Modeling%2BWeather%2BGeographies%2Busing%2BScikit-Learn.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "# Modeling Weather Geographies using Scikit-Learn\n",
    "\n",
    "In this notebook, we are using aggregated `TMAX` data by global weather stations in order to create a machine learning model to predict the daily maximum temperatures at any given latitude and longitude. Our model will take 3 continuous predictors: latitude, longitude, and elevation, and provide an estimated `TMAX` for a given day of the year.\n",
    "\n",
    "Finally, we will show how to save this model to your DSX Local filesystem to be used for online scoring."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<div class=\"alert alert-block alert-info\"> Note: You will need to install the Basemap and GEOS libraries to dynamically produce output maps. For your convinience, the maps have been pre-rendered in this sample notebook.</div>\n",
    "\n",
    "## Table of Contents\n",
    "This notebook contains these main sections:\n",
    "\n",
    "1. [Import Libraries](#Import_Libraries)\n",
    "2. [The Data](#The_Data)\n",
    "3. [The Model](#The_Model)\n",
    "4. [Data Visualization](#Data_Visualization)\n",
    "5. [Save Model to DSX Filesystem](#Save_Model_to_DSX_Filesystem)\n",
    "6. [Predict on New Data](#Predict_on_New_Data)\n",
    "7. [Summary](#Summary)\n",
    "\n",
    "<a id='Import_Libraries'></a>\n",
    "## Import Libraries\n",
    "Run the cell below once to install the `tqdm` library"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Collecting tqdm\n",
      "  Downloading tqdm-4.19.5-py2.py3-none-any.whl (51kB)\n",
      "\u001b[K    100% |████████████████████████████████| 61kB 316kB/s ta 0:00:01\n",
      "\u001b[?25hInstalling collected packages: tqdm\n",
      "Successfully installed tqdm-4.19.5\n"
     ]
    }
   ],
   "source": [
    "!pip install --user tqdm"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Now restart the kernel to ensure that the `tqdm` library is loaded. You can also comment out the above cell in order to avoid reinstalling the library on each runthrough."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import matplotlib.pyplot as plt\n",
    "import matplotlib\n",
    "#from mpl_toolkits.basemap import Basemap, maskoceans\n",
    "\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "\n",
    "from tqdm import tqdm\n",
    "\n",
    "from sklearn.ensemble import GradientBoostingRegressor\n",
    "from sklearn.metrics import mean_absolute_error\n",
    "from sklearn.model_selection import train_test_split"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='The_Data'></a>\n",
    "## The Data\n",
    "The dataset was created using data from the Global Historical Climatology Network. We have averaged the `TMAX` over the entire history of each weather station for the spring and autumn equinox, and the summer and winter solstice."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "df_data_1 = pd.read_csv(\"https://raw.githubusercontent.com/IBMDataScience/DSX-DemoCenter/master/weatherGeographies/data_assets/seasonal_data.csv\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Let's pull out the data from Boston's Logan International Airport:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>id</th>\n",
       "      <th>latitude</th>\n",
       "      <th>longitude</th>\n",
       "      <th>elevation</th>\n",
       "      <th>country</th>\n",
       "      <th>name</th>\n",
       "      <th>state</th>\n",
       "      <th>21-Mar</th>\n",
       "      <th>21-Jun</th>\n",
       "      <th>21-Sep</th>\n",
       "      <th>21-Dec</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>27094</th>\n",
       "      <td>USW00014739</td>\n",
       "      <td>42.3606</td>\n",
       "      <td>-71.0106</td>\n",
       "      <td>3.7</td>\n",
       "      <td>US</td>\n",
       "      <td>BOSTON LOGAN INTL AP</td>\n",
       "      <td>MA</td>\n",
       "      <td>7.690244</td>\n",
       "      <td>25.507407</td>\n",
       "      <td>21.975309</td>\n",
       "      <td>3.892593</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                id  latitude  longitude  elevation country  \\\n",
       "27094  USW00014739   42.3606   -71.0106        3.7      US   \n",
       "\n",
       "                       name state    21-Mar     21-Jun     21-Sep    21-Dec  \n",
       "27094  BOSTON LOGAN INTL AP    MA  7.690244  25.507407  21.975309  3.892593  "
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_data_1[df_data_1['name'].str.contains('BOSTON LOGAN')]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "As shown above, we have the lat-lon coordinates, and elevation data for the station. The last four columns determine the average daily maximum temperatures over the history of the station in Celsius. For example, the average `TMAX` on the 21st of June over all the recorded history for Logan Airport's weather station is 25.5$^{\\circ}$C.\n",
    "\n",
    "<a id='The_Model'></a>\n",
    "## The Model\n",
    "Let's first split the data by columns into features and response variables, and then further into training and testing sets."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "x = df_data_1[['elevation','latitude','longitude']]\n",
    "y = df_data_1[['21-Mar','21-Jun','21-Sep','21-Dec']]\n",
    "\n",
    "x_init, x_test, y_init, y_test = train_test_split(x, y['21-Jun'], test_size=.25)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "We will fit a **Gradient Boosting Regressor** model. We should first tune our model for the hyperparameter `n_estimators` to discover what is the maximum number of estimators we should use."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "x_train, x_val, y_train, y_val = train_test_split(x_init, y_init, test_size=.25)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Now we can iterate over a specified range to determine the best value for `n_estimators`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 200/200 [02:00<00:00,  1.66it/s]\n"
