmk0zPzMwMWYYk6XyGOkcPfLSq\njiX5XWBfkh/1zqyqSlJzDayqncBOgKmpqTmXkSQNb6gj+qo61r2fBL4NrAdOJFkB0L2fHLZISdLC\nLTjok7w7yXvPtoFPAS8Ce4HN3WKbgUeHLVKStHDDnLpZDnw7ydnP+XpV/UeSHwB7ktwBvAbcNnyZ\nkqSFWnDQV9VPgA/N0f8z4IZhipIkLR7vjJWkxhn0ktQ4g16SGmfQS1LjDHpJatywd8aOhcltj426\nBEkaW00EvaQ2jOqg7ciOm0ey3ovFUzeS1DiDXpIaZ9BLUuMMeklqnEEvSY0z6CWpcQa9JDXOoJek\nxhn0ktQ4g16SGmfQS1LjDHpJapxBL0mNM+glqXEGvSQ1bsmeR59kA/CPwDLgX6pqx1KtS5KGMco/\nXnQxnoW/JEf0SZYB/wR8GlgL3J5k7VKsS5J0YUt16mY9cLiqflJV/wPsBjYu0bokSRewVKduVgKv\n90wfBf6gd4EkW4At3eRbSV5ewHquBn66oAovLutcPJdCjWCdi63ZOvMPQ63v9wdZaGR/M7aqdgI7\nh/mMJNNVNbVIJS0Z61w8l0KNYJ2LzTqHs1Snbo4Bq3umV3V9kqSLbKmC/gfAmiTXJPktYBOwd4nW\nJUm6gCU5dVNVp5P8DfBdZi+vfLiqDi7BqoY69XMRWefiuRRqBOtcbNY5hFTVqGuQJC0h74yVpMYZ\n9JLUuEsu6JP8eZKDSd5OMtXTP5nkV0kOdK9/Hsc6u3l3JTmc5OUkN46qxnMluTvJsZ5teNOoa+qV\nZEO3zQ4n2Tbqes4nyZEkL3TbcHrU9ZyV5OEkJ5O82NN3VZJ9SV7p3q8cZY1dTXPVOVb7ZpLVSZ5M\ncqj7//xzXf/YbU+4BIMeeBH4U+CpOeb9uKrWda+/vsh1nWvOOrtHQWwCrgM2APd3j4wYF/f1bMPH\nR13MWZfgYzU+3m3Dcbqm+hFm97le24D9VbUG2N9Nj9oj/GadMF775mngC1W1FvgwsLXbH8dxe156\nQV9VL1XVQu6ivaguUOdGYHdVnaqqV4HDzD4yQhfmYzWGVFVPAT8/p3sjsKtr7wJuuahFzeE8dY6V\nqjpeVc917TeBl5h9IsDYbU+4BIO+j2u6X+v+M8kfjrqY85jr8RArR1TLXO5M8nz36/NY/NrZGfft\n1quAJ5I82z3qY5wtr6rjXfsNYPkoi+ljLPfNJJPA9cDTjOn2HMugT/JEkhfneF3oCO448HtVtQ74\nO+DrSX57DOscqT41PwBcC6xjdnveM9JiL10f7fbDTzP7K/0fjbqgQdTstdbjer31WO6bSd4DfBP4\nfFX9snfeOG3PkT3r5kKq6o8XMOYUcKprP5vkx8AHgCX7MmwhdTLix0MMWnOSB4HvLHE583HJPFaj\nqo517yeTfJvZ005zfac0Dk4kWVFVx5OsAE6OuqC5VNWJs+1x2TeTXM5syH+tqr7VdY/l9hzLI/qF\nSDJx9kvNJNcCa4CfjLaqOe0FNiW5Isk1zNb5zIhrAqDbMc+6ldkvlMfFJfFYjSTvTvLes23gU4zX\ndjzXXmBz194MPDrCWs5r3PbNJAEeAl6qqnt7Zo3n9qyqS+rF7H/ko8wevZ8Avtv1/xlwEDgAPAf8\nyTjW2c37EvBj4GXg06Pepj11fRV4AXie2R12xahrOqe+m4D/6rbdl0Zdz3lqvBb4Yfc6OE51At9g\n9rTH/3b75h3A7zB7dcgrwBPAVWNa51jtm8BHmT0t83yXOQe6/XPstmdV+QgESWpdM6duJElzM+gl\nqXEGvSQ1zqCXpMYZ9JLUOINekhpn0EtS4/4Pvc2z+Ka9T7cAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x12c76a510>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAD8CAYAAAB5Pm/hAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nXpIaZ9BLUuMMeklqnEEvSY37P/eNa7qYOhamAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x12c8099d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "cutoff = 200\n",
    "x_full=x_full[Y_full < cutoff]\n",
    "T_full=T_full[Y_full < cutoff]\n",
    "true_Q_full=true_Q_full[Y_full < cutoff]\n",
    "Y_full = Y_full[Y_full < cutoff]\n",
    "\n",
    "plt.hist(Y_full); plt.title('Y')\n",
    "plt.figure()\n",
    "plt.title('T')\n",
    "plt.hist(T_full)\n",
    "Q = true_Q_full\n",
    "plt.figure()\n",
    "plt.hist(true_Q_full)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Reserve half for out of sample evaluation\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "n = 1000\n",
    "x_test = x_full[n:]; Y_test = Y_full[n:]; T_test = T_full[n:]; true_Q = true_Q_full[n:]\n",
    "x = x_full[0:n]; Y = Y_full[0:n];T = T_full[0:n]; true_Q = true_Q_full[0:n]\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Optimal treatment tends to bounds"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "\n",
    "# import pickle\n",
    "\n",
    "# pickle.dump(x_full, open(str(datetime.datetime.now().strftime(\"%Y-%m-%d_%H-%M\")) + 'x.p', 'wb'))\n",
    "# pickle.dump(T_full, open(str(datetime.datetime.now().strftime(\"%Y-%m-%d_%H-%M\")) + 'T.p', 'wb'))\n",
    "# pickle.dump(Y_full, open(str(datetime.datetime.now().strftime(\"%Y-%m-%d_%H-%M\")) + 'Y.p', 'wb'))\n",
    "# pickle.dump(true_Q_full, open(str(datetime.datetime.now().strftime(\"%Y-%m-%d_%H-%M\")) + 'Q.p', 'wb'))\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXoAAAD8CAYAAAB5Pm/hAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x120a82b90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYgAAAEKCAYAAAAIO8L1AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzs3VmMZNl54Pf/uTdu7PuSEbnvtWXW0t3F7maTzaZESpQt\nYSQYnjEFjE0MDAiwBdgvHkN6mAcbFjB+MDyAYRmmxw+EBwNBwFgQofHI0NASRDbZS1V115pZue8Z\n+77Hvff4ITNr6cpiV7E7Kqs6zw8oZGTUjbrnRqPji3u+c75PSClRFEVRlM/STnoAiqIoystJBQhF\nURTlWCpAKIqiKMdSAUJRFEU5lgoQiqIoyrFUgFAURVGOpQKEoiiKciwVIBRFUZRjqQChKIqiHMtx\n0gP4IuLxuJyYmDjpYSiKorxSrl+/npdSJj7vuFc6QExMTHDt2rWTHoaiKMorRQix+SzH9XWKSQix\nIYS4LYT4VAhx7fC5qBDib4QQy4c/I48c/8dCiBUhxH0hxPf6OTZFURTll3sROYhfk1JekVJePfz9\nj4CfSClngZ8c/o4Q4gLwfWAO+C3gT4UQ+gsYn6IoinKMk0hS/y7wo8PHPwJ+75Hn/0xK2ZFSrgMr\nwJsnMD5FURSF/gcICfx7IcR1IcQfHD6XlFLuHz5OA8nDx8PA9iOv3Tl8TlEURTkB/U5Sf1NKuSuE\nGAD+Rgix+OhfSimlEOK5GlIcBpo/ABgbG/vyRqooiqI8pq93EFLK3cOfWeAvOJgyygghBgEOf2YP\nD98FRh95+cjhc5/9N38opbwqpbyaSHzuKi1FURTlV9S3ACGE8AkhAkePgd8E7gA/Bn5weNgPgL88\nfPxj4PtCCJcQYhKYBT7q1/gURVGUX66fU0xJ4C+EEEfn+ddSyr8WQnwM/LkQ4j8HNoF/BCClvCuE\n+HPgHmACfyiltPoxMMvsUdrfIzI4jO54pbeCKIqi9E3fPh2llGvA5WOeLwDfecpr/gT4k36N6YjZ\n7VLJZkBK4mMT/T6doijKK+lU1mKShpu89FJIZ+g0Gyc9HEVRlJfSqQwQPcum5o6QafTIbW0g5XMt\npFIURTkVTmWACLgNxhJB6p4YxWKFWiF30kNSFEV56ZzKAAEwFfcRiMXJtAWFnW0ss3fSQ1IURXmp\nnNoA4dA1ZgcC2JFBspUmxd2dkx6SoijKS+XUBgiAVMhNPBqiqPkpZNK0G/WTHpKiKMpL41QHCIBz\nqQDOaJJMwyK/qRLWiqIoR059gPC5HIwnAjQ9cYrlCtVc9vNfpCiKcgqc+gABMBn3E4jFSXccFHa3\nMXsqYa0oiqICBKBrgjNJPzKSIldpU9zd/vwXKYqi9NFGvkG52T3RMagAcWgg6CYRDVJw+ClkMrTq\ntZMekqIop1Sp0WUlWydb65zoOFSAeMS5VABXNEm2aZNXO6wVRTkBli25t1/F69SZTvhPdCwqQDzC\n63QwkQjQ9CUolqoHBf0URVFeoJVsnVbX4sJQEF0TJzoWFSA+YyLmIxiNkukZBwnr7snOASqKcnqU\nGl22i01Go17CXudJD0cFiM86SFgHkOEUuWqbws7WSQ9JUZRT4NGppZmBk51aOnIqA0TX6nJr/xY9\n6/jlrImAi2QsSMkRpJjL0axWXvAIFUU5bY6mls4PnvzU0pFTGSDKtTK3N29zc/fmU485mwzgjA6Q\nadjktzaRtv0CR6goymny6NRSxHfyU0tHTmWASIQSjAfHWU4vk60fv3Pa49SZTARo+RMUy1XK2fQL\nHqWiKKfByzi1dORUBgghBG9MvYFTOLm2cQ3LPr719UTMRygaJWs6Ke7u0uue7JpkRVG+el7GqaUj\npzJAAHjcHq4MX6FUL7GQWTj2GE0TnEkFIDJIttahsK0S1oqifHle1qmlI6c2QABMpiYZ9A1yd+cu\nlfbxiei430UqFqBshCjl8yphrSjKl8KyJQv7VTwv4dTSkVMdIDRN4+rkVbDh442Pn7pz+kwygCuS\nIN043GGtEtaKonxBK9k6za7FhZdwaunIqQ4QAEF/kPnUPJlyhpX8yrHHuA2dqYEAnUCSQrlGKb3/\ngkepKMpXycs+tXTk1AcIgHMj54i5Y3y69Sn1zvFd5caiXoLRMFnLTXFvl16n/YJHqSjKV8GrMLV0\nRAUIQNd1rk5cxeyZ3Ni+cewxmiY4lwoiIimy9Q757c0XPEpFUb4KVnMv/9TSkVMbICzTfOz3eDjO\n2fhZtvPbbJWOX60U9TkZigWoOCOU8gXqxcKLGKqiKF8R5WaXrUKTkajnpZ5aOnIqA0SrXmPz9idP\nrEiaH5snYAS4tnGNrnV8kb6ZAT+uaIJMWyO/tYFlqu5ziqJ8PsuW3Ns7nFo64TLez+pUBgiX14vD\ncJLbWMe2Hm6SczqdXB2/SrvT5pPtT459rdvQmUkE6IZSFGpt8ltqqklRlM/36NSSQ381Pnr7Pkoh\nhC6E+EQI8VeHv0eFEH8jhFg+/Bl55Ng/FkKsCCHuCyG+168xaZrOwMQ0Zq/7RLXWwdggk+FJVjOr\nZGrH94MYjXqIRoLktQDlfJ5GudSvoSqK8hXwqk0tHXkRYey/Bh7dqvxHwE+klLPATw5/RwhxAfg+\nMAf8FvCnQgi9X4Ny+/2Ekymq+exjU01CCK5MXMGtuflo4yNM23zitUIIzg8GcUQSZFqS3NbGY3ci\niqIoR17FqaUjfQ0QQogR4LeBf/nI078L/Ojw8Y+A33vk+T+TUnaklOvACvBmP8cXGRrGcHnIbaw/\nlrT2uD28PvI61UaVO3t3jn2t3+U4KOYXSFGqNlXfCEVRjnU0tXT+FZpaOtLv0f4L4L8FHt16nJRS\nHu00SwPJw8fDwPYjx+0cPvcYIcQfCCGuCSGu5XK5LzS4g6mmqWOnmsaSY4wERljYW6DUPH4KaSLm\nIxQOkhM+ipkMrVr1C41HUZSvlkenlqKv0NTSkb4FCCHE7wBZKeX1px0jD2pbHF/f4umv+aGU8qqU\n8moikfiiwzycahqkVsjRrJQfPK9pGq9PvI4DBx9tfIQtnyyvoWkHU01aOEm2JcltrmM/pTKsoiin\ny6s8tXSkn3cQ3wD+gRBiA/gz4NeFEP8KyAghBgEOfx41ZNgFRh95/cjhc30XHRrG6faQ3Xx8qing\nC3Bx8CL5ap6l7NKxrw15DMbjfhq+AYqVOqW9FzJkRVFecke1ll7FqaUjfRu1lPKPpZQjUsoJDpLP\n/5+U8h8DPwZ+cHjYD4C/PHz8Y+D7QgiXEGISmAU+6tPY2Cu3sO2DmxehaSQmprB6vSemmmaHZkl4\nEtzavvXUMhxTCT/BcIis9FJM79NuHH+coiinQ67WeVBr6VWcWjpyEmHtnwO/IYRYBr57+DtSyrvA\nnwP3gL8G/lBK2Zf5mlKzx72dLdbyDz/I3T4/4dTBVNOjy1aPynDYls3Hm8dXfNUPp5pEOEm2YZLb\nWFMVXxXllGr3LO7tV/G7Hcy+5LWWPs8LCRBSyr+TUv7O4eOClPI7UspZKeV3pZTFR477EynltJTy\nrJTy3/VrPH6jhtSW2UivUGk+3AkdHRzG6fGS29p4bKopGopyLnGOveIeG8WNY//NiM/JSMxPzZug\nVKlTSu/1a/iKorykpDxoH2rbkvnhENpLXmvp87yaE2NfUJUQ+85BCtYed3b2sR6Zaho4nGr6bDG+\n+bF5wq4wNzZv0DaPr+Q6O+DHF46QMV0U93bpNJt9vxZFUV4eW8UmxXqX2aQfv8tx0sP5wk5lgOg0\nGmznbdKah0JjkfvphxvlXF4fkcEh6sXHd0g7HA7eGHuDbrfLja3jK746dI3zg0FkJHUw1bS59tQm\nRIqifLVU2z1Wc3UGgi5GIt6THs6X4lQGiKTfy5zDQb7pJqtJdnJL5OudB38fSQ0dTDVtrj9WjC8V\nSzEdmWYjt8Fe9fgppLjfxVDUT9Udo1yuUsmk+349iqKcLNOyubNTwTj8kvhVcSoDhMNq8W5rmRHT\nYq/rIW/nuLuzS9c8SCwfTDVNY5nmE8X4Lk9cxqN7+Hj9Y3rW8ZVczyQDeCIxMj0nhd1tem3VXEhR\nvsruZ2o0uxbzQyGMV3RJ63G+OlfyHMxqA2Nrj3eraxhdgz3clFr3ubf3cErJ5fUSGRymXipQLxUf\nPu9ycXX0Ko1W46nNhZwOjXOpAFY4Rb7RI7u5rqaaFOUrKl1ps19uMxH3vVKF+J7FqQwQWjCB9J9l\nOJfhjU6WastDVuhkivfZK7ceHBdJDeLy+J7o+zAyMMJUeIrVzCrb5e3jTkEy6CYZ8VNyRqiUSlRz\n2WOPUxTl1dXqWiykq4S8BlNx30kP50t3KgOEcGg4Z69iyyivZdaYMrvsdtzkZInFvS1a3YPtF0cb\n6GzLemyqSQjBa5Ov4Xf4+Xj9Y1q91rHnOZsK4I7ESXcdFHa26HU7xx6nKMqrR0rJnb2DBS7zQ6/+\nktbjnNoAYQz5Mc58C6Nu825lBX/PYE96KHdWubNbeDAldDDVNHQw1fRIi1GXy8VbE2/R6XSeuoHO\nbejMJgOYoRT5eof85saLukRFUfpsLd+g0uxxPhXE4+xbZ4ITdSoDhFWp0L59DT0RQKTeIJ4r8vX2\nPrWWm7TQyVcW2Cw83MMQTg09mGoyew+nmpKxJOcS59jOb7NWXDv2XMNhD7FwgKIzQrlYpFbI9/36\nFEXpr1Kjy0a+wWDYTSrkPunh9M2pDBCaz4cwDMzsKs7ZOaRjiLn0OhesFnstDzkarGbWqbUPgoEQ\ngoHJKWzbJr+18di/dXH8IhFXhBsbN55aq+nCYBBnKE66Lchvbz4WZBRFebX0LJs7exU8hs7ZZOCk\nh9NXpzJACIcD99wcmCayuY8x9y1oOfhGYYm45WC766Lc3eT2Tu5BQT+nx0tkaJhGufjYVJPD4eCt\nybewTZsPNj44tiy4x6kzMxCgGxqkWGtT2FZ9rBXlVbWwX6Vr2swNh17ZKq3P6qt9dU8hpU3XUcI5\nOYFVLqB7bfTxrxPK1/hmc4tez8OudFCq3WM5W3vwunByEJfXR+4zU02xcIz51DyZUob72fvHnnM0\n6iEaDpB3BCnlc48tnVUU5dWwU2qSrXaYTvgJeYy+nqvarWKdcH+ZUxkgTLNCrb5EN9rBEY3Qy2zg\nGBtHBqaZ3t/islUl0/KSEV22MqsUG13gaKppGmnb5DfXH/s3z42cI+lNcmvr1rEd6B70sQ7FybQE\n+c11NdWkKK+QRsdkOVMn6ncyHutvKY2W2eJW7hb3S8d/4XxRTmWAaBLgVslJvr4BMwmEw8AubWCc\n/zr0/LydWWZECrbbLkr2Lnd29ulZB1NHTreH6NAIjcrjext0XefNqTfRpMYH6x8cG/n9LgdTA0Ha\noUGKjTa5jeMT24qivFxsW3J7t4KmCS4MBhGif0tabWlzN38XgMnQZN/O8yxOZYCgUae5a7KSbVBt\nLWOcnUC2mgiqOGa+iafY5t36JrrlYdsyqDYXWdx/WNAvlEzhCYTIb2/SbT1c7RT0B7kydIVircjt\n/dvHnno86iUU9JN3RKmUilSymb5frqIoX8xKrk69bXJhMIjb6O+S1uXSMvVenXPRc3gcnr6e6/Oc\nygARCoe5MJSiUvCyWq7QMvYxRoaxCvsYqSjE5xnZ3eGqWaLY9rKPyX5hmUz1oKbS0VSTputk1lYe\n60M9PTTNiH+Ehd0FsvUnd09rmmB+OIQjFCXTNSjsbD0WZBRFebnk6x22Cgfd4RIBV1/PlW6k2W/s\nMxYcI+6J9/Vcz+JUBgi7WcJfW2XM5SST1titZTCHNPSAn152Hee5K0gR5+ruClNItltuilaWe7s7\ntHsHwcBhGAxMTtNttyjsPCy3oWkaX5v+Gi7h4sO1D+nZT+YZ/C4HswMBOsEUhWaPzPqq6kCnKC+h\nds/i7t6L6Q5X79ZZKi0RdoWZDJ7s1NKRUxkg2pZGpdEmKYv4bT9re22KjQ302SGEtJGNfYwL30Kv\nmnyrsorH9rBlOqi1l7m7V3qwa9obDBFODlLNZR7rHeH1eHlj9A2qzSqfbH9y7BhGIh7iYR8lV4Jq\ntUZxb/eFXLuiKM/mqDucZdt97w5n2iZ3C3dxaA4uxC70NcfxPE5lgPAGQrhGL9Hu9jjjqmM1Qyxl\nylStNYzZSex6Bd0n0EavMrCb4eu9LNVugF3bIl9aYqf0sPZSdGgEl9dHdmPtsVpLY8kxpsJTrKRX\n2K08+eEvxEGyyxUKk7E8FPf3aNWqL+T6FUX5fBuFg+5wZ5KBvneHu1+8T9tsMxebw6m/PBVhT2WA\n2Cg3+Yuck6JnkE6zxjm3RbngZb1coO0v4RgYwMxvY0zMYLtHubi9yjk67LR95O0CS/ub1DsHPauF\nppGcnAEpya6tPri7EELw+uTr+HQfH61/dGybUrehcz4VwAqlyLcl2fW1x3phK4pyMvL1DqvZOqmQ\nu+/d4bZr2+RaOaZCU4Rcob6e63mdygAxFHAT8jn5SIzSNoI4OiWGHS5292C/uosc86C5XNjlTZxz\n30Br6rybWyGCi82uQbW7xp2d/INd1obbTXxsgnajRmn/4d2Cy+XizYk3abfbXNu8duxYBoJuhqM+\nqr4klXrziVIeiqK8WM2uyZ3dCn63o+/d4SqdCmvlNeKeOKPB0b6e61dxKgOEIW1+U9RAE1x3TdOV\nGkMU8VgBVveaFNtrOM6OIrsdoIY+801C6SLvtfdpmUG2TUm5tsjSI7usA7E4/mic0memigbjg5yJ\nnWEzt8l6Yf2Y0cCZpJ9AMEBWBCnn89SKqqCfopwEy5bc2jlY0n55JIzex7xD1+pyt3AXt8PN2ejZ\nvp3niziVAcJuNPDs7vENWaEg3ax6Z2j3TM44a3TrIZYyJepiG+fEGHYljyMVR0bOMr25yhWapDs+\nCrLGTm6NdOXh1FFibALD5SKzvvpYg6HLk5eJuCJc37h+bEE/h64xNxSCUJxsVyO/uaF6RyjKCVjY\nr1Jvm8wPh/pawltKyb3CPUzbZC42h6H1t2zHr+pUBoiu7qGoDzKcLXFWb7MgQ+Q8g/TaTc64upTy\nHtbLWTqxFno4jJXfxHn+DYQZ4Ot790lhsNZyUbe2ube3R+MwH6HpOsmpGWzTJLfx8G7B4XDw5uSb\nmKbJR5sfHds7IuQ1mEz4aQVSlJrdx/IZiqL033axSbrSZnrAT9zf3/0OG9UNyp0ys+FZ/M7+Lp/9\nIk5lgHD7DAKTQ9REmDeL+yR0m2tilJYRwtWrkNSd7OxaZGrbMBNF6Bo00zguvIsn3+C9xhZSBlnt\n6LQ697m1U8CyjxoM+YgOj9KolB7bJR0Px5kfmGe/uM/93PH1VabiPiIhP3kjSrVSoZzZfyHvh6Kc\ndqVGl6VMjUTAxUSf6ywVWgU2q5sM+gYZ9A8+9bjFRotc92TrtZ3KACGEIDEawD0+SrXu5tfraYSA\na8YUHWkwTBGnGWR5r0q5vYpxZhy71UD3G4jBy4xvbvENUaXYC7Nr9qjUFriffpiPCA0k8QbDFHa2\n6DQf7pI+P3aehCfBzc2blNvlY8c1NxTECEbJmC6KO9t0mo0X8p4oymnV7lnc3j3o73BhqL91ltpm\nm4XiAj7Dx0x45qnHbbU6bLW6lHuqmuuJELpgYCqMPjKGLFm8axYp42LZM0PHlJwxqrRrYZayeRrO\nfRxDg1jFPZwz55GOAa6s32XekGy2gxSsKnv5VfbKB/sjhBAkJqbQdJ3s+sNSHLqu89bUWwhb8IvV\nX2DaTy5p9TodnEkF6AWTFNo2mbXVx0p5KIry5TkqwmfZkkujYYw+9newpc3dwl2klMzF5tC143Mc\npZ7JYqNN3Olgxtvfqa7P07d3QwjhFkJ8JIS4KYS4K4T47w6fjwoh/kYIsXz4M/LIa/5YCLEihLgv\nhPhev8bWbvTYWTjoxxCfimHGR0jma1wQDZYIknMPYXU7zBhdCjk3m6U05tBBJzqrtIUx/y20quDd\n9D2SDjcrHRd1e5vFvb0HXegeK8WxvfXg3KFAiNeHX6dYK3Jj+8ax4xsOexgI+yh7ElRr9cdKeSiK\n8uVZytaoNHtcGAr2fTPcSnmFWrfG+dh5vMbx01hty+ZmrYlX17jo9574jup+3kF0gF+XUl4GrgC/\nJYR4G/gj4CdSylngJ4e/I4S4AHwfmAN+C/hTIURflhFIw2K1vcTueh5vyElwfICGK8ZrxRwDwuSa\nNkzDCOE1aySEk829Dtn6BmI2CbaJoI42+038e0W+29pBl2GWW9qDfIR5WBr8oBTHENV89rEGQVND\nU0yHp1lJr7BR3Dh2jOcHg7gDITLSRznzeCkPRVG+uL1yi51ii/GYl2Swv32lM40Me/U9RgOjTy3C\nZ0vJp7UmloQrAS9GH5fYPqu+BQh54GhNp3H4RwK/C/zo8PkfAb93+Ph3gT+TUnaklOvACvBmP8Zm\nYSIjbZbq90lvVIgN+fCOj1Dvevm12j46cN2YpoPBiChjdEMs71eoWusYU2PY9RLOwSgkLjK4vMK7\nWo2KHWW716VeX2DxkXxEdGj4oAvd5vqDpauapvH61OuEjBAfr39MpV15YoxOh8bcUBA7NECuAznV\nYEhRvjTVdo/FdJWIz8lMn4vwNXoN7pfuE3KFfml/h4VGm6ppcdHvwe/ob0nxZ9XXHIQQQhdCfApk\ngb+RUn4IJKWUR8tz0kDy8PEw8Ohcys7hc186v9PPXPICItxjpbhKKdNkYDKEY3Qcq6LxbidHRRrc\n98zQtSRnHFUalSDLuSztYAk9HsPMbeG6cAmMQeaXb3HJCVvdEHmrTLqwyk7pIDktNI3k1JOlOJxO\nJ+/MvIM0Je+vvn9s1deY38VYzE/Nl6Rcb5PbVA2GFOWL6po2t7YrOHWdi8Ohvk7jmLbJ3fxdHOKg\nCJ8mjv/I3Wp12G13mfS4GHC9PHsi+hogpJSWlPIKMAK8KYSY/8zfSw7uKp6ZEOIPhBDXhBDXcrnc\nrzy2pC/JbGqKuqfAWnqbdqNHYjKKPThGvNRi3q6xQpC0ewTb7DGtdchmDLbKu8hxN5rbjVncwHnl\n22hNB9/cucWoy89y00vN3OT+/i7Vw3yE4XITH3+yFEckGOGNkTco1UpPzUfMDPgJBP3kHCGqxce7\n2CmK8nyklNzZq9C1LC6OhHA6+rtOZ6m0RMtscT52Hpd+fML5ZUpKf9YLWcUkpSwDf8tBbiEjhBgE\nOPx59Im3CzxajGTk8LnP/ls/lFJelVJeTSQSv9p4TJvuTo1J7wQjqRT7bLK1lcHh1IiMx2n7klwu\nF0nJDjfEIA1HCL/dJCadrO02yTXX0c+OICwLenkc57+NZ7fCr1XXMRwRllo6rc4it7YLD1qVBqJx\nArEEpf3dx0pxTA5OMhOZYTW9ylrhyTsEXRNcHA6hBWNkeg7yO5t0260njlMU5fOt5uoU613OpoKE\nPP39pr5T2yHbzDIRmiDijhx7zMuWlP6sfq5iSgghwoePPcBvAIvAj4EfHB72A+AvDx//GPi+EMIl\nhJgEZoGP+jE2adpY9R693ToXIheIp0Ksdu+zt14gFPfgmxiiZgd5r5bGwOZjxxRt4WSUCo5OmKW9\nwkE+4swUdr2KHjLQRt4gtbTCt2WZGgm2Oz3q9Xvc3X2YX4iPjmO4PI+V4tA0jdcmXyPsDHN94/qx\n+yMCboOZgQDtQIpio3cwVaUaDCnKc8lW22zkmwxHPAyH+9vKs9AqsFpeJe6JMxYYO/aYlzEp/Vn9\nvIMYBP5WCHEL+JiDHMRfAf8c+A0hxDLw3cPfkVLeBf4cuAf8NfCHUsq+bADQ3A70lAe7ZUK2y6WB\ni/jjTu7XFslsVUiMBnCNj9FrOHmvlaUmDBbd03QRzGgVqpUQy7kMLU8GY2QIq5jGmJlB+iY4d/8m\nb7hstswouV6JfHmNrcJBPuKgFMc0tmmSXV97PB8x/Q5Y8PPVnx+bjxiLeomFvBTdcarVKoVdtfRV\nUZ5Vo2Nyd79KyGtwNhno77l6De4V7uE1vJyLnnvqXcG9euulS0p/Vj9XMd2SUr4mpbwkpZyXUv73\nh88XpJTfkVLOSim/K6UsPvKaP5FSTkspz0op/12/xtZsNtnM7dDxS6xKB0/VwVzqPBwmrauFNgOT\nYcTQOMFKj0tmmTURYt81grBMJkWHTMbBVnmP3qBED4cxcxs4L7+D6Hp5e+1TJlw+Vjp+Kp1NltO7\nVJoHH/our4/46DjNapni3s6DMYWD4V+ajzjYZR3CGQiTxU85k6ZeLPTrLVKUrwzTsrm5U0YXh9O1\nfa7Qejt/G13TuZS4hEM7fm/FVqvDXqfHlPflSkp/1qncSe106nRa22RaOUyPoJdtkrCjzKYmqbrz\nrO1vY/ZsYhMxzMgwc+Uqw1aDT7QhqkaUoGwSw2Btu0Omug4zETSngazvYcy9hzvb4NeKy3iMCMsd\nB53OAjd3cnTNg2mhYGKAQCxBOb332P6Go3zEyv7KsfkIt6FzLhWkF0hQ6GnkNtdVPkJRfgkpJXf3\nqrS6FheHQ7iN/n1Tt6XNnfwdulaX+fj8MyWlpz0vV1L6s05lgGjXG9itMtXSChmriHQKunt1Jtxj\njKRS7LHB1lYad8DAP56k4YjyzWoOl93lmjFJW7gZE1Wc3QD3t8sUWis4zo8je100Zw9t6m0GVjb4\ntlWgoSVYa3dpNe5yZ7f8YFopMTaBy+cnu7764ENeCMHrU68TdUWfmo9IhdyMRH1UvCkqbZPM6rIq\nxaEoT7FRaJKrdZgdCBDx9beV52JxkWq3yvnoeYLO4xsNvexJ6c86lQHCF4wQDl3EYbcpl1bJ6zUQ\nYO42uBA+TzwZYrW9xN56keigD/f4CL2Oh2+3cjSlzl33DG1L46yzRq8ZYXE3R9VexzkzgV0v4xga\ngPAss/du8TWnzb6dINMuUayssVF4uD8iNTWL0DTSK0sPWo0ahsHXp7/+S/MRZ5MBQgEvOVeceq1B\nbnPjRb59ivJKyFTbD9qGjvW5QutGZYNsM8tkaJKE9/jVla9CUvqzTmWAwJYkPAMEfVOIboVieZOS\nq4U0JaS7zCfm8CUMlqqLFHbrDIwH0UfG8dUkVzp5tgiw453A7Jmc1+tUykGWsmmavjyOoRR2cR/j\nwmsIO8SmZSlVAAAgAElEQVTVpetMerysWGEq7XXWMruUGl0AHE4nyakZep0O2Y2Hm+iO8hHlWpnr\nW9efGL6mCS6NhDF8QTIiSDWfV/sjFOUR5WaXu3sVwl6DC31uG5ptZtmobpD0JhkPjj/1uFchKf1Z\npzJASNOGlsWAawK/e5Bec59CdZeGp4vd6OEpHSStZajDUn6ZZqVHbDKClRjlTLXFqFnjJglKniGE\n2WFa75LJONgo7WAOSfRQELu6h/Pie7iLXd7bX8DvjLHUddJq3+PWTo6OeTAt5AkEiY+O0ayUKe3v\nPRjj5OAk05FpVtOrrBZWn7gGj1NnbihEzx8jZzrIb2/SbjzZrU5RTptm1+TmTgW3Q+fSSLivSelK\np8JicZGQK/RL24a+KknpzzqVAUJzO3AO+3HYkPKdx+uM0Kisk22k6XhtrFKbeCfMzOAkVVeetb0t\nNE0jNJag5Rngm7UiAavFh4zQcMfw202S6GzsdNmvrMNsFKHrIMs4Zt8hvr7De509mlqStZZJu3GP\nW9tl7MMmQ6GB1OEmup0HSevH8hHr1ym3nsxHJAIuJhM+6v4UlY5NZnXlwVSVopxGXdPm062D/1eu\njIX7ulO6bba5m7+LU3cyH5t/ahmN4iuUlP6sUxkgkBLdzmIkvThtQcp/CZ/TR620SqaTxXRBL9Ng\nwjHKcCrJrr3B5uY+gZgb7/gQbcvPdxtZhDT5yDFNR/eTooa35ztIWjdX0M+OILtttIgXkvPM3LnN\n20aPtEiSbhUoVVZYSD/cUR0fG8fl8ZFdX6XXPuhzbRgG78y8g7AFP187Ph8xnfATC3rIuweoN1uP\nTVUpymli2ZJbO2XapsXlkRBeZ//Kd5u2ye38bSxpcTF+EUM//q6gZdnc+hWT0qVGCdM62S98pzNA\nNAuQvo2jt4oeceOVBgnfJTwOjUpxlYxdwtbB2mtyIXSeWCrIamuJvc0i8dEAxugYouXkvWaWugU3\nXbP0hJMprYxsRVjYzVITWxjTY8h6CefULJoxwGv3rjHr9rAqozS66+wXttksHHSM0zSd5MwsQgj2\nV5awrYMpqFAgxNWRq0/NRwghmB8O4fL5yWhhasUS5fTeE8cpyleZlJJ7e1XKzR7zQyHC3v6tWJJS\nslBYoNlrMhefw2f4jj3OlpKbh0np154zKV1pVvj7O3/Ph8sfflnD/pV8boAQQmhCiNeEEL8thPh1\nIcTAixhYX/nitN3jUNnGMPbRfAYh4SPsnsOQnYOVTY7aQRXBvQ4XE3N4Ew6WKouU0g0SE2EYmiDc\nkLzZKbBnuVhyz2BJnbNahWolyGJ6l1agiD4Qx6qkMebfwVOzeXfrFmF3nHtdP932PZb30+RqB2XA\nDaeLgakZep022Y2H+yAmBid+aT7C5TioSikDUbK2i+LeLs3qkyXEFeWraiVbJ1NtM5v0M9Dn3g6r\n5VUK7QIzkRmi7uhTj7vzSFLa9xxJ6Va3xc8WfgY2zI3OfRlD/pU9NUAIIaaFED/koC/DPwd+H/gv\ngX8vhPhACPFPhHjKpNtLrlXrsl9OUO4OIArLOIM1hEsnZsQIemegV6FY3qDobCK7Fu6cxlzqPHao\nw1JuhW7bIjoeoZcYY6La4my3yH0ZZNczAbbNGZrksi7W8luYowI94EN2CujnvklsO8O3a5vYriEW\nOwZW+ya3d7PUOwe3kt5giNjIGI1y8UHl12fJRxzVtW/5k5Q6kF1fxex2X+TbqignYrvYZLPQZCTq\nYTx2/Lf5L8tefY+d+g7D/mGG/U/vRrDcaJPu9Jjxup8rKW1aJu8vvE+73WHSewHq/d278Xl+2Qf8\n/wD8K2BaSvk9KeU/llL+x1LKS8A/AELAf/oiBvll8wSc+D0dStYo9V4YkbuDM9pD1wVJ1xgB9xC9\nZoZifZe6t4dd7xFvBJkZnKDizLG6s4Xh0gmPxeiEhnmjXmOoW+NTcbCyyY3FsG2xtdtjr7KBOBND\naKB5QAxfZuLuPd4VdUqOEVZbJmbjNp9uFR/stA4nU/gjMYp7OzQrB8Hg0XzE+6vv07OezEeMx3wM\nhLwUvUlqzQ6ZtRVV1E/5SsvVOixlasQDrr7XWCq1SyyXlom6o8yEZ5563Farw3qrw4jbydRzlO+W\nUvKL+7+gXC8z4T1DQEQw3Ce7HPapAUJK+ftSyr+Xx2Q8pZRZKeW/kFL+6LjXvvQaeeL2bTwyT96a\notXzoOU/xUiALgQp9zn8riiNyga5RpqOx8IstJlghJHBJDv2BpvrafwRN77RJE3XAN9uFvF1m3yk\nj9IwwsREm0DPzdJ2mVxr9SBp3WmiDw8h/GPM3fiAt32wpw+z1yrTrN3l9u7DlU2JiUmcHi+ZR5LW\noUCIr41+jUq9wgebH2DLJz/8LwwF8fl9ZJ0x6qqon/IVVm33uLNbwe9yMD8U7Ouu5Gavyd3CXbyG\nlwuxC089V7bTe7Bi6bzv+aa6rq9fJ1PKMOyeJKqniPgceFwvaYA4IoT4h0KIwOHjfyaE+L+FEK/3\nf2h95I0hgoMkXJsYVoWsOUPX1NFLn2LEdZxCkPRcxOf0Uyutku7k6LkkVqbNBd9Z4qnAg57W0SEv\n7pEhWnaY7zZz0OvxsTFDR/czKmporSALWxkqYgtjYhiaJYzzr6PLAG98+nPmQkFWxBCl1i7F8uqD\ndqWappOangUgvfowaT2eGud8/Dxb2S1u799+4tIMXePiSAjhC5GR3oOifo/0w1aUr4J2z+LTrTKG\nrnF5NIxD799sd8/qcTt/G4FgPj7/1AJ85Z7JrXqLoEPncuD5Viwt7i6ysb9BzJkkZYwTcOu4OhZm\nsf0lXcWv5lne1X8mpawJIb4JfAf4P4H/rb/D6i+zZ/LxnTKmM0DStYawOmS6ZzC7Fo76bRwRA69m\nkPBcxOPQqJZWydpFbCFhv8vF2By+AQeL9UUymxUS4wGco6OIno93mxmqpuSmawZbdzNDmVY9xOLe\nDq1IBT0exW5kcV5+D1fF4p2FXzAWSrAo4zTqi+wVdh+UBzdcbpJTM3TbbXKb68BBPuLS5CVGA6Pc\n27nHRnHjiesLug3OpoJ0/IdF/TbWHtyFKMqrrmfZfLJVxpKSK2Phvhfgu1u4S8fqMB+fx+M4vo9E\nw7L4pNrEJQSvBb3ozxEcNvOb3N24S8gRYcw4i8+p4TVtNJ+BkepvTuXzPEuAOKoE99vAD6WU/xY4\n2czJF7S/uc3y3QX+/sM0uNwkjWUsyybbm8Vu1XH0FtH9DoKal8hnVzbZNu6M4NLARYyYxWJ5geJe\ng4GJENrIOOGWg6udLLu2iyX3NGgG07JKoeBmObuBOaaje73IbgHH/HsE9kp8a+tTAoERFmSYXuMG\ny+k0+frByiZvMER0eIR6qUA5fdDKW9d13pp5i5AjxEdrH5Fv5J+4xuGwh6GIl4o3RbVtkl5TRf2U\nV59tS27tVGh2TS6PhPG7+rfXAQ5ahpY7Zc5GzhJyhY49pmPbfFI9+FL3esiLS3v2u5lcNceN5Ru4\nNS+Tzgt4nToBKdHdDpwjAcQJ12t6livZFUL878B/Avw/QgjXM77upZUcDzM17yWzv8PPP63i9DgY\n0JfpmgZZcxqaBQx9Hd1jEHVECblnoVejWDlY2WR3LIIlFxdS57BCbRYL96mXO8THw9ipcaZqNmfa\nBe7bAXbdY7iEYNTssLNnsVNeg7MxhKYhHB30mbdJrmzwXmUdyzPKQs+DWb/OrZ38g5VNkdQQ/kiM\nwu72g+WrLpeLb8x+A93WeX/lfRrdxhPXeS4VJBjwknXGqdfq5Lc2X+j7rChftoV0lVKjy/nBINE+\nV2ddr6yTbqQZD46T9CWPPcaSkk+qTVqWzWtBLz792e9mqq0qP1/8OZqtM+2+iNtpEBQgDB3nWACh\nn3wxv2f5oP9HwP8LfO+wt3QU+Kd9HVWfGUaE829dYWTWy9bmOh/f7+H1SqJimZbpp9AbR9T3cXrS\n6E6dAdcIQdfI4cqmHeruHlaty0A9wtnBaVreMkvpVcyuRWQsSjc+yhvNJoOdCp+IJEVXkqBmE+lq\nLG+VybaWcVwYA6uHHvYghi4xcfMW37RLlF1jrHctzOonj61sSoxP4nS7yayt0Os8TFp/ffLrtNot\n3l97H9N+fNelrgkujYRw+A+K+lVyOap5VdRPeTWt5ersl9tMJXwM9bll6HZ1m83qJoO+QSZDk8ce\nIw83wlVNi8sBL2Hj2e9m2r02P7v3M6yuzYznEj6Hm4gm0HXtIDj0sUTI83iWUQwC/1ZKuSyE+Dbw\nD+lTr+gXRQiBT5zhynuvkxgyWL6/xK0tnaCvQ1iuUbNilM0UorqGM1BF1zWS7ln8rhiN6hbZ5v6D\nmk0j3SSTQ6MUjH1WdrbQDUFwNE7bN8y3WlX87RofO8ZpOiMMai1cbS+LW1lK5jLG2Qlkp4FjZBgt\nNMmF6x9x1WWz6xhnv12iWb3zYGWTpuukps8AkH6kB8RQYojXBl8jV85xbevaE2U2vE4Hc4NBer4Y\nBcsgv7VJp/nk3YaivMz2Ky3Wcg0Gw26mEv7+nqu+z2pllYQ3wZnImacet9Bok++anPM9/16Hny38\njGarxaTnAn6Hn4hDoGsC51gAzXl4F1LZgc7JFuB8lgDxbwBLCDED/BAYBf51X0fVZ2a5Qf3vPsFV\nGeWN37hCMGpz5+Y9lvMewt4qfnOTUm+IuhVDqyzgDHdwahpJ1zx+I0CttEamk6PnBSvfZoYJRocO\nGg1tbO3jDbnwjSbp6jF+rV3CarW5ZkxjOgKMUadTDXBve4+asYMxPYFsVjDOzGHoEd64/jPOB32s\n6qNUmlsUSg9XNhluN8nJabqtFtm1hzWXZkdmORM5w2p6lYXswhPXOxB0Mx73UfUlqXRs0qvLWOaT\n+ygU5WVUbHS5t1cl4nNyPtX/0t1LpSWi7ijno+efuhJprdlhp91lwuNi7DkK8Ekp+XD5Q0qVEmOu\nGSKOKFGHQBcCYySA5j68C6nsQvo2FJ/sLPkiPUuAsKWUJvAfAf+LlPKfcnBX8coSuk2vU6L90W0C\n3TO8+R9cxu3vcu2ju+w0/MT9BdzdPfK9cdq2H716GyNs4XUYxN0X8Ro6leIy6W4Oyy2wMi3Ou86Q\nGoyyYS6xtZEmlPTiGhtBt4N8u5Oj2rW56Z5BGF5mqFIp+7m7u0kzkMUYHUI2ixhzX8Pb1Pj67fcZ\nDg+wIAZp1++wW9h5sLLJGwoTHx2nUSmR3z7IKWiaxmvTrzHoG+Tm5k12KjtPXPNMwk804CHvTtJo\ntkmvLKtNdMpLr94xubVTxut0cGmkv/2kC60Ci8VFgq4gc7G5p1Zn3Wt3WWm2SbkMZp9jIxzAJxuf\nsJffI+UcI+EcIuLQcAhwDvvRfYd3IbXMQXDwxiA5/0Uv6wt5lgDRE0L8PvCfAX91+NyrU9D8GG1p\nUQlkqDfzNH9xixBzvPVb8+hGg5//fIFcN8iAP42jmyfTm6FrOXA0b6P7JUGHh6hrDpcwqZRWSFt5\nbCfIdId5/0Fhv+X2IvtbeWLDfhwjY4S6Ht7o5Ng1XSy5JtF1JzN2lXzezf39DdqJJno8hmyX0efe\nIZSp8o3VawRCo9wjRrf8McvpzIOVTaGBJOHkINVc5sHKJofDwdszb+PX/Xyw+sET5Tg07aio38NN\ndNnDpbOK8jJqdk1ubJbQhOC1sTBGH/c6VDoV7hbu4nF4mI/Po2vHJ5vzXZM79RZRw8G83/N8ex32\nF1nbWyPqSDLsmiSkC5wCjEEfeuAw4d4owP6n4A7RG7iIPOGWpM/yjv8T4OvAn0gp14UQk8D/1d9h\n9ZcuSzRcd2hF1mnWc7R+fo+Y9zJvfu88tl3mpz9bpC79JD1biE6FTO8sVk9imPfQvRoRI0LIfQHj\nsGVpmjJSl2j7JpfCcwSTLhbrC+T2qsTGgjA0zkRTY7aZ474Ms+cew6U7mDAb7KUFq9lVzHGBHgoC\nTRxn3mZofYd3MovYgUmW8dItf8CtnQKNw5VN0eHRw5VNW9SKB8tcvR4v35j5BtKU/HTlp3SszmPX\n7TYOivpZ7iA5PUStkH9Q70lRXiatrsX1zRISeH080te9DrVujVu5W7h0F5cTlzG047//1kyLm7Um\nfv1gI5z2HB/e24Vt7q7dxU+ICfc5ArrAowkcA14c4cMd160S7F4Hp49e8jJ31hfZ3j/ZSgifGyCk\nlPeA/wa4LYSYB3aklP9j30fWR01niHJsgrJ3m2Zok2YlTfuDZZLRN7j6nTO02wX+9v11urqPpHMd\nq9MmY55Bdls4uY/u1BhwDhByn0W3G5TLK2REFalJnGnJxegcrhgslu9RLbSIjh0sf32tbZJsFvhU\npCh5R/HrGiO9Lps7PTYLq9jTfnSPG+HV0EYuMXP3Hm83M5S9M2xbNmbpYz7dKtE1bYQQDExM4fYH\nyW2s06od9JaIhqK8Pf429Uad91ffx/rM3oeIz8m5wSBtT5QCHop7O9SLhZP4z6Aox2r3LG5slbBs\nyetj/d3r0Ow1uZW7hUNzcDlxGad+/NLZlmVzo9rEIQSvB5+vdHe+lufa8jWc0sOkdw6/JvDpAkfM\njRE/XI3VrsLOdXC46CWvcHdzicXyXYqcbBWEZym18W1gGfhfgT8FloQQ3+rzuPoq6okyOfa7tAZm\nqfi3aAa3aOV36X64ycjIm1z+1iS1apq/+8UOwu1kwLFKtyPJmbPQKeN0bqDpgoRriKBzCtGtUqqs\nkTfqSCnxZx3MJ84jw10W8ov02hbB4QhmZIR32028jTIfykFqnkFCDkj0JKtbdXarS4izCYQm0BMR\n9NgMc9ev8ZposeOeItsr0yjffLCySWgaqelZHE4X6dVluu0WAKPJUS4lL7Ff3OfGzo0nrn847GE8\n5qXqSVC1DbIba6pdqfJSaPcsbmyW6Fo2r41FCLj7N5vdNtvczN1EILicuIzbcXztpJ4tuVFtYErJ\n60Ev7ueY6qq1a/xi4RfQ05jxzBPQDAK6wBFy4xjwHhzUbcDOx6BpmIOvsbi9zkLpDsOeKhPek53N\nf5Yr/Z+A35RSviel/BbwPeB/7u+w+kuaNqlqlNHB36aZmKAS2KLh26KZ2ab38T5Ts28x99YQxeIO\nf/dxAadXEBUrNDsuivYkWjuD07ePQxck3eP43WPIToFiaZ2Cs45t2UQKXs6nZuj4qiyml9F0gX8k\ngeVJ8Z1eFa1V5wNtjKZrgIRuEeho3N8skm4soZ8fBruHPjaO0xXjjeu/4IzHw4oxTrWxST6/yr39\nKlJKdIeDwdkzB42Glu4/KPF9fvw8U6EplveXWcotPfEezAz4GQgeJq1NSK8s0et2njhOUV6Ujnlw\n59AxbV4fjRDy9O/DsWt1uZm7iSUtLiUu4TW8xx5nS8mntSZNy+ZKwEvgOfo6dHodfnbvZ3Q7Pabd\ncwQ0NyGHQA84MQZ9B/mLXgu2PwIk1tAbLO3ucK94myF3hdFMB23zZPu6PEuAMKSU949+kVIu8Yon\nqaVpYzd7DFUSDKZ+m2Z8iGp4k4Znk+buOtYnFc5ceoeZSwNk0pv8/HaDQMAmZC9T7YQoW8NojS2c\nvhy6Lkg5p/E7hzHbWQrVLUquFrJnM1iJMj04TsnIsLK3gdvvwDM6iKbH+I1uEbPZ4CPHJF1XjGHd\nxF13c28jTa63jOPsKJht9Jl5vD2Dtz/5GYOhARadw5j1m+zltrmfOVz+6nIzOHsWy+qRPuxGp2ka\nV2euknAnuLFxg3Qt/dh7IIRgbihIwOcm4x6g2Xn4WkV50Y7qK3V6NldGw4T6+M25Z/e4mbtJx+pw\nMX4Rv/P4fRVSSu7UW5R6JvMBD7HnaGHatbr8dOGn1BsNJpzniTiChB0Ch9fAOXxYQsPsHNw52Cb2\n0Bus7Ke5XbhJwlVkNNfGXYxj+Ka/rMv+lTxLgLgmhPiXQohvH/75P4Br/R5YP2luB86xIFiS0VqK\ngcHfoR5NUo1u0nBu0Vy/j7zbZv7NrzM6E2RrY41ryz0iwQ6+ziqlbpK6GERrrOH05DEcGin3WfzO\nFN36HsX6NlV3G7tlMtEYZGxokDTbrG/vEoy5MUaGccgwv94t0Gi2+dCYpueMMOpoo9V93NvYpaxv\n45geQcg2jtnXiJbbvH3vA3yhce45BpCVX7Cd2WM1dzA15PL6SE3N0mk1D/pASHnQQ2L2HdzCzfsr\n71NtVx97Hxy6xpXRMC6Pl6wrQaPeILO+onpaKy9Uz7K5sVmi2TW5NBIi0scSGpZtcSd/h5bZYj4+\n/9T6SlJKFg6b/sx63Qy6nn1MPavHTxd+SqlSZtx5hrgRexgcRg9LaFi9g+DQa2MPvc5qNs/N7CfE\nnUXG8x3chRhy2k8vfrKVD54lQPwXwD3gvzr8c+/wuVea5jkIEtKSTNRHiA39h9TDMarxTWrGNs37\ni7AoeO29t4kPO7m/sMSdbZ1EuI67tUG+NUxTH0RvreL05nHqgpTrPH5XnFZ1m1xth5qni2yYnOlO\nMDgYY9NaYWs7QzTlwzEyil+G+FY3T6XV44ZrFssZYlxr0K16ubOxTt2fRR8eQOgm+szrjO5keHPr\nDnb4LEuOGLL8Pmv7abaLD/dIJMYnaFbLD6q/+v5/9t48SJLzvM98Misz677vvu/umZ77wk0QgChK\nlCjSPCRrqSUl8ZAtUZbWG7H2riSLkkK7Dq9WZqy81i7XWluOAO2gJIomJR6iSACDATBHz9nT931U\n133fVXnsH9UYYAgMMANOE6TQv4iK7s7Kur7syie/733f32uz88jwI2gtjXMr52hpt3eZs8gmjvR4\nMMwOMpKXSr5AdnvzB3489vXOlKrpXNsqUG2pHO724HfcW13BvUg3dG5mb1JqljjgO/CG7ULnqw22\nGy0GrWYG76HWoa21OTd/jlw+R795lJASwSMJyBYJ88sWGrrWyVZqVdGjx9jIlriWvIJHydKfrWPJ\nuGHIhXkggN1558ZEPwjdTRZT0zCMPzYM40O7t39rGMabLlYLgtArCMIzgiDMCoIwIwjCb+xu9wmC\n8G1BEJZ2f3pf9Zj/WRCEZUEQFgRBeO/399HeXKJVwtzrxNAMhqoDeLp+krLXTTm0SVnapDY/i7Ru\n5/R7HsDpN7h+fZaVrJmQO4dS3yRV66YuhnchkcNsEgmbD2JXPNRK66SrO9SsKkJZY1IcJRT1stJc\nIL6Tw9/jROzuw6/ZeaiVJlXXuG4eBcXNoFGlWrAws7FCI1jFFPAimAXE3sOMLSxwOrtK2TPJssmF\nnjvLfCxFotjxZ3IFQngiXZSz6VsprEFvkNO9pymUC5xfe22jIbdVZrLLRcPiISPYKaYSlNL7nk37\n2ltpusH17QKleptDXW6Czr2Dg2EYzGXnyDfyjPvGCdqCd9x3rlJnq9Gi32pm9B6a/qiayosLL5LJ\nZehRhgnJXXhEAbN513xPNoGuQ+wK1AsYkSNsFetciV/GKWcYzDewZjwIg17MAwE8vpOYTHvrOfVm\neqOe1NOCINy40+0unlsF/kfDMA4CDwK/JgjCQeBfAt8xDGMU+M7u3+ze94+BSeAngH8vCMKet1MS\nbTLmXieoMFofwRF5LxW3jWpog7KxQXV6Bks8xIM/+SBWe4tLl2bZKVsJu1LI9W2S9V4apjCmxjKy\nNYfVJBNWDmOXnVQLq6QacRpWFTGvc0gewx22sFCZIx0v4u9xIHT306XaONFIE2sIzFlGkCxOBrQ6\nhZzMfGyJdq+ByeVA9NiRgqMcunqFk9U4Wc8hNiUbRvZZbm6lSJc73PZ39+LwBcjtbFPOdmokBqID\nHAwcZCuzxfWd668Zh5DLwnDIQcUaoKCbSW+u33KO3de+7rf0XTgUam0mu9yEXPfWfe1eZBgGC/kF\n0vU0w55hIvbIHfedr3bg0GdVGL8HOOiGzotLL5LKpuiWh4gqPXhMdODQ60Q0S2AYnSK4WgYjfJBY\nReNK/AoWKcVwvo4t7YJ+L+ahIB7/KQTV9La7HbzRDOKngfe/we0NZRhG3DCMK7u/l4E5oBv4APBy\nq9I/Bz64+/sHgP+6O2NZA5aBM/f6ge5WlXzj1lq7aJMx93UgMd6YwBr9cUouM9XoJhV1ndq1GzgL\nPTz4UycQpSovXpgn13YRcSSQa9sk6n00TEGk1jKyJY9dVggqR7DKNsqFFRKNJE2zhpyDw/YJbCGR\nueIs2UQZX48TuvoZaps53Eix2pRZNI9gNtvpaVVIpmAxsYg2YsVks2CKhpFcvRy98BLHmzlS7sPE\nTAp65iw3tjIUap0lpFD/IFani9T6KrVSEUEQODx4mD5nH3Pbc8yn5l8zJoMBO1GPlZw1REUXSa4u\n30qd3de+7pd03eBGrEiu0rHtjrj3Dg4AK4WVW7bdvc7eO+63UG2wWW/Ra1GYsN/9lbtu6Ly48CLJ\ndJKoNEiXubczc1BMKD0ORJvcgUNiGipJCE6QaMpc3rmMLCYZKdaxpp3Q58U8FMDtOwkNgytf/3u2\nLt/Ntfje6Y0AsWkYxsadbgDCXdaZC4IwABwHLgBhwzDiu3clgJeN1ruBV5cNbu9uu++ql1uktypk\ntiq3QULpdSJqAgdakyjRpyi7JKrRLcrNTWpTN/A2x3nwJ46h6wWeOzdHETdhZwKpskWy3t+BRHsJ\nSSngVCwElMNYTTLlwjKJVpqWomPLShx2HcAcNJgpzlDIVPF1uzC6BphoK4zV0sy3LKxahrFbHHTV\na2zvNFjLLGCMeRBlE3L/IGZbiBPnz3FQrbLtOkTSZKA9+39z5akPULbYEYaGCF+Y6liEryzRrNUw\nmUw8OPYgIUuIq+tXWcu91mrjYNSF12EhbQ5Ra+sklhf3jf32dd9kGAY3d4pkyk0mos49t+1eK66x\nXdmmx9FzR9tugMVqg416k16LwgHHvcHh/PJ54uk4EamPHksfbkG4NXMwOXaD26k5KMXAP0pStTG1\nNYUgxBkpVbCnrYg9PpShAB7/SYQmXP7G37Ny/SzpzUvf7xB8X3ojQDwjCMKvC4LQ9+qNgiAogiA8\nKW9ND3QAACAASURBVAjCnwOfeLMXEATBQccR9jcNw7gtjcbonJ3vKWVGEITPCIIwJQjCVDqdvpeH\n3pLVqeCN2KgUmqQ3yxh65y2Y7LuQUEUm1eNI0ScpugxqXRuUamvUL04TlCc58+OTtFpZvvvsLCXD\nTcSVQKzESNb7aZqCyNoiJqmAW7Hhk49gEUVKhSWSWgZV0nFmzRz2HkAJ6Mzkb1LMV/F2uzCiAxxV\nRfpraW6qTrYtg7isNkL1FqtbRbZLC4jjYRA1TIOjWBUvpy88x5igsZGykL9xnrYzwZXoGLWdBKZ/\n8k+IXL6KYDIRX+7USMiyzKPjj+KW3FxYucBOaee2sRFFgSM9Hqw2G0klSK3WILGy/LZPdff1oy/D\nMJjZKZEqNRkLO+nxvn7twf3SamGVjdIGEXuEYc+d00WXqg3W6016LAoT97CsZBgGF5cvEkvECJl6\n6bEM4hbAYtmFw8vme5klKGyAd4A0bqa2ptCJMVouY09aEKNB5JEA3sBJhKbA5W/8PWvTL9IdNDH5\nwGPf7zB8X3ojQPwEnXaj/0UQhJ3dYPMqnarqnwc+bxjGf3qjJxcEQaYDh6cNw/jy7uakIAjR3fuj\nwMvR0BgdK/GX1bO77TYZhvEFwzBOGYZxKhi8c6DpjVRX6yzpM5gDBtVii9TrQMKkikxqJxC7Hqfg\nVGl0b1Esr9J4aYYu9wkeeO+BDiSem6Wou4m64oiVGIl6Py2TH1lfRDKV8CoOfPJhzIJGsbBEwsih\nCjqutJlD3gNIAY2Z3AzlYh13lwsjMsgZVSBaTXNN9ZC0DuC1WPHVNBY2ssSbS0gHehFEHdPgODas\nnDn/DP3f/CZL8gjlbjv1IZGrXRM0mm3kf/W7REfGMTSN+PICuqZhsVh4bPwxbKKNF5ZeIF29HbSK\n1GkEL1hsJGUf1WKR9Ob6Wxrrfe0LOifT2XiJRLHBSMhBn3/v4GAYBkv5JTbLm0TtUca943c01Vup\nNVirN+m2KBywW+7afM8wDC6uXmQrsUVQ7KLXPIgHsFg72Uq34JBbg+wyuHvISWGmNqdoGTGGK2Xs\nSTNiOIA86scbOIXQFLn89b/rwCEkceqDH8fSM3GfRuWt6Y6AMAyjYRjGvzcM4xGgH3gKOGEYRr9h\nGJ82DOPqGz3x7vLTnwFzhmH88avu+iqvzDw+Afy3V23/x4IgmHcNAUfZo8ZEhmFQq9VYNRZQgjq1\nUovUxvdAoseBpJqYVM9A12PknC0a0U2KhRXqL8zR5T/DA+89SKuV4btnZylqLiLOOGI5RqIxSFvy\nIxsLSKYSPsWNVz6ErLcpFJZIiR1zP0/GwiHPBCa/ymxuhlqliTPiwggN8qiq469kuawHyFp6CZgt\nOMoGc2s7pLUV5AMDCCYdafgQbtXEQys36ErlWHRN0uySKfVLXOsap70dw2yzER4epVWvk1jp2Hw7\n7U7eNfouTLqJ5xefp9C43f3VYZY40u1GtbjImFyUMmnyiZ3XG8597etNNZ8o3+oGNxCw3/8XePpp\nGBjAEEUWHhom9jdP0+vsZdx3Zzis1pqs1Jp0mWUO3iMcLq9fZnNnE58Qpc8yglcUsNo6WZHiy0V+\nhU1Iz4MjTNHSy6X1S9T1bUYqeVwpBTEYRBkP4Q2eRmiKTP3Nt1i7eb4Dh3/0CSzdby8c4C57SxuG\n0d4NOhfefO9begT474EnBUG4tnt7H/CvgfcIgrAE/Nju3xiGMQN8iU6dxTeBXzMMY0/Keo2Wga/h\no16os6otIAdUauUWyY3SK5BwKCg9DhRN5pD+EEb0YfLuOo3IJqXcKo0X5nchcYj2LiRKuouIKw6l\nDiRUyYvMIiaxhN/sx60cQNIa5AtLJIQihsnAm7VxyD2O4Gszk5uh2Wh3IBEc4F1aG3sxw2XClCxd\nhMwSSlFgZm2LrLGCNNGPIBmYhg7j1yQezm7gT1WY8x1G74V8r4Xrxx5D0w1sLjfB/kHq5SKpjTUM\nw8Dj8vDo8KNobY2zi2df09fa7zAzFnZSs/rJYyUX26KSf3vNw/b1o6fFZJlYvs5AwLY33eCefho+\n8xn0zQ1m+8wkmhkG/tXnGf7b83d8yFqtyXKtQdQsM3kPtt2GYXB14ypr22t4CTNoHcUrgNXWqau6\nBYfcGiRnwB6i5Bji0toUFW2T4WoOV8qMyR/EPB7CEzy1C4dvsD7zMhx+EXN0lGysQq3UeuM3tMcS\nfpSrZk+dOmVMTb21ou5UKkUqmyKhJZDdMoPSGFpaxuqQCQ24bjUm0SotWtsVGmKTaeNZTPHzBHIO\nrOlBnIEhrA9PEitMceGb08iynyceG8clVkmUuxBcXUQsy0jtIi19nLbuINmIUVSXQXHhdg8TwYOo\nCuS8FabzCwh5M5P+SWRFpryTQ8uu8e3lDfTzUzwcexElYmXn0XdjnBnkyNAAPn0IdX4D45nnUb/9\nZWJuB88HBikHZI7mp9FP/QLRD36coz0eRFEgH4+R29nGFQgR6BtAEARiqRjn1s7hdDh5auIpzKbb\n89EXEmU2sxWC9Tguk0bX2AEs9r1t+7ivH30ZhsF8ogOHPr+NsbBzb15oYABtc4PZAStZt8RwrEFv\nug39/bC+/prd1+tNFqudhj+H77Gnw9XNqyxvLOMhxLBtAo8ANofSaRX6cje4zDJkl8ARpuIa4dLK\nZbLtVYbrGfxJBdETRDkYxRs6hdiSduFwke6I0oFDZJjMdoVKvoknZMUbuf8zLkEQLhuGcerN9vvh\n6Iz9NigUChEOhImaorSLbdbURcRgi3qlTWq9hP7qmUS3A4tuZlJ4HLXrFFlfmXp4k1J2ldrz03S7\nTvLQ+47Qbmf47vPzFHU7EccOemmHRHMETXGjCAtIQoWQuRuXNITYLlHYjUlokoEvZ+eQZwzD22Q2\nO4eqqtgjXkxzMZ74wp9gVEtcsI2hJit0ffvrCBfXubG6Tt60hjTWi/DEY5h+7AN0N1UeLGxh1ZzM\n/NSnkQ4rxGMLt8z9vNFuPJEuSpnUrY503aFuTvecplgu8vzy87T127OWxsIO/M6XM5sgsbSwn/66\nrzeUrhtMx4q7Mwf73sEBULc3mR7qwGFsaxcOAJuvdQTY2IVD+C3A4cbWDVY2V3DpAYatE3gwXguH\n1HwHDq5uqu4xrqxdI9NcZbCWxpuUEV0BlANhvKGTHTh87eusz1ykJ2rh1Ic/iTk8TGarAwdv2LYn\ncLgX3Y3d96+/utr5H5KCwSCRUKQDiUKbdXUJIdCgXmmTXHsVJJwdSNg0Mwf0x2lHTpDzFGmGNinl\nlqidmyZsPcZD7zuK2s7yzNk5irqtA4linHhjBM3sxCwuIgtVwpZenPIoolqhUFgmYeTRZAN/3sFB\n1wiap85MZg5d07B98T9hblh4Yv46TbOVi+Zx9KJO1/PfgZzE9ZU18vIG0mgP4pPvRvzN32bgPe/l\nzLtPIh59gjl7P3L9KltbyywmO75N/u7eWx3pXobEUPcQx6PHSeVTvLT20m19JARB4HC3G4fNStIa\noanp7CzO0242ftCHbF8/AlI1natbhVvZSiOhvZtttvU2Nx4epOiQOLBepyv7qoubvtsSMNmoN1mo\nNggp9w6Hm7GbLG4sYte8jNgPdODgNKP078LBMCBxE/Jr4OmnbB/g6uo1Es0VBppp/GkZye5HORjB\nGz6F2JKZ+trfsD7bgcPJD/0y5uAAqc0ylUITb8SGJ7y3WV53o7uZQYSBS4IgfEkQhJ+429qHHxX5\n/X6i4ShRKYqaV1lXlyHYoFltk1wtomud9M4OJJw4DTsTPEErepKsr0Kza5tibp76i9OELcd56H3H\nULUczzw/R0mzEbHH0As7JJpjaIodxbSATJWQ0oVb3m049NyXiH/iZ1B/8imCH/nnHDy/guaqMpud\nw0gksDbK2HUHjy/doGJ1MKUMQ6JCt2wg5MQOJJQtpJFuTC47Qs8BRre3OTl3iZZ7jCVHH1L1Iqsb\nq6zumvv5e/pwhyIUU4lb3ktjvWMcCh5iK73Fpa1Lt1lyyLvGfpLZQlyJ0GypxJcWUNv7NRL7ekVN\ntdMJrlBrMdnt2tNspZbW4lrqGpVPf5zJJIQL6it32mzwh39468/NXTgEFZkjTus9dYOb3Zllfm0e\nm+pmzHYQrwE2txml39WpkNZ1iF+H4hb4hilaurm8doV4a4XBRopASsJk9aEciuIJn0RsKUx97Wus\nz07R02Xl5Ec+iTnYT3KjRK3Uwt9lxxN6++EAd+fF9Nt0Mor+DPhFYEkQhP9VEIS314f2Psrn89EV\n6SIqR9EKGhutZYxgjWZN7cwkXoaEqwMJl+HgoPAkrfAJst4mjd4dSoUF6i9cIyId4aH3HUfV8jxz\nbo6SbiXs2EHN75Bsj2OYbR1ICDVC5iiuqSLyN79G0VEjrlRp76wS+qMvceDCCm1XlbkTPQi6hrlR\nw92y8MjaLHmHh6nQaQx7hC5ZQMzCjeUV8uZtpKEuJJ8bMTrGwY01ji9dpewaY80RxVR6kcX1zVvm\nfoHeflzBMIVknGxsC1EUOTR4iHHfOCvxFa7Frt3m7GpVTBzv82AoZpKWEI1Gk/jSPJqqvu647uud\npXpL4/J6nlpL42ivh6h774rgmlqTa6lrHVfWj/46gT/6007MQRA6P7/wBfjYxwDYarSYrzYIKBJH\n7xEOC4kFZldnsakuxqwH8RgCNs/LcNj1VopfhXIcAmPkpSBTa1Nk1Q2Gmmn8KQHJ7MN8uBtP5CSm\ntpmpr36N9dnL9HTbOPnhT6H4+0iul6iX2/i77bgCb6//0qt1t1lMBp2q5wQdjyUv8JeCIPybPXxv\nP1B5vV66u7rpUrrQChprrWXUYIVmTSWxVkK7DRIOHLqNQ8YTaOHT5Lwa9d4disVZqi9dI8wkD73v\nBKqW57vPz1HWrUTsO7RzcRKtCQyzDbM4j6QXCT/9t7hXRGSLRGnQTkKp0a5kCP+/f8eEc5DWRx5m\ndtyLqLUwN2uEqgKPbi+QffdTXBKH0O1RIrIJIQ/Xl1fIW3eQBiKYAj5MwWGOLC9wZH2GnGeSmDOI\nWDjLzOr27ZAIhCgkdsjtxBBFkePDxxlwDTAfm2c2NXvbOLksMsd7PaiShZQSpF6r3aqv2Nc7V+VG\nm6mN3G4nOA+BPXRlrat1rqau0tSaHA0e7biyfuxjnYC0rnd+7sJhu9FirlInIEscu8c+0rPxWaZX\npjG3nYxZJ/EKJuzeXTgopldcWSspCB0kY7i5uHaRorbJUC2FPyEgKwGUIz14IicwtS1MfeUrrM9d\npqfbzskPfQrF10NytUij0ibY68Dl78DB0A1q61na+bc31nc3MYjfEAThMvBvgBeAw4Zh/FPgJPDh\nPX5/P1C53e4OJMxdUIDN1irtQIlWXSW5+mpImFF6ndgEK4e0xyDwIFkf1PtTlEozVM9fI6wf5KH3\nnUTTCx1IGBbCthjtXIJk+wC61Y1ZWkQqrhOoing2LB1IDNmJW+q0d1aJln2MP/VzND/x48wf78Wk\ntZB9XsIf/SSPj3ZRKBS5YBrEcHQRkSRMOZ3rS0sU7Amk/gimSAiTb4Dj8zMc3Fwg6TlC0unDlHuO\nmZUt1jNVBEEg0DeA0x8kH98mH49hMpk4M3qGiDXCjY0brGRXbhsnj03hSI+HpmwjYw5Sr1RIrCzu\nV1u/Q1Wotbi8kQfg1IAPj23v+jnU2jWupq6i6ipHQ0fv2M8BOqmss5U6flnimOvu4WAYBlc3rzK7\nMotNdTNhO4RXELH7XgWHl/s5fP2v4B/9ComJ01z65PupXf8Gw5U0vqQJyboLh+gJTG0rU3/916wv\nXKOn286pD30K2ddNYq1Es6YS7HPi8HaquA3NoLqSYmH6BbY3X2us+YPU3bRI8gEfetl/6WUZhqEL\ngvDTe/O23j65XC4EQUDYEYgX4my51+n290PWTXK1SHjQjUkSMdllzP0uhE2BSfUh5vwSOfE8XtKw\nrmNc0AmdPspD7xN46etTfPf5OZ589AAh2zapLCT8Y0Tsayi+NOR1AlUnrFsp9NcpDTswKnaiikZX\nNYDx1M+ycOYB5mseJnxjGC0dfybL49UYZ3M65739nHEIhCox0vkm15aWODYK7r4QGAai2ubkzA3a\nksxy1xHMXMNfeI65lUfRjT6Ggg6C/YMYhkFuZxtBEPFEojwy/gjPzj3LpdVLyCaZPs8rQb+Aw8yh\nbjfT25BRDChmSK4uEx4evafg375+tJUuN7kZK2KWRI73ebEqe2fAXGlVbvWQPh46jl1+/QwfwzCY\nrzbYarSImGUOOe5+WUk39E6F9M4WbiPIsGUcN2D3WTH371p2q60OHP7uaxi/8yfsmJxcPeBDM7YY\n+eomztMfRDz9IMpEBE/4eAcOX/5r1hev09Pj4NRHPoXkjpJcLdJqaAT7ndjdnRmXoeqUlmIsL1yi\n0igT0H7IO8oZhvG73wuHV903d//f0tsvp9NJT3cPUUsUsSgSa23Q8OVpNTQSq0U0tXOlLFoklAEX\nFrOVyfYZFPejZAMWqoMZKtWbVC5cJtQa4eH3nUHTi3z33BxV3ULItk07u0W8Noj68X+GYlpFshYI\n1MG7YcdssVD++I8T/85f0vyln6P7PZ9i9LO/Q/3mt1jILWG2mpCDfnzOHh4X69RyeS7Qh+bsISgr\niFmV60uLFF1ppN4QYk8vVkeQB6avMJCIse47SdbtRi6dZWF5jeVUGUEQCA0M4fD6ycY2KSQTmM1m\nHh17FIfJwfmV8yQrydvGKeyycKDLRVVykFN8VAp5Uuur+x3p3iGKF+vc2C5gU0ycHNhbOBSbRa6l\nryEKIsdCx+4IB90wuFF5xbL78D3Aoa23eWHxBbZ2tvAbUUbN43gEAYff2rkYlE3QbsDWeWhV0P/4\ni2wpbqYOeMBTYcRQcdV8SNdvYJnsxhs5swuHL3fg0Ovk1Ec+jeSKkljpwCHU73oFDm2N/Pwmi7Pn\nqTTKBL1HcUQH7tcQviW9Y+sgtDc5iTkcDnp7eumydmEqmoi1Nqn5srSbGomVIlp7FxKKCXO/C7Pd\nykHtJHbHY+SCTspDOSqNWcoXLhGsD/DwTz2Aphf5zguzVA1LJwU2u0b8oV+i9dnfR3HlkG15Ag4f\nvic+i0UQKX/zz4lXtmi0i/TGZEb+3dNUp7/JQnoJ2SyhBPz4XH28W2xQz2e5YPSiOvsJKZZdSCxQ\ncGeQ+kKYegewmj08MD1FXybBhvckOY8fpXKOxaUVFpO7kBgcxu7xkd3eoJhKYrfZeWz0MRRD4fnF\n58nVb6+k7vZYGQ07KMsuCoqHSi5Ddut1ryf29Q9Im9kaM7ESHpvCyX4vZmnv4JBv5Lmevo4syhwP\nHccmv36GT1s3uFyqkWy2GbNbmLDffSprQ21wdvYs8WSCEL0MykN4TAKOgGUXDiK0ah04qE30rpOs\nNwwuTziRXSWGVA1HK4RoErBszOANPdCpc/irv+zAoc/JqQ9/CpMrTHy1SLulER50YXN1luP0pkb6\n5jKLc+epqw3CgdN0nxjDG3176yDekZXUZVXjcqnKpMNKUHnj5ui1Wo3t7W3i9TiqSyVs7sKRDSIp\nJsKDLuTdqyZDN2jHKrRKdZaZodA8hyedx7nuxiZN4Dx2gqxzmxe+fh5RdPLEu47jNZVI5r3oniHC\n/hrm8jztihNVGSb3G58lb1+njo5jrUqkImOV3Wz0qqz8n7+P9bklxv7DVyCdpdnVS/n97+bZYw9g\n8vo4I8RRyuukF26iL13myNQ13Hkf6kc/htoTpqrVeOnYg6yHuhko3yCYT1FXTjE0OsZExAmGQXJ1\nmWoxT7BvEFcwRLaQ5dnFZzGZTTw18RRO8+2FT8upCuuZKr52DrdawhvtxtfVc8/HZl8//FpOlVnP\n1Ai5zBzqct9yHdgLZeoZZrOzWCUrR4JHXlPl/7Iams6VUo2KpnHYab2nHtLVVpVzc+coFkr0iEN0\nm6J4rBIWvwWle7eHdLMC2xdB19G6TrCWynH9f/oYdnWTgbaIpR1AkjSs8U08JR/61DWmvvJXbK7M\n0dPn4dSHP4nJESK+WkRXDcKDLiy7hn56QyU5Pc/a6jQaOuHQabqO9ePwvuqzPv00/NZvdYr/+vo6\nKby7gfi3orutpDZ97nOfe8sv8nbrC1/4wuc+85nP3PPjDAMybZWNegtJEPDIdw7FyLKM3W7HqBvU\ny3WKpjyyU0Cu2qjmm1jsMpJsQhAERJeCoIKn4aMh2slZswhKFlOuhJZo4PWMEzzqZnNxldX1DJ5I\nH92eMvV8iZIWQvEGsQgxqBUx/+evIpStqJ42NZ9Cq97A3GgTrNqRGiXSf/v/kZVlfEUVuVRCWtig\nO+Rg1e0lbgkRnr2B69t/S72tk4i48VRj2P7+EsbIQSSvn0hslYrdznpgAlmq4asukMxJqGYXQacF\nh9dHs16jmEogKQpeXxCf2cdqepWt4hZdni7M0iv/wD67QkvTSbZkJENFL2URRRMWx95V0O7rByvD\nMJiLl9nK1en2Wpnscu0pHDZLmyzkF3AoDo4Gj6KYXv+kX1U1pko1mobBcZedsPmNL/perUK9wLnZ\nc1SKVQakcbqFEF6HjCVsR+l2IIgCNIodOABq1wmWEyluZK7h8dToX9jE2vQiSyqOrQ08qzqt3/19\nXpy+zs7WMn2DPk5++FOI9iCJ1SK6bhAZegUOWrXNzrVpNtZvoosi0e6H6DnRd2vZCbjlNUWm0x2S\nYhG++U0YGIAjR97S2P7e7/1e/HOf+9wX3my/d+QSk8UkcsplI6jILFQbzFXqb7hubrFY6O3tJeqI\nYilbSDbilP1JDMEgvlKkWuy0+hQEAaXLgRKyM2I6SFh8N/lQhNxQkToLlK5exJcO866feReS0uDc\n85dYTMlEgnWU0gLpJJTtR5HdbeSeNl61TWDdhVUTqAw6SLpUajad3v92jQMFK5q+zcyQQksSUSoF\n7F97jqdkASOX5eILazSyECw3EZsy0xPDlAZ05L/4ItLAKHazm4dvXGRoa4UNx2ES/m4c+lVW529y\nM1bEQCAyNIrV6Sa9sUY5myESiPDI4CO0Gi2eWXjmNQ6w42EnEbeFtOynItrJxjYpZfZ7W/9DkK4b\n3NguslOoMxi0cyDq2rNkBN3QmcvOsVpcJWQLcSx4DNn0+if9QlvlYrGKjsFpt52Acjd5Nx2lK2nO\n3jxLtVhnSJqky/DhdStYe5woUXvn89XzsHUJBBPt6EnmYztMZ67hNWL0BQawnP4ZJJ8Nx+o67nKA\n6h/+b5xtVknH1xid6OHkR34FweonsVLE0A0iQ27Mu4Z+WrnF5uUpNjdmQVTo7nuEnuM9WJ3fA8Lf\n+i2o1W7fVqt1tu+x3pGAqLVrXE5eok+u02dV2Gq0uFquoep3CYmKhWQtTta9hahAaqNMMf1KvrIc\ntKFEHQzIY3TxbsrBHvKDJariAoVrF3DG3Dzxoffi8ApMTV3m6kqbUKiFtTpPNt6gaD2K/LPvQXZl\ncBstgltubG2BypiHhLNNtZUl3A5wuORHbm4zMyBQlyUsOxvYg4M8aRYQKkUuuieolRT8pRZCQ+HG\nwWFK3ixyfwRpYBirxcPDN6cYWV9kw3aIhL8PJzdYm7/Oja18BxIjo7dal5ZzGbpD3Tw6+CitRotn\nF56lUH8FEoIgcDDqIuiykFL81EUr6Y01Krnsnh7Pfe2t2prO1a086XKT8YiT4b1wZN3Vy9XRyVqS\nAdcAB/0HMYmvH99It9pMlWrIosBptx3XPcRBdoo7vPD0/0Xrj/6Usd/4U6L/6v/AO30Ba78babcW\ngUqqAweTTDN8nNntDWbzNwjqMXoTNcw1O/L7n8D1//xHXLMVcl/5a55tlCkVEkyenODIBz8FsrsD\nBwwiw27M1g7A2vk6KxdfIra9jElx0j38GN3HordmFrfpdTyl3nD7fdQ7EhCCICAKIjczN7HrGSbs\nFjItlUulKg3tzrn8ZrOZ3t5eupxduGouctUsO84VBLtGLl4lG3ulhanktWDuddJvHqHXeJJycIDc\nUIWGvETpxkuY5wye+MD7CfbYmJm5wbkbZXxBcDQXye8Uyb73N5E++bMo0RoumgRqI9jf+0GqAYOk\nz6DcSuJvOThcjmKv7zDbr1Ps60Jx2rAFh3gyk0ISDS66JqgVO5AQGwrTp49S8uWRhrqRB0axWH08\nOHOVsdVZNi3jxENDeEyzbCxc4/pWHhCJjIxhcThJra1SyeeIBqM8NvQYalPlmYVnyNfzt8ZIFDu+\nTT6HmYQSpCFYOn2xi/fiFL+vHxbVWy9bZ7Q51O2m17d3FhCVVoXLyctU2hUm/ZMMuAfuuO92o8XV\nUg2HSeS0247ddPdwWMuu8dJ//QLa3zzHgVmNiOjCuz6N9bd/FdPX/qKzU3alUwSn2GmEjjKztcp8\ncZqItk13oo656UA+PIB76CgOxxjxF5/n7N98hXazwPFHTnHgfb+Ihpn4aqcffHTYg7Jr6NfKVFm6\n+ALJxAZmi5fu0YfpPhy8df9r9D2eUm+6/T7qHRmDkEWZsC1MXa2zXdlGoc2AI0CsqZJotvHJEmbx\n9dlpMplwOBzoTR3qUGgVqNtKuBQnzTy0Gho2l9KBkNmEyS5jr9uQmn7S9hKqNY1craDGSphrLoYe\nPU21EWdjeYNUUWJ8xI2pnqZUlmkf/TEcH3wA8T0PIL3nA0jR06gXvkXNqdPUNKiU8OoOfJqLspgg\n9ksfxRIawm53Ilrc9DzzLbY9LrYsIQKlEm5Zp3HyDAmzhjOsYPf2IWgSplqDSGyFhiiwFpxAtIqE\nGsskEg2qkouIx47T56NRLlNMJVCsVnz+TkxiM7PJemGdsDuMVe5ceYmCQMhpJldTyRkKNq1Oo5DB\n4nQhK3tXYbuv+6tspcmVzTyqbnC0x0PIdfftOO9VmXqGG5kbiILI0eBRvJY7+4Ou1jq+Sn5Z4oTL\njnKH7+rraS4xx/Wl60hf/AYHl00ERDPuwibm2E3EWgkuT8HPPQn5dXBGqHkPcHNjkZXKHN3NuiH6\npgAAIABJREFUGNFkGzNulCPD+AZOYlW6WPu7v+XiuWeRZZVTP/YU/Q+/n3pVI7lewiSJRIbdyOYO\nwBrxIgtT58jnEtgdUaIHHiA67kOSbwecbuhc2biBruu4+oY6MYdX+57ZbPD5z+95DOIdCQgAURAJ\n2UIIgsB2ZZtGu8C4K0S6bbDVaOGSTNhMd4aE0+kEDcS6SLFRpGIt4rDa0Ism6uUWNpcZ0SQgyCZE\np4KtYcHc9JO2lWlaM5i1BupmCjErM/jAQ2jWEhuLa2wn24yMhrCoGUoFaNqGcLibiI04siojmfrR\n5qdoeqFhElArFdyahcCP/TzVn3uc7VocRbXgPHAYkztEz4UX2bFKbIRG8b/7CTxjAzSKVRLVHFaf\ngSs8CJqEUG8S3VmjaeisBifALhFpLpNKNKiYnEQ8dlx+/y4k4kiygj8YJmAJsJHZYCO/QdAdvJWC\nKIoCQaeZTFUlhwWbWqORy2Bzu5Hkvau03df90Xqmyly8hE2RONHvxW3du2P26mD0sdCxO6axGobB\n3G7/6KhZ5ojThukug+SGYXB9+zrza/OYG04O/pcpAga4cuuYd2YQtTaYAGcVfvanwT9KydrL9PoM\nm7V5emvbhNM6suTFfHQEX98ZFMPN7Fe+xPWrUzhcEmd+8gNEjz9OMdMgG6titklEB91IignDMKhv\n51i4/ALFYga3Z5DIweNERryYvuc809bavLh4ic34DjbBTuSJJzsB6cuXoVTqeE19/vPfVxbT3QLi\nHZnm+r3K1DPMZecwCSZGfAdYbshUNI0DDiu9ljt/MQzDIJ1Ok86mSbaT4Iao1IM158MkiYQHXbem\njUZbp7VVIlNIsiKexVRbxZ8UUTasOP1HsD9wnJX8Za4/P4Ni9vDkI5NY9DqZahAlMkTYvg3pHVqN\nELWpVdLf+TPqJDFEB673fozwUx/GaKjMSSvEK0W66Cdii6A1VerJDZ5vVClZPJx0VPFVV8lXs9Rd\nIsODffTYJ9DnErRWl2gXU1waHGNu4jg9xOhLL5Ap9xIZPc7J4TAmoZMCWysWbqWyZvIZzi6fRTfp\nPD7+OEH7K73CG+3OEkWz0aSrmcBigujYAcy2Hw63yn3dLlXTmYuXSZYahF0WDna57vokfK/SDZ2F\n3ALJWpKQLcS4d/yO8QbNMLhRrpNutRmwmhmz3/1sRjd0Lq9fZm1rHUfTy4Q8hvd3/xfs8xeRU8sI\nGKAA3SaIhjG+cYlMU2ZmZ4ZMa43ecoxAVkC2+bEcHsEbPY1QM7jy5S+xvraIL+Ti9Ps/iqt3guxO\nhXKuid2tEOh1IooChm5QWU+xeO1FKvUSPv844YmDBPucnSypV6narHJu/iLFUoXD0UkmhgZfs8/9\n0N2mue4DYle1do3pzDRNrcmAe5iM7iHTVum3mhmzmd8wY6NQKJBIJkg1U7SdbQLmEO58BAGRUL8T\nq6MDGUPTaW1XyOfSrHAetb2IP93AtmzBZpvAcfIUSfMGF759EQwbjz54hIC5SarkQQqPEnZnEVNr\ntCo+mrY+Uu0YdWMNVbLgcA4RMgWQm7DAKluNHAEtSq+tF03VaCa3OdcoUVCcHHPrhOsr5IoJag7o\n7e1i0H8IYz5Ha2medjHOVN8YcweP0yUm6U/PkSl1ERw6zunRLmSTsJvZlL7VmS5XzHF26SyaSeOx\nsccIO8KvjG1LZWo9j95u0tVIIpsgOjq+35Xuh0y1lsr1rSK1lspIyEG/f++KtFpai5uZm5RaJQbd\ng/S7+u+4b1s3uFqqUlA1xu0W+q13v0yp6irnV84Ti+3gaYUZNw3i9ZqxXXsW+dd/uZMNZBcgagLZ\njPYv/i2xp97LTGqGphGnp7CDJwOyO4zt8Cje8GnUbJGLX/4S8Z0tugbCnHz/z2P2d5NaL9OotvGE\nrHjCNgShA4fC8hbL1y9Qb9fxBw8TPjiCv8vxmnNKtprjhdlLtJsaZwZO0NsTecvj+2baB8RbUFtv\nM5edI9fIda6+lV5iTbXTYMRpxfQGkKjVasRiMdL1NFVbFY/VS6DUi9A24e9x4PTtGnHpBu14lWqu\nyFLzElVxBm+miHNZwcIAzkMPUO6t88LXn6FZFzl18giDPoNk1o4QHCUSqCNlFmgV7LRtA2S0CmVj\nnpYAducQQSWEtSGxbGyw2U7jaPkZsAxi6AbNTIyXagUyspODXhO9jRXy+R1q1ibh7gij0SMIC2Va\nSwuo2W0u9w0zc/A0USXNYGqOdD5EYOg4p688j/l3fotsPkthcAD7pz5N+Fc/S75U4Lml51BFlcfG\nHiPifOUfvNJUmVrPIWptos0kEjqRkTGsTtd9O377euvKVDqeSgCHu93499CNtdKqMJ2Zpq23OeA7\nQNAWvOO+dU3nSqlKTdM57LQRuYcah1q7xvml86STGYJaNyNCL96AFWu/C8lj7tQX/Ot/Ce0EuEK0\nfu0PWHv0EWazM5iFJD25LI68gBSMYj84hid0mvrGOue/+tdkc2mGDgxy5H0/j2jxkFwrorZ1Aj2O\nV5nu6WTn1liZmaKpNQlGTxKdHHzdXg9buRiXFq4hGQqPjJ3BH3BjqCqYTHuSTrwPiLcowzBYK62x\nWdrEbXbjsI2y1tBwSyaOuWx3DF4DtFottre3yVaz5JU8druDaG0QoSG/presmmvQiJdYq94kI1/F\nWUjiWZFQKiFcYw/QmnTywre/RSnXYvLAJIcHzKTSZnTfKOEuE+b8TVopaJsHyF+4SOGl/0CzUsDa\nsOH7yKdxP/bTbKhbrOpx5LqLYesogg6tbIIr9RwbhoVRv5VRdYNibpO6XMMZ8XOg/wjyskpzYQ41\ns8WVnkGmD54haskykpkjdTaJ+z9/jVMrN3C06hTtVjLBAJb/4Z8T+dXPUqqWObt0lpbQ4tGxR4k6\no7c+c7He5spmHpOmEm0lMekqkeFRbG7PfT2G+7o3rWWqrKQqOCwSR3s8e+qp9HJltCRKHA4cxqnc\nuZAy21K5Ua6hA8ecNvz3UuNQTXNx4SKVQo0ebYA+MYwvYsMy4MZklztW3YnpTh8HZ5Sqc4il7VWW\nyvP49ATRTBlrRUbq6cE+PorHf4LizRu8+M2vU60VmTg+yYEf/yiaZia5UUIAQgOvqo5uaaSmF1hb\nvI6KRrj7DNFDfbcuFF+t+fgiN1YXcMtuHjlwCofThl6vU78xjRTwYx6+/4Z9+4D4PpWqpZjPzSOJ\nEmHXBGtNCbMgcNxlw/EG+daaprGzs0OmlCFDBsWp0K0OYqrYcHjMBHoct9YU9VqbxnaZWHGZmOkC\nSm2HwIaAknDg6j+NeHqAF5//FulYkcH+YR445CWdMqG6Rgn0ubHXZmh/9VnUL36XYkUg11WkYdWR\n8xreX/hn+B/7GXbaSRaNTYSqlRHrOCYk9EqO2dIOsy2ZqN/NUXGHSnaNupFHCXuZHDyCdVuiMTeL\nmtjgWnc/0wdPEbJVGP+Df0pyXUHJwOmNefz1EhWLmdT4CPLTXyQ6OkGlUePs4lkaQoNHRh+h29V9\na3xKjTZXNwugqURbKSStRWhwGIfXtyfHcV93lqrpzMZLpEpNIm4LB6J7F28A2ChtsFZcw6k4ORQ4\ndEfbDOhYdS/VGthNIsecNuz3UOOwlF5iemUarSIwLIwRFp14uxxYBtydJj/tBuxc6VRIB8bIiz7m\ntubYbqwQae4QybSQW1aU8SGcA6M4XAdIv/QsLz13Fk1rcvjh0ww/9jPUKjrZWKVjuzPgupWp1C41\n2bp6hfj2KoYkEul9gOhk9+3V0exmKq3dYDW2SdQe5czkMcxmGTWfp3HzJgCWyUkk3/3/buwD4j7o\n1VPhiHOEHc2BbsDRN7ma0XWdVCpFOpcmoSaQ3TJhvQdL2YPFLhMacN7KXDBUnVasQjoTY11/CV1f\nI7jdxrym4Agdx/bQES7NPMPGQpxIpJfHT3RRyELDMoCrJ4Lv/ZNogkhL7qNasZAJVWi4W4Ad9//+\nHwnpbrLNLPPGOq2qiRHLBGbRjFGvsFHY5nLdwOl2ccKcRyssU2llEIMODg4dxpV20Lw5h7qzxPXo\nANcnTxP8d7/N4dR1EhU/WtbMsc1lektpamYzyZfOIZokomPj1Nstzi6epW7UeXj0YXrcr/gyVZoq\nVzfzqKpKdzuNqd0gNDCE0x/Ys2O5r9v1g4w33Eswuq0b3Kx0gtFhs8yk3Yp0l9BSdZWrm1fZ2NpA\nqlsZN43jU6y4u+wdwz1J7EAhdhk0FT1ymGQVZhIzlNRNussxfDkD2eTGPDmKu+cwViXKxt99hctT\nlzEpAseeeDd9p56imG5QSNWxOmSC/a98n+uJIiuXz1MoJJEtToJ9Z+g6EMLiuH1prKW2eGlxikQ2\nw7BviOMTBzGZRFrb2zSXlhBtdtTuUSx+563iuvupfUDcJ7W1NjPZGQrNAkFbNwUhQlXXOeiw0vMG\nGU4AuVyORDJBspVEd+kExSjOYhDZLN12xWEYBmq6TmEnzWr9JerSEr5UGceChM0xgfOBh5jOTLEw\ntYzbE+KpB4dpVVRKWgjLz32cUGMJety0nKPUy1Yy7ga1QB3td/4Al2OIkOCjVi9zU1+m1oBRZQKL\nyQrtNsnsJudrTUS7ixPOOkphiXItjh4wMzF8CH8xQHN6HnVzjuuRXm7MbuJNznI8fZlM00M1bWc8\nscOEFVqzM8SXFsAwiIyM00Ln7MJZqkaVh0Yeuq2fRK2lcmWjQGsXElK7RqB3AHcofKfh3Nd90svx\nBkEQONTl2tN4Q0NtMJudvatgdEXVuFauUdP0ew5GV1tVzi+fJ5PM4mmGGDUN4HFbcPY5kUK2zqy9\nnID4DTDJqJGjbKYLzGRmQI/Rk0viLAoonijWgyOdDnCqxOLX/oIbM3PY3BZO/cRPEZo4TWarTLXY\nwukzd4LNu5lKhZUYqzcuUW2UcLp6CQ4fJTTkfk0BXKVZ5YW5ixTLFY50HWJ8aAAMg+biIu2dOCa/\nj6qnn0pRxekzE+i5/35m+4C4j9INnZXCCrFKDKfZgyoPUtSg26IwYbe8YfC6XC4Tj8dJ1pI0nU28\nih9vsQsTEoFex23TTq3corKVY6VwiaIygzufxzMvYKYX14l3sS6uc/WF61gsbh5/6DB26mT+xZ9g\nWkoRzC+gRARawQO06k5SYTfV3/vvaCN1MpykEFq9wbS2SKmp0m8axiV5EA2dUm6bc+UyddnGYY+B\nr7pMsbSN7oHhkUkirW6a1xZR124yo5m4nqyhVHc4kzhPVTWTK4To+cQvc/xXfgGj3WJnaR5NbRMZ\nGkWXJP5/9t48RpIEve77Rd4Rkfd9VGVm3dXd1fdMz7Ez3OWKS1P0GhZJyAIEmxIoaG2YPugDlgwd\nlP0PbUswZIAQbIKUBIMrWpZhk4QoagkuORzu7Bw9fXfXXVmZWVfed0Zm3P6jenrn2O6dme3eHXLn\nAQ10d1ZVFjKOL77vve+91zZfY2yPeXHhRQqR79wgprrJzUqXqWaQM9u4tBHR3CyRdPaZHs8fZfwg\n+YaG0mC7u41t26xGV59IRp+oGg9GU1zCaYceeYKB5odRH9W5vnWdcXfCrFlkxpkkkpSQikGcwYfX\nV2sX2jvgCzONnWXv+JDN3jp+44hcu484duDJFfGvrBCKX8Q4Oeb27/0O+4dHhJMRrn31ZwnmlqmX\nB6gTg2hGIpQ4JZtt3aR2d5Py9j0MWyeaWiM2P08yH8DxoR2H1qjNG+vvYmgm1+avMptNYWka0/v3\nMXt9nDN5eo4Y6khFtCrEimnc0ae/Mf15gXgGqI1rbHe3cTs8SNIydcOJ7HRwMfBkXmI6nXJ0dERj\n3GDoHRKUQ6RGRdCcBGM+ohn5O7yEZjKp9ii37tNy3MKn1IjugHcYIXj2izQzE66/9iam7uTSxQss\n77xB4+t/hNmwidVL+CMDtOJ59J/5m7Sv/RgD5yaqYCDKcyS8KdwTWNe3aZoKUStNxpnD4RTQ+k3e\n6jep2x5Wo25mtDK9ThkzoJGfX2XWtYBxaxd99y6lqca7Eyd6t8E1fQPHCy9xnPlxYvllXlibx4VF\nbWcLdaKQLM4j+Hy8vvU6A3PAC4svMBeZe/TZqIbJzUoPRdXIWT3c6oBwOkssN/usD+ePFH6QfINh\nGez2dqmNawQ8Ac5Ezzx2+c2ybbYepr+FXU4uBCR8j1lQ/TBs22arucWDvXWsvsASy6S8QUKzfnz5\n9+VG1+6edg/BLAOpwNbhDmVlh4RSJdma4DN8eJaXCC2cx+9fon/rW7z92uv0Bn2y8zNc/qmfxRPM\nUC8PMA2LRP47CXDGSKV8/V3qx/vgdpLIPU9yMUso+dEsioP2Ee9s38Zte/jCyjVisRDmcMj03j1s\nXUcoLtFRfJhTBZ+xTWPYwJc9y/Lqp9uWfhI+LxDPCH21z4P2AwzLICrP0bJDGDackX3knjBy0nWd\n4+NjmoMmbWcbKSCRMvK4hzJe0Uki/76Rk2WjHY84OtnhSH8bwTokvm/gOxEJzL+MeSnDt1//Jv3m\niNnZOV7sbDD4nT9hUrMIANFf+hmMM+fRVZmeO0nXs8fUGuOVc0TFHEHNR0ktc2A2cesBCs5F3C4X\nhtLnTueYXV2gGJVY4pB+aw9LGpLKLzIXOIt1p4qxc5cTn8jbYpzexhaXK98k7NCoXvorSK/+NC9e\nPUvA46S2t81kOCCWy+MOBnl963X6Rp/Lc5dZji0/uoB00+JWtcdgojFj93BP+4SSaeKzjx9HfI6P\nj7FqcPfwlG9YSgbIx57dkuJAG7DR3mBiTMgH8xSDRRzCd7/hT02Lu0OFnmGSFz0sS75PlP52o3KD\navUA30hm1bFENCQTnAviTj184FJHcHIb1CF2fIWWFWDjeIOmWiLbrxLt6XhcYcRzZwjnr+C1/VT+\n4He4dXcD02Gycvk8K1/8Kobpo1kd4nAIJOeCjziBSWPA7ttv0Os38UhhksXnSS3GHoUAvR8bR9vc\n298i7AnxhXPPI8sier2BurkBLhdWYZVO18ap9RG0dWrjCVukuLi6wtoz2If4vEA8Q2imxmZnk860\ng+wJo3sKjEwHGa+bM08g1SzL4uTkhHavTdNuIgQE4o4UgX4SB46PjJyMzpRm5ZCy8i00R4X40Rix\n5MKfuoz00nPc3Hyd/Y0q/kCUV6+t4NYNepMw3twiyYQG++toPQdDd5a295iJ3cbpSxD2F4hbAerj\nGrvmAarmoOhaRnRIYE7Zah3wYGKSCEqc87YYtXawXG1Cs3lWUpcR1tto/+9v0tvd4q2Fi5wIIiu9\nW8wPKuz/2C8gXPkKzz13nlRIolEuMeq2CSXTiNE4b+6+SWPaYGVmhUuZS4/ISsO0uHPYozvWydLH\nO+kSiCVIFOY+z7j+lLBtm4POhN3mEKfDcWqiKD8bywzbtjkYHrDf38ftdHM2epaw7/Hy5Y5+KmE1\nbFjzi59ov2GoDnl7921a9Q4JJcWCK08kLeOfD+N87+bcq0JjExwOzMQ5jgY6G80NplqZbOeI0MDG\nE87iXztHOH0Fu9Xm/u/9NjuVI6RIgItf/BK5Sy8z6mh0jsd4RCfJYhCX23nKN5SO2Lv9DpPpkECk\nQHz+Aqn573guvQfLtrhRusP+8QGZQIYXzl7C7Xah7e+jlSs4QkGmyQUGHR23dow62eFIdbLrSOH1\nezk7k+VyLvNdPoXvD58XiB8AjkfH7PZ2EXDg8c3Rx4/0cOQUeMzIybZt2u02rXaLltpiKk/xewMk\nxrM4VM9HR06KTm+/QanzBmP3LuFWj8CmgNdXJHjlVQ69J9x8/Tq25eLKpXPkw07aXS9CfJFk3o/7\n5B5aTWXsSNISR0yEE2y3H79cJO6MoCsTHqjb9A2drDBHmAhOp8VB+4AboymiJHI5oDBtbmBxgpjN\ncHbueVwv/RVUbcikkObt1SvsSUmK3Xtc0A+o/PVfZmIlOX9pjYVsjPZBhX6zjj8SI5yd4d29G5SH\nZfLJPC8WXnzk9W9aNncOe3RGGplv/ja+3/hV/NUDksEwwveZoPWjholmsn7SpzvWSQS8rGYCzywW\nVDVVNtobp0IOMcFydBm34/E3/Mrk1GxPeihhfdJ49sM47h9zY+cGo9aUBX2OrJQgOhvEV3w4UjI0\nqN87teqW4mixFconNTY663j0CplmC1kB38wSgTNrBCNrjO9d551v/jGtQZ9UIcelr3yVYHae9tGY\nUU9FCnpI5B/aZhgWx7cfUN2+j4lBNHOe+Pw88Rn/R/gGzdD49tZ16p02i7EFLq+cQbAtphsbGM0W\nznSagS/DpD/Bo5UYKEeUTT9Hrih+yU1BsFjOpIjN/DnkIARB+KfAV4GGbdtrD/8vCvxLoAiUgf/A\ntu3uw9f+e+BvACbwX9i2/Y3v9R4/7AIBp9uam51NBtoAjyfOxJnDwsHq9/BxUhSFk5MTupMuXWcX\nj+whZcziHQU/OnIyLJSDLnsn1+lyH2naIboD7oGIvPgSxmqCN771TYYdhUJxjudX4nRbAoY8S3R+\nBr+yjVZuM7XCNL1OJp4KusOBT8wT9SURVSeb4x1qZp+gnSRlzeD1OWl3Tnhz0MN2ebgQMXH2ttG1\nA9zJIEv/9f9IcEdD84po+QS3zlziQaxItF/iC3//r3FcGtEbxZhbXeXi0gz9+jGd40PEQIh4YY71\n6gbr7XXikTivzL/yaEZtWTb3fuNf0vzH/4RU7wDJVpBUlfRUR/i1X/u8SHwMHHYVduojEE5DnLJh\n8Zm9V1NpstXdwrItlsJLZPyPf9o1LJsH4wl1VSfhcXPe//ElrLZts1HfYGNvA7vrYNVeIhmJEF4I\nfWekNG6d8g2mDvFl+o4I+ydldkebRMZlUu0hPsOLuHyO8NIVJFeSkz/+Xd69uY5qGSxeXOPsj38V\nnBKN6hBTswinpEd8gqFo7L/1No2jCoLHRSJ/jcRC+rtuRrdHHd7auslYmXAxt8byXBF7MmFy7z6W\nMsZRWKCrBzBGA1zqNu1Jmy0rwVgMEvQ6WEAn7PGQnFtACoY+5dF5PD4LBeLHgBHwf76vQPwvQMe2\n7f9JEIS/DURs2/5bgiCcBX4LuAZkgT8Elm3bNp/0Hp+FAgGnJ291WKXcL4PgxvLOoQvS99Rxm6ZJ\nrVajO+jSMBoQgJAQIzJMn6qcZvzIYe+j99AaY8rluzSNmyB0iB+qeEoGUuwM4gvXuLH3NtXtQ4Kh\nCK9eXcZSLBQ7hn9uiaivhbFVQlO9dHxxRt4KmqDg8CQISDPEzQDVcZWydoJgyuSFeXwekfG4xVud\nBgPbybmYg5BSYjquIrzxe+Tv7ZG5O0A3vei5COtnLnB76Qqen/85viAeM6p2qTVDJGeLvHh5hUm3\nTbOyj9vrJTm3QLV9zK2TW4iSyCuLrxAVTxeC7GKRB1MXNX+M5KSJbI8RNY20P4SjXP4BHdU/e5jq\nJusnAzojjYjs4Vw2iM/9bLoG0zLZ7e1yMj75nkQ0nMaC3nooYV2WfRQ/gYRVN3Wul69TPTgk0Auy\n7JojngkTXI7gDHjAsqC1dWrR7ZExU+epdRX2Wns0tAqpXol4V8XrCiGdv0QkfxVXf8SDf/M7bO0f\n4w3KnH/1VWYvv8K4p9M9UXC6HSTygUeb0Uqjx86bbzDot/EGIiTmniO9EPtI+ptt22yd7HF/fwO3\n4OXawhUyqfgHlt/s4gqd334d/q9/iq1vU09n2fjKf4Tz6mUiLlgUDAKBIOHiPD6v94kqyU+LH3qB\nePhLFIF//b4CsQV8ybbtE0EQMsBrtm2vPOwesG37Vx5+3TeAf2Db9ptP+vmflQLxHobakM3OJiNt\njO1JozlT+F0uLgSkJ6Zd9Xo96vU6bbXN2DdG9vmJKzO4NfEjIydzqFHbK3EweBvNXSc0GOLf1PAY\nYYIXvsyB3OX2WzcQbDfPXzpLUnbRHUp4skskMk7YuYcxsBm6EnSlAVOhgen0IUtFEu4Y41Gfjcku\nU9NBlnkCziCWpXCjdcihZrMY8ZJ31Onf+WMce98mctxk/lYdoe1Gn42w/zM/y/Uf+2m0XI4vBGu4\nG03KhzJyJM3L187jNDTqeztYlkmyME9Pn/B25W1ww0sLL5ENZsHhwLZtNhNFjoJJ4tM2AWuITzdI\nn9Rxup7+4tCfddT6UzZrA2wbFpN+ZiIfVdE8LQy1Ievt9VMiOpCnGHo8EQ1QV3XujyY4gAuf0DKj\nN+3xzu47tI67zIyzFMQcsbkw8nwIwe38ABFNOM9YmqVaO2J3sIvDPCbZPCIw0PFFsgQuXCWcuMR0\n8ybvfvNPqHX6xGYzXPmJnyY0s0TzcMhkqCOHPMRm/Didp+dhb++A3ZvXmU7HBBJF4vNrpOZCuD8k\nEVZ1lXd2bnPcrpOSk1xbvYQk+T6w/KZnF+n9f99E+Of/K7q3xYEvyl5oHtFpk/3yq8y+8NypzDuZ\n5lZ9i4QU5WLs6cu+P6sFomfbdvjh3wWga9t2WBCEXwXesm37Nx++9hvA79u2/f886ed/1goEnJJS\npV6Jw9EhhiCiuQu4nSLL32PxR1VVjo+P6Y67tIU2br+bmJ5BHkfxii6ShfeNnHSL4VGb8tG79O1t\nXPSI7U7x1ATkmSsYa0W+/fZrDHtTFuaLXJqP02m7IFIkuZDC09hEP+oyMWXaopeJ7xAdC484Q9iX\nxjuF9dEmbX1K0pEnYiVwekweNA/YGqtkwzLL0gjl3T9AKH8bT6fFUk1DeuWvovk91ASLtxYv0c0v\ncTXSJD1qs7PnwuGJ8NKLlwhLXuqlXabjIaFkGmSZN/ffRLEVLhcvs/zcT0KlAsBObJZKOENE7RKO\nefD8i98ivbj8efDQQ2iGxWbtVL4altyczQaRPsEN+JPAtm0Oh4fsD/ZxOVyciZ55YrCPbtlsjScc\nqzohl5OLn0DCatkWO80d1vfX0Zomi3qRTDhFdDWC972RUrcCzU1wOLGSazQnAqVGiaPpPlHlgERn\niE8Fb26J8Lmr+MU8zT/9N7zz7gMmpsn82lnWvvzvYjtkmgdDLNMmmpUJPowctQ2Lo1u60pHGAAAg\nAElEQVT3qe7cx7ItorMXSBTniM34cXxoKtAYNHl76xbTqcbZzCpn5hcQ+ODy2yhYYNxVcP63f5WB\ndkJJnKHhT+LXFeb2d0i6HCRee526x8v64ZtY9S2WonOcXfvpT3/QHoPPfIF4+O+ubduRT1IgBEH4\nGvA1gHw+f7Xy8EbySaAZFncPeywk/ESekaqjN+2x0dlAMTQmriwud4KU18OaX8T9BJVTs9mk1WnR\n0BtYfouAI0x0mMUteD4wcgIwelMOSpvUhtcxXG3CrSHSpoZPzOG7+iq3ju9wWDoiHIryyqU51LED\n3ZsmvLREUGiib++ij6DnCtMPtNGEIYIngl+cJWr5KQ1LHE3biI4YKSOPKDrZ7x5xf6Tgcjk5F3Xg\nGu2iD/dxRaBQvERCSaK+8Sa9YYe3FtY4Lq5xJjZgxW6zs22hmwHOnjvDUiFN+7BKv1nH5w/iT6V5\nu3ydttbmzMYhF37xH+AYnwa1lyJZSpkFIr/4C4RfPIfT6SS9sIzP/yNoF/71r5+G1VerNJbX2Pil\nv4P5pS+xkPCTj0rPrGv4CBEdWX4kLvhuaGo666MpqmUxJ3pZkLwfW8I6UAfcLN+kdtzA3/Oz4CiS\nzEQIn4mdjpQM7dRob3xKRE8jSxw2Wuz0dtD1I5KdKqGBhtcTPuUbilfwjnW2vvG7PNg7xuWXOPfy\nS8w990UGbZ1+c4Lb6ySZD+B5T8LaHlJ6+226rRpOr4fE3DUScylCiQ/yObZts364zYPKNqJT5IWl\nKyTjUazxmOn6OuZwhGNmlh4xtH4P52SL9t/9b9jwL6B6JQLKgJXtB0RGIyKjMferexwdfAt/+5BC\n20V29YsELl369AfuMfisFojPxIhpMNW5d9hnopmkQz6WUv5novDQLZ293h61cY2R7cf25gm5RS4E\nRMJP2BQdDofUajXakzZD7xCfz0dcmcWnn9qGx7LfGTnZhkW3WmP/6E3GdhmfMSSyoeAZ+Qgsf4Fq\nWOfezds4BA/XLi0TdvkY6X58uUViORmhsoF+0mOkS3SDMPXUMQQ3olwg5orRG7XZHZUxEckZCwT8\nEgO1y7udBn3VZiHiJe1sMG5u4hJ7xGdXKMhnMf/0FqP6Pu/OrrA7f5FCwuR5X4vqnkGn5yWZzfD8\npTPooz7Nyj4Ol4vITJ57JxtUhhXmtio8/z/8Gu5yFfJ5Kn//V9h58cuE3BbxSQ3bMEgU5n60/Ju+\n/nX42tfQpypb8QK1QJyAYHLu7/1X+H/+2ZH3TaXJdncb0za/JxGtWzab4wknqo7sdHD+e4xX3w/L\nttht77JeWkdt6MyqOXL+NNFCGP/Cw5HSuAUnd8AysOPLdKwA5XqFirJHUKkSa7eRVBsxvYh/9Tyh\n+Dm0ndvc+MM/4bg9IJRNceUn/h1CMyu0DkaoyqmdRTR72hXYpk3jwQ6V9TuoxgQplCKWv0RqIfIo\n1+U9KNqEt7dv0ui2yQazPL9yAZ/Pg3Z4hLa3Cw4nQmGR9siN1a9hqTuc6AZb//z3cY8HRDpNlve2\niIzHOC2Bd/+9Vxj84l8i02qTV7MEgytIF1ZwZ56+2/FntUD8Q6D9PpI6atv2fycIwjngX/Adkvqb\nwNKzJKlNy2a/NabaGeMQBBYSz25u+94F1jdMJq4ZAp4Yi7KPouh57FOVruunKqdhl4bVwOl3EjFT\nBMcJfKKbxOx3nnYAtN6E/Z0btMZ3sR1dokdjfPs6UmwRbe0cb915C2U4YXmhyLlCgl7HiR3MEl2c\nQzbq6Nsl1JFN2yMxDrbQBA2PmCXkzeKY6mwMdhgaJhkWCLnD4DJ40D5iZzghKXpYDqlMWhs47Sr+\nVIqFmedxXT9isnOL28lZ1hevEk6L/IVwm15do1wBj1fm6pXzRPw+anvbmLpOOJNjf3jCVmeLZDTJ\ny3MvPyI/j3oTNk8G+ByQNVpY0xHhVJZobuZHY1eiWKTd6LKemkdzuil2j5nrHOMo5OEZkPeKrrDb\n26Uz7eB3+zkbO/tEIrqh6myMT7uGecnLvPjxu4aRNuJG+QYnh3UCbZk5d4FEKkp4JYIr4kOw7fcR\n0X60+FmOO0NK3RIDtUKiWybUn+BzBZCWzhNeuILPEum+/Ye8fX2doW5RPLPM+Z/4Krbgp3U0ApsP\ndOVqX2HvrbfoNk7A5SA6s0Y4M0siH8D1Ib7huFvj+vZtdM3k/MxZlgtFbENH3djAaHdwRSNoqTm6\njSn0y6hmlarho+yIIe5vMfvr/wezlRLx/pBmPMG9Ly8hfPEShUCKnHsVX2geZzKNJyXjij19FdoP\nvUAIgvBbwJeAOFAHfhn4beD/BvJAhVOZa+fh1/8d4BcAA/gl27Z//3u9x9PgIMaqwWZtSHesEfC5\nWM0ECYkff2nn40IztVNHy0mbrh1G9OWIeHyc9YuP9Z15b2ei2WrSUBtosobsCBIfzeDBSzBxmlzl\neF830ShXqBx/C9WuISkDQutTvFYA3/lXuNXZ4+igRiwS5cULC1hTmBh+fDPzxHIB2FlHb/Tp6V4G\nUZWpe4DtDhAQCwRMkd3BLo3pgLAzS9RI4wt4OB41uNPpgAUrUSduZR9b2UaMOMkXrxEqg3rjDdb9\nEe4uX8GVSfNKrEFQ19nYtJlOHcwvFjm3VKBVLaMMesiRGF1B407jLrIs8+riq4+WrjpjjbuHPbBt\n8kIfe9RFDkVIzi3gcD47X6EfNgzTYjezwGEwiaRPOFcvEVLHpy8KwqmS5ynBtEwqwwqHw0MEBIqh\nIjl/7rFEtGZZbI6n1FQdv9PJWkD82F2DbduUOiXul+4zPVGZnebIhTLEFiLIxdCpA6s6PO0a1CGE\nC/Q9aSr1Q0rDXXzTColWHXlqIibmCZy9Qih+Bqv8gI3X32DnqI0gipx94RrzL/w4g5bGsKPilVwk\n8gHcntPFt9bWPvt3b6KqE6RwnMjMRWKzYYLxDz40WrbFveoGW9U9/B4/Ly5fJRoJYTSbTDe3sE0D\nV3Gevh1i0hkgDHcYmk12zAgdbwjZDUtOi+RbbxL91f+dBxGRg1fzBJbyLMTOEpPWcMfzuGJhPFk/\nDunp34vgM1AgfhB4miR1fTBluz5E1S1yEZHFpB/3xyTUPglORifs9HboGQITZ5qgN05e8rEs+R7L\nTby3M9FW2nSdXXx+kZiaRlIiuL1O4jn/B+R2k+6I3e1vM5hs4rCGREoDfHWQ8heoxv1srj/AsmB1\npchSKsKg58QOZIkuzSNP62g7JcZDi57sQAn00J1OfOIsEWec1rBOVTnGNH2k7CIBbwDVMeFW84TG\nRKMYcJH2dpm2HiCKPeL5s+S0POa3vsU+8O7SZcaZBS4leqx6xlSqArUTi1A4yPNX1zBHfbonR3hE\nCUMWuVG/jeAWeHnx5UfhQ4pmcLvaY2qYzLgmOPsNPKL455a8bgyn7NRHTH7+r1PYvstC+xAH77tu\nC4Wn1kE0lSa7vV1UUyUlpZgPzz8xt6Gh6qyPJ2iW/Ym7hrE+5lb1Fof7xwTbMgV3ntRMnPBqFFfQ\ne1r0emVo7YDDiZk4R31ss9fco6GWSHT2CfVHSA4JeekS4cXn8Kk2rbf/mFv39umpBrFcmrVXv0Rk\ndpVWdYSmmoQSIpHUqcOrOpyw/9Y7tGuHCC6BcPYs4WyR+Kz/u7iwjnhr8ybtQY98eJbnVs7jcoC6\nu3tKRAf8WDOLdDomVr8OeomWNmHdTmCIMkGnyRm3QDwWx5Wd5Z3qLUb1W2TbPebtOSR5EXd6Blfc\nf7rb4Xx2XfHnBeJTwDAtSq0xBx0Fl9PBUtL/TJaMpsaUUr/EybhBy/QiuDMkxCirfh8Z73cnzd/b\nmej0OzSNJnbARhYCREYZPJaIP+wlmpFxur+TM3Gwt8lx7dvodpdQp4+8PcHnSyFcfolb5fs0ai2C\ngQDPX5jHY3iY6BLi7CKxtB9rewO11adrOBjGx6guHZc3TtA3i0u32ent0lMVQkKWqJXG43exPaix\n3R8REkwWohZ6ex2vo0IglWM+fAleu0lr1OZG/gxH+TVmogJfiHVRegJbOzqC7eT82jKpSIDG/h6C\nIOCMhLnZvI+KytW5qyzETtO1dNPi3lGfzkgj7TGQhic4BIHU4jKi/+nbI/8wMJjq7NSHdMc6stfF\nmW99g/B/+jdPc5TfgyTBU1ggVHSFnd4O3WkX2S2zHFkm5H38gtaHu4bzAfGx7gEfhm3bVLoV7pTu\noh2qZKdpZqI5EqtxxJnAd5beGuugjcGfZOyf46DRYK+/izAtE2tUCExNxGiRwNo1IrFFjJ3brH/7\nXXbrA5yyzOLFNZZf/HE0zU3neIzDKZCYDSAGPNi2TWe3SunWu6hTBW8oQmzmCtGZ8Hc32msdc333\nNrYBlwrnmZ+ZwRoMmK6vY02muPJ5Ru44o9YQYbSPatWoWSI7Vhi36CbhsFj2uojPzNLy+bhT+hNc\n3RJzXZOcZw1PqIArmcKTlR850A6nOm6n45nssnxeIL4PDKc6m7UhfUUnLLlZSQcI+J5+q9dX++z2\ndqlPhrSsIAExQ14KcsYvIj2me+n1ejQaDbpql4FrgEtyEdLjhJQkLqeLaEb+QKzhsNNjZ+s1FHUf\nlzoiutPHM3DhL1zgOB7k3sZ9DM1goZhjZSbFeOBECKSJLC8iDuvopX0GQ5NeSGMqTzBdIrJUIOAI\n0hrVqQ6OMHUvSXuOkBykbQ253aqhawYLIQdevYpzsoUUdlGYeR7pTptJ6QH34jk2Fy7hiiV4NdEk\n7rDZ2IZBTyebSXDhzAKdwwradIo3HOH+YI+u3uXs7FnWvnkX59/9e1jVA7bPv8Dhf/JfEv7KF4kp\nJ9iGQTxfJBh/vLX0Zx1T3WSvOeKkN8XtcrCQkMmFH9603qdiIp+H79OC5LuNk2b8T+Z06qrOxniC\n/rBrmPsEXYOiK9w5vMPB3gH+pkTeO0u6kCZyJopTcoM+gcYGjOrglrDiKzQnAuVGmcPxDtF+iXCv\niyyIyAsXiSy+gE+ZUHvrdW5tHjEyLJKFGdZe+TLh3BydkwnKQDsN9pkN4HQ70BWV/Tev0zypgEMg\nnFkhnJv/CK8HYJgGd8oP2DuuEPQGeXH1KiG/jFauoFXKOHw+KCzR6QuY3RpMS3TtKSUrTs8t43GZ\nFF2Ql0Ricwvc7dc5OPgWkW6DxWmYqPcs7lQBVzyIJ+dHcDuxbZtyW2G/NSIZ8LGW+3O4Sf2DwLPc\ng7Btm+P+lN3GCMO0mI1KzMdlXE957GTbNnWlzl63xJFmoAhRknKaM/4AhceQ2Jqm0Ww26Q16tPU2\nqqjidYtElTSyEUb0u4nl/I/2JkzDYH/3Fo3adSxzQKilIpWGeJx+XKvPcW/U4PjgCJ/o47lzRWRH\ngKnhQ5pdJJIMYG1tMGkO6AgG45iC6RFwehLIvgxOzaLU26enKATsLFFHCkF2cL97zNFQIeu2SPvH\nGJ27+OU+qcIlkoM4xlvfpoyDWwtrDDIrXIiNuBCacFz3UCkpiD4vly+sIEzHjLpt3JKffa3J4fVv\nkP1nX+fazSPk6enc/SBVYPuX/2d8f+FLZPU/u+S1adlU2mMqbQUbm3xUohCTn8moEz46TloIL+Bx\nPl72rVkWG+MpdVUn4HKy5v9kXcNB/4A7u3eZViZk1RQziRmSa0l8afmUhO7uQ3vv9BtiCwzdcWrN\nBqV+CV3ZI9LcJzjRkMJ5QmsvEYkU0Tauc/ed+1TaYzyhICtXrzB/5WWmipNeXcG2bSJp6RGX0N0/\npPTudRRlhBgMESlcJpqNEn4vVOh96I66vLN9m954yHysyKXlszg17VS+OhjiSqdQgjMM6gMY7KPZ\nDWq2lz07glN0Iztsllw2yXAEVybHO9VbKM275DpDFpxL+MQF3Jks7nQAV8yHIAiMVIP14wGDiU5G\nFlhMh/B6n/7Y9PMC8ZSgGRa7jRHHvQlet4PlVIBU8KPB498vDMugOqxS6h9S1Ry43SmKgSTnA/Jj\nJbHD4ZBGo0F/0qdDB0EWkMwgUSWLV/ASTkqEEuKjE7/TqrO7/U00tY7H0AmW+3jrKr5whn5hkTvl\nHSbKhFwmzlpxFnXsRggmia4s422foJUr9EYao+gUTTYxnB68viyyK0p30OCwf4yheUnYc/gDAQ70\nLpudNqKpUQxZ2MNNREeVcCLJTOwK7je36DarXM8sclC8SDrq45VkF1t3s7Fpok80ludnyMRCdE+O\ncLjc9P/zr7ElDfGqE56/d0KurQPQXj7Hvd/9o4fk9QB71EEKhUnNLX7myWvbtqkNTh9GVN0iFfSx\nmPQ/szCfTzpOgtOuYX00wbBtFqQnK/A+jIkx4c7hHarbB/gbPmbFWTKLWaJnojg8Lhg1T8dJugL+\nFNPgHM3eiIPeAbVplVB3l1C3iYwH/9xloovX8A16HLz9Jvd2Gyg4yS7kWXvly/jCWTrHY7SpiRh4\n+KDkcWKoOpW33qV2UAIBQpllwrlF4jP+R3Ya70E3de5Xttg53scluLg6d4FCNod+dIS6uwsOB0Jx\nkZ7iRWudYE/K9FHYs2P03TIet03GYVP0uIjlcjRcLu7sv463V2Fh6CDrWcMTKeJKxR8R0bZtU+0o\n7DVHOB0Oznh6+O/9Mc5sEe/Vn/y0h/qx+LxAPGX0FZ2N2oDR1CDq97CaDjyTjdWJMaHUK7EzbHFi\n+IiKaS6Ekyw+hsS2LIt2u02n06Gjdhi6h7h8LsLTFCEtgc/nJva+i0DXdcr7t2id3MIyx4hDFX+p\nh3ds45lZYcvtoHJ0hMvt5PzyLDFvDM0SkQuLhMMy5s4WSntAHxUlNsbwOrFdISRfDqcpUOlW6A3H\n+K0cMXcaRTS536kxVhQKko3oaOJWNpDlKYnsOeJ1Gf3OTe5KYbbmL2KnZngl3iErw3bFS/t4SCwk\nc265wKjVwPyLfxHT5WBjIcjEY3Km1ORcaYDLFlCm2vvI6ynOfh2Pz3dKXnufflF/GuiONbbrQ4ZT\ng6DoZjnlJyw9m+VNwzKoDqocjg5xCA6KwVN10pO6rLFpsj1WaWqnXcN5v/ix3VdNy6TcK7Ne2kAv\nTUjqCWZTeZIX04gJCTQFmhunzqtuCTO+SmsCx61jDpQy7skBodYB/skUKZQjdPYVIuEU0/s3uX1j\nm+OBihQLs/L8C+QvPMewbTDsqLhcDqI5+ZF1fq96wt4776CMB4iBAKHCZaKZGJG0/JGN6MPWMbdK\n91HUKfnILBcXz+BzOFA3NzFabZyRMFpynl5jiNUpo9l16njZI4LD50ISLFY8AvFAkHC+yI3aPifH\nbxLtNVk20oS9Z3ClZnGnQrjTp0S0op12DT1FJyE7WGrfRX/wDqo9wHvtK4RWv/zpD/pj8HmBeAaw\nbZvD7oTd5gjLskmHfMzF5WdSKHrTHpvdHXbGE8ZCgLw/w5Vw7LG++aqqnnYToz4tvYUhG3gQiY2z\nSAQIRH1EMtKjcPXReMDu3hsonV3QVfytMfJ+H49TQp1Z4Xa3yXA8JhkLsrZQwJ6KOAJxosvLeIZ9\n9PI+w57CwDtgGjUwPF7c3hQ+V4LhuM1R7wR95CbhmMcbCrA3aVDu9kk6DdIhC2uwhUiZcDRMRl7D\ne32HijLg1swK3cJ5zoY0LicVOoMgu9tDXJbJmaU84s/9JSbdDrYgcJALcBJ3kmr2eGHgw79VRjct\n7h726Y4/2+S1ohnsNkY0Bipet4OlZIBU0PvMRmINpcFubxfN1EjLaeZD808cJ6mWRUlROZhqOAWB\nedFLUfR8rN/Ptm1q4xr3Kw8YlLr4exJpMcPMmQKRlTAOB9Apnf5BwI4t0BPC1FoNDsYHqEqVULtE\nYDxEdEjIhfNEF67i6zbZf+dd7ld6aA4XMyuLrH3hxxHcEbonYyzTJhD3EUlJOJwOtLHKwY3b1Ksl\nLNsimFkgMrNMYiaAz//B62ikjrm1e5/jTp2A28+VhQukE/EPyFedxXn6ZoBp/RhzVGEojCkRo+OW\n8TpN0k5Y8HmIz8zSdXu4tf8OxmCH2d6IBedZvNIc7kwGTy6AM+R9lNex1xwhCLAqjgk+eI3pyQ6j\neJDdsy+ST6ywGPo8D+JT4YflxaQaJpW2wmFXwbYhFTwtFLL36RaKRxdaZ5/dqYnXHeVsOMeFYPCx\nJPZgMKDRaNCb9OgKXQRZwK9GiE0z+Dy+DzxZAdS7h5R3/xRjXMM5NQlV+ngbCp5gnKNQjK1WEwFY\nXUiTltIYiEizc4TTaTg6Ynp4QH+sMA520YJODLcfnzeH03Zz1Duk1x8jqxki/gxtt8p2t4GlquRE\nG9Hdwzl6QFDsE08sEj/wMizt8G4kS3XuErFomFdTXbxukfVtgXG3z+zuBvl/9Csogo3hcNAPSewt\nyLh/5t/n+Z/7j5kNzWJZNtuNIYedCRGPTfT95PU3/uCpkryfFLppUW6NOegqp6RwXCYflZ5J/Kdt\n27SnbSqDCkNt+LHGSYZlU5mqlCcapm0z6/MwL3nxOj4eD9Kddrl/cJ/mTh1fy03ckySTzZG8kMIb\n9sKwfto16BMIpFH8BertPoeDQ1qTMqFOCX+/hV9wIiWXCS+/iOx0oWzc5vbtfeojE38yxtmXXyY1\nv0avrjEd63glF/GcH4/owtRNave3ONpeR51O8PplIvnLRLMJImnpA7kNlm2xfVTiQXUL27RZySxx\ntriIoGuou3sYzSYOv4yRW6LXnGC2S6hmg7rDw54dRvC6kASTVZ+TVDiCmM5y82iXRus2gXGLxalE\n0nkWd7yAOx3DnfXj8Dg/kNcRl10sDdYx7ryBanSpra6xlblE3Y5xORTgC5Gn/2DzeYH4AUA1TKpt\nhcPuBPNhR1GMy/ifcqHQLZ1Kv8Kt3gnHmpOElORyOM2CLOH5LheuaZq0Wi063Q5trc3EO8Hl9hAe\nJwlZcaSgh2hafqTYsGyL8sk9apV3sCYjxJ5KoNTDMzUx41nWDYvWWCESEjk3l8djh7BdMv5ikUAg\nhFWpMD6p0zd6TEJ9dL8E3jg+b4rh639E/Y0/Qh/JxNU4rp/8EqVzcxy2u4RNlWTQwqGWkcxdImGR\nlGMB9519Hggu1gtnMTNLvBwdUIhBpRHiuNzFc+cGC//qN5GPD9CyWZS/9h9SenEJxamyklvhQuYC\nLoeLg47Cdn2IzwlZo431+/+a4D/8R8QaTRzvnfZPSSb6vfBe91lqjdENi2xYZD4hPxMJo23bNJQG\n1WGVsT7G5/JRCBRIy+nHdgC2bXOo6uwpUzTLJulxsyR7kT8mf6PoChu1dQ43D3HUbKJClHQ6R2Yt\ng5SWT+WqjQ0YN8HjR48u0RxbHHeOOVL2Efv7+LsH+A0LKVIgtPQS4WAE4aTCzo0HbBwPMd0+iudW\nWXnxixi6yKA5QXAKRDMy/shp99XeqVK5e4fJuI/T4yaQXiKQKJDIBz5izd0adLixe5feeEDSn+Dq\n4nkCkohWqaJVKwiCgGsmz8ARYXxyjDGsMBJGlIQoHaeE22WRdQksil7is3mq6pT1g+s4lEMKikrB\nMYdXKODOzuLOBnE9JMo/kNcha0Q2XmNysM44LLFz7iXKvgUmiMxLXi4H5U+Utvdx8XmB+AFCNUwO\nOgoH3QmmaZMMepmLy09dGqvoChvdPe72e/QtFwkxypVIlkXZ/135iel0Sr1epz/qn+5O+G08hkx8\nmkUUZOSQh3BKerQQNDWm7JS/zaD2ACYa/mMF+aCH2+WlFYqxMZ5i2TaFmRjFRBpblxCkEKHiHJLg\nQSuXGTQajNxN1PAUbfcAzz/7beyRTT0VoBdM4J+kiLzyUwxeukBp0KA/HJESNALyFI/ygKC3RSKQ\nJVYWqLV6vJvM0ylcYDns5LnMGN0RZ7cEg2aDqORhNhUCQ0NTNY7NHifOHulYmmuFawS9QdojlXtH\nfQAKf/krWK0aHsMg2R3gNYzTD+opLpp9GIZpcdKfUu0oTDSTiOxmKRUg+Axk05ZtUR/XqQ6rTIwJ\nkksiH8yTklJPHA01VJ1tZYpiWoRdTpZl3xO9wt4PzdTY6+yxt76DdWAQsoOkYzkyF2YIzgROw3s6\n+6cKJcGBFV2gY/k5adU5HFexB3v423v4tSkBOUNg4Xki8RncnROOHmyzXu7S02xC6QTnX/0ioeQC\n3doEQ7fwRx7u/rgcDE5aVG7eZNBpIjhAjucJZpYJJfyEkuKj0SqcJr3d3d+gVKvgcXq5WDhLMTuD\n0Wig7u5iqxquVBIjkadXG6I1SqhGk4bTSYkotseBJFisii6ysThCJMY75TuMRtvERz0WrTghewF3\nJIcrFXtERL8/ryMqu1id7GHceo2p0uBk9RzbuaucECfi9rAme8kMuwQCAeTw4x1zPy0+LxA/BGiG\nRbWjcNBVME2bRMDLXEJ+6jeDvtpnu3fAvUGXnukk7gtzJZplxR/6SKGwbZt+v0+z2aQz6dBz9HDJ\nLkQtSGSawidI+MNewinpkSy2PW6wt/8naJ1DnCOTwP4AX2sEkp+SV+Z4ouF2OynMRJgJZbFNH65Q\njHB+Ds/EQKmW6XcbKP/2n6ANjrF6Jp6myUgWacRD6OEFAj/1NcRonKZLZb/fxDUeExcNvBwjG1vE\nAgLJSQp229zwRynPniOYzvLFSI9wxE1znKK018Ma98mEfUQkN1g2TaVP1dXB4/dxtXiVQrjAWDW4\nc9Bj+lM/zWzvCLdLw3I4iA6GhMaT05vnU7SqgNNdhsPuaXdpmDZhyU0hJpMIPH3JommZ1JQa1UEV\n1VTxu/0UggXiYvyJhaGnG2yPp/QME8npYFnykfyYT6umZXIwOGBrcwO1pCDpMqlIhtxannAhhMPW\nT32TelWwDAhmGf7b69R+7dc5FHoMc14CL68gJ8MEPCEC+StEc4t4h20am3s8KLVpT2y8IT9zZ8+w\ncPllxgMHykDD43U+El5MB2OqN27ROj7AtkzkcBJ/bpVQPEI4LX0ks6HaOOJW6T6qrlGM57k4fwaX\nrqFub2P2+jgDfoTZeXpDgWn9EGN0xIgxJUeEtkvC7TTIuhws+SViuRwb7Qb7raSTHp4AACAASURB\nVLtISo051UmORdziLO5UGlfKjyt2qiI86k3Yrg/BhqWgSXz7dZTSHUayh93zL1OWl1EQWRC9nBF0\nhNoRhqYSTmWJzcx+P6fHd8XnBeKHCN20OOgoVDsKhmkTD5x2FE/b40nRFTZ7VW73W3QNiHiDXI1k\nOReMfSTFzjCM0yLR69DTeyhuBYfXgU8LEp4mkB1+/FEvoeTpRWXbNtXuFoflN7AGA8SWRnCvi3uq\nowbDlAUXtamBx+ukmIuTCSQBCU8sSThTQOgMGP3SLzEM2Sj+Y/SJiqOuIwwFuhGRzt/629D1EiQD\n8SgHdp96t01InxASp0jGFiF3jYQrRKQKO1Mn9zNzaLPnOe/XOZdUEMQI+/UgjeoxPlsnG/TgdYKi\nqZTNJhPZZjm3zMXsRbCd3Lv2ZTr9CTGlS1xro3lciKpGMhjGtb//VI7JYKpTbSvUB1MAkgEf+ahE\n6Bl46hiWwcnohIPhAZqlEfKGyAfyxMTYE79vbJrsjFUamo7HIbAg+Zjxuj82AV0f19nYuc9ou49X\n8RILppg9VyS2GMVpa6cdQ/8AbAsCaabyDI1/9Tsc//r/RkNSkCI6Ulwk4PQSuPhl4j/5l5G0MZ2d\nEg92W9THJi5JYnZ1maVL18AZpFefAJxGgMZFTN3g6PZ9TkrbGJqOGAwRzJ1FjsaIZuSPSFeHkxE3\ndu5S77cIegM8t3SBuBxA3d9HPz5GcLlxFecYEWB0eIQxOEIXBtQEL/t2EMvtQHJYrEpuZhJJhg4n\nt0/uY072yU4mzNk5JHsOT2oGV/JUoeTwOJnqJhsnA9ojjYjkYtU4wLzxh0xHJ9QWVtkqXuOEBBG3\nh7M+J6luE7Xfxe0VSRSKiIHgpz9BnoDPC8RnAIZpcdCdUGmPMUybqN/DfFx+6jJGzdTY6h9yo1On\nbZgE3RJXIxkuhFO4PsRRKIpCq9ViOB7S03uM3WOcHic+LUBYTSI7Ty3FQ0kRl9uJZmjs1t+hc3wH\nelPkIxX/4QCXqaOIIlXBRcMU8Igeitk4KSmB4JSRUlkCf+M/wzQFBsk4o3CPqbeJ2ddxm2H0X/nH\ntEdN+p0Rjq6fkCfDMCJxpHaYDPtE7AmSt0XQ2CTuU0l0/IxbFndCKWrZFQKxDC/5B2RSMCTLbtlg\n3Dwh5rH+f/bePFi6/C7v+5zzO/vpfb378s77vvPOjJYRI7QgWWazg3Fs7BA2U4CdGJzCGGJccUJI\nYlyk4pSxK1EFQ8DEFEXJSVy2CdhQwgYBAgvQLo1me/e73+7b+9m33y9/3NGgQdtopAEJv09VV9+u\ne7rrVHfXefr7fL/P96Hj6KALTqIxEz+l1x/ypr030fiX/5aj/+Z/4LbXw5QlO6sTqoaN/v0/QP+v\n/Zcvu5RXSnERZhzNYuZRgRAamy2X7bb3ingZCllwEpxchlLJkrbTZre++8JCw0+HT5xM0jWNfddi\n17Ffci70MlvyzJ2PMX1mjL7SaXk9th+5wtrDAwwtvTS5rU4vD25ukrjrzMKMi+UFxz/83RjVCG9o\nUzdtGkvo3J5TNzTCv/cjPPXsOaeBRHNstq4+xLUn3ojpdFlOUspc4jUsuhs+utAYPXOH42c+RpZE\n2I5LbfMate4WnTX/RXkpcFnpPHt0m2eOb4HSeHTzYR7e3qc6PyO/dw9Vlpibm2T1dRYnF5SzQwq1\nYG7oHKg2K83A0AvWLcGNho/f7fPk6IBxdJNmNOVq5dGrrmM2tzDWB1hrPuL5VR6ny5Rbo+Ay5a8J\n/Tu/Q3zzA8SOxq3XvIV79UeIcbniWjycR+gX55fGvvUNWsN1tJc4GPBy8IAgvohQVpLjecLBLKYo\nJW3fYrfr0fVf2tjgS0UlK26uTnnv7JRJXuAbFl/WWud17XVM8WJNOY5jptMpq3D1AlHolsDJ6rSz\nAb5Ro951aPU9hKmzSuc8d/pu0vE9RADeaYF3PMfIUwLL4Fg3mGJhew77G1267gDxzB3qP/vzeKEi\nHW6wHPrE7QnV215N8ao3IrxtkkowC6aEkxw7aOF6Qy4aGqfhFDsOqBkJde0ObXHCsNBpjC3uVy5P\n9zaI1m+w3/R5fXuF2/E5jzY4PJiigildo6Bm6iyLmGNjgdmt88TeE+z98n8g/JH/mSdLm3hjm83v\n+mb8tzxOkSU0+kO6W9vo+kud81ecLhKOZjFxXuGYgu2Oy0bLfUXcz3mVcxwccxKeUKmKnttju779\nWU1upVQcpjn3kuxlTSbFRczNw2c5feoINZU0rDab1/bYevUWlhbD7A4E56Dp0NwmsgfMgoTJasIo\nOaVa3sF7x9vxXZtmpNM5imidnJEZFk+39zj8c9+EMkzWruxy/fVvwvb6BNOcspDYrkFr6OHWTRYH\nZxx85ENEqwXCENQGe9QGD9Feu0yB+0QntFKK+6Njnjm6SZBGrDWGPHHt1ThZRnbzFjKKMDptqrU9\nlhch2eiAPJ8RGBWHeouZZiNEhaPDDddks9fjPEt5bv4MRnLCdlKxq/axzT3MjY+7oS/PYRpm3ByF\nRFlJyzN5RBsh3/9Okvkxo92HePbqV3CmDWiZFo8KRXd2jkxi3HqT/s4epvPKe3YeEMQXISqpOJkn\n3J9G5KXENnXWGg7rLfcLOvmklOJOMOb3pseMsgRXFzzeGvBEZxvb+EOhJ59IFPklUWi2wP71j9D5\nuXdRO57QqGk0/7vvQ3zHt3MeHHAwei/Z4hR9JfFOJP7BHHF6xOr4HqeVYu42cK49xO7rXkPn/gTr\n3/8mjbunmIM9om/8JsL9Nql5ROXFlE4b/H0iqTOZT8kuFG7cRWv0OHELVsEMP1tRs+e01HP0rYj+\nSqAuBE96fY5624i1qzzhl1y9/VtUP/Xz3D62mG/v4Hztn6L1+A2kDifVlOz0Ofb+xS/x+HtuYa9t\nc/OH/xdOvuKrqVk6m1pAvrjAclwG+1exvU+fe3DZX0g4nl9KiA3XZKfjMajbn2S8+kIgzEPOojPO\nojOkkvS9Prv1XWrWZ07Ui6qKwyTnNCuolKJvmVz3bPyXaHSbxTMOj+4yun1GcVHgiTrre7vsPb6L\na8QwvXuZ6qYbqOY2gdFltoqYRBMu4hOY38GZ38UtIurv/Rite0t698+oKsVzrW3u+WuUjsvgr3wb\n1778TXj1IeGsfBExeA2LcDzn4IMfZDE5R0PD76zTWL9BY1CnNfQQxh8QnVKKg9EJTx8/R5BE1Gyf\nV28/wlar/Qdjq66DvnOFVaITH98nC8fEIuXUbHOOC3qFLRR7jsFuzQfH46OTe8TpXbppwJWyS0te\nxRzuYK13MIYeuiUI0oJb45BZmONagqttQevOe4ieeg+xIbn9mrdwu/UqYhyu2iZXowViMUUYBt3t\nHeqdP7rwqwcE8UUMKRWTKONskTIJM5SChmuy3nRYazpf0F+f94Ipvzc94TRdYesar2l2+bL2NjXL\nf9FxSZIwnU5ZBksW734n0T//CYhXOFWHdtyhUZk0/u730vgvvhkhdCbRKfcn7yOa3Uf/6G3cd92k\nNtfQi5SlyjmzLVZXb+Du77Mz6NH2Bjhei8ZwB4FPdH5BtDwiEfepvCWl1UD6u6wqwXS6pLgQeGWP\nvNHmxAiR4RxHhjStc1oc0FMhvYuKi6zJ081NpnqD4Uee5k2H76c9eorl2OJu7SHSb/vPqb/qCs4H\nf5/pb76ThZfiJQGvuXnBlaXG5Mf/KU+/+WtRCnZ8iZidIquKzuY2reHai96jIC04nF32F6SEft1m\nt+u9Is7nrMoYRSNG8YioiNA1nYE3YKe+8xlDe5RSTIqSoyRnUpRowNA22XUsmi9hMkkqyWhxzv3b\nd1jdn1OEJa5Zo7u2zv7jV2i6yaXBLZ6CMJHNXZZak9kyYBJPmIaHiNktrMV9XFlQ97o0JgXd//sX\n0EcX3G5ucau2SaYJeqs511/zZdT+/j8iWspLYvAMWgMPt2YSjGacPfM0s/MTZFXhNdrUNx+l3u3Q\nXvdetI5bKcXh+JSnj55jlYTUbJ9Htq6z21ujPD7+g7HV7R0is0VwcEi2OCXRE0ZWnRPpUQqFJSSb\ntsmVmovv17i7mnEU3cZLRuzmBhv5Q1jNPczNNaz1GqJukZUVd8YRZ8sEoWtc6bisB3dIP/QbJLND\nxpu7PHv9T3NqDGmZFg9XKd3ZCK0qafSHdDa2EMYrkyP+6fCAIL5EkJUVo2XG6TIhTEt0HXo1m/Wm\nS6/2hZOgDuMlvz854SheYGiSHcfhseaA7doAx/iDkjZJEmZvfSvLKGRe1wnNCK1IsWWHjnOVxk//\nHI2uQ63jYFqCeXLBwbe+kUU9Q3M7eFmf2spBzzPmNowfe4TQr+PW22wPBrT9HrbXoNZewzBaJNOA\ncH5Ipt+hsKcUlk/pbTErTRYXMUwsHG3Asu5xxhwvCbC1hJpxRIsjhumKxoXk1pNj7ne2yawOj12c\n86qzD2KeHXLivorT/+0nUf/936ExPaCyBOOuRW6mbFzM+LLQxf3oHZ46XTGPcnqeoJtNyMMlXqOF\nv7HDLJGMVilBWiJ0jfXWZeP5C+2gL2XJJJlwHp2zyBYANKwGQ3/IwB18xgzoUipOspyjNCeuJJau\nse1YbDnWS5KSsiLj9OyQo9sHROchKgPHa7C+s83m9XXqVog2uwvpEgybqrnLQtWYLpZMkgnz5V3M\n6XNYwTGuBo3agFbvGp2nb6P/+E9wz+3ybG2bRDdpBUuu3j+ilQri3j7l9/9tnK//WlpDD9sRzA5O\nObv5LMF8glISx61TW79GrbdGZ8P/pOjPw9EJTx/dZJkE+JbHo1vX2e0OqUbn5EfHqDxHDPqUvS2W\nh6ekkxPiKmBquxxpDVJNIfSSNdvkmu9S81xOopDD9Bg9O2OQJuwWG9TFVayNLczNJkbXRcKLlitu\ntT125Zjyw+8iOblJ4lncufFGbnVfQ6IcrpgaV5YT7CTEcj36O/t/bLnqDwjiSxBBWnC2TDlbphSl\nxDL0F6qKL5Sn4iSJ+Mh8xP14TlImNIRi3/W51uix5g8vw2F0ndSymHY6LBt15nWdyIxRZYb9M79A\nu+hTN5p4dYt6x8HreEQu3H/M5eKKh+a3cYshflBHXH+YmUyYGRDVG7itLhvdHk2viW01qDX7WG6P\nNKgIZydk2nPk1phSWKTOGpPSJhyXaAsfw+gR1G0maoGdBJgqpGEc0uGUjXf+MlVg80znYcadbfzc\n4g3nz7E5eZrs//1l7r7pW1h5HpaeYmkZQcNh3pDYZcijP/evuNG7wdmy4M5FSCUVXhExPjsnqaAx\nWGPYbzOo21/w/oJSink25zw6Z5JMkEriGA5Db8jQG37GagE+WUZqGoId12JomS9pmV4YrDg6vM/Z\nnRPieYKumdSabXau7bO+XcPOJ7A8vlykZ7oUjR3mlcdsuWAcj1jNbmFNn8FKL/A1g2Z9jU7/IZq1\nDvlsyeGP/u/ckXUiYVFPQh46OKIbayTeOqVh4aQrWnUN88MfYHT7HqP7t0jCJZqm49ba+L093NaA\n1pr3ghnu4/hEYvAsj8e2rrPT7lOdnVKcnKDKCqPTRg22WExWJGcHRNmKhSU4Ek1CTUdoBS1T8LDv\n0HQcRlnCYXYG5Tm9LGEj8+mUV7D7+1jbfYyhh2bqnC1T7lxcLlccNGwe8nPEx36L6M6HyFTG6bXX\ncm/ryzmhSVsYXE8DeqsZQofOxjbNwWf2przSeEAQX8L4VBJU3THYaLkMGw6W8flfoJJKcidc8Uww\nYZwsKKuUrim54tXZ/XP/Gf2nDrBKRWpZzLpdFo06870+4T/6H9EdgZYZ+FmLtt7D++t/g9qdp6iH\nI0qRcf+Gy+iag1rbxP7a76R+K0Ubz5nmIUtbkdVrCL9Bq92jV2/hOy3qtR6O2yfPbYLFGZl6ltw6\no0AjsrqMC5d0bGBEdXTRJPIcFlqAypeYckn7yV+kXx4zHF8w1Ybc6lwldvrslIo3fMefx/t7P8js\noxOO7DVi28I0UrAVk60W6bf9p7T2rrLVfzVR4vP06YqslOy1TLbkgobI6a+tfU4N7M+GMA85j84Z\nx2NymWPoBn23z5q/9lmbzp9KRlqzTXZeqoxUSqaTMcf3D5gcX5AGGcJ26A7X2L26Qb+WIaJzSOaX\nT3A7ZN6QWW4xXy0YxWfEF09jTZ/FKpfUhEWrvkW3f4WaVWd2OuHuScjpsqD4tX9PLQ7ZPTljEEgy\nd+0FYmguj9Edl7NHH+Hib/9NijTFEAZua4DX28dtNmn0LsOwPrEBfTg65Zmj51g8TwyPbl1jt9Gl\nPDmmOD8HpTD6AxhuEs5DwsP7RPGCpag4MRrMNQtNz6kZgquuRd+2mBQZx+UFqjqnncZsZh6dYger\nuY21sYa5WUfULGbPL1cMn1+ueK0j8O+/j+jJ3yZPl4x2r3Fv/82cmmugCbZVye5ijFfm+M023Z3d\nL4rUwwcE8ScEeXkpb5wuEoLnJaiub9OtWXR86/OWOT5+sbkdBtwJpyyzJcYH30P/Z36S3ftHrM9z\n+osSWWsw/cf/mNVb38IyWxJrMZmZAQL73U9S/6lfoDUr8ZMV9WCMaSQc/p8/xNkTfao4xDytqD+b\nI84DVumCpSjJHEFVr+M1OrTbfdr1NjWnjecPKFWLaDkl4ya5uE+GZCXqTPMa2dJBrhyE1qKyPVYi\nJTt4EvP2++nWLxiUY1rLgHv1q5zdeCPalcd53fiEa7/4s1j3bjOZOxy7Q2atNquv/7NM+i0uygi9\n6fHww9f4iquvJs0NJmFG3RZsaiuy+WUDu7e7/7KX/mVVxjgacx6fv9BX6Dgdht6Qrtv9tJnPH8fn\nKyMVccbo7ITju4csJkuyrMKq+WzubLG35tIQ88skNyXB8qn8NULRJEgK5sGc0eqIbPxRzMVtnCqm\nbjt0m3u0e1cxc8nx8ZK7ZxGLtEKzTXob62z9gx+jcTwld4fPE8OSZnhO0Whx9qqHWWwOqOpN7D/1\nNtzWOm53F69do9lzP2k1xtH4jKePnmMRr/Asl0e2rrHrtSiPjykvLkDXMNfXke0Bq9GUZHRGEq0I\nKDh1G4yVhdIKPENn3zVZM3RWUnIkpyg5opVErKcOvXIXq7mDudbHHNQQbYe4qLg1DpkEGY4puNpz\n6U6fIf7Qr5MtzpgP1rn78Fdw5O6TKoN1XXElnFOPAmzbprez+4o4ol8uHhDEn0B8XIIarVKy4tL5\n65iCtm/S8S3anvV57fbJpOQ0LbgdrTj7rV8leucv4J/eoS8qNr7lOxj+5e+kLurEQcxqtSLOYlbF\niljEyPf9B9S/+je455KOs0njW/4qtW/8OpymYJw+y/H8g+TJEvOiwjsU2AcR6WLBMl+QmFDVLYx6\nE7/do9cZULfb1L11pGgRJxkpt8i12xQUBJXBtHIIgjpl4CNkC82oE10ckT73PjQxZtBZsV6TaIbL\n7fpVgv4VuiE89ru/iZyOmeku46/+JiY3XoeRhtTm91DhXRI9oLmzzhtf/zU0nB2ePQ9RwK4rEYtz\nyiKj1unR3drBMD+z7FfJimW+ZJEtWGZLltnlyo+X2leA5yWosmKUFZ+zjKSUQmUV0WLF+PyM48MT\nglVEoSlqjTZXtjpsdSrsfAayAGEha2tERosg1wiCgGWxZLk8Jh99BDu4hy1zmm6dXusK3e4e8Szl\n7uGSo2VOVhS47TZrV/bZ2LuBYTRJf/ldqB//cZzlBY14RjToc/7YVaJ+G6ULnKTAf+JtWF/59dR7\nPo2e80lZ0Mfjc54+eo55vMSzXG5sXmPH9qmOj6nmCzTTwNjYoPBbBMdnJJMRcRoTGYqpXeO0sijI\ncUydbVuwqUOiaZywpFJjWvGKYWLTL3exW7uY6wPMYQ3RtMml5N4k4mSeoOsa+x2PrWpE8oF3kp7e\nJqzVufvwmznoPsJKOfSR7IdzOmmIJXSa/TXa6xsvK5MkzEOELnCNL3zs8QOC+BOOKCuZRTnzOGce\nFxTlJWF4lqDtX1YXLc/EfokjjX8Ys6LkOM05iALm2YKiXNDQUnqmRsuuUzfrWJWFShVZnBFmIYEM\nSIzkMlAlM6kXHbpmn0bNw+9YLLjP6eLDJNkFMi+xzxT2fYV+EBAGMyIZUTo6NFzcVot2d0Cjtk7D\nHaIbXZJSkOknFOI+lZwRVBmzVGcW1kijJsg2QmshNZ1EmyE5o5HdpbacMdKbHDZ2SesbtIXP61zF\nlYHFcKtH6K5zvFBEFyPiux9mtrpLYRVsXL3GGx7/s4xDh3lUMKhZDFgRT8ZomkZ7Y+tFWnIpyxeI\nYJEtCPIAhUJDo2bVXqgWPltfoVKKaV4yzgsu8pJCqRemkXYc6zPuSVKlpIoKwvmS+XTCeHLBchkQ\nZSnYFr1WnetDm56XImQGmkDV+sRml1VhEEYRy3zJcnFCNnsasTrGyCY4KFpem/7wEepml/E44c7x\nimlcogxBezhk48ZjtJtrlKVDVSp0oeE3LOx3/luW/88/Z9T1SHttdKXhXSxxlxLnL34jjW/+89S7\nzovGVZM05d7ZEfcvjgiyENdyuLFxlR3hIk+OqYIQzbYwt7bIdJvV0THJbEpUZiwNk4ntMZcGhcox\ndMnQNthWFdIUnIuYUo5oJEsGsU2/3MFt72Ou9zHW6oiGRVpIjuYxJ4sEKRWbbZd9O6H66K8R3/4w\nidA4vP4EtzeeYKZqtKnYC+cMkhDHFDT7Q5rD9c/6I+JTIasybk1vcfPsJlutLd6w+4bP+TU+Gx4Q\nxH9EUEoRZiXzqGAWX5JGVV1+rr5tXFYXvknbsz7nBmshFadZzklaMM4CwmyFoSJMFVHTK2wdTM3E\nKi20VIPi0tCViIRcz5FSx8lqtLQeHbtDo+0inYhZcZtpdIssXyCTHPtUx7pTkB+sCKIphSjAN3Ca\nPn67Q7u7QcNZx7TWKXWXVCsojVNKcURSzLiIU84XJotVi6SskasOqfLIVIjSpnSL+/TzGdI2yept\n6Gyx7vZ4ra3R6bgY/QFze43jQLE6O2Z2+/dYxkeYnsHDj72eja03cxYpdE1jt2FgBSPC5ZxCVIhB\nk9gsCPPwBUJo2A2adpOW3aJhNTD0zywF5lJykZeM85JpUSKVwtA0epbBwDLpmcandDwrqZBJiYwK\ngvmC+WLKeDZhEQZEeYoyBL5jsdO02Wkr6naFpungdUmsNivpEkQJYR4yXx6RTJ5Gn99DFJek4NkO\nndo2rfY+ovI4OF5x7yIiyQuseoPB7h5ruw9jmS3K4vK75XgC14A8XjEdHRHMRhRZiilMvFoPp7OL\n02nT7Hsv6i/ISnI6HXP3/IDRcoxE0fZb7A+22FYG1ckpMk7QPRdjY4u4KAmOTohXS8KqYG67jIRL\nBFQqx7Z0errGmiwwTYORlZKrCxrRnH5o0pe7eO0rmJtDzGENvW6yTC5HmS+CDLhc5b/X0DBuvYfo\nyd8myyNO9x/j5t6buTC6eLJiO5yznobUDYNGf0BrbeNlEUMhCw4WBzxz9gxhGDKwB7xq81UM+8PP\n+bU+Gx4QxH/EUEqxSkvmUc4szlnGBZW8/JzrjkHLs/AsgWsJPEvgGOIlGbyWRclJVjDJS+KqIikT\nlEwwVYwuVzjkUFbkcY6e6WhSo1QlylFouoBC4BctGqJFw2zg1xxyc8ZC3WGZ3qWsYuQywTnWUbcS\nkpMlcbYCW2LWbOqdJs1WE68+RBM9cnpEymYmCmZqQqQuSIoJSVRSxjaVrGHoHTytiaElaNox7fQ2\nXnLBXJnMpYlMBZuRxmuuXqH9htdhDvrM3TWOAsX04Bbje79Llk+otxo8fP0NhNYaF0WBEiktfY42\nn0BZ0eoNWd+9SrfWp2k1ES+hmR1Xkou8YJyXzIvLzbK2rjOwDPqWSccUn1JCknmFDAuqf/lLhO94\nB9Mi4WJ3k/nrHyfcXMPUJW1LsdlwGNR0GjUDIQTYdTKnx0rVWMUZUR4xXxwQj5+C2S1EOcNG4js+\nncY2rfYejugwm6fcO5wzCjJKFK3hkOH+DTrddZTyUUrD1MHWoShWLKfHBMsJRZ6gK4Ft2ji1AU5n\nB6/XoNl3XzSqugpC7p4fcjg5JilTLMNkp7fJbr1HPU4pTs8uR1UbdbThkDiMCI5PCaOYQClmjs9I\ns0hlgS4ktoCeLFnTJabjMrYKMjWiFs7ohoJ+tUut+xDm1iUxaJ7JOMg4nMWskgJDaGy1XbYaBuL4\no4Tv/3dkqzHj4RVuX38LJ+4WopJsRnO24oC2ZX5exCCV5Hh1zNNnTzNfzWkZLW4MbrA52MSyXpl0\nwQcE8QAvQErFKi1ekKRWaflChQGgaZe9DNcSuKb4BPIwcE3xKcNsorJiUlz+2p0VFVIpSllgqPSS\nMKqAPJqTRglZnFHIAqUrlK0wdAtdmviyQUO1aJgt6laN3B6zUncJqyNkmVJOErT7kNwMySYhJQk4\nOspxEJ6N7dZxzC61SGLdOsWaLjBqGdrXPMTiWovz2YKLqUGYekjVwpZ1HGGiLe9Rv3gSwYqlbxLa\nDkiD7e4Oj27vUWv7lL02p6bP3SBgcvdZgvPnoIrwGzX85jZa/Tpurcu19TX2TEU2vbiUndY3L2Un\nXYd3vOOTwolW3/KtjPOCcVYSVhUANSEY2JeVQuMPSYJKKVQuUVlFFRVUYUYQrpj8zrsY//o7mZsV\nuVHhlhG9MmXjbW+m85bX49c8hOUinTa54RMpl1VaEaURs/ld4vOPIGe3EeUcW5f4Tp1OY4d2ex+t\najKZJZyNlkzigqIsMbwa3e1thtvXccwW6DZ6KbF0SZEHhMEZweqCqkjRlMAxPWyvjeX3EbU29f6L\n+wtFWXJ0fsb98SEX0RQN6Dd77LaGrJcgL6bIKAINjE4H1W4TzWYEZ6PLprmmMbM9Jhjkd55FfOi9\neBfnDLOY4Ve+jeyrvoqVm1MUY7zlhE6o0Zd71HvXsLbXMAY+0hGczBOO5jFZIfEswXbHY92tkPc+\nQPz0e8hm58zqXW4/+jaOW9fJC8laMGc3DehZJq3BkNZwHeNlXshH0YineKUUVgAAIABJREFUzp5i\nPB/jaR7Xu9fZW9/DeYXXbTwgiAf4jMjKiiSvSIqKOL/8O84r4rykrF78nbBN/ZI0TAPXEhi6hq5r\n6BpoaIAikJJZUbEoK+KqunSt6uBoOZaMqcIpy9UFcRwTFSlxWZBQEMsCpIWqLKyyjq3VsHEwrBDM\nc3R9ikGFMcvR7udwlqNHGZpMMQ2Fr+XUqoSuYVHTbbTCRWYCeeNhqsf3CWsLzlYjTs8zVrFBWjqo\nsyVmbuBS0NQWGCJh4RmErqDq1BiYFo8pHbPbRQ16LPwup6uCxfFdsvkJtlbR9uv49T0KcxO/2eTh\n9TYNFZGsFliOS/f334/9Az9AgMbKrxH4dabDNdIf/EH4yq+kbRoMLIOeZeALgZIKlVeorEJmFSp/\n/r6QlGVBlEdMwhln0QVhNEG9819QD5b08pDeYkl7McMNIkgUKS5xZZJuXyH57u8me+ubWc1uEZ5+\nmGp+B1EtMYXC9+p0mnt06nsUhc94FHE6jQiyAomGVW/Q6PVpr+1Q9zsIvYZWaoiqpCxWxOk5UTil\nKjOEEtimh+10sGoDjHobr+PgNS3cuoWua5cJd4sFd88POZ6dUMgSz3bYaa+xozvYi4BquQJAtJqI\nbocSSXh+wfJixiotmQmLie2yVJJCZoh7t2n86q+weXpEe7UkGjQJ9xqIr3sLznBIK3XpVpvUetex\n99YwBzUSHY7mMWeLlEoq2r7FTsejq4Vkz/4OyXPvo4iWzJt9jvef4F7/VQSlTi+csx+vWLMt2oPn\nK4aXSQyLdMFT509xMj3BkAZXmle4tnkN3/c/+5O/AHhAEA/wslFU8gXSuCSQ8gUCycvPnptQKElQ\nSYKiIqgkhXp+4krXsZVE5gkyT6jyhKJIyauEXCWUekKhpWgoTCXwpYcvBLaxQnfGCGuFoST6qsSY\nKKqTkuJDx8hURzcFwhbUHJ2u69Jwapj9TQrNofAlYSvn2C6YphXBMyekep1Ms5HKxikFa1WGp6Ws\nOi1Sx8BwFHuWzmuFiT/Yxdq6RlBb4+4q4O79jzKd3sQoU5qOj2VtUrlDLN9n2LHJy5zoH/4YahVQ\nj1OEUphFQStY0vd8eu/+bYxSvYgMVH75HlWqIswjgiImyOasogvCbI7KQrQioImkYbv0/o9/QnM6\nw1ykZNImxiG1PFLfIqlrBG0ofMhbBurKOqLpYBgS32vRbu7QcneIYofRKGQcZGR5ibJs/Fabem+d\ndm8d126g46CXoLKcsliRVmOSdE5V5gh5SQqOfUkKot7C6zr4TRu3Zr7QW0jSjMPzE+6ND1lmK3Rd\nY6PRZ8f06cQFcrkEBXrNx+i2qXRJPFsQXiyIspJlLpmZDmPDIpYFVZVgaoquKtj9sX+IWEwJN2sU\nAwvTUNQLQSvr0PprP4TT3cDc7GEOfOZlxeEsZhrm6DqsNVy22w5+MiL+2G+Q3n2SvEiZ9Pc42n8D\np809VoVGI5hxJV6x6Zj0BsPPixiiIuLZi2e5N7qHzCU79R1ubNyg2Wz+kRrnHhDEA7wiqKSilBKl\nQCmQSiGVQnE5Pv+Jj6VSKAVBWTHLS6Z5ybIo0YWOKTQMQ8dCYVYFep6h5RlamVIVIVkVEKmItMoo\nygqVC4QETQWgrdD0ECVS9H/9rxGVAYmPjGvkmYfMbQxTYHkOvU6LbquDa3lggmbnFMc3mVQJZ7rN\n3GoQ6w65ZiI1C8+yccuC0PUv8yJkwd70jNdMRngPXcN44m3knQ1u5hofGN/hbHVKUVUYpo8jOhhG\ng55tsvF//QS1LKMThmzOFvSCBAwLZVjwi78IgKxS4mJBkE2JshlBPCNOZ1R5hF6mCKlhCQvXsHFt\nl2ZvE6e7huY0Sf7yN5NUBXnDIGxA1NApXY3CBc3U0B0NIRR2qeFoFvW3/Bk8sUEQGpxfZMyTkkqB\n7tWoNTs0euu0/A4WHhomVZohZUalr8hkQJGvkFWBIQ1sw8M221j1AUajiddz8Zs2jn9JCrKSTBdL\nzmdjxssJs2SORNJwfXbsBpuVjlgEINVlw7nTROoV8TIgmCwJk5JVobMUFpFtMakq4iJBkym2pti2\nDdYbDaqBR/DD34deV9iaSXNl0p4a1MclZrTC/PDvodVMzuOco1lM+Ku/hvVzP8vWrY+x2bIR/+3f\nIN5xSY/vkKJzvvUw93dez8TbIIljWtGKnTRkw7EYDIe0Pw9iyKqM29Pb3Dq/RRInbHqb3Ni4Qa/T\nQ38F13p/OjwgiAf4ooRSikQqoqoiriRhJYnKiqiSFEpRFAVZlpFnGXqRo5cpqorRtYxKxgglQYIG\nyLKg+qdvB7lC+QppSaxCokkDmfkUqUeRe1R2F9ut03Rc6q6Hp8BKQwyRgRGwECVj02BUa5GaNhkV\nRp6gZEXkdEmcJhYag6RgqDQs10MzDTTbRekGk2LCMhthZhF1adLQezQ/8Azt+RhbJqDAzDO8NKLq\n+0R/868SJROiNELKiqoo0ZWGIWxM08c8HuP/7nuxRmOMWgP9L3wD1ZveTJouyMMx4cfeS/Lkhygd\nRWWCboBuKgw0nELhxQXupEJfaiSFYFU5LL7xO4nzCilM7FoTv96h2Rji6z5Ct6iKkpKAQg+oZAx6\ngZIlmtQxlIElXCyzg+X3MJpN/L6H17Qug3kUBGF8SQiLCRfhlFzlaEDNcekbLhvSoBlmqEpejqe2\nfaShSIOQ1SRkFVesCsFctwgsk3lVkJcxeVVg6BUNobFf8+mu94jsjHR+F2M2wf///h2t45L2aYGz\njLFWIwinzPcfYvzrv80kzCgrRe23f4Od/+nvsjY9plgTxJseed0heOtXcvKnv4GDjS9jonnoUUAv\nXLCrK4auQ6ffpzlce9nu51KW3F/cf2EyqW/3eWTtEYa9IcYf8YK+T8SXLEFomvZ1wNsBAfyMUup/\n/XTHPiCIP1nIpSSqLm9hVRGVFfMkZZGkL5BGXhWUsqCUJVIVqJvPUr37XVRVAVTgFii/Qnckwsox\nCzBwEP4GagYyEchCIDIHqzSxDA/L9tA8B+HoZHZBqBcs5ZJQzylFimYoEqeG0n0cKRkUMfvZimtp\nSE3TMUyHwvQYaSXzKiBZTijjFFkqdJlhVRE6GpVuULQHyM4Qw61hOy6O7WE7NpZtIgSI27cQ7/kt\nchSFoVMYGqWuIx/ao2r4SHL0wzsImWNKhZ1KzAjECrLUIixsAulRYCE1HYXAtH3sv/ANeH6Xmt3A\n1CwqvaLSQ0otAlGgi+IFMjA0EwsfofsIt45u1zBrPt7gslKwPYM0zhnNp4wWY8arCXEVowG2Iejp\nDj3NoluBnVxOZ2mGwGg5KBPSKCKYpyyCinlhsNQtFhrMi4SsSihViWVo1Eyddd9lvd3G6HusgkPU\n5BB7mdCoWrTdXRo3z7Hf/mMwOWYmDMa1NtP2kOpv/S2Mr/4q+vXLxZet114j1cekA4fcFkysFsfG\nGkdrjzH7/r+DFy3ZSGN2HJPNbpdGr4/fbL3s0J5CFpysTnhu9Byz5YyW0eLh/sNsDbc+p8kkpdQr\nIj29VIL446OwTwFN0wTwT4A/AxwD79M07ZeUUk//8Z7ZA/xRwNJ1LF2n/YmTgs0alVLElSQoSqI8\nJ81zsqIkLXLy7haZ9Eh/5ZfJohWZaZBZNpmhk1kQNqCoK9QTV9GKFaQpIs4wkxVGlEGk0FMdO7Ww\nQwt75dOufPqJDaJGbkNoZQRmRGAfE3oGdzyPO7UW7/YbdLKQzWRJK13iSQspHQq1S+gqygr0SsOu\ndAQFXsPFsyz0VYZaFRhmAMaSXCvJtRK0AnVwB9VxQZPoegW6QtMlYnmI1X8UQQs1dijjNklpsyw8\nysqi0gQYAlNoCKXhILDQMIWO/qYvh9aAykjJzDMKUaJrYGoGVmVgag0M5WHaDTS7hu54WE0bu2lh\nuya2ZwCXDeZnj48YLy9YZEuUkghV0dFtdnDoS41aLNA0DU1IhGegDxyUJknjlNkoYh5UTHKTeSWY\nScWyisnlDE0obNukV7fZ6Q4YdFroNYMwPie6eAbjyTn13KKlb9DtXsXZ2oJenelbJRcNm9k/+3nk\nZILVbbP217+Lwbf+JdquiQrOiD/yb5huBeSizrnT5761yak9JHI9mrM5r52fs92osbF7lXq3/7Jl\nJLh0Px8sD7h3cY8gDPB1n8d7j7O/sf+SJ5OUUsTLBYvzM7xmk/b65ss+n88XX1QVhKZpbwZ+RCn1\nnzz/+IcAlFL/4FMd/6CCeICPQ0pJcf06xfk5uWlSmCaFZZGbJtnWJulP/jhJmVLJiFJF5FVIXkXk\nRUhRBORZQB5FlKsYFZaYZ3MMPExZw5I1NOkCJkqTlBTMXZ2ZKwgdjUov8MqAZjannqywJQipU2qC\nEg2t0jFLMIX9/AW8RC8zEBLN1FC2g25bKE1HWyxRSkfLdWQl0HOBKnVUISiuvoa8sJB37qIVGbqu\n0DWJicLQCgwkOgWmptC1Ejwb8epXI3b3MHSBqQxMHAx1WR3odu2yOmjY2C0b2zcRpk5ZlazSkDAJ\nWSUhYRyyyJeUVQ5FTh1BTzfpVzpt3UWgoZsVwhXoFlRAnheksWQVSVaJ4iLRmeaSWVUSyJhSVBiW\nwPNshu0W290OrbZHVS1YTQ6Q8xF6lGKXAl816TSu0Ni8Cmtt5jqM44J5lKPU5Yj2oGEzqNs0HQMZ\nnpHeez/5/Y9RXJwTYXL6a+/ndtVh7PYASW92wdXD++wpRe897/m8sp+VUkzTKXcmdziZnZClGR2z\nw15nj63+1kueTFJSEswmLM7PKbIEw7Rpb2zS6PVf9rl9OnxJVhDAJnD0CY+PgTf+MZ3LA3wJQdd1\n7L//97G/53tgsfiDf3ge6kd/lOL6oxRFgZTy0lvwh24AUpZUVUpehsTf+g1EJAQNi1XDJ220qbw6\n6C6aXafb7NMrC4pMY26bzN1N5vVNlhT4xZx6McGWGZpSSK0i0koguxwLzgWy0KEUyEpHJjpIH+HU\n0UIdCsVll0VDV5fubeEZmDWJZZe4XgPn1i3srMBWYCgQwsR4/Zsx9q+jYz5/M9A1CzQDzbDQ7Bpm\nzUavmWi2oNAykiplkV0QTELCk4ikjCiKFFVVaLJEoPA0g6HU6WHQMxvYqkIIhfB1NDOlqCRJprG8\nKJitJKtMsMohKCFWklAWpHqBNBW2Z9Nqtdnu9Rh2G7hmQjw9Ir14jvm9CFFAXfr49hqNxia1/iaq\n22QmBM9kBYtJdPmxWoLdrke/7tB0DMrVMdnN9zG//zHKyQW51Jg0Nhhf/Wru1PeYuW/G+9VfYe/k\nHlcPD9g/O6WOjvipn4KXSQ6FLDgNTrlzcYfJcoJWaQydIVd2rjDoDl6ylFSVJauLEcvxiKossF2f\n/t5DZFYN9TJX5Xyh8MVGEJ8VmqZ9D/A9ADs7O3/MZ/MAX1T49m+/vP9D5jTt278dCz43V+p/9SPw\nXd8Fz5vZABSAaaL+2c9QffO3oKqK2TLi/HTJ5N4ZR4ennJ+eMG+sMbEMLFVQz0JqeYBQOUroyOrS\nOSKUwqrAqsCQCj1TaIWP6TewVwlummOXFUiFEqD2r4G/hq509I6Jtr2Heu4OKoxRpkXxqtdSdh6C\nVAdbIBwdbAPN0hGOidQq4nJCnKyIFgFZGSOLAlWVCCosoKYEawhqQlAXOjXdvKyGTB3NqlBGSlmt\niCuT1VwyC2AeG6wKnbDQSFAk0wn52QGyjNAcA/3aVZyHdtlor7Ez6NOwS0Q4Ip1+jOxsSplJLGXR\nN3rUGg/T6O5QtTtElskFkjtpSby6XHtRcwz2+z6Duk3dNiiWB2TPvZ/p/acopxNSqTNtbHK6/9Uc\n1zZZahZlEtNcrXhif5PrX/FGNn/st7Dv3X/hu/HCd+ZzQFREHCwOuHtxlyAM8HSPK7UrXBlcodls\nXrrWXwKKPGM5Omc1GaOkxK03aQ7WWGLz1DQmypasNR2am595/fsriQcS0wM8wKfDO94BP/ADMJ1e\nPu524e1v/7QXFaUUs+/7rzl81+/w5Ktfw8H2LmG9RmlZWKaObSqc6TlKK0i1klyvUKpAlxVuDq5l\noZUVWp6hFyWOBK9U2H4Lc30HzdCeby6XVFpBQYXUSiSXY8UoRSUvt0EpNDTt8h7t0tQoZIGn6biA\nr2nUdIWv6XiGgSV0dB00AUo3kLqJxKRUJrnUCVKDReawyg2CUieVOqnQKA2FNCX4BmIyxnvv71EL\nVrTCgFoWYtsS8zu/Fe3KOsVihJ4VCKVjixa+v4bT2kK11ggdk6VUxEXFxy9JtqnTcExankm/buMa\nOsXyHtnBB8juPUU1mxFLwai2xXF7j1N/gxUCvcxpItm0TPY6bbb7PTqd7stuOH/8s/10MtJmbxPf\n919yMzmLYxbnp4TzGQC1The/N2SSaS+4umuOwW7XY1h3XpGc8y/JKSZN0wzgJvA1wAnwPuCvKKWe\n+lTHPyCIB/iixPd+L/z0T1MqxVPXb/DkX/xL3H/oGmklKWczDFkhVIVBhlblpJqi1CTSMDFziTuP\nEGWOUgWKDIMEt1fDaXdwLAPb0LEtHVfoOKaOb5nYQmAbBqauIwDt48aUUoKUaAoc4YBpkkmD/7+9\nO4ux5CoPOP7/6tR29769TM/mmbGDLQIkMg5xCATbKCYsLywiiAjFKIpkI5FNUR6IUBSDsUAJEZEi\nQpRICIhsEC8oViALRkp4AJQEhYDN5m1mPEvP9HL79t1qP3mo2+O2uWNPz3ane76fVKq61T0936nu\nW989S50zsh6xDRjlLlFqiFKHKIXo6GnSnzxNmmQklRr5kZvIFxZJHUviWAhA6h5SC6jVK1SrARWv\noJJFVD79KdzRBl4oUHHBdUEMEoa4v3IXfrgHr7EfZvYzqNbp2ZzCljc/z3Vohi7Nikcz9GhWXALX\nYIuUZP0ZkmP/S3LscbLVDhuFy+nKfo63DnO2toceLqEUNBFubDU4MjfLwbnZbd20z+e8zUgL22tG\nAhhudFlfOs2o18VxDM2FPQTteU71c06uj8hzy2zd5/Bslbn6lV1UaEcmCAAReRvw15TDXD9rrX3w\nfN+rCULtBNZa4pUVjj57jKP/8Q1Off//6DRbpH5A6nt4RsgW5xn5BfHaMuQGkzk0IqiMHCQ1WCtQ\naWAdFxwfKy6IjzUGxMUxIA64RjDGwXiC5xk8t3wgURCSDNJcyAvIEfLNBxoFCsfC2jLyxI+xeYK1\nBUiGGDC//EsEN/8cgZdTcTJq2YggG+AnG3jxCMcWFBbst7+NpA4yKpBBjtPPkH4KvYToC18kdR0Q\nwRihGXq0Ku44GXiEnsHanCxaJu2eIFt+huzscfLVs+S9ISuZx6lwL8ebh1itzDHyAqoOzIchL5ud\n4cjcHPvbM7gXse7CC6V5ymq0yqnuKU6tnzrXjHSgfmDbzUi2KOh31lg/c5pkNMR4fjlXV73NiW7C\nmY0IKGeNPTxXvWxLC7+UndpJjbX2a8DXph2HUpeLiBAuLPDyhQVefttrsF/4At0HP8ZRz+Pkz9/C\nmV99LXF7jqzIKZZ+yiAs6FYsS80U2w6oWaE9hKA6j+n0IClAPKi3IKxSFIJ1PKy4xOJS4ELuYTMH\nbPksRJk9LOJacCzGKzDG4roFgWdxxOKu/Bj34AjvXA0nw1iLOfNtzHwHzxaYwpI7htzUGLp1bGUv\n+DNItY33g+/gnDxV1lwAr8jwsxR/YY69iw2a44RQ9ctkkKddssFRkmeeYXT2OPnKEmm3x6if0iGk\n67VYCw9ytjlLtzZHHFRpeh4HG3VuWZjlxrk55sLgsjwnMMpGrAxXONU9xZnuGYajIU7hMOPPcNue\n27bdjJRGEb21VXory2RpjB9WWDh8E4lf5+lORGdtA2OEG2arHJqtXtJCX1fSNZcglNrt5J57mLnn\nHm4FbgVsmrK0vMwzp05y+l8eZSUIaIWGYdgg8lMSZ8DAxNjVp2jHA/zU4OQBTuRjX/ULcMONFDjY\nJIc0RvIRNsvAQCGCiIN1HBwjGMfBiOCMJ1l0cHCsAxiclePYAmwuSGGxWbl4UZ5Z5M5XkdTahI05\n6tUWtcDHNw6B6+Bvbr/zbvw//AO8/gZ+nuFgoVql+PifkM+mZKOTpEvHWTt7lOT0SZLVLsNeQrfw\n6LoN1rwmneoRBvvaJGGTxA8IwoBGEHBLs8HLF+Y41GzQuEwje3pJj+XBMifWT7DaWyWKInx82n6b\nW+ZvYXFmkXq9jneBU3incUS/s0Z/bZVkNASg0mgye/AQPSfkh2sj+tEGgefwsj11DrQr216f5Wq7\n5pqYtkObmNSu89BDZPfdx7FWm6OHb2Bp3yKn9u1jWHeJnYJCUry8Ty3p0Yx6zBQps3fewULgU51t\n4zWa5GGVJHOJI0iijKwQwGAcFyMG1wvxPR/fr+D7AYEX4PsB7t1vwn32WUxhkdwi1iIW7Pw8Tq2K\nd/QZ5NAh8o99hOI97yLPY/IspigS8nRIHvUpHv0q+Rf/kaLfIZ9vk991J8WNR8jW1ol6MeupYcOp\nsRG06Fbb9GttstoMab2JX61RDwJalZC9tSoH6nXmgnIK9MpluJFaa1mP1zk7OMuJzgm6gy5RFFEz\nNWaDWQ7MHGC+NU+tVtvWSKRBp0N/bYV4WA7BDWp16u05guYMZ4flBIFxWlALyo7nvc0r0/G8HTu2\nD2I7NEGoXekF60gMPvoAR+9/gOOLe3n2wAHOzs/Qq4dkHojEmBv24dqcWp4zIwVtFxarPvPtKjPN\nADcwZMalsOVki0VuKYrxlhdkeUGRW/jRj5B//zokKViQHHAMTl4g5OSuYF0H63vw2tfCgYOQJtgk\nJc8y0sIhLQwpDikuieORGp/ICehXWgyb82TNOdLaDH6zQa1apV2psLcSsC/wmPNdGsYQXsZP1VmR\n0Yk6LPWWOLl+kv6wTxInNL0mc+EcB9oHmG3OUq1WL3jSvCxNGXRW6a+tEQ16AASVGvXZOYLWDOsJ\nLHUjOsOEooB2zePwXI25mn9VZ2x9MZoglNpNjhyBY8cAKBBWqi2e2HuQn77qNpbvu4f+YJms6JEk\ng3LyvzSlUghVx9A0LjOuwXcEIxaXsnZgKDBCubdFOTj2+DHkh4/hDIcQhpAmSBZTiCERn8S4JCYg\nrTaI77ibxAtIvQqZF5J7IUUQkHoBhV/BBhUcz8MLfCrVGnO+YY/vsc/3mfVdmm45tcq2TViIife9\nj7Qol33tJ306ow5rgzW6wy5xFFPkBS2vxWJtkf3t/bQaLcIwvOAbdp6lDDodemurRP1yzQq/UqXe\nniVstulmDmc2IlYHMUVRPt292AxYbIU0r1LH83bs2E5qpdQEDz4I994LwyEOlj3DdfYsJdz+kQ9x\n9nWv4/T6kOVOnzjqkxQd+kWXtXiN7qBPJ844ah18E5aL+gQ1PL+KZ0IcY8CWy14WFuyh18Pry2RB\nUcDDD7P5YEKBYHMwWY4bFZg3vJ3Q92l6LlWnHHZbNQ5Vx6HqlvvQcfAdoWYM3uVoVnnoIbj3XpJk\nRL/usBGdpvOJP6KTPcXGbb9IlmUkSYKLS82t0TRN5lpz7G/vp9FoEAQXPnw0jSJG/Q36a2uMehuA\nxQsqtPcdoNJqs5Ebnt2IWHm2R1GUz20cbJfPLjQr7jVTW7gUWoNQaqc4zyfnTaMkZ2kj4nR3RH+U\nkKcRPiPybIVhtMIg2mAQbZDmCcZxcIyhUq3TaszRbu+h3dxDJajjOhWMCSkQkrveSLa0hFsU1EZD\nwiTGT1O8/fvxnnziit8ErbXEeUw/7bMRbdB5x1vojLr0K+VcW6nn4WdCtdqm+pGP0660adfbNCoN\nwjAkCIILajoqipx4MCAa9In6PeJ+nzwvZ6J1/YD67BzV1iw9azjTjVnpx+SFxXcdFpshi82AVsXb\nMUlBm5iUuo51RylL3YiljYg0KxCb0/Cg5oJjh/T7y3S6y3QHHfpxj9gm5JLjhxXCapVKrU67OcvM\nt75L44FPUNkYYApwCosJKjh/+UnMu9+DcQxGDI445/aOOOe9UaZFSpqn5X48bXuap8R5TJzFxOl4\nP96SLCHLMtI0Jcsygj97gGpkqY0KZjdi2t0R9VFMEMf4w+EF9yOkcTROBn3ifp94NGQ8mQpeEBLW\nG4S1Ol6lRt+6nNmIWO7H5LnFcx32NAIWmyHt6s5JCltpglBKURSW1UHC2V7E+jBllJRzS7lGaFd9\nWoEhdDKK4QYb6x1WO2foDNYYZiNim5J7lvzpn2K+8y3czjqmUcfc/SZ49avP3Yy37kVkvB65WyaP\n8ZbmKXEWk+c5RVH8zIYFV8ajrBwXV8rNczwCE9CutWnX2tTe+OsETz6Jn6Y877Z8+DAcPXqea5AT\nD4fnagbRoE+epQCI4xDW6gS1OmGtjqlUGaSwPkrpjrc8t7hG2NMoawqz11Bn88XSPgilFI4jLDQC\nFhpl23uU5nSGCZ1ByvowYblXToTnmgoz7RYHD9zMKwKDlw7pdTr0uh361VtIX/lm8iInL3KsCPQc\nyhWOBFynPGeEwhEstuzToCCzGSkpDg41p4bruAReQOCONy8gdMtht8aYidvzagUf/nDZF5Om507l\ntRrZ/feTrXfI05QsTciSpDxOYpIoYmvtoNJsnUsKqfHZiDLODFPWVxOGcTkTsAjUApe9zZCFRsBs\n1Z/60NRp0BqEUtexKM1ZH6asDRLWhwnDcQ3DGGGm4pW1jNAQklNkCWkck8URaRKTxTFpEmOL4nk/\n03EMxvdx/QDH83E9D+O6GGMQR8opz8c1DSifNGfLJ/LN1+X3lf0QeZKQjW/4+SOPkH3mM2Qry+R7\nF7G/fQ/cdefzYjCuh+v5uL6PF1YI62XtoJ9yrmawWTuAci6oVsU7tzVDF/caf4jtUmgTk1Jq2zYT\nRmeY0BkmDOPnpjsPPIeqbwg9Q8UzVPxy71Hg5ClZOk4acUyWJGRJebz56f1yEMfB9XzMOPG4foDr\ne+PXPo7rkoshySHOcqK0oBeXyWCzLJu1g1alnCm2VfGo+tdXY4qbGYrvAAAFnUlEQVQ2MSmlti30\nDHtbhr2tcnnMOCsTRj/OGCU5UZqzNkiI0xfWGhgnjiqVsEGlUSaPGc8pE4i1ZSVBAGvHI2fHizVZ\nNicsH3/t+cdlUvDIxSVDiLOCOCsYpHl5HBdE/ZwkG5Fkz48Lnqsd7GtVrovaweWkCUIpdV6Ba1hs\nGhZfcL4oLKM0L7dx4tg87o5Ssvz8tYatLUoicq6zWcbrVgAIzx1bcuIsovjZez+eW84HFbjl2hGh\n5xB45ty5wDX4riaDi6UJQim1bY4j1AKXWjD5FpLmBaM0J0rKZp5882G7c7WHsm/BUj6HZynPbz1m\n/DURxp3ahsB77sYfuM512XF8NWmCUEpddp5x8IxzTU4zoS6c1r2UUkpNpAlCKaXURJoglFJKTaQJ\nQiml1ESaIJRSSk2kCUIppdREmiCUUkpNpAlCKaXURDt6sj4RWQaOXcKPmAdWLlM407IbygC7oxy7\noQywO8qxG8oAV64ch621Cy/1TTs6QVwqEfmfC5nR8Fq2G8oAu6Mcu6EMsDvKsRvKANMvhzYxKaWU\nmkgThFJKqYmu9wTx99MO4DLYDWWA3VGO3VAG2B3l2A1lgCmX47rug1BKKXV+13sNQiml1HlcdwlC\nRH5TRB4XkUJEXrPl/BERGYnI98bb300zzpdyvnKMv/anIvKkiPxERN48rRi3Q0TuF5GTW67/26Yd\n03aIyFvG1/tJEfnQtOO5GCJyVER+ML7+O2axdxH5rIicFZHHtpybFZGvi8gT4317mjFeiPOUY6rv\ni+suQQCPAe8Cvjnha09Za28dbx+4ynFt18RyiMgrgPcCrwTeAvytiJirH95F+dSW6/+1aQdzocbX\n99PAW4FXAL81/j3sRG8cX/+dNET0c5R/61t9CPiGtfZm4Bvj19e6z/Gz5YApvi+uuwRhrf2RtfYn\n047jUr1IOd4OfMlaG1trnwGeBG6/utFdd24HnrTWPm2tTYAvUf4e1FVgrf0msPaC028HPj8+/jzw\njqsa1EU4Tzmm6rpLEC/hxnE17j9F5A3TDuYiHQCe3fL6xPjcTvD7IvL9cVX7mm8S2GInX/OtLPCo\niHxXRO6ddjCXaNFae3p8vAQsTjOYSzS198WuTBAi8qiIPDZhe7FPdaeBQ9baW4E/Bh4WkebViXiy\niyzHNeslyvMZ4CbgVsrfxV9NNdjr06+N//7fCnxQRO6YdkCXgy2Hau7U4ZpTfV+4V/M/u1qstXdf\nxL+JgXh8/F0ReQq4BZhaZ93FlAM4Cdyw5fXB8bmpu9DyiMg/AP98hcO5nK7Za74d1tqT4/1ZEfkK\nZdPZpL66neCMiOyz1p4WkX3A2WkHdDGstWc2j6fxvtiVNYiLISILm525InITcDPw9HSjuiiPAO8V\nkUBEbqQsx39NOaaXNH4Tb3onZSf8TvHfwM0icqOI+JSDBB6ZckzbIiI1EWlsHgO/wc76HbzQI8D7\nx8fvB/5pirFctGm/L3ZlDeLFiMg7gb8BFoCvisj3rLVvBu4APioiKVAAH7DWXlMdRludrxzW2sdF\n5MvAD4EM+KC1Np9mrBfoL0TkVsqmgKPAfdMN58JZazMR+T3g3wADfNZa+/iUw9quReArIgLlfeFh\na+2/TjekCyMiXwTuAuZF5ATw58AngC+LyO9Szvj8nulFeGHOU467pvm+0CeplVJKTaRNTEoppSbS\nBKGUUmoiTRBKKaUm0gShlFJqIk0QSimlJrruhrkqdSWJyBzl5HAAe4EcWB6/vn08V5NSO4IOc1Xq\nChGR+4G+tfaT045FqYuhTUxKKaUm0gShlFJqIk0QSimlJtIEoZRSaiJNEEoppSbSBKGUUmoiHeaq\nlFJqIq1BKKWUmkgThFJKqYk0QSillJpIE4RSSqmJNEEopZSaSBOEUkqpiTRBKKWUmkgThFJKqYn+\nHzA29yRT774qAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x12cb14f10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "cons_T = np.linspace(-15,15,20)\n",
    "\n",
    "plt.scatter(cons_T,  [np.mean( real_risk(np.ones(n)*t, beta_cons, beta_x, beta_x_T, beta_x_quad_T,x)) for t in cons_T ] )\n",
    "\n",
    "def real_risk(T, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x):    \n",
    "    if np.isscalar(T):\n",
    "        risk = T*beta_cons + np.dot(beta_x.T, x) + np.dot(beta_x_T.T, x*T) + (T-np.dot(beta_x_quad_T.T,x))**2\n",
    "    else: \n",
    "        risk = np.zeros(len(T))\n",
    "        for i in range(len(T)): \n",
    "            risk[i] = T[i]*beta_cons + np.dot(beta_x.T, x[i,:]) + np.dot(beta_x_T.T, x[i,:]*T[i])+ (T[i]-np.dot(beta_x_quad_T.T,x[i,:]))**2 #+ np.dot(beta_x_quad_T.T, (x[i,:]**2)*T[i]) + np.dot(beta_x_high_freq.T, np.sin(x[i,0:HIGH_FREQ_N]*FREQ)*T[i])\n",
    "    return risk\n",
    "def real_risk_scalar(T, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x):    \n",
    "    return T*beta_cons + np.dot(beta_x.T, x) + np.dot(beta_x_T.T, x*T) + (T-np.dot(beta_x_quad_T.T,x))**2\n",
    "\n",
    "bounds = [(min(T)/(0.5*d) , max(T)/np.abs(0.5*d)  ) for i in range(d) ]\n",
    "bnds = tuple(tuple(x) for x in bounds)\n",
    "\n",
    "def get_oracle_treatment(beta_cons, beta_x, beta_x_T, beta_x_quad_T, x): \n",
    "    res = minimize(real_risk_scalar, np.mean(T), args =(beta_cons, beta_x, beta_x_T, beta_x_quad_T,x))\n",
    "    return res.x \n",
    "\n",
    "oracle_T = [ get_oracle_treatment(beta_cons, beta_x, beta_x_T, beta_x_quad_T, x[i,:]) for i in range(n)]\n",
    "plt.hist(np.asarray(oracle_T))\n",
    "\n",
    "fig=plt.figure()\n",
    "for i in range(50): \n",
    "    plt.plot(cons_T, np.asarray([ real_risk(t, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x[i,:]) for t in cons_T ]), alpha = 0.3 )\n",
    "\n",
    "for i in range(50): \n",
    "    plt.scatter(T[i], real_risk(T[i], beta_cons, beta_x, beta_x_T, beta_x_quad_T, x[i,:]),color='r')\n",
    "plt.xlabel('T')\n",
    "plt.ylabel('y (loss)')\n",
    "# plt.axis('off')\n",
    "\n",
    "\n",
    "fig.savefig('outcome-model.pdf')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Constant policy "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'disc_res' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-6-0ccc84649c8a>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mreal_risk\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdisc_res\u001b[0m\u001b[0;34m,\u001b[0m  \u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      2\u001b[0m \u001b[0mdisc_res\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'disc_res' is not defined"
     ]
    }
   ],
   "source": [
    "# real_risk(disc_res,  )\n",
    "# disc_res"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "LAMBDA = -3\n",
    "\n",
    "def pol_eval_wrapper(beta, args): \n",
    "    data = dict(args)\n",
    "    t_lo = data['t_lo']; t_hi = data['t_hi']\n",
    "    tau = np.clip(np.dot(beta, data['x'].T) , t_lo, t_hi).flatten()\n",
    "    data['tau'] = tau\n",
    "    return data['sgn']*(off_policy_evaluation(**data))\n",
    "\n",
    "def pol_eval_l2_wrapper(beta, args): \n",
    "    data = dict(args)\n",
    "    t_lo = data['t_lo']; t_hi = data['t_hi']\n",
    "    tau = np.clip(np.dot(beta, data['x'].T) , t_lo, t_hi).flatten()\n",
    "    data['tau'] = tau\n",
    "    return data['sgn']*(off_policy_evaluation(**data) + LAMBDA * np.linalg.norm(beta,2))\n",
    "\n",
    "\n",
    "def pol_eval_l1_wrapper(beta, args): \n",
    "    data = dict(args)\n",
    "    t_lo = data['t_lo']; t_hi = data['t_hi']\n",
    "    tau = np.clip(np.dot(beta, data['x'].T) , t_lo, t_hi).flatten()\n",
    "    data['tau'] = tau\n",
    "    return data['sgn']*(off_policy_evaluation(**data) + LAMBDA * sum(abs(beta)))\n",
    "\n",
    "def off_policy_evaluation(**params):\n",
    "    \"\"\"\n",
    "    Takes in a choice of kernel and dictionary of parameters and data required for evaluation\n",
    "    tau is a vector of treatment values (assumed given)\n",
    "    If y_samp, T_samp is present, use that instead. \n",
    "    \"\"\"\n",
    "    THRESH = params['threshold']\n",
    "    y_out = params['y']; x = params['x']; h = params['h'];Q = params['Q']; n = params['n']; t_lo = params['t_lo'];  t_hi = params['t_hi']\n",
    "    kernel = params['kernel_func'];kernel_int =  params['kernel_int_func']\n",
    "    if ('y_samp' in params.keys()):\n",
    "        y_out = params['y_samp']\n",
    "    if ('T_samp' in params.keys()): \n",
    "        T = params['T_samp']\n",
    "    else: \n",
    "        T = params['T']\n",
    "    if ('x_samp' in params.keys()):\n",
    "        x = params['x_samp']\n",
    "    loss = 0\n",
    "    tau = params['tau']\n",
    "    clip_tau = np.clip(tau, t_lo, t_hi)\n",
    "    for i in np.arange(n): \n",
    "        Q_i = Q(x[i,:], T[i], t_lo, t_hi)\n",
    "        if (abs(clip_tau[i] - t_lo) <= h):\n",
    "            alpha = kernel_int((t_lo-clip_tau[i])/h, 1)\n",
    "            if alpha < 0.5: \n",
    "                print alpha\n",
    "        elif (abs(clip_tau[i] - t_hi) <= h):\n",
    "            alpha = kernel_int(-1,  (t_hi - clip_tau[i])/h )\n",
    "            if alpha < 0.5: \n",
    "                print alpha\n",
    "        else:\n",
    "            alpha = 1\n",
    "        loss += kernel( (clip_tau[i] - T[i])/h )*1.0 * y_out[i]/max(Q_i,THRESH) * 1.0/alpha\n",
    "    return loss/(1.0*h*n)\n",
    "\n",
    "\n",
    "def off_pol_epan_lin_l2_grad(beta, *args):\n",
    "    \"\"\"\n",
    "    Compute a gradient for special case of Epanechnikov kernel and linear policy tau\n",
    "    \"\"\"\n",
    "    # THRESH = 0.001\n",
    "    d = len(beta) \n",
    "    params = dict(args[0])\n",
    "    #! FIXME x vs xsamp\n",
    "    tau = np.dot(beta, params['x'].T)\n",
    "    params['tau'] = tau\n",
    "    params['beta'] = beta\n",
    "\n",
    "    THRESH = params['threshold']\n",
    "\n",
    "    [f, g, nabla_f, nabla_g] = f_g(**params)\n",
    "    # compute gradient vector via quotient rule\n",
    "    if g < THRESH: \n",
    "        g = THRESH  \n",
    "    \n",
    "    return params['sgn']*(np.asarray((g*nabla_f - f*nabla_g) / g**2 ) + LAMBDA * 2*beta)\n",
    "\n",
    "def off_pol_epan_lin_l1_grad(beta, *args):\n",
    "    \"\"\"\n",
    "    Compute a gradient for special case of Epanechnikov kernel and linear policy tau\n",
    "    \"\"\"\n",
    "    d = len(beta) \n",
    "    params = dict(args[0])\n",
    "    #! FIXME x vs xsamp\n",
    "    tau = np.dot(beta, params['x'].T)\n",
    "    params['tau'] = tau; params['beta'] = beta\n",
    "    THRESH = params['threshold']\n",
    "    [f, g, nabla_f, nabla_g] = f_g(**params)\n",
    "    # compute gradient vector via quotient rule\n",
    "    if g < THRESH: \n",
    "        g = THRESH  \n",
    "    l1_sub = np.zeros(d)\n",
    "    for i in range(d): \n",
    "        if beta[i] < 0: \n",
    "            l1_sub[i] = -1 \n",
    "        elif beta[i] > 0: \n",
    "            l1_sub[i] = 1\n",
    "        else: \n",
    "            l1_sub[i] = 0 \n",
    "    return params['sgn']*(np.asarray((g*nabla_f - f*nabla_g) / g**2 ) + LAMBDA * l1_sub)\n",
    "\n",
    "def f_g(**params): \n",
    "    THRESH = params['threshold']\n",
    "    y_out = params['y']; x = params['x']; h = params['h'];Q = params['Q']; n = params['n']; t_lo = params['t_lo'];  t_hi = params['t_hi']\n",
    "    kernel = params['kernel_func'];kernel_int =  params['kernel_int_func']\n",
    "    if ('y_samp' in params.keys()):\n",
    "        y_out = params['y_samp']\n",
    "    if ('T_samp' in params.keys()): \n",
    "        T = params['T_samp']\n",
    "    else: \n",
    "        T = params['T']\n",
    "    if ('x_samp' in params.keys()):\n",
    "        x = params['x_samp']\n",
    "    BMI_IND = params.get('BMI_IND') # propensity score for warfarin data evaluations \n",
    "        \n",
    "    loss = 0\n",
    "    g = 0 # also keep track of normalized probability ratio quantity \n",
    "    partial_f = 0 \n",
    "    partial_g = 0 \n",
    "    tau = params['tau']\n",
    "    clip_tau = np.clip(tau, t_lo, t_hi)\n",
    "    Qs = np.zeros(n)\n",
    "    for i in np.arange(n): \n",
    "        if (params.get('DATA_TYPE') == 'warfarin'): \n",
    "            Q_i = Q(x[i,BMI_IND], T[i], t_lo, t_hi)\n",
    "        else: \n",
    "            Q_i = Q(x[i], T[i], t_lo, t_hi)\n",
    "        if (abs(clip_tau[i] - t_lo) <= h):\n",
    "            alpha = kernel_int((t_lo-clip_tau[i])/h, 1)\n",
    "        elif (abs(clip_tau[i] - t_hi) <= h):\n",
    "            alpha = kernel_int(-1,  (t_hi - clip_tau[i])/h )\n",
    "        else:\n",
    "            alpha = 1\n",
    "        Qs[i] = kernel( (clip_tau[i] - T[i])/h )/max(Q_i,THRESH)\n",
    "        loss += kernel( (clip_tau[i] - T[i])/h )*1.0 * y_out[i]/max(Q_i,THRESH) * 1.0/alpha\n",
    "        if abs((clip_tau[i] - T[i])/h) >= 1:\n",
    "            partial_f += 0 # don't add anything to partial derivatives \n",
    "        else:\n",
    "            partial_g += -1.5 * ((clip_tau[i] - T[i])/h ) * 1.0/max(Q_i,THRESH) * x[i,:]\n",
    "            partial_f += -1.5 * ((clip_tau[i] - T[i])/h ) * y_out[i]/max(Q_i,THRESH) * x[i,:]\n",
    "    norm_sum = np.mean(Qs)\n",
    "    return [loss/(1.0*h*n), 1.0*norm_sum/h, partial_f/(1.0*n*h**2) , partial_g/(1.0*n*h**2) ]\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def oracle_eval_wrapper(beta, args): \n",
    "    data = dict(args)\n",
    "    t_lo = data['t_lo']; t_hi = data['t_hi']\n",
    "    tau = np.clip(np.dot(beta, data['x'].T) , t_lo, t_hi).flatten()\n",
    "    x = data['x']\n",
    "    return np.mean(real_risk(tau, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x))\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "## Train GPS on your simulated data\n",
    "def norm_T_Q(x,t,t_lo,t_hi): \n",
    "    return norm.pdf( t - np.dot(x, TRUE_PROP_BETA), loc = 0, scale = T_SIG )\n",
    "\n",
    "trainind = np.random.choice(range(n),size = int(round(0.4*n)),replace = False)\n",
    "train = x[trainind]\n",
    "test_mask = np.ones(n, dtype=bool)\n",
    "test_mask[trainind] = False\n",
    "test = x[test_mask,:]\n",
    "t_lo = min(T)\n",
    "t_hi = max(T) \n",
    "d = x.shape[1]\n",
    "train_data = { 'n': train.shape[0],'h': 4, 'y': Y[trainind],'Q': norm_T_Q,'x_full': x[trainind], 'x': x[trainind],'x_samp': x[trainind], 'T_samp': T[trainind], 'd': d, 'T': T[trainind],'t_lo': t_lo ,'t_hi': t_hi  }\n",
    "train_data['kernel_int_func'] = epanechnikov_int\n",
    "train_data['T_samp'] = np.array(T[trainind],order='F'); train_data['y_samp'] = Y[trainind]\n",
    "train_data['kernel_func'] = epanechnikov_kernel\n",
    "train_data['inds'] = trainind\n",
    "train_data['threshold'] = 0.01\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "\n",
    "def discretize_tau_policy(**params):\n",
    "    '''\n",
    "    Discretize the treatment vector 'tau' according to uniform binning.\n",
    "    '''\n",
    "    x = params['x_samp']\n",
    "    T = params['T_samp']\n",
    "    n_bins = params['n_bins']\n",
    "    t_lo = min(T)\n",
    "    t_hi = max(T)\n",
    "    bins = np.linspace(t_lo, t_hi, n_bins)\n",
    "    T_binned = np.digitize(T, bins).flatten()\n",
    "    bin_means = [T[T_binned == i].mean() for i in range(1, n_bins)]\n",
    "    tau_binned = np.digitize(params['tau'], bins).flatten()\n",
    "    return tau_binned\n",
    "\n",
    "def off_pol_disc_evaluation(policy, **params):\n",
    "    THRESH = params['threshold']\n",
    "    y_out = params['y']; x = params['x_samp']; h = params['h']; Q = params['Q']; n = params['n']; t_lo = params['t_lo']; t_hi = params['t_hi']\n",
    "    n_bins = params['n_bins']\n",
    "    if ('y_samp' in params.keys()):\n",
    "        y_out = params['y_samp'].flatten()\n",
    "    if ('T_samp' in params.keys()):\n",
    "        T = params['T_samp'].flatten()\n",
    "    else:\n",
    "        T = params['T'].flatten()\n",
    "\n",
    "    t_lo = min(T)\n",
    "    t_hi = max(T)\n",
    "    bin_width = t_hi-t_lo\n",
    "    bins = np.linspace(t_lo, t_hi, n_bins)\n",
    "    T_binned = np.digitize(T, bins, right = True).flatten()\n",
    "    bin_means = [T[T_binned == i].mean() for i in range(1, len(bins))]\n",
    "\n",
    "    loss = 0\n",
    "    tau_vec = policy(**params).flatten()\n",
    "\n",
    "    i=0\n",
    "    #! FIXME need to establish whether policy returns discrete bins or means\n",
    "    treatment_overlap = np.where(np.equal(tau_vec.flatten(), T_binned))[0]\n",
    "    n_overlap = len(treatment_overlap)\n",
    "    Qs = np.zeros(n_overlap)\n",
    "    for ind in treatment_overlap:\n",
