notebooks/01-eda-train-test-data.ipynb

{
 "cells": [
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Versions:\n",
      "  Python: 3.8.2 (default, Jul 16 2020, 14:00:26) \n",
      "[GCC 9.3.0]\n",
      "  pandas: 1.1.1\n",
      "  numpy: 1.19.1\n",
      "  seaborn: 0.10.1\n",
      "  sklearn: 0.23.2\n"
     ]
    }
   ],
   "source": [
    "%load_ext autoreload\n",
    "%autoreload 2\n",
    "\n",
    "import os\n",
    "import sys\n",
    "\n",
    "\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import seaborn as sns\n",
    "import matplotlib.pyplot as plt\n",
    "import sklearn\n",
    "import scipy\n",
    "\n",
    "%run constants.py\n",
    "\n",
    "%matplotlib inline\n",
    "print(\"Versions:\")\n",
    "print(\"  Python: %s\" % sys.version)\n",
    "for module in [pd, np, sns, sklearn]:\n",
    "    print(\"  %s: %s\" %(module.__name__, module.__version__))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Problem description\n",
    "We are provided with daily historical sales data (in a day by day format) and the task is to forecast the total amount of products sold in every shop for the test set, which is dated to the next month.\n",
    "\n",
    "Evaluation will be done with RMSE. One thing caught my attention: True target values are clipped into `[0,20]`."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# But first, some conventions\n",
    "\n",
    "- First it's worth noting that all data is under `../.data`. \n",
    "- We will use a Makefile to describe the execution steps and dependency graph of the whole analysis. The goal is to be able to run `make final-submission` to generate the final submission from scratch.\n",
    "- Notebooks are only used for exploration and documentation. The final result is coded separately (inside `../src`) from the experience gathered through notebooks."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "import zipfile\n",
    "with zipfile.ZipFile(os.path.join(RAW_DATA_DIR, 'competitive-data-science-predict-future-sales.zip'), 'r') as datasets_file:\n",
    "    item_categories = pd.read_csv(datasets_file.open('item_categories.csv'))\n",
    "    items = pd.read_csv(datasets_file.open('items.csv'))\n",
    "    sales_train = pd.read_csv(datasets_file.open('sales_train.csv'))\n",
    "    sample_submission = pd.read_csv(datasets_file.open('sample_submission.csv'))\n",
    "    shops = pd.read_csv(datasets_file.open('shops.csv'))\n",
    "    test_set = pd.read_csv(datasets_file.open('test.csv'))"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Training/test sets"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "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>date_block_num</th>\n",
       "      <th>shop_id</th>\n",
       "      <th>item_id</th>\n",
       "      <th>item_price</th>\n",
       "      <th>item_cnt_day</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>2.935849e+06</td>\n",
       "      <td>2.935849e+06</td>\n",
       "      <td>2.935849e+06</td>\n",
       "      <td>2.935849e+06</td>\n",
       "      <td>2.935849e+06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>1.456991e+01</td>\n",
       "      <td>3.300173e+01</td>\n",
       "      <td>1.019723e+04</td>\n",
       "      <td>8.908532e+02</td>\n",
       "      <td>1.242641e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>9.422988e+00</td>\n",
       "      <td>1.622697e+01</td>\n",
       "      <td>6.324297e+03</td>\n",
       "      <td>1.729800e+03</td>\n",