     ]
    }
   ],
   "source": [
    "n_est = list(range(1,201))\n",
    "val_mad = list()\n",
    "for n in tqdm(n_est):\n",
    "    val_model = GradientBoostingRegressor(n_estimators=n)\n",
    "    val_model.fit(x_train,y_train)\n",
    "    val_pred = val_model.predict(x_val)\n",
    "    val_mad.append(mean_absolute_error(y_val,val_pred))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Let's plot our estimators vs. error plot to see what a good cutoff for n_estimators should be."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAlYAAAJCCAYAAAAC4omSAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\nAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbb7f996110>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plt.rcParams['figure.figsize'] = (10,10)\n",
    "plt.plot(n_est,val_mad)\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### Finding the Elbow\n",
    "In order to find the optimal value of `n_estimators`, we'll calculate the equation of the line between our minimum and maximum value of `n_estimators`, and find out the value of `n` in which the validation error is the furthest away from the line."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The best value of n is 33\n"
     ]
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAlYAAAJCCAYAAAAC4omSAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz\nAAALEgAACxIB0t1+/AAAIABJREFUeJzs3Xd81eX9/vHrPid7kABhhBH2iiSsgAraatXWitSBFdxK\nnXWPWr+1VdtqtcNR3NS9UJGqtWqt1D0xYe89wgiBkHlycnJOPr8/klB+lBHgJPc55/N6Ph48SpLT\n5CJ/mCv3fX/et3EcRwAAADh8HtsBAAAAYgXFCgAAIEwoVgAAAGFCsQIAAAgTihUAAECYUKwAAADC\nhGIFAAAQJhQrAACAMKFYAQAAhEmcrS+clZXl9O7d29aXBwAAaLGioqLtjuN0OtDrrBWr3r17q7Cw\n0NaXBwAAaDFjzPqWvI6tQAAAgDChWAEAAIQJxQoAACBMKFYAAABhQrECAAAIE4oVAABAmFCsAAAA\nwoRiBQAAECYUKwAAgDChWAEAAIQJxQoAACBMKFYAAABhQrECAAAIE4oVAABAmFCsAAAAwoRiBQAA\nECYUKwAAgDChWAEAAIQJxQoAACBMKFYAAABhQrHCvr38stS7t+TxNP7vyy/bTgQAQESLsx0AEerl\nl6XLL5d8vsa3169vfFuSzjvPXi4AACIYxcqFAsEGvfrdBhXvrFWntER1Sv/vn5q6oBZvrtT4G3+h\n9s2lqpnPp9pbb1MyxQoAgL2iWLnMJ8u36XfvLNGa7TVK8HoUCDXs9XXnlm7d6/sTN2/SL2bM1+3j\nhygQbND26oD6dkpVUry3NWMDABAVKFYu8sLX63TH24vVJytVz148WscN6qSquqBKq+pUWlWnbVV1\nSvB6NLR7O5npPaUNG/7nc1R1ztbMOcWaUVS86305HVJ035l5Gts/qw3/NQAARB7jOI6VL1xQUOAU\nFhZa+dputGRzpU5/9EuN699RT15QoIS4Azy3sOcZK0lKSZGmTdOC74/Xx8tK1SEtQcnxXj368Sqt\n3V6js0b10NXH91efrNTW/ccAANDGjDFFjuMUHOh1rFjFKMdx5K9vUHKCV7WBkK6dPkeZKfH6y0+H\nHbhUSf89oH777Y0rVzk50j33SOedp3xJ+T0yd7301PxsPTRrpZ7+Yo3eKCrW8YM66ZJxfXTsgCwZ\nY1rnHwgAQARixSpG3ff+Mj3x6Wr16piitMQ4LdlSqRenHKljBrTedt22Kr9e+XaDXvpmg7ZX16l/\n5zRdMq63JhX0VJyXyR4AgOjV0hUrftrFoLfmbtITn67W8YM6aUjXdqryB3XjiQNbtVRJUuf0JN1w\n4kB9edvxeuDsYUqO9+r2Nxdp0rRvtGGH78CfAACAKMeKVYxZWFyhs574SsN6ZurlS49UvMWVIsdx\n9I/5m/XrtxbJcaTjB3dWSaVf5b6AjurbUePzslXQu4O8HrYLAQCRraUrVhSrGFJTF9QPH/yssdBc\ne4yy0hJtR5IkFe/06fY3F2nVtmp1y0xSSkKcvl27Q/76BvXqmKKbThqoCfnd5KFgAQAiFIfXXeiJ\nT1drU3mt3rjy6IgpVZLUo32Knp8y5v97X01dULOWluiJT9fo+lfn6YlP1+jisb3047xstUuKt5QU\nAIDDwxmrGLGpvFbTPluj04Z3U0HvDrbjHFBqYpxOG95d7157jP46ebgCwZB+OXOhRt89S7+YMV+V\n/nrbEQEAOGisWMWIP76/TJJ068mDLSc5OB6P0WnDu+snw7ppfnGFZhYV65XZG/Tt2jI9dt5IDe2e\nYTsiAAAtxopVlPPXh/T2vE36x/zNuuJ7fdU9M9l2pENijNHwnpn6/elD9foVR6k+1KAzH/tKt74x\nX/9evFW+QNB2RAAADogVqyjlrw/plhnzNWtpya5D4Fd8v5/tWGExqlcHvXfdsbrnvaV6f+FWvV5Y\nrMyUeD123kiN7ce1OQCAyMVTgVHqkY9W6i//XqHzj8rRSblddWSfDjF5EXIg2KDZa8t01zuLtW57\njf5wZp7OLuhpOxYAwGUYEBrDtlX69dgnq3XyEV119+l5+v7ATjFZqiQpIc6jYwZkaeZVY3V0v466\n9Y0F+uO/lqmhwc4vBAAA7A/FKgr95d/LVR9q0P+dEl0H1Q9HRnK8nrl4tM49MkePf7JaV78yR7WB\nkO1YAAD8fyhWUWbRpgrNKCrWxWN7q1fHVNtx2lS816N7Th+qX48fon8t3qqzn/xaSzZX2o4FAMAu\nFKso89CslcpMjtc1PxhgO4oVxhhdemxfTbugQBt3+jT+4c9128wFKq2qsx0NAACKVTTZWuHXR8tK\ndM6YHGUku3s6+Um5XfTpLcdryrg+eqOoWD966DP9Z2mJ7VgAAJejWEWRGYUb1eBIk0bzVJwkZaTE\n6zen5ur9649Vl3ZJ+tnzhbrz7UWqC3L2CgBgB8UqSjQ0OHqtcKPG9e/ourNVBzKgS7re/PlYTRnX\nR89/vV4XPD1bZTUB27EAAC5EsYoSX6zaruKdtZo0Osd2lIiUFO/VHRNy9dfJwzVvY7nOeOxLrS6t\nth0LAOAyFKso8ep3G9Q+JV4/OqKL7SgR7bTh3TX9siNV7Q/qjEe/1JerttuOBABwEYpVFCje6dOH\nS0p05sgeSoyLzUGg4TSqVwe9dfU4dc1I0kXPzNb02RtsRwIAuATFKsJV+ut16fOFSorz6qKje9uO\nEzV6dkjRG1eN1bj+Wfq/vy/U01+stR0JAOACFKsIFgg26OcvzdGqbdV6/PxRyumYYjtSVGmXFK+n\nLypovPrn3SV6f+EW25EAADGOYhXB7n53ib5YtV33TczXMQOybMeJSnFejx6aPFzDe2bqhtfmqWh9\nme1IAIAYRrGKUBvLfHr52w268OheOmtUD9txolpSvFdPXVig7Iwknf/UbD335VoucQYAtAqKVYR6\n8rPV8hqjnx/X33aUmNAxLVGvXn60juzbQXe9s0STp32jjWU+27EAADGGYhWBtlX59XphsSaO6qGu\nGUm248SMrhlJevbi0frzWflaurVS46d+rg+XcA0OACB8KFYR6Okv1ioYatCV3+9rO0rMMcbopwU9\n9e61xyqnY4oue6FQf/rXMjkOW4MAgMNHsYowFb56vfT1ek0Y1o2ra1pRTscUvXHlWE0e3VOPfbJa\nj3y0ynYkAEAMOGCxMsb0NMZ8bIxZYoxZbIy5fi+vMcaYqcaYVcaYBcaYka0TN/bNKNqomkBIV36/\nn+0oMS8p3qt7z8zTmSO66/4PV2hmUbHtSACAKBfXgtcEJd3sOM4cY0y6pCJjzIeO4yzZ7TU/ljSg\n6c+Rkh5v+l8cBMdx9EZRsYb1zNSQ7Ha247iCMUb3TcxXSZVfv5y5QNkZSRrbn9EWAIBDc8AVK8dx\ntjiOM6fp71WSlkrqvsfLTpP0gtPoG0mZxpjssKeNcYs2VWrZ1ir9lPEKbSohzqPHzx+lPlmpuvqV\nOSreydOCAIBDc1BnrIwxvSWNkPTtHh/qLmnjbm8X63/LFw7gjaKNSozzaMKwbrajuE67pHg9ecEo\nBUOOrnypSP76kO1IAIAo1OJiZYxJkzRT0g2O41QeyhczxlxujCk0xhSWlpYeyqeIWXXBkN6ev1k/\nOqKrMpLjbcdxpb6d0vTApOFatKlSv35rEU8KAgAOWouKlTEmXo2l6mXHcf6+l5dsktRzt7d7NL3v\n/+M4zjTHcQocxyno1KnToeSNWbOWbFO5r54p65adlNtF150wQG8UFWvqf3hSEABwcFryVKCR9LSk\npY7jPLCPl/1D0oVNTwceJanCcRxuvD0IrxduVHZGksZxcNq6G04YoIkje+jBWSv04jfrbccBAESR\nljwVOE7SBZIWGmPmNb3vV5JyJMlxnCckvSfpFEmrJPkkXRL+qLFr1bYqfbqiVDecOEBej7Edx/U8\nHqP7Juap3BfQHW8vksdI547JUePvGAAA7NsBi5XjOF9I2u9PFKfxMMrV4QrlNtM+W6OkeI8uOKqX\n7ShoEu/16NHzRuqyFwp1+5uL9OGSEt17Zp6yM5JtRwMARDAmr1tWUunXm3M36eyCnuqYlmg7DnaT\nFO/V85eM0V0TcvXtmjL98MHP9N26MtuxAAARjGJl2TNfrlWowdGlx3AvYCTyeIwuHtdHH9zwPXVK\nT9RFz8zW7LWUKwDA3lGsLKr01+uVbzbolLxs5XRMsR0H+5HTMUWvXnaUsjOSdPGzs/XNmh22IwEA\nIhDFyqKZRcWqqgvqiu9xL2A06NwuSdMvP0rdMpN10TOz9fGybbYjAQAiDMXKoplzipXXPUN5PTJs\nR0ELdU5P0muXH6UBXdJ02QuF+sf8zbYjAQAiCMXKkhUlVVq0qVJnjuTmn2jTMS1R0y87SiN7tdf1\nr87VjMKNB/4/AQBcgWJlyd/nbJLXY7gXMEqlJ8XrhSljdEz/LN06c4FmFhXbjgQAiAAUKwtCDY7e\nmrtJxw3spCxGLEStpHiv/nZhgcb1y9Itb8zXW3P/5xYnAIDLUKws+Hr1Dm2t9OvMkdwLGO2ay9VR\nfTrqlhnz9dWq7bYjAQAsolhZ8Pc5xUpPitMJQzrbjoIwSE7w6skLR6lvp1Rd+VKRVm2rth0JAGAJ\nxaqN1QZC+tfirTo1P1tJ8V7bcRAm7ZLi9fRFo5UQ59Elz83W9uo625EAABZQrNrYJ8