    "        Q_i = Q(x[ind], bin_means[T_binned[ind]-1], t_lo, t_hi) * bin_width*1.0/n_bins # BUG FIX: this is going to have to be integrated against \n",
    "        Qs[i] = 1.0/max(Q_i,THRESH)\n",
    "        loss += y_out[ind]/max(Q_i,THRESH)\n",
    "        i+=1 \n",
    "    \n",
    "    norm_sum = np.mean(Qs)\n",
    "    if n_overlap == 0:\n",
    "        print \"no overlap\"\n",
    "        return 0\n",
    "    return loss/(1.0*n*norm_sum)\n",
    "\n",
    "def pol_disc_l2_wrapper(beta, args): \n",
    "    data = dict(args)\n",
    "    t_lo = data['t_lo']; t_hi = data['t_hi']; pol = data['pol']\n",
    "    tau = np.clip(np.dot(beta, data['x'].T) , t_lo, t_hi).flatten()\n",
    "    data['tau'] = tau\n",
    "    return data['sgn']*(off_pol_disc_evaluation(pol,**data) + LAMBDA * np.linalg.norm(beta,2))\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'n_bins' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-11-f65c0f1f54e5>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      1\u001b[0m \u001b[0mt_lo\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmin\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mT\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      2\u001b[0m \u001b[0mt_hi\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmax\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mT\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 3\u001b[0;31m \u001b[0mbins\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mlinspace\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mt_lo\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mt_hi\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mn_bins\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m      4\u001b[0m \u001b[0mT_binned\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdigitize\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mT\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbins\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mflatten\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0mbin_means\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mT\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mT_binned\u001b[0m \u001b[0;34m==\u001b[0m \u001b[0mi\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mmean\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mn_bins\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mNameError\u001b[0m: name 'n_bins' is not defined"
     ]
    }
   ],
   "source": [
    "t_lo = min(T)\n",
    "t_hi = max(T)\n",
    "bins = np.linspace(t_lo, t_hi, n_bins)\n",
    "T_binned = np.digitize(T, bins).flatten()\n",
    "bin_means = [T[T_binned == i].mean() for i in range(1, n_bins)]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 432,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "5.3017425743851954"
      ]
     },
     "execution_count": 432,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "data = { 'n': nn,'h': 5*pow((n_0*1.0/nn),.2), 'y': Y[trainind],'Q': norm_T_Q,'x_full': x[trainind,:], 'x': x[trainind,:],'x_samp': x[trainind,:], 'T_samp': T[trainind].flatten(), 'd': d, 'T': T[trainind],'t_lo': t_lo ,'t_hi': t_hi  }\n",
    "data['kernel_int_func'] = epanechnikov_int\n",
    "data['T_samp'] = np.array(T[trainind],order='F'); train_data['y_samp'] = Y[trainind]\n",
    "data['kernel_func'] = epanechnikov_kernel\n",
    "data['inds'] = trainind; data['sgn'] = 1\n",
    "data['threshold'] = 0.02\n",
    "\n",
    "data['n_bins'] = 10 \n",
    "data['pol'] = discretize_tau_policy\n",
    "data['tau'] = T[trainind]\n",
    "off_pol_disc_evaluation(discretize_tau_policy, **data)\n",
    "beta_d = np.random.uniform(size=(d,1))*4\n",
    "disc_res = minimize(pol_disc_l2_wrapper, x0 = beta_d, bounds = bnds, options={'disp':True,'gtol':1e-1,'maxfun': 1000}, args=data.items())\n",
    "data['tau'] = np.dot(disc_res.x, x.T)\n",
    "disc_pol = discretize_tau_policy(**data)\n",
    "disc_T= np.dot(disc_res.x, x.T)\n",
    "np.mean( real_risk(np.dot(disc_res.x, x.T), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x) )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 412,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([    0.,     0.,     0.,     0.,     0.,  1000.,     0.,     0.,\n",
       "            0.,     0.]),\n",
       " array([ 1.12933997,  1.22933997,  1.32933997,  1.42933997,  1.52933997,\n",
       "         1.62933997,  1.72933997,  1.82933997,  1.92933997,  2.02933997,\n",
       "         2.12933997]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 412,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXcAAAD0CAYAAABgk2Y8AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x115749590>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 362,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "opt coef bounds for beta: \n",
      "[(-2.9558302454279373, 3.6139946196115438), (-2.9558302454279373, 3.6139946196115438), (-2.9558302454279373, 3.6139946196115438), (-2.9558302454279373, 3.6139946196115438), (-2.9558302454279373, 3.6139946196115438), (-2.9558302454279373, 3.6139946196115438), (-2.9558302454279373, 3.6139946196115438), (-2.9558302454279373, 3.6139946196115438), (-2.9558302454279373, 3.6139946196115438), (-2.9558302454279373, 3.6139946196115438)]\n",
      "200\n",
      "[-0.11307281 -0.87557897 -0.2221516  -0.01761266 -0.2977354   0.24907352\n",
      "  0.11314573  0.0313166   0.0857681   0.45362848]\n",
      "2.53700237051\n",
      "[ 0.15080832 -0.80299567 -0.03155778  0.02964014 -0.1331966   0.20039607\n",
      "  0.03042686  0.11666065  0.03919754  0.81400155]\n",
      "2.60750146004\n",
      "[ 1.30859075 -1.65956808  0.79758927 -1.08244053  0.58893735 -0.40808881\n",
      "  2.22151267  0.62148064  1.46108363  0.38589725]\n",
      "-16.2167815295\n",
      "[ 0.32919023 -0.6009378  -0.01417606  0.63285094  0.61394999  0.39117909\n",
      "  1.17146372 -0.01390529  0.24757144  0.99251226]\n",
      "-10.3189152287\n",
      "[-0.31151171 -0.89586549 -0.46109056  0.0387649   0.13485948 -0.17541403\n",
      "  0.12764604  0.39442471  0.58198597  1.49747   ]\n",
      "-0.286563889634\n",
      "[-0.03791903 -0.79242874 -0.30422346  0.10410075 -0.12139534  0.22705067\n",
      " -0.00402944  0.63875141  0.95179445  1.43240986]\n",
      "-0.207601711412\n",
      "[ 0.22890899 -0.94391466 -0.0622236   0.2145947  -0.11045147 -0.1824121\n",
      "  0.33522837  0.57835308  0.47927675  0.63098552]\n",
      "-19.6540647019\n",
      "[ 0.52207615 -0.77923198  0.64548707  0.49693571  0.12016456  0.04155987\n",
      "  0.12493455 -0.11542606 -0.45057347 -0.06795489]\n",
      "-17.7384139117\n",
      "[ 0.09226392 -1.57387576  0.08628132 -0.33118931  0.15797601  0.12363565\n",
      "  0.00448363  0.39638729  0.94873587  0.31586896]\n",
      "-0.392636234862\n",
      "[-0.35473886 -1.74092101 -0.20617296  0.15914535  0.17841355  0.43307152\n",
      "  0.35723106  0.19698076  0.75069356  0.41069591]\n",
      "-0.753451800093\n",
      "[ 0.09371498  0.44380631  0.09842995  0.8132626   0.18461151  0.46268645\n",
      "  0.76090821  0.26323676  0.26477966  0.47576975]\n",
      "-11.9628311956\n",
      "[ 0.59044975 -0.44111256  0.02633594  0.24218006 -0.20963517  0.05441798\n",
      "  0.44822399  0.78773122  0.57474896  0.19921609]\n",
      "-12.3357171128\n",
      "[-0.66001366 -2.62111699  0.03958665  0.67019471 -0.52959764  0.86192121\n",
      "  0.00345274 -0.21558553  1.07918004 -0.01808281]\n",
      "-1.86774977036\n",
      "[-0.48509896 -1.58144316  0.02208971  0.52790696 -0.59867552  0.6355532\n",
      "  0.07414454  0.2939967   1.15571588  0.65209055]\n",
      "-1.35269720396\n",
      "[ 0.67426695 -0.70723523  0.65945697  0.03875602  0.01019091  0.76781459\n",
      "  0.14275101 -1.13009103  0.18127375  0.71745077]\n",
      "-18.3044174768\n",
      "[ 0.23003264 -0.94427643  1.07083845 -0.21468248  0.35990188  1.15732388\n",
      " -0.31153064 -1.29805098  0.39170043  0.28933049]\n",
      "-17.4295414795\n",
      "[ 0.19162796 -1.24215507 -0.24005383 -0.06433405  0.0895494  -0.21502978\n",
      "  0.01623308  0.60725217  0.49164725  1.0189341 ]\n",
      "0.334801112092\n",
      "[ 0.22412518 -1.33171147 -0.25149784 -0.1006383   0.1404546  -0.20794982\n",
      " -0.05382985  0.69091147  0.44223237  1.11754894]\n",
      "0.259239733455\n",
      "[ 0.21078826 -0.05949267 -0.25334209 -0.18283157  0.47718658  0.04931158\n",
      " -0.59633634  0.14681725  0.36446938  1.02260364]\n",
      "-14.9318471726\n",
      "[ 0.1655292   0.2233692   0.41519769 -0.10374993  0.6281875  -0.15883556\n",
      " -0.42702911  0.44922835  0.29601863  0.81723047]\n",
      "-15.1559020998\n",
      "400\n",
      "[-0.09972316 -1.8556408  -0.01967033  0.10151733  0.26798275  0.58991177\n",
      "  0.12760975  0.12169098  0.86716513  0.5373843 ]\n",
      "-0.0525370431056\n",
      "[-0.21511252 -1.88176757  0.11608009  0.10855363  0.12463581  0.60588134\n",
      "  0.15466299  0.1165353   0.84425939  0.56716428]\n",
      "-0.0614849056645\n",
      "[ 0.72099486  0.44395618 -0.03632855  0.72567368  0.30117554 -0.01503884\n",
      "  0.42736014  0.41250279  0.35404035  0.87787659]\n",
      "-15.0838464401\n",
      "[-0.09800275  0.41144466 -0.54672592  0.35358643 -0.03271996  0.07174221\n",
      "  0.04397075 -0.0348431   1.19659407  1.97084619]\n",
      "-15.7106206108\n",
      "[-0.07444565 -1.63196962 -0.33253005 -0.26853976 -0.0858875   0.4209846\n",
      " -0.22867153 -0.0401592   0.94752866  1.09031694]\n",
      "-0.73985233924\n",
      "[ -4.97997700e-01  -1.30910645e+00   4.14329151e-04  -8.03894497e-02\n",
      "  -3.98364778e-01   3.15935948e-01  -6.77948660e-02   3.01517724e-01\n",
      "   7.85368958e-01   8.53925105e-01]\n",
      "-0.490092616463\n",
      "[ 0.44635722  0.34704944  0.78752632  0.96567909 -0.11654779  0.15436936\n",
      "  0.85298174  0.30026249  0.38111886  0.62092025]\n",
      "-13.405207208\n",
      "[ 0.46170966 -0.13414465  0.44945308  0.79240895 -0.27451362  0.62817393\n",
      "  0.92246077  0.84261522  0.3925641   0.83373531]\n",
      "-14.4566054857\n",
      "[-0.41600284 -1.25998505  0.04094674 -0.03650546 -0.40533438 -0.0728618\n",
      " -0.21688222  0.26420304  0.72636367  0.62990971]\n",
      "-0.10566997137\n",
      "[-0.38309124 -1.83098428  0.03029175 -0.05182346 -0.25246286  0.26377061\n",
      " -0.3087169  -0.29082243  0.42714584  0.75154887]\n",
      "-0.406365260204\n",
      "[ 0.19826045  0.3229281   0.04825191 -0.0241417  -0.25190324  0.60318713\n",
      "  0.22916522 -0.09117743  0.73516101 -0.2361212 ]\n",
      "-10.093898332\n",
      "[-0.00377138  0.68040541  0.07568751  0.06176823  0.03174391  1.18099655\n",
      "  0.46318685  0.62348159  0.95715676  0.06554788]\n",
      "-9.59348036783\n",
      "[-0.32054436 -1.26957999 -0.17576272  0.21467075  0.19288951  0.63306312\n",
      "  0.16582396  0.68014419  1.07480495  0.43286063]\n",
      "0.298777298482\n",
      "[-0.39807039 -1.35639496 -0.00258329  0.1991101   0.16592442  0.69343659\n",
      "  0.10421556  0.81794516  1.12620934  0.4219811 ]\n",
      "0.235067529671\n",
      "[ 0.01093956 -0.23241832  0.96151975  0.47591023  0.33865494  0.14552701\n",
      "  0.79682278  0.22228186  0.64718075  0.62573383]\n",
      "-17.1206624617\n",
      "[-0.22076386 -0.75932234  0.39103528  0.48404827  0.04630062  0.53459796\n",
      "  0.24854423  0.4578074   0.82844305  0.35676584]\n",
      "-16.1946095555\n",
      "[-0.25666197 -1.32654156 -0.21040586 -0.32188933 -0.17898978  0.51869612\n",
      "  0.19045919  0.05412875  0.56994512  0.5445776 ]\n",
      "-0.346768385182\n",
      "[-0.34366646 -1.27237209 -0.0710565  -0.32754781 -0.1026565   0.47840534\n",
      "  0.17917864  0.1834949   0.54524174  0.45081345]\n",
      "-0.313651867306\n",
      "[-0.09577363  0.21193933  0.15493809 -0.36654305 -0.28274736  0.38464674\n",
      "  0.03637606 -0.09213616  0.25478531  0.89128848]\n",
      "-10.9883331496\n",
      "[-0.49371096  0.40320024  0.69974543 -0.68656233 -0.281333    0.99006782\n",
      "  0.02572104  0.21736896  0.07568758  0.81668891]\n",
      "-12.1019056182\n"
     ]
    }
   ],
   "source": [
    "import cvxpy as cvx\n",
    "np.random.seed(2)\n",
    "\n",
    "n = len(Y)\n",
    "SAMP_N = 5; \n",
    "ns = [200, 400]#, 800, 1600 ]\n",
    "n_0 = ns[0]\n",
    "cons_evals = np.zeros([len(ns), SAMP_N]); mean_evals = np.zeros([len(ns), SAMP_N])\n",
    "zero_evals = np.zeros([len(ns), SAMP_N]); oracle_evals = np.zeros([len(ns), SAMP_N]);DR_oracle_evals = np.zeros([len(ns), SAMP_N])\n",
    "oracle_coefs = [ [None]*SAMP_N ] * len(ns)\n",
    "DR_oracle_coefs = [ [None]*SAMP_N ] * len(ns)\n",
    "bounds = [(min(T)/(0.5*d) , max(T)/np.abs(0.5*d)  ) for i in range(d) ]\n",
    "bnds = tuple(tuple(x) for x in bounds)\n",
    "# bounds = [(-np.mean(sampled_T)/max(np.mean(X_sim_std[:,i]), 0.25*d) , np.mean(sampled_T)/max(np.mean(X_sim_std[:,i]),d*.25) ) for i in range(d) ]\n",
    "# bounds = [(-max(sim_dose)/(.25*d*np.mean(subframe[:,i])) , max(sim_dose)/(.25*d*np.mean(subframe[:,i])) ) for i in range(d) ]\n",
    "print \"opt coef bounds for beta: \"\n",
    "print bounds\n",
    "\n",
    "n_restarts = 1\n",
    "for ind, nn in enumerate(ns): \n",
    "    print nn \n",
    "    for k in range(SAMP_N): \n",
    "        trainind = np.random.choice(range(n),size = nn, replace = False)\n",
    "        sub_T = np.zeros(nn)\n",
    "        \n",
    "        # Learn GPS from data\n",
    "        lr = LinearRegression(); lr.fit(x[trainind,:], T[trainind])\n",
    "        T_hat = lr.predict(x[trainind,:])\n",
    "        beta_T_gps = np.dot(x[trainind,:],lr.coef_)\n",
    "        resid = T[trainind] - T_hat\n",
    "        # Assume normal noise\n",
    "        mu_resid = np.mean(resid); sigma_resid = np.std(resid)\n",
    "        def norm_T_Q_est(x,t,t_lo,t_hi): \n",
    "            return norm.pdf( t - np.dot(x, lr.coef_), loc =  mu_resid, scale = sigma_resid )\n",
    "        \n",
    "        data = { 'n': nn,'h': 3*pow((n_0*1.0/nn),.2), 'y': Y[trainind],'Q': norm_T_Q_est,'x_full': x[trainind,:], 'x': x[trainind,:],'x_samp': x[trainind,:], 'T_samp': T[trainind].flatten(), 'd': d, 'T': T[trainind],'t_lo': t_lo ,'t_hi': t_hi  }\n",
    "        data['kernel_int_func'] = epanechnikov_int\n",
    "        data['T_samp'] = np.array(T[trainind],order='F'); train_data['y_samp'] = Y[trainind]\n",
    "        data['kernel_func'] = epanechnikov_kernel\n",
    "        data['inds'] = trainind; data['sgn'] = 1\n",
    "        data['threshold'] = 0.02\n",
    "        betas = []; vals = []\n",
    "        \n",
    "        k_ind = 0\n",
    "        while k_ind <= n_restarts: \n",
    "            k_ind += 1\n",
    "            beta_d = np.random.uniform(size=d)\n",
    "            LAMBDA = 0.5\n",
    "            oracle_res = minimize(pol_eval_l2_wrapper, x0 = beta_d, jac = off_pol_epan_lin_l2_grad, bounds = bnds, options={'disp':True,'gtol':1e-1,'maxfun': 1000}, args=data.items())  \n",
    "            betas += [oracle_res.x]\n",
    "            print oracle_res.x\n",
    "            vals += [oracle_res.fun]\n",
    "            print oracle_res.fun\n",
    "        oracle_coefs[ind][k] = betas[np.argmin(vals)] \n",
    "        oracle_evals[ind, k] = vals[np.argmin(vals)] \n",
    "        zero_T = np.zeros(nn); data['tau'] = zero_T; zero_evals[ind,k] = off_policy_evaluation(**data)\n",
    "#         oracle_coefs[ind] += [ betas[np.argmin(vals)] ]\n",
    "#         oracle_evals[ind, k] = vals[np.argmin(vals)]\n",
    "        mu_T = np.mean(T)*np.ones(nn); data['tau'] = mu_T; mean_evals[ind,k] = off_policy_evaluation(**data)\n",
    "        \n",
    "        clf = RandomForestRegressor(n_estimators=5)\n",
    "        clf = clf.fit(x[trainind,:],y=Y[trainind])\n",
    "#         data['rf'] = clf\n",
    "        y_hat = clf.predict(x[trainind,:])\n",
    "        data['y'] = Y[trainind] - y_hat\n",
    "        betas = []; vals = []\n",
    "        k_ind = 0\n",
    "        while k_ind <= n_restarts: \n",
    "            k_ind += 1\n",
    "            beta_d = np.random.uniform(size=d)\n",
    "            LAMBDA = 0.5\n",
    "            oracle_res = minimize(pol_eval_l2_wrapper, x0 = beta_d, jac = off_pol_epan_lin_l2_grad, bounds = bnds, options={'disp':True,'gtol':1e-1,'maxfun': 1000}, args=data.items())  \n",
    "            print oracle_res.x\n",
    "            betas += [oracle_res.x]\n",
    "            vals += [oracle_res.fun]\n",
    "            print oracle_res.fun\n",
    "        DR_oracle_coefs[ind][k] = betas[np.argmin(vals)] \n",
    "        DR_oracle_evals[ind, k] = vals[np.argmin(vals)] \n",
    "        \n",
    "        data['n_bins'] = 10 \n",
    "        off_pol_disc_evaluation(discretize_tau_policy, **data)\n",
    "        \n",
    "        ## Best-in-class policy via regression\n",
    "        # find best in class policy \n",
    "#         d = x.shape[1]; beta = cvx.Variable(d)\n",
    "#         # Implement a thresholded loss\n",
    "#         cost = cvx.sum_entries(cvx.square(x[trainind,:]*beta)) # minimize y^2 \n",
    "#         constraints = [beta >=min(T)/(0.5*d) ]\n",
    "#         for i in range(d): \n",
    "#             constraints += [beta[i] <= max(T)/(0.5*d) ]\n",
    "#         prob = cvx.Problem(cvx.Minimize(cost), constraints)\n",
    "#         prob.solve('SCS')\n",
    "#         print beta.value\n",
    "#         prescient_beta = beta.value.flatten().T\n",
    "#         prescient_beta = np.ravel(prescient_beta).T\n",
    "        \n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "$$ \\beta^T X + \\beta_{x,T}*X\\circ T + (\\beta_{x,T^2}^T x - T)^2$$\n",
    "\n",
    "Suppose we replace T with the linear policy $\\beta_{\\tau}x$: \n",
    "\n",
    "$$ \\beta^T X + \\beta_{x,T}*X\\circ \\beta_{\\tau}X + ((\\beta_{x,T^2} - \\beta_{\\tau})^T X)^2 $$ \n",
    "\n",
    "If we derive wrt $\\beta_{\\tau x}$: \n",
    "\n",
    "$$ \\beta_{x,T}X +2(\\beta_{x,T^2}^T X - \\beta_{\\tau}X)X = 0 $$\n",
    "\n",
    "$$\\beta_{\\tau}^T X)X =  \\beta_{x,T}X +2\\beta_{x,T^2}^T X $$\n",
    "\n",
    "$$\\beta_{\\tau}^T  =  (\\beta_{x,T}X +2\\beta_{x,T^2}^T X)(X^TX)^{-1} $$"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 336,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "[-0.00435006  0.01131169  0.02045897  0.00788028 -0.00518324 -0.01321837\n",
      " -0.00207648  0.00110226  0.00821719  0.00261471]\n",
      "-5.13967400903\n"
     ]
    }
   ],
   "source": [
    "def oracle_eval_wrapper(beta, args): \n",
    "    data = dict(args)\n",
    "    t_lo = data['t_lo']; t_hi = data['t_hi']\n",
    "    tau = np.clip(np.dot(beta, data['x'].T) , t_lo, t_hi).flatten()\n",
    "    x = data['x']\n",
    "    return np.mean(real_risk(np.dot(tau, x), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x))\n",
    "beta_d = np.random.uniform(size=(d,1))*5\n",
    "best_in_class = minimize(oracle_eval_wrapper, x0 = beta_d, bounds = bnds, options={'disp':True,'gtol':1e-1,'maxfun': 1000}, args=data.items())  \n",
    "print best_in_class.x\n",