       "      <td>2.618834e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>-1.000000e+00</td>\n",
       "      <td>-2.200000e+01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>7.000000e+00</td>\n",
       "      <td>2.200000e+01</td>\n",
       "      <td>4.476000e+03</td>\n",
       "      <td>2.490000e+02</td>\n",
       "      <td>1.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>1.400000e+01</td>\n",
       "      <td>3.100000e+01</td>\n",
       "      <td>9.343000e+03</td>\n",
       "      <td>3.990000e+02</td>\n",
       "      <td>1.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>2.300000e+01</td>\n",
       "      <td>4.700000e+01</td>\n",
       "      <td>1.568400e+04</td>\n",
       "      <td>9.990000e+02</td>\n",
       "      <td>1.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>3.300000e+01</td>\n",
       "      <td>5.900000e+01</td>\n",
       "      <td>2.216900e+04</td>\n",
       "      <td>3.079800e+05</td>\n",
       "      <td>2.169000e+03</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       date_block_num       shop_id       item_id    item_price  item_cnt_day\n",
       "count    2.935849e+06  2.935849e+06  2.935849e+06  2.935849e+06  2.935849e+06\n",
       "mean     1.456991e+01  3.300173e+01  1.019723e+04  8.908532e+02  1.242641e+00\n",
       "std      9.422988e+00  1.622697e+01  6.324297e+03  1.729800e+03  2.618834e+00\n",
       "min      0.000000e+00  0.000000e+00  0.000000e+00 -1.000000e+00 -2.200000e+01\n",
       "25%      7.000000e+00  2.200000e+01  4.476000e+03  2.490000e+02  1.000000e+00\n",
       "50%      1.400000e+01  3.100000e+01  9.343000e+03  3.990000e+02  1.000000e+00\n",
       "75%      2.300000e+01  4.700000e+01  1.568400e+04  9.990000e+02  1.000000e+00\n",
       "max      3.300000e+01  5.900000e+01  2.216900e+04  3.079800e+05  2.169000e+03"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sales_train.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "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>date</th>\n",
       "      <th>date_block_num</th>\n",
       "      <th>shop_id</th>\n",
       "      <th>item_id</th>\n",
       "      <th>item_price</th>\n",
       "      <th>item_cnt_day</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>02.01.2013</td>\n",
       "      <td>0</td>\n",
       "      <td>59</td>\n",
       "      <td>22154</td>\n",
       "      <td>999.00</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>03.01.2013</td>\n",
       "      <td>0</td>\n",
       "      <td>25</td>\n",
       "      <td>2552</td>\n",
       "      <td>899.00</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>05.01.2013</td>\n",
       "      <td>0</td>\n",
       "      <td>25</td>\n",
       "      <td>2552</td>\n",
       "      <td>899.00</td>\n",
       "      <td>-1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>06.01.2013</td>\n",
       "      <td>0</td>\n",
       "      <td>25</td>\n",
       "      <td>2554</td>\n",
       "      <td>1709.05</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>15.01.2013</td>\n",
       "      <td>0</td>\n",
       "      <td>25</td>\n",
       "      <td>2555</td>\n",
       "      <td>1099.00</td>\n",
       "      <td>1.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         date  date_block_num  shop_id  item_id  item_price  item_cnt_day\n",
       "0  02.01.2013               0       59    22154      999.00           1.0\n",
       "1  03.01.2013               0       25     2552      899.00           1.0\n",