u3yRcIaUI+\nh9ZjTc8OKfrbhQUqrarTBU/PVoWv3nYkAEAbo1i1sXcXblHH1ASN6dPBdhS0ghE57fXkBQVava1a\nFz47W1V+yhUAuAnFqg3VBkL6z9Jt+tHQrorz8q2PVd8f2EmPnDtCizZV6GfPFao2ELIdCQDQRvjp\n3oY+Wb5NtfUhjc/Lth0FreyHR3TVQ5OGq3B9mS5/sVD+esoVALgBxaoNNW8DHsk2oCtMGNZNf5yY\nr89Xbtc1r8xRfajBdiQAQCujWLWR2kBIHy1jG9BtflrQU78/7QjNWrpNP3u+UJWcuQKAmMZP+DbS\n/DQg24Duc8HRvXXfmXn6atV2nfX4V9pY5rMdCQDQSihWbeTTFaVKT4pjG9ClJo/J0fNTxmhrhV9n\nPPaV1m2vsR0JANAKKFZtZPa6Mo3u3YFtQBcb1z9LM68aq1BDgy58Zra2VfltRwIAhBk/5dvA9uo6\nrSmt0ejerFa53YAu6Xr2kjHaXl2ni575jjNXABBjKFZtoHBdmSRpTJ/2lpMgEgzvmaknzh+llSVV\nmvLsd6quC9qOBAAIE4pVG5i9dqcS4zzK655pOwoixPcGdtLD54zQ3I3luuTZ2ZQrAIgRFKs28N26\nMg3vmamEOL7d+K8f52Xr4XNGaM6GxnJVQ7kCgKjHT/pWVl0X1OLNFZyvwl6dkpetqZOby9V3lCsA\niHIUq1Y2d8NONTjSaMYsYB/G52frr5OHq2jDTl3yHOUKAKIZxaqVfbe2TB4jjczhfBX27dT8bnpo\n0nAVrW8sV74A5QoAohHFqpXNXlem3G7tlJ4UbzsKItyEYY3lqnBdmS55lnIFANGIYtWKAsEGzdtY\nzvkqtNiEYd304KTh+m5dmaY8951qAyHbkQAAB4Fi1YqWba2Uv75Bo3oxvwotd9rw7npw0nB9u7ZM\nV71cpECwwXYkAEALUaxa0byN5ZIaB0ICB+O04d31hzPy9MnyUt08Y75CDY7tSACAFoizHSCWzdtY\nrqy0RHXPTLYdBVHonDE5qqit133vL1Naolf3nJ4nj8fYjgUA2A+KVSuat7Fcw3tmyBh+GOLQXPn9\nfqr2B/XIx6vkr2/Qn8/K5yJvAIhgFKtWUlFbrzWlNTpzRHfbURDlbvnRICUnePXnD5arui6oR84d\nocQ4r+1YAIC94FffVrKguPF81TDOVyEMrj6+v+6akKsPl5Tot+8ssR0HALAPrFi1knkbGotVfg+K\nFcLj4nF9tKXSryc/XaOROe111qgetiMBAPbAilUrmV9crn6dUpWRzGBQhM8vfjhIR/ftqNvfXKgl\nmyttxwEA7IFi1Qocx9G8jeVsAyLs4rweTT1nhDJT4nXZC4XaWOazHQkAsBuKVSvYVF6r7dUBjaBY\noRV0Sk/U0xeNVnVdUJOnfUO5AoAIQrFqBc2DQVmxQmsZ2j1DL196JOUKACIMxaoVzN9YroQ4jwZ3\nbWc7CmJYc7mq8tfromdna2dNwHYkAHA9ilUrWFFSrQGd05QQx7cXrWto9ww9ddFoFZfV6rIXCuWv\n59JmALCJn/ytYHVptfp1SrMdAy4xpk8HPThpuArX79T1r85VXZByBQC2UKzCzF8f0qbyWvXtlGo7\nClxkfH627pyQqw8Wl+jiZ75TRW297UgA4EoUqzBbu71GjiNWrNDmLhnXRw+cPUzfrSvTT5/4Slsq\nam1HAgDXoViF2erSakkUK9hx5sgeen7KGG0u9+uiZ2azcgUAbYxiFWZrSmskSX2y2AqEHeP6Z2na\nBaO0dnuNrnyxSIFgg+1IAOAaFKswW11are6ZyUpO8NqOAhcb2z9Lf5yYr6/X7NBtMxfIcRzbkQDA\nFbiEOcxWl1arX2e2AWHfmSN7qHhnrR74cIUGZ6fr8u/1sx0JAGIeK1Zh5DiO1pTWqB9PBCJCXPuD\n/jolr6vue3+Zvli53XYcAIh5FKsw2lrply8QUl8OriNCGGP057OGqX/nNF0zfQ5X3wBAK6NYhdHq\nbY0H11mxQiRJTYzTkxcUKNTg6LynvlXxTsoVALQWilUYNY9a6M+KFSJMn6xUvTBljMp9AU168hut\n215jOxIAxCSKVRitLq1WemKcOqUn2o4C/I8ROe31ymVHyRcIatK0r9kWBIBWQLEKozWlNerbKVXG\nGNtRgL0a2j1Dr15+tGoDIU15jqtvACDcKFZhxOXLiAaDuqbrifMbB4he/fIc1YcYIAoA4UKxCpOa\nuqC2VPiZYYWoMLZ/lv5wZp6+WLVdt81cqIYGBogCQDgwIDRM1m7nKhtEl7MLempzea0emrVSXo90\n35n58njYxgaAw0GxCpMNTQeBe3VMsZwEaLnrTxigBkea+p+VMjK698w8yhUAHAaKVZis29G4YtWr\nIytWiB7GGN144gDJcTT1o1VKTvDqzgm5PIABAIeIYhUmG3b4lJWWoLREvqWILsYY3XjSQNXWh/S3\nz9cqMyVeN5w40HYsAIhKtIAwWbejhtUqRC1jjH51yhCV++r10KyVykiO1yXj+tiOBQBRh2IVJht2\n+HRU3462YwCHzJjGM1aV/nr99p0lykyJ1xkjetiOBQBRhXELYeCvD2lLpZ8VK0S9OK9Hf508QmP7\nddQtMxZo1pIS25EAIKpQrMKgeKdPjiP1zuKJQES/pHivpl1YoKHd2unqV+bomzU7bEcCgKhBsQqD\nddsbRy3kdKBYITakJcbpuUvGqGeHFF36fKEWbaqwHQkAogLFKgzWN82w6s1WIGJI+9QEvfizMcpI\njtdFz8zW6tJq25EAIOJRrMJg/Y4apSfFKTMl3nYUIKyyM5L14s/GyBjpwqdna3t1ne1IABDRKFZh\nsH6HT707pjJUETGpb6c0PXvxGG2vrtPPX5qjQJBLmwFgXyhWYbB+R41yuMoGMSyvR4b+/NNhmr2u\nTHe8vUiOw6XNALA3FKvDFAw1qHhnrXpTrBDjfjKsm64+vp9e/W6jXvp2g+04ABCRKFaHaXO5X8EG\nR706cHAdse/mkwbpuEGd9Pt3lvCkIADsBcXqMK0va758mRUrxD6Px+iBs4erQ2qCrn5ljqr89bYj\nAUBEoVgdpnU7GkctMHUdbtEhNUEPnztCxTtrddvMhZy3AoDdUKwO04YdNUqK96hzeqLtKECbGd27\ng37xo0F6d+EW/emD5bbjAEDE4BLmw7R+h085HVLk8TBqAe5yxff6akOZT49/slqd0xN1ybg+tiMB\ngHUUq8O0uaJW3TOTbccA2pwxRr8/bai2V9Xpd/9coqy0RE0Y1s12LACwiq3Aw7S53K9uFCu4lNdj\nNPWcERrdq4Nufn2+vlq13XYkALCKYnUYagMhldUEKFZwtaR4r/52YYH6ZKXq8heLGMMAwNUoVodh\nS0WtJKlbZpLlJIBdGSnxem7KaLVLitPFz36nzeW1tiMBgBUUq8OwudwvSeqWwYoVkJ2RrOenjJG/\nPqQrXiySvz5kOxIAtDmK1WFo/q2crUCg0YAu6Xpw0nAt3FShX73JjCsA7kOxOgybK2pljNQ1g61A\noNlJuV1044kD9fc5m/Tsl+tsxwGANkWxOgyby2vVOT1R8V6+jcDurv1Bf/0wt4vueW8pTwoCcBUa\nwWHYXO5XNuergP/h8Rg9MGm4+mal6upX5mhjmc92JABoExSrw8BwUGDf0hLj9LcLCxRqcHT5i0Wq\nrgvajgQArY5idYgcx9Hm8lpGLQD70TsrVVPPGaEVJVW65NnZqqFcAYhxFKtDtNNXL399A1uBwAEc\nN6izpk4eoTkbynXJc9/JF6BcAYhdFKtDxKgFoOXG52froUnDVbiuTFe8WKRQA2MYAMQmitUhai5W\nnLECWmbCsG76wxl5+nzldv3l38ttxwGAVkGxOkTNxSqbM1ZAi00ek6NzxuTo8U9W61+LttiOAwBh\nR7E6RFsq/EqI86hjaoLtKEBUuesnuRrWM1M3vz5fK0qqbMcBgLCiWB2iTeW16paRJGOM7ShAVEmM\n8+qJ80cqJTFOlzz7nbZV+W1HAoCwoVgdosZRC5yvAg5FdkaynrlotMpqArrs+ULVBriwGUBsoFgd\noi0VfooVcBjyemRo6jkjtGBThW54bS5PCgKICRSrQ1AfalBJpV/duHwZOCwn5XbRb8bn6oPFJbr3\nvaW24wDAYYuzHSAalVT61eAwwwoIhynH9NGGMp+e+mKtenVM0QVH97YdCQAOGcXqEGwubzxsm02x\nAsLiN6fmqninT3f+Y7H6dUrT2P5ZtiMBwCFhK/AQbCjzSZJyOqRYTgLEBq/HaOo5I9Q7K1W3zlzA\nnYIAohbF6hBs2FEjj2HqOhBOKQlx+uPEfBXvrGUyO4CoRbE6BOvLfMrOSFZCHN8+IJxG9+6gC4/u\npee+Wqei9TttxwGAg0YzOATrd/jUqyPbgEBruPXkwcpul6RbZsxXSSXDQwFEF4rVIdhYRrECWkta\nYpymnjNC2yr9mvj4V1q3vcZ2JABoMYrVQaquC2pHTUA5HVJtRwFiVkHvDpp++VGqqQvqrCe+5k5B\nAFGDYnWQ1u9o/O2ZFSugdeX3yNSMK4+Wx0hTnvtOZTUB25EA4IAoVgdpww5GLQBtpX/ndE27sEDb\nqup09ctzVB9qsB0JAPbrgMXKGPOMMWabMWbRPj5+nDGmwhgzr+nPHeGPGTnWN8+wYsUKaBPDe2bq\n3jPy9PWaHbrnXa69ARDZWjJ5/TlJj0h6YT+v+dxxnFPDkijCrd/hU/uUeLVLircdBXCNiaN6aOmW\nSj31xVoNyU7XpNE5tiMBwF4dcMXKcZzPJJW1QZaosLHMp5yOHFwH2tptPx6sYwdk6ddvLVLRev6T\nBCAyheuM1VhjzAJjzPvGmCP29SJjzOXGmEJjTGFpaWmYvnTbWl9Ww/kqwII4r0ePnDNS3TOTdcWL\nc7SlotZ2JAD4H+EoVnMk5TiOky/pYUlv7euFjuNMcxynwHGcgk6dOoXhS7et+lCDNpf71YtiBViR\nkRKvv11YIH99SOdM+0Ybm848AkCkOOxi5ThOpeM41U1/f09SvDEmJq+m37SzVqEGh4PrgEUDuqTr\n+SljtNNXr4mPf6VlWyttRwKAXQ67WBljuhpjTNPfxzR9zh2H+3kjUfMTgaxYAXaN6tVeM648WsZI\nZz/xtdYynR1AhGjJuIXpkr6WNMgYU2yM+Zkx5kpjzJVNLzlL0iJjzHxJUyVNdhzHab3I9mxg1AIQ\nMQZ2SdeMK8bK4zG65pU58teHbEcCgAOPW3Ac55wDfPwRNY5jiHkbdtQoIc6jLulJtqMAUOMvOX85\na5gufaFQ97y7VL8/fajtSABcjsnrB2H9Dp9yOqTI4zG2owBocmJuF112bB+9+M16vbtgi+04AFyO\nYnUQNu6sZdQCEIFuPXmwRuRk6tY35mvVNi5sBmAPxeoglFT61TWDbUAg0sR7PXrsvJFKTvDq8heL\nVOWvtx0JgEtRrFqoLhhSWU1AXdtRrIBIlJ2RrEfPHan1O3y66fX5amiIyWdoAEQ4ilULbauskySK\nFRDBjuzbUbefMkQfLinRL95YoGCowXYkAC7TkkuYIWlrpV+S1IWtQCCiXTKut6rrgnrgwxWqrqvX\n1HNGKDHOazsWAJdgxaqFtlY0FitWrIDIZozRdScM0B2n5uqDxSW6+uU5bAsCaDMUqxaiWAHRZcox\nfXTXhFzNWrpNUz9aaTsOAJegWLXQ1kq/kuI9apfM7ikQLS4a21sTR/bQX/+zUh8v22Y7DgAXoFi1\n0NZKv7q2S1LTtYgAooAxRvecMVRDurbT9a/O1apt1bYjAYhxFKsWKqnwqwvbgEDUSYr36onzRykh\nzqNJT36txZsrbEcCEMMoVi20leGgQNTK6Zii1684WolxHk2e9o2K1u+0HQlAjKJYtYDjONpWWcfB\ndSCK9e2UphlXjVXH1ARd9MxsLd/K