    "print best_in_class.fun\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 281,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXQAAAD0CAYAAACLpN0/AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nEAa6JBXCQJekQhjoklSI/we2I9ONToCTcgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x115adb810>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXEAAAD0CAYAAABtjRZ7AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nRgEGjsNSAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11e6e8cd0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXEAAAD0CAYAAABtjRZ7AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ncUkq7P8Afir3I6wpdPwAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11e651250>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXEAAAD0CAYAAABtjRZ7AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nCjPiklSYEZekwoy4JBVmxCWpsP8Dgxt78Ea5wC8AAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11e4e6d90>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "data": {
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11721b8d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "def real_risk(T, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x):    \n",
    "    n = len(T); risk = np.zeros(n)\n",
    "    if np.isscalar(T):\n",
    "        risk = T*beta_cons + np.dot(beta_x.T, x) + np.dot(beta_x_T.T, x*T) + (T-np.dot(beta_x_quad_T.T,x))**2\n",
    "    else: \n",
    "        for i in range(len(T)): \n",
    "            risk[i] = T[i]*beta_cons + np.dot(beta_x.T, x[i,:]) + np.dot(beta_x_T.T, x[i,:]*T[i]) + (T[i]-np.dot(beta_x_quad_T.T,x[i,:]))**2#+ np.dot(beta_x_quad_T.T, (x[i,:]**2)*T[i]) + np.dot(beta_x_high_freq.T, np.sin(x[i,0:HIGH_FREQ_N]*FREQ)*T[i])\n",
    "    return risk\n",
    "\n",
    "zero_T = np.zeros(n); \n",
    "# pred_T = np.dot(oracle_coefs[1][0], x.T)\n",
    "dr_T = np.dot(DR_oracle_coefs[1][0], x.T)\n",
    "colors = ['b', 'g', 'r', 'y','orange']\n",
    "EVALS = [zero_evals, mean_evals, DR_oracle_evals]\n",
    "\n",
    "# use scipy minimize (1D optimize outcome wrt T )\n",
    "def get_oracle_treatment(beta_cons, beta_x, beta_x_T, x): \n",
    "    res = minimize(real_risk, np.mean(T), args =(beta_cons, beta_x, beta_x_T, beta_x_quad_T, x))\n",
    "    return res.x \n",
    "\n",
    "POLICIES = [ zero_T, mu_T, dr_T ]\n",
    "lbls = ['zero treatment policy', 'mean treatment policy', 'DR treatment policy'   ]\n",
    "\n",
    "def plot_evals(evals, POLICY, X, c, lbl, ns, SAMP_N): \n",
    "    plt.axhline(y = Y_OFFSET + np.mean(real_risk(POLICY, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x)),color=c)\n",
    "    plt.scatter(ns, np.mean(evals,axis=1),label=lbl,color=c)\n",
    "    error = 1.96*np.std(evals,axis=1)/np.sqrt(SAMP_N)\n",
    "    plt.fill_between(ns, np.mean(evals,axis=1)-error,np.mean(evals,axis=1)+error, alpha=0.3, edgecolor=c, facecolor=c)\n",
    "# plt.axhline(y = np.mean(real_risk(oracle_T beta_cons, beta_x, beta_x_T, x)))\n",
    "\n",
    "\n",
    "plt.figure()\n",
    "for j in range(len(ns)):\n",
    "    for i in range(SAMP_N): \n",
    "        plt.hist( np.asarray(real_risk(np.dot(DR_oracle_coefs[j][i], x.T), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x)).flatten())\n",
    "\n",
    "plt.figure()\n",
    "[ plot_evals(EVALS[i], POLICIES[i], x, colors[i], lbls[i], ns, SAMP_N) for i in range(len(EVALS))]\n",
    "for j in range(len(ns)):\n",
    "    for i in range(SAMP_N): \n",
    "        plt.axhline(y = np.mean(real_risk(np.dot(oracle_coefs[j][i], x.T), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x)),color='r')\n",
    "    for i in range(SAMP_N): \n",
    "        plt.axhline(y = np.mean(real_risk(np.dot(DR_oracle_coefs[j][i], x.T), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x)),color='r')\n",
    "plt.axhline(y = np.mean(real_risk(np.dot(oracle_res.x, x.T), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x)))\n",
    "\n",
    "plt.figure()\n",
    "plt.hist(real_risk(np.dot(oracle_res.x, x.T), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x))\n",
    "# plt.axhline(y = np.mean(real_risk(pred_T, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x)),color=c)\n",
    "plt.figure()\n",
    "plt.hist(real_risk(zero_T, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x))\n",
    "\n",
    "\n",
    "# def plot_evals(evals, POLICY, X, c, lbl, ns, SAMP_N): \n",
    "#     plt.axhline(y = np.mean(real_risk(POLICY, beta_cons, beta_x, beta_x_T, x)),color=c)\n",
    "#     plt.scatter(ns, np.mean(evals,axis=1),label=lbl,color=c)\n",
    "#     error = 1.96*np.std(evals,axis=1)/np.sqrt(SAMP_N)\n",
    "#     plt.fill_between(ns, np.mean(evals,axis=1)-error,np.mean(evals,axis=1)+error, alpha=0.5, edgecolor=c, facecolor=c)\n",
    "# plt.axhline(y = np.mean(real_risk(oracle_T, beta_cons, beta_x, beta_x_T, x)))\n",
    "# [ plot_evals(EVALS[i], POLICIES[i], x, colors[i], lbls[i], ns, SAMP_N) for i in range(len(EVALS))]\n",
    "\n",
    "import pickle\n",
    "plt.figure()\n",
    "for i in range(len(EVALS)):  \n",
    "    pickle.dump(EVALS[i], open(str(datetime.datetime.now().strftime(\"%Y-%m-%d_%H-%M\")) +lbls[i]+ 'EVALS.p', 'wb'))\n",
    "    pickle.dump(POLICIES[i], open(str(datetime.datetime.now().strftime(\"%Y-%m-%d_%H-%M\")) +lbls[i]+ 'policy.p', 'wb'))\n",
    "    pickle.dump(oracle_coefs, open(str(datetime.datetime.now().strftime(\"%Y-%m-%d_%H-%M\")) +lbls[i]+ 'oracle_coefs.p', 'wb'))\n",
    "    pickle.dump(DR_oracle_coefs, open(str(datetime.datetime.now().strftime(\"%Y-%m-%d_%H-%M\")) +lbls[i]+ 'DR_oracle_coefs.p', 'wb'))\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 328,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([  10.,   45.,  124.,  184.,  204.,  200.,  135.,   66.,   27.,    5.]),\n",
       " array([-13.49061212, -10.5721903 ,  -7.65376848,  -4.73534666,\n",
       "         -1.81692484,   1.10149698,   4.0199188 ,   6.93834062,\n",
       "          9.85676244,  12.77518426,  15.69360608]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 328,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXEAAAD0CAYAAABtjRZ7AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\ngMH+HxLVX6jzePSdAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11c6a0190>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(T)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 452,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Calibrating clip to  6.93507159806\n",
      "Calibrating variance regularizer to  774.172130334\n",
      "PRM(Self-Normalized-ERM)  Best. Parameter =  {'varpenalty': 0.0, 'l2reg': 0, 'clip': 6.9350715980596425} Time:  0.0185790061951  ClippedDiagnostic:  2.81444380731e-06 UnclippedDiagnostic:  2.81444380731e-06  Performance:  None\n"
     ]
    }
   ],
   "source": [
    "# Impute POEM \n",
    "import sys\n",
    "sys.path.append('./POEM-norm/')\n",
    "import DatasetReader, Skylines\n",
    "\n",
    "def norm_T_Q(x,t,t_lo,t_hi): \n",
    "    return norm.pdf( t - np.dot(x, TRUE_PROP_BETA), loc = 0, scale = T_SIG )\n",
    "\n",
    "t_lo = min(T)\n",
    "t_hi = max(T)\n",
    "bins = np.linspace(t_lo, t_hi, n_bins)\n",
    "bin_width = (t_hi-t_lo)*1.0/n_bins\n",
    "T_binned = np.digitize(T, bins,right=True).flatten()\n",
    "bin_means = [T[T_binned == i].mean() for i in range(1, n_bins)]\n",
    "Xtrain = x[trainind,:]\n",
    "ttrain = T_binned[trainind]\n",
    "\n",
    "gps = np.asarray([bin_width*norm_T_Q(x[i,:], bin_means[ttrain[i]-1], 0.5, 0.5 ) for i in range(len(trainind))])\n",
    "# all 0-1\n",
    "# gps = integrated_propensity_scores\n",
    "ytrain = Y[trainind]\n",
    "Xtest = x\n",
    "\n",
    "mydata = DatasetReader.BanditDataset(None,False)\n",
    "mydata.trainFeatures        = np.hstack((Xtrain.copy(),np.ones((len(Xtrain),1))))\n",
    "mydata.sampledLabels        = np.zeros((len(ttrain),max(ttrain)+1))\n",
    "mydata.sampledLabels[range(len(ttrain)),ttrain] = 1.\n",
    "mydata.trainLabels          = np.empty(mydata.sampledLabels.shape)\n",
    "mydata.sampledLoss          = ytrain.copy()\n",
    "mydata.sampledLoss         -= mydata.sampledLoss.min()\n",
    "mydata.sampledLoss         /= mydata.sampledLoss.max()\n",
    "# computed on training set \n",
    "mydata.sampledLogPropensity = np.log(gps)\n",
    "#ones_like vs ones_line? \n",
    "mydata.testFeatures              = np.hstack((np.ones_like(Xtest),np.ones((len(Xtest),1))))\n",
    "mydata.testLabels                = np.array([])\n",
    "mydata.createTrainValidateSplit()\n",
    "pool = None \n",
    "coef = None\n",
    "\n",
    "maj = Skylines.PRMWrapper(mydata, n_iter = 1000, tol = 1e-6, minC = 0, maxC = -1, minV = 0, maxV = -1,\n",
    "                            minClip = 0, maxClip = 0, estimator_type = 'Self-Normalized', verbose = True,\n",
    "                            parallel = pool, smartStart = coef)\n",
    "maj.calibrateHyperParams()\n",
    "maj.validate()\n",
    "Xtest1 = np.hstack((Xtest,np.ones((len(Xtest),1))))\n",
    "rec = Xtest1.dot(maj.labeler.coef_).argmax(1)\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 375,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "[-11.981815426936143,\n",
       " -9.0064932552158528,\n",
       " -5.5881381267090555,\n",
       " -1.8942891665267862,\n",
       " 1.6293399730948344,\n",
       " 5.3301456637254274,\n",
       " 8.7249863215911514,\n",
       " 12.268056421501164,\n",
       " 15.814595553523269]"
      ]
     },
     "execution_count": 375,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": []
  },
  {
   "cell_type": "code",
   "execution_count": 453,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "100.353197267\n",
      "79.9704574287\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "(array([  11.,  312.,  201.,  171.,  152.,   88.,   33.,   24.,    6.,    2.]),\n",
       " array([ -86.73588356,  -28.25840323,   30.21907711,   88.69655744,\n",
       "         147.17403778,  205.65151811,  264.12899844,  322.60647878,\n",
       "         381.08395911,  439.56143945,  498.03891978]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 453,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXEAAAD0CAYAAABtjRZ7AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11e429fd0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "POEM_pol =  [bin_means[rec[i]-1] for i in range(n)]  \n",
    "# print POEM_pol\n",
    "POEM_risk = real_risk(POEM_pol, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x)\n",
    "print np.mean(POEM_risk)\n",
    "print np.median(POEM_risk)\n",
    "plt.hist(POEM_risk)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 295,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[ -1.51635123,  -0.76968095,   0.0332445 , ...,  -0.58029878,\n",
       "          0.68329363,   3.17487636],\n",
       "       [ -0.09633913,  -3.02051254,   0.62410646, ...,  -0.16831035,\n",
       "         -1.88969581,  -0.93700002],\n",
       "       [  1.28295978,   1.55285965,   0.70334812, ...,   0.33922146,\n",
       "          0.32607988,  11.66598013],\n",
       "       ..., \n",
       "       [ -0.84835847,  -1.78056669,   0.28752181, ...,   0.45480389,\n",
       "          0.97962469,   0.69258063],\n",
       "       [ -0.41519852,  -0.65276462,  -0.31912961, ...,  -0.38795049,\n",
       "          1.01675513,   3.16505215],\n",
       "       [  0.32207699,  -2.34564014,   0.25879004, ...,  -0.92527222,\n",
       "         -0.46216755,   5.78096866]])"
      ]
     },
     "execution_count": 295,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_w_dose"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "DM with random forest regressor"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 377,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "0.876501283585\n",
      "[-0.14869785 -1.94284229  0.29220843  0.07710959  0.6036456  -0.02156021\n",
      "  0.00812235  0.23521679  0.38506636  0.28939633]\n"
     ]
    }
   ],
   "source": [
    "from sklearn.ensemble import RandomForestRegressor\n",
    "\n",
    "X_w_dose = np.column_stack((x, T))\n",
    "X_w_dose.shape\n",
    "\n",
    "clf = RandomForestRegressor(n_estimators=10)\n",
    "\n",
    "clf = clf.fit(X_w_dose[trainind,:],y=Y[trainind])\n",
    "data['x_full'] = X_w_dose[trainind,:]\n",
    "data['rf'] = clf\n",
    "print clf.score(X_w_dose,Y)\n",
    "\n",
    "def dm_wrapper(beta, args): \n",
    "    reg_lambda = 0.5\n",
    "    data = dict(args)\n",
    "    t_lo = data['t_lo']; t_hi = data['t_hi']\n",
    "    x = data['x_full']\n",
    "    clf = data['rf']\n",
    "    tau = np.clip(np.dot( beta, x[:,:-1].T ), t_lo, t_hi).flatten()\n",
    "    counterfactual_X = np.column_stack((x[:,:-1], tau))\n",
    "    return (np.mean( clf.predict( counterfactual_X )) + reg_lambda * np.linalg.norm( beta, 2 )) \n",
    "\n",
    "dm_res = minimize(dm_wrapper, x0 = beta_d, bounds = bnds, options={'disp':True,'gtol':1e-1,'maxfun': 1000,'eps':np.ones(d)*1e-1}, args=data.items()) \n",
    "print dm_res.x\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 378,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.56422025353847105"
      ]
     },
     "execution_count": 378,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "np.mean(T)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 379,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.19847593763428459"
      ]
     },
     "execution_count": 379,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXEAAAD2CAYAAAAgRbdwAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11cce6b10>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "\n",
    "dm_T = np.dot(dm_res.x, x_test.T)\n",
    "\n",
    "plt.hist(real_risk(dm_T, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test))\n",
    "np.mean(real_risk(dm_T, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 380,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "def get_oracle_treatment(beta_cons, beta_x, beta_x_T, beta_x_quad_T, x): \n",
    "    res = minimize(real_risk_scalar, np.mean(T), args =(beta_cons, beta_x, beta_x_T, beta_x_quad_T,x))\n",
    "    return res.x \n",
    "n_test = x_test.shape[0]\n",
    "oracle_T = [ get_oracle_treatment(beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test[i,:]) for i in range(n_test)]\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 402,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([   1.,    6.,   15.,   68.,  142.,  178.,  194.,  107.,   31.,    7.]),\n",
       " array([ -5.83865989e-08,  -4.84677142e-08,  -3.85488295e-08,\n",
       "         -2.86299449e-08,  -1.87110602e-08,  -8.79217548e-09,\n",
       "          1.12670921e-09,   1.10455939e-08,   2.09644786e-08,\n",
       "          3.08833633e-08,   4.08022480e-08]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 402,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXEAAAEACAYAAABF+UbAAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11e406350>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.hist(np.dot(disc_res.x, x_test.T))"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 429,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(749, 10)"
      ]
     },
     "execution_count": 429,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "x_test.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 433,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array([   0.,    0.,    0.,    0.,    0.,  749.,    0.,    0.,    0.,    0.]),\n",
       " array([ 1.12933997,  1.22933997,  1.32933997,  1.42933997,  1.52933997,\n",
       "         1.62933997,  1.72933997,  1.82933997,  1.92933997,  2.02933997,\n",
       "         2.12933997]),\n",
       " <a list of 10 Patch objects>)"
      ]
     },
     "execution_count": 433,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAXEAAAD0CAYAAABtjRZ7AAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n4gBgMCIOAAb7P/Ejxc6M2V8nAAAAAElFTkSuQmCC\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x11383f410>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "data['x']\n",
    "data['tau'] = np.dot(disc_res.x, x_test.T)\n",
    "disc_pol_binned = np.digitize(np.dot(disc_res.x, x_test.T), bins).flatten()\n",
    "disc_pol = np.asarray( [bin_means[disc_pol_binned[i] - 1] for i in range(x_test.shape[0])] )\n",
    "plt.hist(disc_pol)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 437,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "\n",
    "def cons_pol_wrapper(beta, args): \n",
    "    data = dict(args)\n",
    "    t_lo = data['t_lo']; t_hi = data['t_hi']\n",
    "    tau = beta*np.ones(data['n'])\n",
    "    data['tau'] = tau\n",
    "    return data['sgn']*(off_policy_evaluation(**data))\n",
    "beta_t = np.random.uniform()*4\n",
    "bnds_t = tuple([(-15,15)])\n",
    "cons_res = minimize(cons_pol_wrapper, x0 = beta_t, bounds = bnds_t, options={'disp':True,'gtol':1e-1,'maxfun': 1000}, args=data.items())\n",
    "\n",
    "\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "plt.style.use('seaborn-white')\n",
    "plt.figure()\n",
    "\n",
    "# Use the same simulated loss function to estimate differences\n",
    "TRUE_ORACLE = real_risk(oracle_T, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test)\n",
    "oracle_loss = real_risk(np.dot(best_in_class.x, x_test.T), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test)\n",
    "zero_pol = real_risk(np.zeros(n_test), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test)\n",
    "mu_pol = real_risk(np.ones(n_test)*np.mean(T), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test)\n",
    "DM = real_risk(np.dot(dm_res.x, x_test.T), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test)\n",
    "DR_oracle = real_risk(np.dot(DR_oracle_coefs[1][3],x_test.T), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test)\n",
    "off_pol = real_risk(np.dot(oracle_coefs[1][1],x_test.T), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test)\n",
    "disc_pol_norms = real_risk(disc_pol, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test)\n",
    "best_constant = real_risk(cons_res.x*np.ones(n_test), beta_cons, beta_x, beta_x_T, beta_x_quad_T, x_test)\n",
    "\n",
    "\n",
    "original = real_risk(T, beta_cons, beta_x, beta_x_T, beta_x_quad_T, x)\n",
    "\n",
    "norms = [TRUE_ORACLE, off_pol,DR_oracle,DM, disc_pol_norms,best_constant, mu_pol, original ]\n",
    "\n",
    "plt.figure(figsize=(4.5,3))\n",
    "plt.ylabel('Thresholded absolute policy loss')\n",
    "flierprops = dict(linestyle='-',color='black')\n",
    "\n",
    "# prefer to turn off mean line\n",
    "medianprops = dict(linestyle='', linewidth=0, color='firebrick')\n",
    "f=plt.figure(figsize=(4.5,2.5))\n",
    "bp_dict = plt.boxplot(norms,sym='', showmeans = True, meanline=True,medianprops=medianprops)\n",
    "plt.ylabel('Outcome (loss)')\n",
    "for whisker in bp_dict['whiskers']:\n",
    "    whisker.set(color='#000000',linestyle='solid')\n",
    "for box in bp_dict['boxes']:\n",
    "    # change outline color\n",
    "    box.set( color='#000000')\n",
    "for ind,line in enumerate(bp_dict['means']):\n",
    "    # get position data for means line\n",
    "    xx, y = line.get_xydata()[1] # top of means line\n",
    "    if ind == 7: \n",
    "        plt.text(xx-1, 30, '%.1f' % y, horizontalalignment='center',size=13) # draw above, centered\n",
    "    # overlay median value\n",
    "    else:\n",
    "        plt.text(xx+0.2, -35, '%.1f' % y, horizontalalignment='center',size=13) # draw above, centered\n",
    "# top = 60\n",
    "# bottom = 0\n",
    "# plt.ylim(bottom, top)\n",
    "# plt.title('Boxplot of distances between policy-recommended doses and therapeutic doses',y = 1.2)\n",
    "# plt.axhline(y=np.mean(mean_dose_pol_norm), xmin=0,xmax=1,color='g', ls='dashdot',linewidth=0.7)\n",