       "2  05.01.2013               0       25     2552      899.00          -1.0\n",
       "3  06.01.2013               0       25     2554     1709.05           1.0\n",
       "4  15.01.2013               0       25     2555     1099.00           1.0"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sales_train.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "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>shop_id</th>\n",
       "      <th>item_id</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>107099.500000</td>\n",
       "      <td>31.642857</td>\n",
       "      <td>11019.398627</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>61834.358168</td>\n",
       "      <td>17.561933</td>\n",
       "      <td>6252.644590</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>30.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>53549.750000</td>\n",
       "      <td>16.000000</td>\n",
       "      <td>5381.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>107099.500000</td>\n",
       "      <td>34.500000</td>\n",
       "      <td>11203.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>160649.250000</td>\n",
       "      <td>47.000000</td>\n",
       "      <td>16071.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>214199.000000</td>\n",
       "      <td>59.000000</td>\n",
       "      <td>22167.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "                  ID        shop_id        item_id\n",
       "count  214200.000000  214200.000000  214200.000000\n",
       "mean   107099.500000      31.642857   11019.398627\n",
       "std     61834.358168      17.561933    6252.644590\n",
       "min         0.000000       2.000000      30.000000\n",
       "25%     53549.750000      16.000000    5381.500000\n",
       "50%    107099.500000      34.500000   11203.000000\n",
       "75%    160649.250000      47.000000   16071.500000\n",
       "max    214199.000000      59.000000   22167.000000"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_set.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "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>shop_id</th>\n",
       "      <th>item_id</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>5</td>\n",
       "      <td>5037</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>5</td>\n",
       "      <td>5320</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>5</td>\n",
       "      <td>5233</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3</td>\n",
       "      <td>5</td>\n",
       "      <td>5232</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4</td>\n",
       "      <td>5</td>\n",
       "      <td>5268</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   ID  shop_id  item_id\n",
       "0   0        5     5037\n",
       "1   1        5     5320\n",
       "2   2        5     5233\n",
       "3   3        5     5232\n",
       "4   4        5     5268"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_set.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "About the clipping on the true targets range: let's see if this also happens on the train set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "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>date_block_num</th>\n",
       "      <th>shop_id</th>\n",
       "      <th>item_id</th>\n",
       "      <th>item_cnt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>1.609124e+06</td>\n",
       "      <td>1.609124e+06</td>\n",