1TcAwo9i1QJlNQEFQg1sBQJRrntmsl657CilJHg15bnvtK1p\nPh0AhAvFqgWah4OyFQhEv26ZyXrm4tEqqwno0hcK5QsEbUcCEEMoVi1Q0jx1nRUrICYM7Z6hh88Z\noYWbKnTn24ttxwEQQyhWLbC1oumeQFasgJhxYm4X/fy4fppRVKz3F26xHQdAjKBYtcDWSr+MkTqn\nJ9qOAiCMbjhxoPJ7ZOj/3ly463YFADgcFKsWKKnwq2NqouK9fLuAWBLv9eihScNVV9+gm2fMUzDU\nYDsSgChHU2iBrZV+ZbMNCMSkvp3S9NvTjtCXq3borncWy3G4sBnAoePiuxYoqfSrR/sU2zEAtJKz\nC3pqdWm1nvx0jXI6pOjy7/WzHQlAlGLFqgUap65zvgqIZb/80WCNz8vWH95bpg8Wb7UdB0CUolgd\ngL8+pHJfPVPXgRjn8Rjdf/Yw5ffI0C9nLtg1ZgUADgbF6gCYYQW4R1K8d9dh9ltmzFdDA+etABwc\nitUBND/5ICOQAAAgAElEQVSCzQwrwB36dkrT7eOH6POV2/XC1+tsxwEQZShWB1BS1TgclBUrwD3O\nOzJHPxjcWfe+v0wrS7isGUDLUawOYBtbgYDrGGN038Q8pSbG6YbX5ikQZL4VgJahWB1ASaVfiXEe\ntUtiMgXgJp3Tk3TfmXlavLlSD81aYTsOgChBsTqAbVV16tIuScYY21EAtLEfHtFVkwp66olPV2v2\n2jLbcQBEAYrVAZRU+tWlHTOsALf6zYRc9eyQop+/XKQNO3y24wCIcBSrA9hWWafO6ZyvAtwqLTFO\nT180WvUhRxc/O1s7awK2IwGIYBSrA9hWVafOrFgBrta/c5qeuqhAxeW1uuyFQtUFQ7YjAYhQFKv9\nqKkLqrouyIoVAI3u3UEPnD1Mhet36uH/rLIdB0CEoljtx7ZdM6xYsQIgnZrfTRNH9tDjn67WwuIK\n23EARCCK1X5wnQ2APd1xaq46pibolhnzmW8F4H9QrPajuVh1TmfFCkCjjJR4/eGMPC0vqWK+FYD/\nQbHaj9KmrcDOrFgB2M2JuV00qaCnHvtktd5fuMV2HAARhGK1H0xdB7Avvz3tCI3IydRNr8/Xok2c\ntwLQiGK1H0xdB7AvSfFePXnBKGWmxOvyFwp3rXADcDeK1X4wdR3A/nROT9LfLizQTl+9rniR+VYA\nKFb7xdR1AAcytHuG7j97mOZsKNftby6S4zi2IwGwiGK1H0xdB9ASp+Rl64YTB+iNomL97fM1tuMA\nsIhT2fvA1HUAB+O6HwzQypJq3fv+MnXLTNap+d1sRwJgAcVqH5i6DuBgeDxG9589TNuq/Lrptfnq\nkJqgsf2ybMcC0MbYCtyH/w4HZcUKQMskxXv11IWj1atjiq54oUgrS6psRwLQxihW+/Df62xYsQLQ\nchkp8Xp+yhjFx3l04+vzVB/i2hvATShW+8DUdQCHqltmsv5wxlAt2lSpxz5ebTsOgDZEsdoHpq4D\nOBwnD83WacO76eGPVjKZHXARitU+MHUdwOH67U+OUIfUBN342jxV+ettxwHQBihW+1BS6VfndM5X\nATh0mSkJenDScK3ZXqNrp89VkPNWQMyjWO3DtsrGFSsAOBzj+mfp96cN1SfLS/X7fy6xHQdAK+MA\n0T7sqAmoY1qC7RgAYsC5R+Zo3Y4aTftsjQZ0Sdf5R/WyHQlAK2HFai8aGhxV+uuVmRxvOwqAGHHb\nyYN13KBO+t0/l2jJ5krbcQC0EorVXlT5g3IcqR3FCkCYeDxG9/90mDKT43XN9DnyBYK2IwFoBRSr\nvaiobXx6J4NiBSCMOqYl6qHJw7V2e41+89ZiOY5jOxKAMKNY7UVzscpM4YwVgPAa2y9L1x7fXzPn\nFOvFb9bbjgMgzChWe1FeG5DEihWA1nH9iQN14pDO+u07S/T5ylLbcQCEEcVqL9gKBNCavB6jhyaP\nUP9Oabr65TlaU1ptOxKAMKFY7UW5r3krkGIFoHWkJcbpqYsKFOf16JpX5qouGLIdCUAYUKz2ghUr\nAG2hZ4cU/WlivpZsqdQDH66wHQdAGFCs9qKytl6JcR4lxXttRwEQ407M7aJzj8zRtM/W6OvVO2zH\nAXCYKFZ7Ue6rZ7UKQJv59fgh6t0xVTe9Pk9lNQHbcQAcBorVXlTUUqwAtJ2UhDhNnTxCO2oCuvrl\nOarnsmYgalGs9qKitp6D6wDaVF6PDN13Zp6+XrND97y71HYcAIeIS5j3ory2Xt0zk2zHAOAyZ47s\noaVbKvW3z9dqSHa6Jo3OsR0JwEFixWovKmvrlZHM1HUAbe+XJw/WsQOy9Ou3FqlofZntOAAOEsVq\nL8p9Ac5YAbAizuvRI+eMVPfMZF3x4hxtLq+1HQnAQaBY7aE+1KCaQIhiBcCajJR4/e3CAvnrQ7ri\nxSLVBhgeCkQLitUeKmuZug7AvgFd0vXQpOFatLlC106fqyBPCgJRgWK1h3KmrgOIECfmdtFdE47Q\nrKUluuMfi+U4ju1IAA6ApwL3wHU2ACLJRWN7a2ulX49/slrdM5N19fH9bUcCsB8Uqz3sKlZsBQKI\nELf+aJC2lNfqzx8s18Au6Topt4vtSAD2ga3APVT4WLECEFmMMbpvYr7ye2ToxtfmadW2KtuRAOwD\nxWoPzStWmRQrABEkKd6rJy8YpaR4ry59vnDXf6sARBaK1R7Km1as2lGsAESY7IxkPXH+SG3cWau7\n/7nEdhwAe0Gx2kNFbb1SE7yK9/KtARB5Cnp30BXf66sZRcX6dEWp7TgA9kB72EPjBcxcZwMgcl13\nwgD165SqX/19oarrgrbjANgNxWoPFbUBtgEBRLSkeK/+dFa+NlfU6t73ltqOA2A3FKs9VNTWKyOZ\nKRQAItuoXh30s3F99PK3G/TPBZttxwHQhGK1h4raemUmsxUIIPLdevJgjczJ1K1vLNDKEkYwAJGA\nYrWHcl89M6wARIWEOI8eO2+UUhK8uvKlIs5bARGAYrWHxsPrFCsA0aFrRpKmnjNCa7fX6ObX56mh\ngfsEAZsoVrvx14dUF2zg8DqAqDK2X5ZuH5+rDxaXaOpHK23HAVyNU9q74QJmANFqyrjeWrK5Ug/N\nWqnBXdvp5KFdbUcCXIkVq93sus6GrUAAUcYYo3vOGKphPTN10+vztGxrpe1IgCtRrHZTzgXMAKJY\nUrxX0y4YpbTEOF32QqF21gRsRwJch2K1m/9ewMy4BQDRqUu7JD1xwSiVVNTpmulzFAw12I4EuArF\najecsQIQC0bmtNfdZwzVl6t26L73l9mOA7gKh9d3U+5rXDanWAGIdmcX9NTiTRV66ou1yu+ZqZ8M\n62Y7EuAKrFjtptxXL4+R0pPomwCi3+3jc1XQq71++cYCDrMDbYRitZvy2oAyUxLk8RjbUQDgsDVO\nZh+p9KQ4Xfkik9mBtkCx2s1OX70y2QYEEEM6t0vSI+eO1Poyn/7w3lLbcYCYR7HaTbkvwAwrADFn\nTJ8OuvzYvnrl2w36dEWp7ThATKNY7WZnTb3apzBqAUDsufGkgRrQOU2/fGPBriegAYQfxWo3jRcw\nU6wAxJ6keK/uP3uYSqvrdNvMBVzWDLQSitVudrIVCCCG5ffI1G0nD9b7i7ZyWTPQSpgr0KQuGJIv\nEFJ7ihWAGHbpsX20vKRKD81aqQGd0zU+P9t2JCCmsGLVpPmeQLYCAcSy5suaR/Vqr5tnzNOqbVW2\nIwExhWLVpLlYcXgdQKxLjPPq8fNHKjneq5tnLOA+QSCMKFZNdjZdZ8NWIAA36JyepN+dNlTzN5br\nb5+vtR0HiBkUqya77gmkWAFwiVPzs/XjoV314IcrtKKELUEgHChWTXayFQjAZYwx+v3pQ5WWFKfr\nX50nf33IdiQg6lGsmvx3K5BiBcA9stISdf/Zw7R0S6XueHuR7ThA1KNYNanw1SsxzqPkBK/tKADQ\npo4f1FnX/qC/Xi8s1uvfbbQdB4hqFKsmDAcF4GY3nDhQ4/p31G/eXqTFmytsxwGi1gGLlTHmGWPM\nNmPMXteITaOpxphVxpgFxpiR4Y/Z+nb6uCcQgHt5PUZ/nTxC7VMS9POX53CfIHCIWrJi9Zykk/fz\n8R9LGtD053JJjx9+rLZXzooVAJfLSkvUo+eN0KadtfrFjPlyHO4TBA7WAYuV4zifSSrbz0tOk/SC\n0+gbSZnGmKi7I6GcFSsA0KheHfR/pwzRv5eUaNpna2zHAaJOOM5YdZe0+2nH4qb3RZWdvnpWrABA\n0pRxvTU+L1t/+mC5vl2zw3YcIKq06eF1Y8zlxphCY0xhaWlpW37p/XIcp2krkBUrADDG6L6JeerV\nIUXXTJ+rbVV+25GAqBGOYrVJUs/d3u7R9L7/4TjONMdxChzHKejUqVMYvnR4VNcFFWxwuM4GAJqk\nJ8XrsfNHqspfr+umz+U+QaCFwlGs/iHpwqanA4+SVOE4zpYwfN4203wBMytWAPBfg7u20x/OyNM3\na8p0/4crbMcBokLcgV5gjJku6ThJWcaYYkl3SoqXJMdxnpD0nqRTJK2S5JN0SWuFbS3lXGcDAHt1\n5sge+m7dTj3+yWqNymmvE3O72I4ERLQDFivHcc45wMcdSVeHLZEFzdfZcHgdAP7XnRNytXBTuW56\nfZ7+ee2xyumYYjsSELGYvK7d7wmkWAHAnpLivXr8vFGSpJ+/UsRlzcB+UKykXROGOWMFAHvXs0OK\nHjh7uBZtqtRv31liOw4QsShWknbWNBWrZFasAGBfTsztoquO66fpszdoZlGx7ThARKJYqXErMD0x\nTnFevh0AsD83nzRQR/XtoNvfWqhlWyttxwEiDk1CTfcEprJaBQAHEuf1aOo5I9QuKV5XvTRHVX4u\nawZ2R7FS43U2jFoAgJbpnJ6kh88ZoQ1lPv1y5gIuawZ2Q7GSVF5bz8F1ADgIR/btqFt/NEjvLdyq\nZ79cZzsOEDEoVmraCuTgOgAclMu/11c/zO2iP7y3VPM3ltuOA0QEipWknTUBZlgBwEEyxujPZw1T\n5/REXf/qXNXUBW1HAqxzfbEKNTiq9AeVwVYgABy0jJR4PThpuNaX+XTXPxbbjgNY5/piVRNo/A2r\nXdIBb/cBAOzFkX076prj+2tGUbHenrfJdhzAKtcXq2p/Y7FKS6RYAcChuu6EASro1V63zVyoJZuZ\nbwX3olg1nQlIT+KMFQAcqnivR4+dP1IZyfG6/MVCldUEbEcCrHB9sWoebpfGViAAHJbO6Ul68oJR\n2lZVp2temaNQA/Ot4D4UK7YCASBshvXM1N2nD9VXq3foqc/X2I4DtDnXF6vmrUAOrwNAePx0VA/9\n6Iguuv/DFVpZUmU7DtCmXF+sdq1YUawAICyMMbrnjDylJcbplhnzFQw12I4EtBnXFyueCgSA8MtK\nS9TvTxuq+cUVeuyT1bbjAG3G9cWqqi4oY6TUBIoVAITT+PxsnTa8mx6atUJfr95hOw7QJihW/nql\nJcTJ4zG2owBAzLnnjDz1zkrVda/O1bYqv+04QKtzfbGq9gc5XwUArSQtMU6PnzdKVf56XT99HiMY\nEPMoVnVBzlcBQCsa1DVdd5+ep6/X7NBDs1bYjgO0KtcXqyp/UOmsWAFAqzprVA9NKuiphz9apU+W\nb7MdB2g1FKu6oNK4zgYAWt1vTztCg7um68bX5mlzea3tOECrcH2xqvbXK52tQABodUnxXj123kjV\nhxxd+VKRapoGNAOxxPXFiq1AAGg7fTul6a+Th2vRpgpd/coc1TM8FDHG9cWKw+sA0LZOGNJFd5+e\np0+Wl+r2NxfKcXhSELHD1Y0i1ODIFwgxbgEA2ti5R+Zoa6VfU/+zUkd0y9BFY3vbjgSEhatXrJqv\ns0nn8DoAtLkbTxyg4wZ10r3vL9Xq0mrbcYCwcHWxqqqrlyQOrwOABcYY/WlivpLivbrptXmct0JM\ncHWxqm56IoWtQACwo3O7JN1zep7mF1fo0Y9X2Y4DHDZXF6uqpq1ADq8DgD3j87N1+vBueuSjVVq6\npdJ2HOCwuLpY/feMFcUKAGy6c8IRykiO1y9nLlCQLUFEMVcXq6o6ihUARIL2qQm66ydHaEFxhZ79\ncp3tOMAhc3Wxqt61FchTgQBg26n52TpxSGfd/+FyreEpQUQpVxerKn/TU4GsWAGAdcYY3X16npLi\nvfrZ84UqqwnYjgQcNFcXq+q6oIyRUhK8tqMAACR1zUjS0xcVaHN5rS59/jv560O2IwEHxdXFqsrf\neJ2NMcZ2FABAk1G9Ouivk4dr7sZy3fDqPK68QVRxfbFqx9R1AIg4Jw/N1q9+PET/WrxVL32z3nYc\noMVcXayq6+qZYQUAEerSY/vo+wM76e53l2rVtirbcYAWcXmxCjJ1HQAilDFGf/5pvlIT43T9q/MU\nCDLfCpHP1cWq+YwVACAydU5P0h8n5mvx5krd9/4y23GAA3J1sar2Bxm1AAAR7qTcLrp4bG898+Va\n/XPBZttxgP1ydbGqqqNYAUA0+NUpQzSqV3vd+sYCrSzhvBUil7uLlZ/D6wAQDRLiPHr03JFKSfDq\nipeKVOGrtx0J2CvXFqv6UIP89Q1KZ9wCAESFrhlJevTckSouq9VlLxQyPBQRybXFqqau+Z5AVqwA\nIFoc2bej7j97mGavK9NNr89TqIHhoYgsrm0VVc0XMHPGCgCiyoRh3VRS6dfd7y5Vz/bL9H+nDLEd\nCdjFta2iuVi1o1gBQNS59Ni+WrejRk9+tkYjcjJ18tBs25EASS7eCqzetRXIGSsAiEa/OTVXw3pm\n6pYZC7SmtNp2HECSi4tVlb/xiRK2AgEgOiXGefXYeSMV7zW66qU5HGZHRHBtsWpesWKOFQBEr+6Z\nyXrg7OFaXlKlZ75cazsO4N5i1XzGKp2nAgEgqh0/uLNOyu2iRz9apW2Vfttx4HKuL1ZsBQJA9PvV\nKUMUCDXoL/9ebjsKXM61xaq6rl4eIyXHe21HAQAcpj5ZqbpkXB/NKCrWok0VtuPAxVxbrGrqQkpL\njJMxxnYUAEAYXPOD/uqQkqBfzlzAQXZY4+JiFVQq56sAIGa0S4rXn87K1+LNlbrz7cW248Cl3Fus\nAkGlJLANCACx5IQhXXTN8f31WuFGvTp7g+04cCH3FqumrUAAQGy58aSBOnZAlu74x2LOW6HNubZY\n+QJBpSRQrAAg1ng9Rn+dPELtU+J13fS5qmmaWwi0BdcWq+q6kFIT2QoEgFjUITVBD04arrU7avTb\ndzhvhbbj2mLlC3B4HQBi2dh+Wbr6uP56vbBY78zfbDsOXMK1xaqmLsRWIADEuOtPHKCROZm6beYC\nLdtaaTsOXMDFxSqoVJ4KBICYFu/16PHzRyk1MU6XPl+ospqA7UiIca4sVqEGR7X1IbYCAcAFurRL\n0rQLC7Stqk5XvVSkQLDBdiTEMFcWq9qmibwcXgcAdxjeM1N/Pitf364t4zA7WpUrl2yaH71lxQoA\n3OO04d21bGuVHv9ktQZ3TdcFR/e2HQkxyJUrVruKFYfXAcBVbvnhIJ0wuLPuemeJvlq13XYcxCBX\nFitfoHErkCttAMBdvB6jhyYPV79Oqbp2+lxV+OptR0KMcWWxqm5aseJKGwBwn/SkeD00aYR2+gL6\n0wfLbMdBjHFlsfIFGotVCsUKAFwpt1s7XTy2j16ZvUHzNpbbjoMY4spiVV3X9FQgW4EA4Fo3/XCg\nOqcn6vY3FyoYYgQDwsOVxcrHU4EA4HppiXG649QjtHhzpR79eLXtOIgRrixWNYHmFSuKFQC42Sl5\nXXXGiO56cNYK/XMB9wni8LmzWNU1n7FiKxAA3MwYo/sm5qmgV3vd/Pp8zd2w03YkRDl3FqtAUAlx\nHsV7XfnPBwDsJjHOqycvGKUu7ZJ02QuFKt7psx0JUcyVzcJXF+LgOgBgl45piXrm4tGqCzboZ88V\nqsrPfCscGlcWq5q6oFI4XwUA2E3/zml6/LxRWlVarWunz+VJQRwSdxarQJDhoACA/3HMgCz9/rSh\n+mR5qe7/cIXtOIhCrixWvkCIg+sAgL0698gcTSroqSc/Xc1hdhw0Vxar6rogoxYAAPt0+6lD1LVd\nkm6ZMV/++pDtOIgirixWvrqQUlmxAgDsQ7ukeN03MV+rS2v0IFuCOAiuLFY1AVasAAD7972BnXTO\nmBz97fM1KlrPliBaxp3Fqi7IGSsAwAH96pTBys5I1i/YEkQLubNYBULcEwgAOKD0pHj9cWK+1myv\n0f3/Xm47DqKA64pVfahBgWADW4EAgBY5ZkCWzjsyR099sVaF68psx0GEc12x8tU1LuWmMHkdANBC\n/3fKEPVon6yrX5mjkkq/7TiIYK4rVjWBxguYGRAKAGiptMQ4TbugQFX+oC59vlC1Ac5bYe9cV6x8\nTcUqhWIFADgIQ7LbaerkEVq0uUI3z5inhgbHdiREINcVq+qmrUAuYQYAHKwTc7votpMH672FW/X2\n/E224yACua5Y+eoaV6x4KhAAcCguO7avhvXI0B/fX75rFwRo5rpiVRNoXrGiWAEADp7HY3THhFxt\nrfTriU9W246DCOO+YrVrxYqtQADAoRnVq4N+MqybnvxsjYp3+mzHQQRxX7EKsBUIADh8t/14sIyR\nfvPWIoU4yI4mritWzLECAIRDt8xk/eqUIfp4eanufW+p7TiIEK5btqlu2gpM4YwVAOAwXXh0b63e\nVq2nvlir3lmpOv+oXrYjwTLXtQtfIKjkeK+8HmM7CgAgBvzm1FxtKPPpzn8sVs8OKfr+wE62I8Ei\n120FVteFOLgOAAibOK9HD587UgO7pOvql+do+dYq25FgkeuKlS8Q5OA6ACCs0hLj9MzFBUpJ8GrK\nc99pWxX3CbqV64pVTV2I81UAgLDLzkjW0xeNVllNQFe8WKT6UIPtSLDAhcUqyHU2AIBWkdcjQ386\nK19zN5TrkY9W2Y4DC1xXrNgKBAC0pgnDuunMEd318EcrVbR+p+04aGOuK1Y1AQ6vAwBa112nHaHs\njGTd+Nq8XWN+4A7uK1Z1Qc5YAQBaVbukeD04abiKd/r0u3cW246DNuTKYpXGViAAoJWN6dNBVx3X\nT68XFutfi7bYjoM24qpi5TiOfIEQ19kAANrE9ScMVF73DN3294UqqWQEgxu0qFgZY042xiw3xqwy\nxty2l48fZ4ypMMbMa/pzR/ijHr66YIOCDQ6H1wEAbSIhzqOHJg+Xvz6kW2bMVwOXNce8AxYrY4xX\n0qOSfiwpV9I5xpjcvbz0c8dxhjf9+V2Yc4aFL9B4ATPjFgAAbaVfpzT9enyuPl+5Xc9/vc52HLSy\nlqxYjZG0ynGcNY7jBCS9Kum01o3VOmqaL2BmxQoA0IbOOzJHPxjcWfe+v0wrSrjyJpa1pFh1l7Rx\nt7eLm963p7HGmAXGmPeNMUfs7RMZYy43xhQaYwpLS0sPIe7hqQk0FqtUngoEALQhY4z+ODFf6Ylx\num76XNUFQ7YjoZWE6/D6HEk5juPkS3pY0lt7e5HjONMcxylwHKegU6e2v/27pq5pK5A5VgCANtYp\nPVF/Oitfy7ZW6baZCzlvFaNaUqw2Seq529s9mt63i+M4lY7jVDf9/T1J8caYrLClDBNf84oVW4EA\nAAtOGNJFN580UG/O3aRfvUm5ikUtaRjfSRpgjOmjxkI1WdK5u7/AGNNVUonjOI4xZowaC9uOcIc9\nXLvOWHF4HQBgybUnDJA/GNKjH69WUrxXd/1kr6dnEKUOWKwcxwkaY66R9IEkr6RnHMdZbIy5sunj\nT0g6S9JVxpigpFpJkx3Hibga3rwVyIBQAIBNt/xwkHyBkJ79cp2O6ttBJw/Nth0JYdKihtG0vffe\nHu97Yre/PyLpkfBGC7/mrUCutAEA2GSM0a9OGaLZa8v067cW68g+HdU+NcF2LISBqyavV7NiBQCI\nEPFej/581jCV+wL63T+X2I6DMHFVsfIFgjJGSop31T8bABChcru108+P7683527SrCUltuMgDFzV\nMGrqQkpNiJMxxnYUAAAkSdcc31+Du6brtr8vUGlVne04OEwuK1ZBZlgBACJKQpxHU88ZoSp/kPsE\nY4C7ilUgyNR1AEDEGdglXb8+NVefrijVs1+tsx0Hh8FVxcoXCCmFFSsAQAQ6/8gcnTiki/74/jLN\n2bDTdhwcIlcVq+o6VqwAAJHJGKM/n5WvrhlJuuLFIm2t8NuOhEPgqmLlCwS5zgYAELHapyboqYsK\n5KsL6vIXC+Wv57LmaOOuYlUX4jobAEBEG9glXX+dPEILN1XolzMXKAIvMsF+uKpYVdcFGQ4KAIh4\nJ+Z20S0/HKS3523WE5+usR0HB8FVxcoXCHGdDQAgKvz8uH6aMKyb/vTBMv1nKcNDo4VripXjOI3j\nFngqEAAQBYwx+tPEfA3tlqHrX52n9TtqbEdCC7imWNXWh+Q44vA6ACBqJCd49cQFo9TgOPr9P5fa\njoMWcE2xqmm6gDmVw+sAgCjSPTNZ1/5ggGYtLdEny7fZjoMDcFGxCkoSZ6wAAFFnyjG91ScrVb97\nZ4kCwQbbcbAf7ilWgcZixVYgACDaJMZ5dceEXK3ZXqOnv1hrOw72wzXFyhdo2grk8DoAIAodP6iz\nTsrtogdnrdDyrVW242AfXFOsqtkKBABEuXvPzFO7pHhdN30uU9kjlGuKla/p8DoDQgEA0SorLVF/\n+Wm+lpdU6d73eEowErmmWDWfseJKGwBANDtuUGf97Jg+ev7r9Zo+e4PtONiDa5Zvmp8K5PA6ACDa\n3XryIK0urdb//X2hqv1BXfa9vrYjoYlrVqw4vA4AiBWJcV5Nu6BA4/Ozdc97S/XXWSttR0IT1yzf\n1NQFFecxSvC6pksCAGJYQpxHUyePUKLXowdnrdDY/h01uncH27FczzUto6YuqJQEr4wxtqMAABAW\nXo/R708fqu6Zybpt5gLVBXlS0Db3FKtAiCcCAQAxJzUxTnefMVSrS2v02MerbcdxPdcUK18gqBSK\nFQAgBh0/qLNOG95Nj32ySitLGB5qk2uKVXVdiCcCAQAx6zen5io1MU63/X2hGhoc23FcyzXFylcX\nVCozrAAAMSorLVG/Hp+rovU79TLzraxxTbGqCYS4zgYAENMmjuyuY/pn6Y/vL9PWCr/tOK7knmJV\nF1QaM6wAADHMGKN7zhiqYEODfv3WIjkOW4JtzTXFisPrAAA36NUxVbf8cJBmLS3Rox+vsh3HdVzT\nNGrqQpyxAgC4ws+O6aPFmyv1l3+vUN9OaTolL9t2JNdwxYpVqMFRbT1PBQIA3MEYo3vPzNOoXu11\n0+vztGhThe1IruGKYuULNF3AzOF1AIBLJMV79eQFo5SZnKCbXp+nQLDBdiRXcEmxahzxn8LhdQCA\ni2SlJeoPZw7VipJqPfYJ563agiuKVXVd44oVV9oAANzmB4O76LTh3fTox6u0gqnsrc4VxcpX17Ri\nxVYgAMCF7jg1V2mJcfrFG1zU3NpcUayaV6x4KhAA4EYd0xJ19+l5mr+xXDe9Nl8hrrxpNa5Ywtl1\neC6JnSgAABAOSURBVJ2tQACAS43Pz9aWiiG6+92lSkuM030T82SMsR0r5riiadQ0HV5P5fA6AMDF\nLj22rypq6/XwR6vUvX2yrjthgO1IMccVW4E1TVuBnLECALjdTScN1OnDu2nqf1Zq6ZZK23FijquK\nFVuBAAC3M8bozglHKCM5XrfNXMB5qzBzRbHaNceKw+sAAKh9aoLu/MkRml9coWe/XGs7TkxxRbGq\nqQsqIc6jeK8r/rkAABzQhPxsnTC4s+7/9wqtLq22HSdmuKJp1ASCDAcFAGA3xhjdc0aekuI9uvrl\nOfLXM98qHFxRrHx1ISXHsw0IAMDuumYk6YFJw7Vsa5V++84S23FigjuKVSDEqAUAAPbi+EGdddVx\n/TR99ga9NXeT7ThRzxXFqraeFSsAAPbl5pMGakyfDrr1jQX6evUO23GimmuKVRLFCgCAvYrzejTt\nglHK6Ziiy18o1JLNzLc6VO4oVoEQoxYAANiPzJQEvTBljNKS4nTRs7O1paLWdqSo5I5iVR9SMsUK\nAID96paZrBemjJGvLqgbX5vH8NBD4I5iFWArEACAlhjQJV13/eQIfbOmTE9+ttp2nKjjimLlr2cr\nEACAljprVA+dmp+t/9fevUdJXd53HP98dxdcQJabBJCLXAQMaMsqGm001pBE4cQASU+BRBu0p+Kp\niVrPKbXBcoxp08aaeDlJpLblJFGjtCoXPWliMKntP5E7yFUWuciCIHfI7jJ7+faP+S1dCQts/M08\nM8+8X+fsYffZgf0+PPOb+ezzPPPM915/R2vfOxK6nKJSEsGqLsOrAgEAOF+th4f2q6rUfS+u1onk\nPXdxbtEHK3fnuAUAADqoR5dOenzaOO06VKeHl2wIXU7RiD5YnWxqkSRVshQIAECHXDOst75206V6\naeVuvbp2T+hyikL0waouk33vo67MWAEA0GH3Thip6iE99Y2Fb2v34brQ5RS86INVffKmkhy3AABA\nx1WUl+nJadVyF0cwnIf4g1UyY8VxCwAA/H6G9Omqb00Zq+U7DusHv64JXU5Biz5YNSQzVl07VwSu\nBACA4jW1epCmjLtYT76xVSt3Hg5dTsGKPli17rHiVYEAAHw0j0y5XAN6VOr+Bat1vKExdDkFKfpg\n9f97rKLvKgAAOVVV2UlPTq/WniMNmruYIxjOJPq0UX9qxoqlQAAAPqqrLumlez89UgtX12rR6trQ\n5RSc+INVY/a0WF4VCABAOu65aYSuHtpLDy1ar10HOYKhrfiDVSZ7QCh7rAAASEdFeZkenzZOZSbN\nem6l6jK85U2r+INVI5vXAQBI26BeXfXUjGptfv+YZr+0Tu6cbyWVQLBq4IBQAABy4o9Hf0yzb75M\nr63bq6ff3Ba6nIIQfbCqyzSpvMzUqdxClwIAQHTuvnG4bv3Di/XYL7Zo9S7Ot4o+WNVnWtSlU7nM\nCFYAAKTNzPTtqZerf1Wl/vqldadWikpV/MGqsZllQAAAcqh7ZSf945f+QDX7T+jJN7aGLieo+INV\npomN6wAA5NiNo/pq2vjB+pc3t5X0kmD8waqxmWAFAEAezPn8xzWgRxfd/dxK1R6pD11OECUQrFpY\nCgQAIA+qKjtp/syrVZdp1sz5y3S0rvTeTzD6YNWQYcYKAIB8Gd2/u565fbx2HqzTXzy7Qo3NLaFL\nyqvog1VdYxMzVgAA5NF1I/roO39yhZZtP6R/+9/tocvJq+iDVT0zVgAA5N3U6kG6ZWx/PbH0HW0/\n8NvQ5eRN9MGqgT1WAAAE8c3JY9W5okwPvrxOLS2l8ZY30QcrXhUIAEAY/aoq9Y1JH9db2w/pp8t2\nhS4nL6IPVnUZ9lgBABDKtPGDdf2lF+mRVzdq+Y5DocvJuaiDVUuLZ5cCmbECACCIsjLT979crYG9\numjWsyu162Bd6JJyKupgdbIp+xJPZqwAAAinZ9fOmj/zajW3uO740TIdb4j3fKuog1VdpkmSmLEC\nACCwYRd107zbrtL2A7/V3y1aH7qcnIk6WNUn77DNjBUAAOFdN6KP7pswSovW7NHC1btDl5MTUQer\nhtZgxYwVAAAF4Z6bRujqob300ML12nkwvvOtog5W9ZlkjxXBCgCAglBRXqYnplerrMz0VwvWRHe+\nVdTB6tQeK5YCAQAoGAN7dtHDt47Vql1H9NxbO0OXk6qogxV7rAAAKExfvHKgbhh5kR79+RbtOVIf\nupzURB2s2GMFAEBhMjP9w5Qr1NTSormL18s9jiXBqINVXYZgBQBAoRrSp6se+OwoLd20X6+sqg1d\nTiqiDlatS4FdWQoEAKAg3fnJYfrEsN56aNF61ew/HrqcjyzuYJXMWFUSrAAAKEgV5WV6aka1unYu\n118+v+rUc3exijpYsccKAIDC16+qUk9MH6et+09ozqK3i3q/VdTBqi7TrIoyU6fyqLsJAEDRu2Fk\nX90/YZReWVWrx3/5Tuhyfm8VoQvIpfrGZo5aAACgSNw74VLtPVqvp35Vo77dL9Dt1w0NXVKHRR2s\nGhqbWQYEAKBImJn+fsrlOnAio7lLNmhgry769GX9QpfVIVGvkdVlmLECAKCYVJSX6ftfrtZl/as0\n+6V1OnDiZOiSOiTqYFWfYcYKAIBiU9mpXE9MG6djDU168OXi2swed7BijxUAAEVpdP/umn3zaC3d\ntE8Llr8XupzzFnWwYo8VAADF685PDtMfjeijb766UVveL47DQ6MOVnWZZk5dBwCgSJWVmR6fNk4X\nVlZo1rMrdLS+MXRJ5xR1sKpvbFYlM1YAABStflWVevorV2r34Xo9sGCNWloKe79V1MGqgc3rAAAU\nvfFDe2vurWP0xub9euz1LaHLOauoz7Fi8zoAAHG4/dpLtGnvcf3wv7fp4p5ddNu1l4Qu6YyiDlac\nYwUAQBzMTN+aPFb7jjVo7uL16l9Vqc+MKbzDQ6NdCmxpcZ1samEpEACASLQeHnrFwB76+gurVXuk\nPnRJvyPaYNXQ1CxJBCsAACLStXOFfnjbVWpx16M/3xy6nN9xXsHKzG4xsy1mVmNmD57h+2ZmTyXf\nX2dmV6ZfasfUZbLBiuMWAACIy8CeXTTrU8O1eM0erdx5OHQ5H3LOYGVm5ZJ+IGmipDGSZpjZmNNu\nNlHSyOTjLklPp1xnh9UnwYrjFgAAiM+sG0eoX9UFeuS1jQV1BMP5zFhdI6nG3d9194ykFyVNPu02\nkyX9xLN+I6mnmQ1IudYOaWhMlgKZsQIAIDrdLqjQ7Jsv09r3jmjx2trQ5ZxyPsFqoKS2b9KzO2nr\n6G1kZneZ2QozW/HBBx90tNYOu/RjF6p31845/zkAACD/plYP1JVDemrXwcLZxJ7X4xbc/RlJz0jS\n+PHjczpvN7Jfdy194MZc/ggAABBQWZnpP2Zdp4rywnkt3vlUUitpcJuvByVtHb0NAABAqgopVEnn\nF6yWSxppZsPMrLOk6ZKWnHabJZL+LHl14LWSjrr73pRrBQAAKGjnXAp09yYz+5qkX0gqlzTf3TeY\n2d3J9+dJ+pmkSZJqJNVJuiN3JQMAABSm89pj5e4/UzY8tW2b1+Zzl3RPuqUBAAAUl8JamAQAAChi\nBCsAAICUEKwAAABSQrACAABICcEKAAAgJQQrAACAlBCsAAAAUkKwAgAASAnBCgAAICUEKwAAgJQQ\nrAAAAFJCsAIAAEgJwQoAACAlBCsAAICUEKwAAABSQrACAABICcEKAAAgJQQrAACAlBCsAAAAUmLu\nHuYHm30gaWeO/vmLJB3I0b9dDEq5/6Xcd4n+0//S7X8p912i//no/yXu3vdcNwoWrHLJzFa4+/jQ\ndYRSyv0v5b5L9J/+l27/S7nvEv0vpP6zFAgAAJASghUAAEBKYg1Wz4QuILBS7n8p912i//S/dJVy\n3yX6XzD9j3KPFQAAQAixzlgBAADkXVTBysxuMbMtZlZjZg+GrifXzGywmf3azDaa2QYzuy9pf9jM\nas1sTfIxKXStuWJmO8zs7aSfK5K23mb2SzPbmvzZK3SduWBmo9uM8RozO2Zm98c6/mY238z2m9n6\nNm3tjrWZ/W3yWLDFzG4OU3V62un/P5vZZjNbZ2YLzaxn0j7UzOrb3Afmhas8He30v937eomM/4I2\nfd9hZmuS9qjG/yzPdQV5/UezFGhm5ZLekfRZSbslLZc0w903Bi0sh8xsgKQB7r7KzLpLWilpiqQ/\nlXTC3R8LWmAemNkOSePd/UCbtkclHXL3f0oCdi93/5tQNeZDcv+vlfQJSXcowvE3s09JOiHpJ+5+\nedJ2xrE2szGSXpB0jaSLJS2VNMrdmwOV/5G10//PSfqVuzeZ2XckKen/UEmvtd4uBu30/2Gd4b5e\nKuN/2ve/K+mouz8S2/if5blupgrw+o9pxuoaSTXu/q67ZyS9KGly4Jpyyt33uvuq5PPjkjZJGhi2\nqoIwWdKPk89/rOwFGLsJkra5e64O3Q3O3f9H0qHTmtsb68mSXnT3k+6+XVKNso8RRetM/Xf31929\nKfnyN5IG5b2wPGln/NtTEuPfysxM2V+oX8hrUXlylue6grz+YwpWAyW91+br3SqhkJH8hlIt6a2k\n6evJ8sD8WJfCEi5pqZmtNLO7krZ+7r43+fx9Sf3ClJZX0/XhB9VSGf/2xroUHw/ulPRfbb4eliwD\nvWlmN4QqKg/OdF8vtfG/QdI+d9/api3K8T/tua4gr/+YglXJMrMLJb0s6X53PybpaUnDJY2TtFfS\ndwOWl2vXu/s4SRMl3ZNMl5/i2bXuONa722FmnSV9QdJ/Jk2lNP6nlMJYt8fM5khqkvR80rRX0pDk\n2nhA0k/NrCpUfTlUkvf1M5ihD/9iFeX4n+G57pRCuv5jCla1kga3+XpQ0hY1M+uk7B3teXd/RZLc\nfZ+7N7t7i6R/VZFPgZ+Nu9cmf+6XtFDZvu5L1uRb1+b3h6swLyZKWuXu+6TSGn+1P9Yl83hgZjMl\nfV7SV5InFyVLIAeTz1dK2iZpVLAic+Qs9/VSGv8KSV+UtKC1LcbxP9NznQr0+o8pWC2XNNLMhiW/\nwU+XtCRwTTmVrKv/u6RN7v69Nu0D2txsqqT1p//dGJhZt2Qjo8ysm6TPKdvXJZK+mtzsq5IWh6kw\nbz7022qpjH+ivbFeImm6mV1gZsMkjZS0LEB9OWVmt0iaLekL7l7Xpr1v8oIGmdlwZfv/bpgqc+cs\n9/WSGP/EZyRtdvfdrQ2xjX97z3Uq1Ovf3aP5kDRJ2VcGbpM0J3Q9eejv9cpOfa6TtCb5mCTpWUlv\nJ+1LlH01RfB6c9D/4ZLWJh8bWsdcUh9Jb0jaquyrQXqHrjWH/wfdJB2U1KNNW5Tjr2x43CupUdk9\nE39+trGWNCd5LNgiaWLo+nPU/xpl95K0Xv/zktt+Kbkm1khaJenW0PXnqP/t3tdLYfyT9h9Juvu0\n20Y1/md5rivI6z+a4xYAAABCi2kpEAAAICiCFQAAQEoIVgAAACkhWAEAAKSEYAUAAJASghUAAEBK\nCFYAAAApIVgBAACk5P8Axh41GRlCltoAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fbb6909b4d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "slope = (val_mad[-1]-val_mad[0])/(n_est[-1]-n_est[0])\n",
    "intercept = val_mad[0] - slope*n_est[0]\n",
    "val_diff = np.array(n_est)*slope+intercept-np.array(val_mad)\n",
    "plt.plot(n_est,val_diff)\n",
    "\n",
    "best_n = (np.array(n_est)*slope+intercept-np.array(val_mad)).argmax()\n",
    "plt.plot(n_est[best_n],val_diff[best_n], 'ro')\n",
    "print(\"The best value of n is {}\".format(n_est[best_n]))\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Using the best number of estimators from above, we'll fit the model we'll use for prediction on the test set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Mean Absolute Error: 1.98250334055\n",