    "# plt.axhline(y=0, xmin=0,xmax=1,color='g',alpha=0.3, ls='dashed',linewidth=0.7)\n",
    "plt.xticks([1, 2, 3,4,5,6,7,8], ['best o.o.c.', 'CPO', 'DR CPO', 'DM','Disc. CPO','CPE, cons.', 'Mean','Original'])\n",
    "plt.xticks(rotation=25)\n",
    "plt.ylim((-40,40))\n",
    "ax = plt.subplot(111)\n",
    "ax.spines['right'].set_visible(False)\n",
    "ax.spines['top'].set_visible(False)\n",
    "\n",
    "# Only show ticks on the left and bottom spines\n",
    "ax.yaxis.set_ticks_position('left')\n",
    "ax.xaxis.set_ticks_position('bottom')\n",
    "f.savefig(\"boxplot_risks.pdf\", bbox_inches='tight')\n",
    "\n",
    "\n",
    "\n",
    "\n",
    "plt.show()\n",
    "\n",
    "f= plt.figure()\n",
    "plt.violinplot(norms)\n",
    "plt.xticks([1, 2, 3,4,5,6,7,8], ['Best o.o.c.','OPE', 'DR OPE','DM','Disc. OPE','OPE, cons.','Mean','Original'],size = 13)\n",
    "plt.xticks(rotation=25)\n",
    "plt.ylim((-90,70))\n",
    "for ind,line in enumerate(bp_dict['means']):\n",
    "    # get position data for means line\n",
    "    xx, y = line.get_xydata()[1] # top of means line\n",
    "    if ind == 6: \n",
    "        plt.text(xx+0.05, y+3.5, '%.1f' % y, horizontalalignment='center',size=13) # draw above, centered\n",
    "    # overlay median value\n",
    "    else:\n",
    "        plt.text(xx+0.24, y+3.5, '%.1f' % y, horizontalalignment='center',size=13) # draw above, centered\n",
    "\n",
    "        \n",
    "f.savefig(\"violin_plot_risks.pdf\", bbox_inches='tight')\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true
   },
   "outputs": [],
   "source": [
    "5"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 390,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/Users/az/anaconda/lib/python2.7/site-packages/statsmodels/nonparametric/kernel_regression.py:273: RuntimeWarning: divide by zero encountered in double_scalars\n",
      "  (1 - (np.trace(H) + 2) / float(self.nobs))\n"
     ]
    },
    {
     "ename": "KeyboardInterrupt",
     "evalue": "",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m                         Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-390-e056c2ac4929>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[1;32m      5\u001b[0m \u001b[0msettings\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mEstimatorSettings\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mefficient\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mTrue\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mrandomize\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mTrue\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m      6\u001b[0m \u001b[0mlist_data\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m \u001b[0mx\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mtrainind\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mi\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0md\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m]\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mT\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mtrainind\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m----> 7\u001b[0;31m \u001b[0mkernel_regr\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mKernelReg\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mendog\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m[\u001b[0m\u001b[0mY\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mtrainind\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreshape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mtrainind\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mexog\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mlist_data\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mvar_type\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'c'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mlen\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mlist_data\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mbw\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;34m'aic'\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0mdefaults\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0msettings\u001b[0m \u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;32m/Users/az/anaconda/lib/python2.7/site-packages/statsmodels/nonparametric/kernel_regression.pyc\u001b[0m in \u001b[0;36m__init__\u001b[0;34m(self, endog, exog, var_type, reg_type, bw, defaults)\u001b[0m\n\u001b[1;32m    108\u001b[0m             \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbw\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_compute_reg_bw\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbw\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    109\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 110\u001b[0;31m             \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbw\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_compute_efficient\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbw\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    111\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    112\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0m_compute_reg_bw\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbw\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/Users/az/anaconda/lib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyc\u001b[0m in \u001b[0;36m_compute_efficient\u001b[0;34m(self, bw)\u001b[0m\n\u001b[1;32m    218\u001b[0m                                            \u001b[0mn_cvars\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mix_ord\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mix_unord\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mn_sub\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    219\u001b[0m                                            \u001b[0mclass_vars\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mrandomize\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 220\u001b[0;31m                                            bounds[i]))\n\u001b[0m\u001b[1;32m    221\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    222\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mn_blocks\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/Users/az/anaconda/lib/python2.7/site-packages/statsmodels/nonparametric/_kernel_base.pyc\u001b[0m in \u001b[0;36m_compute_subset\u001b[0;34m(class_type, data, bw, co, do, n_cvars, ix_ord, ix_unord, n_sub, class_vars, randomize, bound)\u001b[0m\n\u001b[1;32m     79\u001b[0m         sub_model = KernelReg(endog=endog, exog=exog, reg_type=reg_type,\n\u001b[1;32m     80\u001b[0m                               \u001b[0mvar_type\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mvar_type\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbw\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mbw\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m---> 81\u001b[0;31m                               defaults=EstimatorSettings(efficient=False))\n\u001b[0m\u001b[1;32m     82\u001b[0m     \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m     83\u001b[0m         raise ValueError(\"class_type not recognized, should be one of \" \\\n",
      "\u001b[0;32m/Users/az/anaconda/lib/python2.7/site-packages/statsmodels/nonparametric/kernel_regression.pyc\u001b[0m in \u001b[0;36m__init__\u001b[0;34m(self, endog, exog, var_type, reg_type, bw, defaults)\u001b[0m\n\u001b[1;32m    106\u001b[0m         \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_set_defaults\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mdefaults\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    107\u001b[0m         \u001b[0;32mif\u001b[0m \u001b[0;32mnot\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mefficient\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 108\u001b[0;31m             \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbw\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_compute_reg_bw\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbw\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    109\u001b[0m         \u001b[0;32melse\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    110\u001b[0m             \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mbw\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0m_compute_efficient\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mbw\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/Users/az/anaconda/lib/python2.7/site-packages/statsmodels/nonparametric/kernel_regression.pyc\u001b[0m in \u001b[0;36m_compute_reg_bw\u001b[0;34m(self, bw)\u001b[0m\n\u001b[1;32m    124\u001b[0m             \u001b[0mfunc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mest\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreg_type\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    125\u001b[0m             bw_estimated = optimize.fmin(res, x0=h0, args=(func, ),\n\u001b[0;32m--> 126\u001b[0;31m                                          maxiter=1e3, maxfun=1e3, disp=0)\n\u001b[0m\u001b[1;32m    127\u001b[0m             \u001b[0;32mreturn\u001b[0m \u001b[0mbw_estimated\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    128\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/Users/az/anaconda/lib/python2.7/site-packages/scipy/optimize/optimize.pyc\u001b[0m in \u001b[0;36mfmin\u001b[0;34m(func, x0, args, xtol, ftol, maxiter, maxfun, full_output, disp, retall, callback, initial_simplex)\u001b[0m\n\u001b[1;32m    391\u001b[0m             'initial_simplex': initial_simplex}\n\u001b[1;32m    392\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 393\u001b[0;31m     \u001b[0mres\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0m_minimize_neldermead\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mfunc\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mx0\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mcallback\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mcallback\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m**\u001b[0m\u001b[0mopts\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    394\u001b[0m     \u001b[0;32mif\u001b[0m \u001b[0mfull_output\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    395\u001b[0m         \u001b[0mretlist\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mres\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'x'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mres\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'fun'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mres\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'nit'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mres\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'nfev'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mres\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m'status'\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/Users/az/anaconda/lib/python2.7/site-packages/scipy/optimize/optimize.pyc\u001b[0m in \u001b[0;36m_minimize_neldermead\u001b[0;34m(func, x0, args, callback, maxiter, maxfev, disp, return_all, initial_simplex, xatol, fatol, **unknown_options)\u001b[0m\n\u001b[1;32m    562\u001b[0m                     \u001b[0;31m# Perform an inside contraction\u001b[0m\u001b[0;34m\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    563\u001b[0m                     \u001b[0mxcc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0;36m1\u001b[0m \u001b[0;34m-\u001b[0m \u001b[0mpsi\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0mxbar\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0mpsi\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0msim\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 564\u001b[0;31m                     \u001b[0mfxcc\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mfunc\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mxcc\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    565\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    566\u001b[0m                     \u001b[0;32mif\u001b[0m \u001b[0mfxcc\u001b[0m \u001b[0;34m<\u001b[0m \u001b[0mfsim\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;34m-\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/Users/az/anaconda/lib/python2.7/site-packages/scipy/optimize/optimize.pyc\u001b[0m in \u001b[0;36mfunction_wrapper\u001b[0;34m(*wrapper_args)\u001b[0m\n\u001b[1;32m    290\u001b[0m     \u001b[0;32mdef\u001b[0m \u001b[0mfunction_wrapper\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0mwrapper_args\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    291\u001b[0m         \u001b[0mncalls\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m+=\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 292\u001b[0;31m         \u001b[0;32mreturn\u001b[0m \u001b[0mfunction\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m*\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mwrapper_args\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0margs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    293\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    294\u001b[0m     \u001b[0;32mreturn\u001b[0m \u001b[0mncalls\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mfunction_wrapper\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/Users/az/anaconda/lib/python2.7/site-packages/statsmodels/nonparametric/kernel_regression.pyc\u001b[0m in \u001b[0;36maic_hurvich\u001b[0;34m(self, bw, func)\u001b[0m\n\u001b[1;32m    266\u001b[0m         gx = KernelReg(endog=self.endog, exog=self.exog, var_type=self.var_type,\n\u001b[1;32m    267\u001b[0m                        \u001b[0mreg_type\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreg_type\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mbw\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0mbw\u001b[0m\u001b[0;34m,\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 268\u001b[0;31m                        defaults=EstimatorSettings(efficient=False)).fit()[0]\n\u001b[0m\u001b[1;32m    269\u001b[0m         \u001b[0mgx\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mreshape\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mgx\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnobs\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    270\u001b[0m         \u001b[0msigma\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mendog\u001b[0m \u001b[0;34m-\u001b[0m \u001b[0mgx\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m**\u001b[0m\u001b[0;36m2\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msum\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0maxis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m)\u001b[0m \u001b[0;34m/\u001b[0m \u001b[0mfloat\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mself\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mnobs\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/Users/az/anaconda/lib/python2.7/site-packages/statsmodels/nonparametric/kernel_regression.pyc\u001b[0m in \u001b[0;36mfit\u001b[0;34m(self, data_predict)\u001b[0m\n\u001b[1;32m    374\u001b[0m         \u001b[0;32mfor\u001b[0m \u001b[0mi\u001b[0m \u001b[0;32min\u001b[0m \u001b[0mrange\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mN_data_predict\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m:\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    375\u001b[0m             mean_mfx = func(self.bw, self.endog, self.exog,\n\u001b[0;32m--> 376\u001b[0;31m                             data_predict=data_predict[i, :])\n\u001b[0m\u001b[1;32m    377\u001b[0m             \u001b[0mmean\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0mi\u001b[0m\u001b[0;34m]\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mmean_mfx\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    378\u001b[0m             \u001b[0mmfx_c\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msqueeze\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mmean_mfx\u001b[0m\u001b[0;34m[\u001b[0m\u001b[0;36m1\u001b[0m\u001b[0;34m]\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;32m/Users/az/anaconda/lib/python2.7/site-packages/statsmodels/nonparametric/kernel_regression.pyc\u001b[0m in \u001b[0;36m_est_loc_linear\u001b[0;34m(self, bw, endog, exog, data_predict)\u001b[0m\n\u001b[1;32m    169\u001b[0m \u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    170\u001b[0m         \u001b[0mM12\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mexog\u001b[0m \u001b[0;34m-\u001b[0m \u001b[0mdata_predict\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0;32m--> 171\u001b[0;31m         \u001b[0mM22\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mdot\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mM12\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mT\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mM12\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0mker\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m\u001b[1;32m    172\u001b[0m         \u001b[0mM12\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0;34m(\u001b[0m\u001b[0mM12\u001b[0m \u001b[0;34m*\u001b[0m \u001b[0mker\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0msum\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0maxis\u001b[0m\u001b[0;34m=\u001b[0m\u001b[0;36m0\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[1;32m    173\u001b[0m         \u001b[0mM\u001b[0m \u001b[0;34m=\u001b[0m \u001b[0mnp\u001b[0m\u001b[0;34m.\u001b[0m\u001b[0mempty\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0;34m(\u001b[0m\u001b[0mk_vars\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m,\u001b[0m \u001b[0mk_vars\u001b[0m \u001b[0;34m+\u001b[0m \u001b[0;36m1\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m)\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n",
      "\u001b[0;31mKeyboardInterrupt\u001b[0m: "
     ]
    }
   ],
   "source": [
    "from statsmodels.nonparametric.kernel_regression import KernelReg\n",
    "from statsmodels.nonparametric.kernel_density import EstimatorSettings\n",
    "\n",
    "\n",
    "settings = EstimatorSettings(efficient = True, randomize=True)\n",
    "list_data = [ x[trainind,i] for i in range(d) ] + [T[trainind]]\n",
    "kernel_regr = KernelReg(endog = [Y[trainind].reshape([len(trainind),1])], exog=list_data, var_type=['c']*len(list_data),bw='aic',defaults = settings )"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 383,
   "metadata": {},
   "outputs": [
    {
     "ename": "NameError",
     "evalue": "name 'kernel_regr' is not defined",
     "output_type": "error",
     "traceback": [
      "\u001b[0;31m---------------------------------------------------------------------------\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m                                 Traceback (most recent call last)",
      "\u001b[0;32m<ipython-input-383-f84d799947a3>\u001b[0m in \u001b[0;36m<module>\u001b[0;34m()\u001b[0m\n\u001b[0;32m----> 1\u001b[0;31m \u001b[0mkernel_regr\u001b[0m\u001b[0;34m\u001b[0m\u001b[0m\n\u001b[0m",
      "\u001b[0;31mNameError\u001b[0m: name 'kernel_regr' is not defined"
     ]
    }
   ],
   "source": [
    "kernel_regr"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 2",
   "language": "python",
   "name": "python2"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 2
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython2",
   "version": "2.7.13"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}