       "      <td>1.609124e+06</td>\n",
       "      <td>1.609124e+06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>1.466479e+01</td>\n",
       "      <td>3.280585e+01</td>\n",
       "      <td>1.068099e+04</td>\n",
       "      <td>2.267200e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>9.542322e+00</td>\n",
       "      <td>1.653701e+01</td>\n",
       "      <td>6.238883e+03</td>\n",
       "      <td>8.649882e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>-2.200000e+01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>6.000000e+00</td>\n",
       "      <td>2.100000e+01</td>\n",
       "      <td>5.045000e+03</td>\n",
       "      <td>1.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>1.400000e+01</td>\n",
       "      <td>3.100000e+01</td>\n",
       "      <td>1.049700e+04</td>\n",
       "      <td>1.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>2.300000e+01</td>\n",
       "      <td>4.700000e+01</td>\n",
       "      <td>1.606000e+04</td>\n",
       "      <td>2.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>3.300000e+01</td>\n",
       "      <td>5.900000e+01</td>\n",
       "      <td>2.216900e+04</td>\n",
       "      <td>2.253000e+03</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       date_block_num       shop_id       item_id      item_cnt\n",
       "count    1.609124e+06  1.609124e+06  1.609124e+06  1.609124e+06\n",
       "mean     1.466479e+01  3.280585e+01  1.068099e+04  2.267200e+00\n",
       "std      9.542322e+00  1.653701e+01  6.238883e+03  8.649882e+00\n",
       "min      0.000000e+00  0.000000e+00  0.000000e+00 -2.200000e+01\n",
       "25%      6.000000e+00  2.100000e+01  5.045000e+03  1.000000e+00\n",
       "50%      1.400000e+01  3.100000e+01  1.049700e+04  1.000000e+00\n",
       "75%      2.300000e+01  4.700000e+01  1.606000e+04  2.000000e+00\n",
       "max      3.300000e+01  5.900000e+01  2.216900e+04  2.253000e+03"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# notice `date_block_num` is actually related to the month\n",
    "sales_train_by_month = sales_train.groupby(by=['date_block_num', 'shop_id', 'item_id'])['item_cnt_day'].sum().reset_index()\n",
    "sales_train_by_month.rename(columns={'item_cnt_day': 'item_cnt'}, inplace=True)\n",
    "sales_train_by_month.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:>"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.distplot(sales_train_by_month['item_cnt'].values, hist_kws={'log':True}, kde=False, norm_hist=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Results near zero are orders of magnitude larger than the rest, but let's keep in mind this is before the clipping. Let's clip it and see how this looks like"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "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>date_block_num</th>\n",
       "      <th>shop_id</th>\n",
       "      <th>item_id</th>\n",
       "      <th>item_cnt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>1.609124e+06</td>\n",
       "      <td>1.609124e+06</td>\n",
       "      <td>1.609124e+06</td>\n",
       "      <td>1.609124e+06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>1.466479e+01</td>\n",
       "      <td>3.280585e+01</td>\n",
       "      <td>1.068099e+04</td>\n",
       "      <td>2.022806e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>9.542322e+00</td>\n",
       "      <td>1.653701e+01</td>\n",
       "      <td>6.238883e+03</td>\n",