      "R^2 value: 0.795319447765\n"
     ]
    }
   ],
   "source": [
    "gbtr = GradientBoostingRegressor(n_estimators=n_est[best_n])\n",
    "gbtr.fit(x_train,y_train)\n",
    "y_pred = gbtr.predict(x_test)\n",
    "print(\"Mean Absolute Error: {}\\nR^2 value: {}\".format(mean_absolute_error(y_pred,y_test),gbtr.score(x_test,y_test)))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Not bad! Remember that we are dealing with degrees Celsius. Our mean error in this case around 2 degrees. We also have a strong $R^2$ value.\n",
    "\n",
    "Now let's fit the models on the entirety of the data, and then produce a grid of predicted `TMAX` values to visualize on a map."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "GradientBoostingRegressor(alpha=0.9, criterion='friedman_mse', init=None,\n",
       "             learning_rate=0.1, loss='ls', max_depth=3, max_features=None,\n",
       "             max_leaf_nodes=None, min_impurity_decrease=0.0,\n",
       "             min_impurity_split=None, min_samples_leaf=1,\n",
       "             min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
       "             n_estimators=33, presort='auto', random_state=None,\n",
       "             subsample=1.0, verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gbtr.fit(x,y['21-Jun'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "And now the model for the December 21"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "GradientBoostingRegressor(alpha=0.9, criterion='friedman_mse', init=None,\n",
       "             learning_rate=0.1, loss='ls', max_depth=3, max_features=None,\n",
       "             max_leaf_nodes=None, min_impurity_decrease=0.0,\n",
       "             min_impurity_split=None, min_samples_leaf=1,\n",
       "             min_samples_split=2, min_weight_fraction_leaf=0.0,\n",
       "             n_estimators=33, presort='auto', random_state=None,\n",
       "             subsample=1.0, verbose=0, warm_start=False)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "gbtr_dec = GradientBoostingRegressor(n_estimators=n_est[best_n])\n",
    "gbtr_dec.fit(x,y['21-Dec'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='Data_Visualization'></a>\n",
    "## Data Visualization\n",
    "Let's pull a matrix of elevation data. The data will contain elevations at latitudes [-90,90], every degree and longitudes [-180,180), every 2 degrees. We'll provide the latitude, longitude, and elevation from this matrix to create a matrix of the same size that contains predicted temperature information."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "z = pd.DataFrame(np.empty([181,180]))\n",
    "elevations = pd.read_csv(\"https://raw.githubusercontent.com/IBMDataScience/DSX-DemoCenter/master/weatherGeographies/data_assets/elevation.csv\", index_col=0)\n",
    "\n",
    "# Make sure that the matrices are indexed by (lon,lat) values\n",
    "elevations.columns = elevations.columns.astype(int)\n",
    "z.columns = elevations.columns\n",
    "z.index = elevations.index"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### Plots\n",
    "We'll use `matplotlib` Basemap to plot our data. First, we should fill our empty `z` temperature matrix with the predicted temperatures. Then we must flip the matrix, as Basemap wants 90$^{\\circ}$S to be the first row in the matrix."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<div class=\"alert alert-block alert-info\"> If you have installed the Basemap and GEOS libraries, copy the code below in a python cell to dynamically produce an output map</div>\n",
    "\n",
    "```python\n",
    "for lon in range(-180,180,2):\n",
    "    for lat in reversed(range(-90,91,1)):\n",
    "        z[lon][lat] = gbtr.predict(pd.DataFrame({\n",
    "            0: [lat],\n",
    "            1: [lon],\n",
    "            2: [elevations[lon][lat]]\n",
    "        }))\n",
    "\n",
    "plt.rcParams['figure.figsize'] = (15,15)\n",
    "m = Basemap()\n",
    "lon, lat = np.meshgrid(list(range(-180,180,2)),list(range(-90,91,1)))\n",
    "x1,y1 = m(lon,lat)\n",
    "m.drawcoastlines()\n",
    "m.drawstates()\n",
    "m.drawcountries()\n",
    "m.drawmapboundary()\n",
    "z1 = maskoceans(x1,y1,np.flip(np.array(z),axis=0))\n",
    "cs = m.contourf(x1,y1,z1, 15)\n",
    "plt.show()\n",
    "```\n",
    "![](https://github.com/IBMDataScience/DSX-DemoCenter/raw/master/weatherGeographies/notebooks/static/jun21.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "This is a fairly predictable distribution of temperatures for the 21st of June. Africa and the southern United States are scortching hot, while Antarctica is frigidly cold. Elevation data has also proved to be important, as the himalayas and tibet are marked to be colder than other regions at the same latitude, just as expected. It would be interesting to see how our map would differ if we used 21st of December data:\n",
    "\n",
    "```python\n",
    "for lon in range(-180,180,2):\n",
    "    for lat in reversed(range(-90,91,1)):\n",
    "        z[lon][lat] = gbtr_dec.predict(pd.DataFrame({\n",
    "            0: [lat],\n",
    "            1: [lon],\n",
    "            2: [elevations[lon][lat]]\n",
    "        }))\n",
    "\n",
    "m = Basemap()\n",
    "lon, lat = np.meshgrid(list(range(-180,180,2)),list(range(-90,91,1)))\n",
    "x1,y1 = m(lon,lat)\n",
    "m.drawcoastlines()\n",
    "m.drawstates()\n",
    "m.drawcountries()\n",
    "m.drawmapboundary()\n",
    "z1 = maskoceans(x1,y1,np.flip(np.array(z),axis=0))\n",
    "cs = m.contourf(x1,y1,z1, 15)\n",
    "plt.show()\n",
    "```\n",
    "![](https://github.com/IBMDataScience/DSX-DemoCenter/raw/master/weatherGeographies/notebooks/static/dec21.png)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "The model has done an excellent job in estimating the temperatures on 21-Dec. As expected, the southern hemisphere is in summer, and thus hot, while North America and Europe are in Winter. \n",
    "\n",
    "<a id='Save_Model_to_DSX_Filesystem'></a>\n",
    "## Save Model to DSX Filesystem\n",
    "We can now save `scikit-learn` models to the DSX filesystem for publishing, scoring, deployment, and evaluations. First, import the `save` function from the `dsx_ml.ml` library. The save function takes a few arguments which are listed below."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "Using TensorFlow backend.\n",
      "/opt/conda/lib/python2.7/site-packages/sklearn/cross_validation.py:41: 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"
     ]
    }
   ],
   "source": [
    "from dsx_ml.ml import save"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "Now we can save both the June 21 and December 21 models."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'path': '/user-home/999/DSX_Projects/dsx-samples/models/June21/1',\n",
       " 'scoring_endpoint': 'https://ibm-nginx-svc.ibm-private-cloud.svc.cluster.local/v3/project/score/Python27/scikit-learn-0.19/dsx-samples/June21/1'}"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save(model = gbtr,\n",
    "     name = 'June21',\n",
    "     x_test = x,\n",
    "     y_test = pd.DataFrame(y['21-Jun']),\n",
    "     algorithm_type = 'Regression')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'path': '/user-home/999/DSX_Projects/dsx-samples/models/Dec21/1',\n",
       " 'scoring_endpoint': 'https://ibm-nginx-svc.ibm-private-cloud.svc.cluster.local/v3/project/score/Python27/scikit-learn-0.19/dsx-samples/Dec21/1'}"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "save(model = gbtr_dec,\n",
    "     name = 'Dec21',\n",
    "     x_test = x,\n",
    "     y_test = pd.DataFrame(y['21-Dec']),\n",
    "     algorithm_type = 'Regression')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "### Model Metadata\n",
    "The model will be stored in the models directory in your DSX Project. Each model is stored as a directory, in which the model artifact and metadata are stored. The metadata is stored as a JSON file, which we can open and display."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Runtime: Python27\n",
      "Model Type: scikit-learn-0.19\n",
      "Algorithm: GradientBoostingRegressor\n",
      "Feature(s):\n",
      "    elevation\n",
      "    latitude\n",
      "    longitude\n",
      "Latest Model Version: 1\n",
      "Label(s):\n",
      "    21-Dec\n"
     ]
    }
   ],
   "source": [
    "import json\n",
    "import os\n",
    "\n",
    "proj_dir = os.environ['DSX_PROJECT_DIR']\n",
    "\n",
    "with open('{}/models/Dec21/metadata.json'.format(proj_dir),'r') as infile:\n",
    "    metadata_dict = json.load(infile)\n",
    "\n",
    "print(\"Runtime: {}\".format(metadata_dict['runtime']))\n",
    "print(\"Model Type: {}\".format(metadata_dict['type']))\n",
    "print(\"Algorithm: {}\".format(metadata_dict['algorithm']))\n",
    "\n",
    "print(\"Feature(s):\")\n",
    "for feature in metadata_dict['features']:\n",
    "    print('    '+feature['name'])\n",
    "\n",
    "print(\"Latest Model Version: {}\".format(metadata_dict['latestModelVersion']))\n",
    "print(\"Label(s):\")\n",
    "for label in metadata_dict['labelColumns']:\n",
    "    print('    '+label['name'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='Predict_on_New_Data'></a>\n",
    "## Predict on New Data\n",
    "\n",
    "Let's make some predictions using new data. Below we have gathered the latitude, longitude, and elevation data for the cities of Chicago, IL and Miami, FL."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "chicago_data = {\n",
    "    \"elevation\" : 200.6,\n",
    "    \"latitude\" : 41.995,\n",
    "    \"longitude\" : -87.9336\n",
    "}\n",
    "\n",
    "miami_data = {\n",
    "    \"elevation\" : 1,\n",
    "    \"latitude\" : 25.7616798,\n",
    "    \"longitude\" : -80.1917902\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "We can call the predict function of our models and print them below:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {
    "collapsed": false,
    "deletable": true,
    "editable": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "On June 21, it is predicted to be 26.9° C in Chicago, and 31.5° C in Miami\n",
      "On December 21, it is predicted to be 1.5° C in Chicago, and 22.4° C in Miami\n"
     ]
    }
   ],
   "source": [
    "new_data = pd.DataFrame([chicago_data, miami_data])\n",
    "\n",
    "jun21_temps = gbtr.predict(new_data)\n",
    "dec21_temps = gbtr_dec.predict(new_data)\n",
    "\n",
    "\n",
    "output_str = (u'On June 21, it is predicted to be ' +\n",
    "    str(jun21_temps[0].round(1)) + \n",
    "    u'\\N{DEGREE SIGN} C in Chicago, and '+ \n",
    "    str(jun21_temps[1].round(1)) + \n",
    "    u'\\N{DEGREE SIGN} C in Miami\\n' + \n",
    "    u'On December 21, it is predicted to be ' + \n",
    "    str(dec21_temps[0].round(1)) + \n",
    "    u'\\N{DEGREE SIGN} C in Chicago, and ' + \n",
    "    str(dec21_temps[1].round(1)) +\n",
    "    u'\\N{DEGREE SIGN} C in Miami')\n",
    "\n",
    "\n",
    "print(output_str)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<a id='Summary'></a>\n",
    "## Summary\n",
    "In this notebook you learned how to create a `scikit-learn` model, create some data visualizations, and save the model in the DSX local environment."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<div class=\"alert alert-block alert-info\">Note: To save resources and get the best performance please use the code below to stop the kernel before exiting your notebook.</div>"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {
    "collapsed": true,
    "deletable": true,
    "editable": true
   },
   "outputs": [],
   "source": [
    "%%javascript\n",
    "Jupyter.notebook.session.delete();"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {
    "deletable": true,
    "editable": true
   },
   "source": [
    "<hr>\n",
    "Copyright &copy; IBM Corp. 2017. Released as licensed Sample Materials."
   ]
  }
 ],
 "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": 1
}