       "      <td>2.577964e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>6.000000e+00</td>\n",
       "      <td>2.100000e+01</td>\n",
       "      <td>5.045000e+03</td>\n",
       "      <td>1.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>1.400000e+01</td>\n",
       "      <td>3.100000e+01</td>\n",
       "      <td>1.049700e+04</td>\n",
       "      <td>1.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>2.300000e+01</td>\n",
       "      <td>4.700000e+01</td>\n",
       "      <td>1.606000e+04</td>\n",
       "      <td>2.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>3.300000e+01</td>\n",
       "      <td>5.900000e+01</td>\n",
       "      <td>2.216900e+04</td>\n",
       "      <td>2.000000e+01</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       date_block_num       shop_id       item_id      item_cnt\n",
       "count    1.609124e+06  1.609124e+06  1.609124e+06  1.609124e+06\n",
       "mean     1.466479e+01  3.280585e+01  1.068099e+04  2.022806e+00\n",
       "std      9.542322e+00  1.653701e+01  6.238883e+03  2.577964e+00\n",
       "min      0.000000e+00  0.000000e+00  0.000000e+00  0.000000e+00\n",
       "25%      6.000000e+00  2.100000e+01  5.045000e+03  1.000000e+00\n",
       "50%      1.400000e+01  3.100000e+01  1.049700e+04  1.000000e+00\n",
       "75%      2.300000e+01  4.700000e+01  1.606000e+04  2.000000e+00\n",
       "max      3.300000e+01  5.900000e+01  2.216900e+04  2.000000e+01"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "sales_train_by_month_clipped = sales_train_by_month.copy()\n",
    "sales_train_by_month_clipped['item_cnt'] = np.clip(sales_train_by_month_clipped['item_cnt'], 0, 20)\n",
    "sales_train_by_month_clipped.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:>"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sns.distplot(sales_train_by_month_clipped['item_cnt'].values, hist_kws={'log':True}, kde=False, norm_hist=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Still not pretty, decaying orders of magnitude until it reaches 20. Good to keep that in mind (maybe we're better off predicting only the extremities well?). Either way, this logic will be available as a script."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Train/Test consistency\n",
    "\n",
    "We need to generate the train set from a list of historical events. The problem with that is we may end up with a train set that doesn't come from the same probability distribution as the test set samples.\n",
    "\n",
    "To test this, let's look at the distribution of item id + shop id pairs on each data set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:xlabel='shop_id'>"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAmgAAAJYCAYAAADWlRrkAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjMuMSwgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/d3fzzAAAACXBIWXMAAAsTAAALEwEAmpwYAAC+DElEQVR4nOz9d3zkd3Xo/7/O9Bn1vmq72m7vuuzaa6+NbQw2xVSTBBIDARJIHBJIublJLvnlXkK4fO9NJ42bxHQMpmNwgoMx2Ma4rbfb3t606r2ORtPfvz9mRqvVqsxI0+c8Hw89Vpr5zGfe0kqfOXPe7/c5YoxBKaWUUkrlD0uuB6CUUkoppS6nAZpSSimlVJ7RAE0ppZRSKs9ogKaUUkoplWc0QFNKKaWUyjMaoCmllFJK5Zm8DNBE5AsiMiQir6TpfBERORL/eCQd51RKqaXoNUwptVaSj3XQROTVgBf4ijHmmjScz2uMKV/7yJRSamV6DVNKrVVeZtCMMU8DY/NvE5HNIvIjETkoIj8XkatyNDyllFqWXsOUUmuVlwHaEh4AftcYcyPwR8D/S+GxLhE5ICIviMg7MjI6pZRanl7DlFJJs+V6AMkQkXLgVcC3RSRxszN+3y8Cn1zkYb3GmDfGP99gjOkVkU3AEyLysjHmXKbHrZRSoNcwpVTqCiJAI5bpmzDG7Fp4hzHme8D3lnuwMaY3/u95EXkK2A3oxU0plS16DVNKpaQgpjiNMVPABRF5F4DEXJ/MY0WkRkQS71TrgduA4xkbrFJKLaDXMKVUqvIyQBORrwPPA9tFpEdEPgS8F/iQiBwFjgH3Jnm6q4ED8cc9CfylMUYvbkqpjNFrmFJqrfKyzIZSSimlVCnLywyaUkoppVQpy7tNAvX19aajoyPXw1BKZdHBgwdHjDENuR5HOug1TKnSkqnrV8YDNBHZDnxz3k2bgI8bY/5hseM7Ojo4cOBApoellMojInIx12NIF72GKVVaMnX9yniAZow5BewCEBEr0As8nOnnVUoppZQqVNleg3Y3cM4YUzTvlpVSSiml0i3bAdp9wNcX3igi98fbmBwYHh7O8pCUUkoppfJL1gI0EXEAbwe+vfA+Y8wDxpg9xpg9DQ1FsU5YKaWUUmrVsplBexNwyBgzmMXnVEoppZQqONkM0N7NItObSimllFLqclkJ0ESkDHg9KzQEVkoppZRSWSpUa4yZAeqy8VxKKaWUUoVOWz0ppZRSSuUZDdCUUkoppfJM3vXiLGQP7eu67Ov37F2fo5EopZRSy9PXrPymGTSllFJKqTyjAZpSSimlVJ7RAE0ppZRSKs9ogKaUUkoplWc0QFNKKaWUyjMaoCmllFJK5RkN0JRSSiml8owGaEoppZRSeUYDNKWUUkqpPKOdBJRao4XVuEErciullFobzaAppZRSSuUZDdCUUkoppfKMBmhKKbUMEflvInJMRF4Rka+LiCvXY1JKFT9dg5ZBujZJqcImIq3A7wE7jDGzIvIt4D7gSzkdmFKq6GkGTSmllmcD3CJiAzxAX47Ho5QqARqgZVjUGLrHfPhDkVwPRSmVImNML/C3QBfQD0waY3688DgRuV9EDojIgeHh4WwPUylVhDRAy5CoMTx3boR/+Mlp/vVn53jkqL7pVqrQiEgNcC+wEWgBykTkVxceZ4x5wBizxxizp6GhIdvDVEoVIQ3QMuRw1wT/+VI/bruV7U0VvNQzwdhMMNfDUkql5nXABWPMsDEmBHwPeFWOx6SUKgG6SSBDjnSPU1fm4MN3bmbaH+ZvfnyKp08P89G7tmT8uRduTtCNCUqtWhdwi4h4gFngbuBAboeklCoFmkHLgKnZEOeHZ7i+vRoRodJt58b1NRzsGmdwyp/r4SmlkmSM2Qd8BzgEvEzsmvlATgellCoJGqBlwEu9kxjg+rbqudteva0BYwyfffp8zsallEqdMebPjTFXGWOuMca8zxgTyPWYlFLFT6c4M+Bo9wSt1W4aKpxzt9WWObimtYqv7euio74Miwig049KKaUWp8tVSptm0NJsZDpA78Qs17dVXXHfjuZKZkMResZ8ORiZUkoppQpFVjJoIlINfA64BjDAB40xz2fjubPtaM8EAlw3b3ozYUtjOQKcHvKyvq4s20NTSimVxxbrPqNKV7YyaP8I/MgYcxVwPXAiS8+bdRdGZmipdlPptl9xn8dho73Ww+nB6RyMTCmllFKFIuMZNBGpAl4N/BqAMSYIFGVBMGMMfZOzXNdaveQx25rK+emJIbyBMOVOm64xUEoppdQVspFB2wgMA18UkcMi8jkRuWx+r1japIz7QvhDUZqrXUses62pAgOcHfJmb2BKKaWUKijZCNBswA3AvxpjdgMzwMfmH1AsbVL6JmYBaK12L3lMS7Ubj8Oq05xKKaWUWlI2ArQeoCde8BFiRR9vyMLzZl3f5CwWgabKpTNoFhG2NVVwZnCaqDFZHJ1SSimlCkXGAzRjzADQLSLb4zfdDRzP9PPmQv+En4YKJ3br8j/WbU0VzAQjnNNpTqWUUkotIlu7OH8X+JqIvATsAv5Plp43q/omZ2mpWnp6M+GalkqqPXYeOzagWTSllFJKXSErAZox5kh8jdl1xph3GGPGs/G82TQ07WfaH6ZlmfVnCTarhddf3UTfpJ+XeiazMDqllFJKFRJt9ZQmx/qmAJIK0ACub6/mmbMjPH58gB3NlfhDEVx2a1rHFAhHECSt51RKKVUYtIxTYdMALU2O9cYyYc1VS28QmM8iwj071/HF5zr5xH8cA8DjsLK1qZzbttSvaSxHuyc41DXO+ZEZ2mvc/NptHWs6n1JKKaWyS3txpsmxvilqyxwpZcG2NlXwlmubee32Bt52fQvlThvv+/w+Hnj63KrHse/8KN880M3oTJDmKhcXR314A+FVn08ppZRS2acZtDQ51jeV9PTmfPOzZTesr+bFC2P8n0dPck1LFa9KMpOWSGMbY/jXn52j0mXj9+7aysWxGb74bCeHu8a5Y2vh1pdTSimlSo1m0NJg0heia8xHS5LTm0tx2qx8+ld2UeOx86XnOlN+/Ct9U/SMz/L6HU04bBbW13gQYH9n0e3JUEoppYqaBmhpcKwvtv5suQ4CyXLZrdx383p+cmKQnnFf0o8LR6M8dmyApkonu9fXAOC0W2mucnHw4tiax6WUUkqp7NEALQ1e7k1fgAbwq7dsAOBrC3bgLMUXDPPQvi7GZoLcs3MdFrm0c3NDXRmHuyYIRaJpGZtSSimlMk8DtDR4uXeS1mo3Hmd6lvS1Vrt5/Y4mvvFiF/5QZNljX+qZ4F+eOMuZQS9vu66Z7esqL7t/Q50HXzDCif6ptIxNKaWUUpmnAVoaHOub4prWypUPTMEHbu1g3BfikSN9Sx7zwvlR7nvgBRD4rTs3cevmKzcVbKgrA3QdmlJKKVVINEBboyl/iAsjM1zbWpXW8966uY6r1lXwwM/PE41e2Q7qmTMj/NoXX6Sl2s2H79xMW41n0fNUue201bh1HZpSSilVQDRAW6NjvbGpw2vSHKCJCL/9ms2cHfLy4+ODl913ftjLb3xlPx11ZXzj/luodNmXPddNHbXs7xzHaN/PVXloX9dlH0oppVSmaYC2RokdnOkO0ADecm0z62s9/OtTZ+eCq0jU8EffPorTZuXLH7yZ+nLniue5YUMNw9MBesZn0z5GpZRSSqWfFqpdo5d7J2muciUVKKXKZrXwW3du4s8efoVnz45y+9Z6Pv/MeQ51TfCP9+2iqTK5umtbG8sBuDAyQ3vt4lOh2rNNKaWUyh8aoK3Ry72TGcmeJfzSDW3840/O8NGvH2JjfRnH+qZ4484m3n59S9Ln6IhvFLg4OgNoR4FMmA1GsFkFu1WT0koppdZOA7Q18AbCXBiZ4R27WjP2HN871Mvbd7Xw4oUxpv1hNteXccP6Gr7+YnfS52iscOKyW+gcTb7wrUpe1Bg+89RZWqvdvPtmzTwqpZRaOw3Q1uBo9wTGkPYSGwttqi9nU335qh9vsQgddWXxDJpKt86RGcZmgkz4gkz5Qytu2lBKKaVWovMxa/C5n5+nxmNn78a6XA9lRRvqPJpBy5Aj3RPYLELUwKGLWm9OKaXU2mkGLQXzF9J3j/l48tQwf/zG7ZSlqYNAJnXUlfHkyWEiUYPVIis/QCXFH4rwcu8k17VVMeELsb9zjFdv03V+Siml1kYzaCkIRaJE4+Uunjg5RLXHzgde1ZHbQSVpQ10ZwUiUgSl/rodSVJ44OUQgHGVXew03baxl3Bfi7JA318NSSilV4PI/9ZMnDnSO8Zf/dRKHzcL2dRWcGpzmj9+4nfICyJ4BdNTFymtcHJlJW1P3YrGWEiMPH+6lwmVjU0MZ0ajB47Cyv1O7NiillFqbwogucuzx44N89KFDlDtt1Jc72X9hDI/Dyvtv3ZD258pUpfoN9bFSG52jPl61JSNPUXLGZ4I8dWqIvRvrsIhgsQrXt1fz4vkxnUpWSim1JhqgreBE/xS/9eABrm2t4i3XtVDutDHlDxGNGv7jaH+uh5e05koXDptFd3Km0VdfuEgoYrhxQ83cbesqXUSMoW9idsmiwEoppdRKdA3aCp4/N0rUwL+/b8/cdGaly061x5HjkaXGYhHW13ro1AAtZYv1MJ0NRvjic53cdVXjZR0dastivxfdY7pjViml1OppBm0FJ/qnqCtz0FSZ/lZO2ZKYNrVZhKPdkzkeTWHpHJnhf//wOABlDhvXt1dz303tfPtgN2MzQT585+bLNgUkArSuMR+vysmIlVJKFQPNoK3gxMAUVzdXIlL464nqyhyMzgQWzQipxX3puU5CYcOu9hpqyxw8cXKI//7tozzw9HluWF/NTR01lx1f5bZjFeGiZtCUUkqtQVYyaCLSCUwDESBsjNmTjeddq3AkyulBLx/IwGaAXKgrdxKKGIamA0k3Wi9l3kCY7xzs4dq2Kt5+fQvGGJ46PczDh3sB+PO37bwicLeIUO2x06UBmlJKqTXI5hTna40xI1l8vjU7PzJDMBzl6ubMtnLKlrry2PRb58iMBmhJePhQD95AmFs3xTpFiAiv3d7IG3eu4+DFce6+qnHRx9WWOXQNmlJKqTXRKc5lfO7nFwDoHPFlrPxFNtWVxdbR6UaBlRlj+PLzF7murYq2msvrxr3zxjb+7y9ei2WJMhq1ZQ4ualstpZRSa5CtAM0APxaRgyJyf5aec80GJmexWoSGisLdIDBftceOzSKcG9YAbSXPnh3l7JCXD9zakfL6w9oyB5OzISZ9oQyNTilVzM4MTfPA0+fwBsK5HorKoWwFaLcbY24A3gR8RERePf9OEblfRA6IyIHh4eEsDWll/ZN+GiucRVNw1CKxYPPUwHSuh5LXIlHDX/3oJI0VTt5yXXPKj58rtTGuWTSlVOoOd03QOerjOwe759oLqtKTlTVoxpje+L9DIvIwcDPw9Lz7HwAeANizZ0/e/DYOTPrZ2lSe62Gk1bpKF6cHNUBbzoPPd/Jy7yT//O7duOzWlB8/v9TGNa1VqxrDYlPqqbSgUkoVJmMM54a9lDttnB708syZEV69rSHXw1I5kPEATUTKAIsxZjr++RuAT2b6eddqxBtgOhBmXVVx9a1sqnRxuHuCSV+IKo8918NZs7X00Uw4N+zl0MVx1td52NxQzgNPn+eOrfW8dRXZM4DaeBFjXYemlErVsDfAtD/ML+xq5czQND8+PsCWxnJatIdyyclGBq0JeDi+jscGPGSM+VEWnndNTvRPAdBcVVy7HRMFd08PTXNTR21Ox5KO4Goteidm+ZPvHOXZs6PYrcLh7gkAHDYLn3rHNauufee0W6krc2ipDaVUys7H1whvbixnZ2slr/RNcXJgWgO0EpTxAM0Ycx64PtPPk25zAVqRlaNIlNc4NZD7AC3X/t+TZznQOc5brm3m5o219E/Msv/iOL9yUzsb6srWdO72Wk9BltrIddCsVKk7N+yl2mOnxmNHRKgtczAw5c/1sFQOaKunJZwd8lLhtOFxFtePqMptp8Jl040CwLNnR7hjawO3bakHYH1dGevryvjlPe1rPvf6Wg+Hu8fXfB6lVOmIGsP54Rl2zOtes67SxcCkBmilSOugLWFoOkClu/DXaC0kImxvquDUIhsFzgxN8/3DvXz6J6f5n99/OQejy57uMR+doz5u31KXkfNvqPPQN+EnFIlm5Pwqe0SkWkS+IyInReSEiNya6zGp4jQw6Wc2FGFz46UM/roqF6PegF5LSpAGaEsYng5QXmTZs4Rt6yo4NTB9WU/OsZkgX3q2k6M9EwRCEX5wpI9oNG821Kbds2djTS1u31qfkfO313qIRA39E/rOtwj8I/AjY8xVxJZrnMjxeFSROjfsBWBT/aXqAesqXRhgaCqQo1GpXCnOCCQNhqcDrK/15HoYGbG9qYKHZrsu68n5Su8kBvi9u7ZyfsTLdw/1cmF0hs0NxVVmJOGZsyM0VTrZ3FDOixcun4pMR9eIxO/OxbEZ1tdl5/dI14+ln4hUAa8Gfg3AGBMEgrkckype54dnaCh3XjZ7sy6+UW1ganZN5+6fnOUTjxzjjq313LalflUlhFR2aYC2iGjUMDoTZEdLcfTgXGj7ugogtlFgLkDrm6S12k1NmYO2SCygONo9UXQB2kP7uogawxMnh9jeVMHXX+zOyPNsiAdlF0d93LE1I0+hsmMjMAx8UUSuBw4Cv2+MuawdR7xDyv0A69drYKxWZ3DaT8eCDUq1ZQ7sVlnzOrTHjg1wetDLl57rpMJp4xu/dcuazqcyT6c4FzHuCxKJmuKd4my6FKBBrNxEz/gsO+MBaUOFkzKHlaPxshPFZmDSjy8YYUtj5oLPpgoXDptFS20UPhtwA/CvxpjdwAzwsYUHGWMeMMbsMcbsaWgozqKiD+3ruuxDpVcgHGHSF5ordJ1gEaGp0rWmnZyj3gCnB7185LWb+fIHbyYQifK9Q71rHbLKMA3QFjHsjc31V7iKb5MAxN6RNVQ45zYK/OiVAQCuaYlVvbeIcG1bFUd6JnM2xkxKrPPIZHbQYhE21HroHNG+pwWuB+gxxuyLf/0dYgGbUmnVMz6LAeoWBGgQW4fWP+m/bN1wKvZdGMMi8P5bO7hzWwN7N9by8zP501ZRLU4DtEUMT8cCtGLNoAFc01LJ48cH2d85xo9e6WddpYv6eU3hr2+v5kTfFIFwJIejzIyzQ14aK5wZ36W7oc6Tlgzaif4pvn+kl2BYd3FlmzFmAOgWke3xm+4GjudwSKpIdcU7jyzMoEFsHZovGJlLHqQiGI5y8OI4O1qq5pa03LG1ntODXiZnQ2sbtMooDdAWMTKXQSveAO0Tb99JXZmD9352HwcujrOz9fL1drvaqglGopzsL656acYYesZn59aIZcpD+7qYDUY4PzzD1164uOrzRKKG/3ipjxcvjPHnj7yy6nfQak1+F/iaiLwE7AL+T26Ho4pR52gs275YgJYIrFZzPX65d4LZUIRbNl0qTJ7o7Xl2KL+v76U+ra4B2iJKIYO2oa6M7/3Oq7hhQzUCXNtyeVPv69urATjaM5H1sWXSlD/MbCiSlR6rtWUOgpEo3kB41ec41jfJhC/Exvoyvv5iN7/79cMle7HKFWPMkfj6suuMMe8wxmgFYpV2F0d9OGyWRV931s3rAJOq/Z3jNFY42Thv88H2pgoaKpycHvSufsAq4zRAW8TwdACX3YLTVtw/nmqPgwc/tJcn/+g1NC5oadVc5aKhwsmRromMPf/UbIhwNLvTdoPxhbbrstDCq648NmU86l1dVQZjDD8/M0J9uYMP3raRq5sr+eFL/fRPrm27vVIq/1wcnaGuzLFoD+Ayp41Kl22uBWGyAuEIPeM+drRUXnZeEeGOrfWcHfIS1ax83iruCGSVhqcDNFQ4V90su5DYrZZF+06KCLvaqzmSoQzakyeH+KsfneRTPzzBV1+4yIUsLaZPbFXPRoCWmKoYm1ldgNY56qN3YpbbttRjtQi/tLsVEXipSDdvKFXKLo75Fp3eTGisdHF2OLWMV/fYLFHDFaU7AO7c1sBsKELfhL7hy1caoC1ixBukody58oFFbld7NeeHZ9K2kDQxNffpx0/z4a8eZF2Vi13t1XSP+fjqCxcZXcUC2FQNTPmpcttxOzJfpLHaY0eA0VUGaM+cGcbjsHLD+hoAPE4bmxrKOdY3qWvRlCoikaihZ2x20R2cCfXlTs4Pz6T0t985OoPAokXXEz2IzwzpNGe+0gBtEcPTAeo1QOOa1ti6tON9qaXVlzMTCPOV5ztxWC28/9YO3rGrlQ/evpFAOMJf/ehk2p5nKQOT/qxkzwBsFgvVHjujM6kHnqFIlNODXm5YX4PdeunPdEdzJSPeIEPT2Wn7UuqLdJXKhv7JWYKRKLVlS7/u1Jc78AbCKe3k7BydobnKtWjXgPpyJ02VTi6OaimgfKUB2iKGvbEpzlKXKFx7rG8ybS/Uz50bYcIX4ldv2UBVvMxFU6WL27fU860DPRy8OJaWsS8mFIkyPB2Ya52SDXVlzlVNcfZNzBIx5oqpiR0tlQhwLI1Bs1IqtxIlNurKl86gJWZ1zg8nF1CFIlG6x3yLLmFJaKxwMbLKNbIq8zRAi0sEHg8+f5GxmeCa22oUg8Q7rHQGAyf6p9lQV0b7gpT7a69qpLnKxf/8/jEiGWrSfn54hogxc1vWs6G2zLGqAO1i/IK9sI9npctOe62H4326Dk2pYtG5TA20hESdymTX6x7rmyIUMXTULx2g1Zc7GJ8JZn2zlkqOBmgLzMRLIpQXcQ20VOxsqeJYmoKBcV+QgSk/VzdXXHGf02blT998NSf6p/jhy/1Jn9MYw9Onh/ncz8+vWFT35EAs0MxmBq22zIEvGEl5HV/XmI+6MseiW+53tlTSN+mnW9tIKVUULo7NYLfK3KzCYqrcdpw2C+eT3Ciw/0JsNmK5mo/15U4Mq9/IpDJLo5AFpuMBWkUR10BLZYpyZ0slPzs9TCgSvWwt1GqcjG8Rv7p58Sb0b722mX954gz//NMzvOXaZqyW5XfRBsOxfnIHu2JlqZ45O8Kv37ZxyeNP9E9jFcnqBpDElEXXqI9r26pWODrGGEPXmI+tS/QK3dlSxX+9MsCPXhngN1+9KW1jVWq+VJcyLDz+PXu1aXyyukZ9tNd6sCxTOcAiwsb6sqSnOPd3jlFb5qBymZaF9fNKATVWZO+Nq0qOZtAW8PpjmY7yIu3DmaqdLZVEoiYtU74nBqapL3cuuQHDYhF+966tnBny8ugKWbRo1PDbXz3Iwa5x7rqqkZ0tlfz8zMhckeHFnBqYoqHCuWLgl06JKYuLY8kvxO0em8UbCF8xvTn/nG01br6xv0t3c6qsG5j085XnO/nT772kvWbTpHPUx4ZFdloutKmhLKkpTmMMBy6O07FCx5TEtXgkCzvoVeo0QFtg2l/8GbRU7Ix3GOhbY3HUKX+IC8Mzi05vzvfma5vZ0ljOPz9xhugya9G++FwnPz05xFuubeZ1Vzfxxh3rCEei/NNPzyz5mJMD01md3oR5Adpo8tORB7viUxO1S68duXVTHeeGZ3jm7MiSx+gOTJVOUWP46YlBPvPkWS6O+vj2gR5e+7dPcd8Dz6+pnVmpM8bQNTqz7GL+hI31ZXSN+QhFll8zdn5khrGZ4IrndDuslDmsGqDlKQ3QFvDqGrTLtNW4qXLb6Z9YWwbt6dPDRIzh6nWLT28mWC3C7929ldODXr7yfOeix5wcmOKvfnSS113dyKs21wGxBbQ3b6zloRe7Fl2jMekL0Z/FEhsJTpuVCqct6WkJgIMXx3HaLDRWLj0Ve21rFfXlDr70bGcaRqnUyvadH+WnJ4e4prWSP3z9Nv7ojdvZ01HLC+fHeO7caK6Ht6J8fcMyOhNkJhhZtFbZQpvqywlHY0sglnM43gEmmXPWlzt1J2ee0gBtgelAGKfNsub1VsVCRNjRXLnmDNpPTwzhcViXnLab7y3XNnPXVY38xX8e55GjfZfd1zsxy+9//QiVLht/+UvXXdbt4bXbGwF4+HDvFedMZKVaazLfg3Oh1hp3Sj1ND16cYP0K61FsVgvv2buBJ04N6TSTyrhAKMITJ4fYVF/GL+9pj7cesvOOXS1cta6CH70yQM+4blpZjUSwtdxi/oRNDbGM2IUV3vAd6R6n3GlLqlxUfbmTkSzVVVSp0ShkAa8/TIVmzy6zs6WSgUl/yuUv5r9b/dnpYTbWly0bdCSO/eb+bu7c1sBNHbX84TeP8C9PnOG/Xu7nn396hrv/7ikujs3wd7+864q1bBUuO3s21PD48cErzv3s2VGcNktS7yjTbX2th7NDXiZ8K79L9QbCnBqYuqIMyWJ+de96rCL8+9Pn8Ycu7WANR6K6Nk2l1c/PjjATjHDPNeuu6On4zhvbqHDZ+PqLXZf9HqrkJHZjJ/M3v6k+tnHo/MjyOzmPdE9wXVvVstfbhPpyB9OBsP7f5SGNRBaY9ocXLW1Qyna2VhKOGoa9gVVNEQbCEcZnguxeX530Y+xWC5//wB5+7Yv7+dsfn567/U3XrOPP3nI1bTWLX8xev6OJT/3wBN1jvssueM+eHWFPR01OMqOJrOHhrglee1Xjssce7honakhqwXBjpYu372rh6y928a0D3VS77cyGIviCEW7fUs+br21Oy/hVaZv2h3jmzAjXtFQu+nfncdj4lZva+fenz/Oz08O8cee6HIyycM0FaDUeDnSOL3tslcdOXZlj2Y0C/lCEk/3T3J/kDu+6eTs5VX7RSGQBbyCc9YXk+S6xUaB/YnZVAdrQVAADNKf42AqXne98+NZY4eApPxaRJUt0JCQCtMePD/LB22MlN0a8AU4OTPPHb9ye8tjToa3ag9UiHLw4vmKA9sL5UawWSWoqGOD//uK1vGHHOl7pneTJU0OUOW2MeAM8f26UWzfVUbNM4UulkvHzMyOEo1HesGPpwGtDXRm72qt59uwIezfWUu3R37tkPLSviydPDVPutC26NGMxmxrKOLfMFOcrvZOEo4Zd7dVJrS1LFMDVjQL5R6c4F/AGQppBW2BzQzkOq4Xu8dWtQxuYim0wWE0FfxGhrtzJzpaqFYMziL1QbGsqv2yaM7GAOdEcONscNgs7mis5eHH5d8cA+86PcV1bFU5bcs3cnTYr91yzjj9643beu3cD79jVyrtubEcEnjg5tNahqxIXiRoOd09wdXPl3Av5Ut6wowlg0SUGxSBTmwzGZ4LLdhBYaKVaaEe6JwDYleSMRV2ZA0EDtHyUtUhERKzAAaDXGPPWbD1vKiJRgz8UpcyR3ItjqbBahLYa96or1w9M+XFYLSlncxa7CCZT/PJ1Vzfx70+fZ9IXospj57mzI1S4bFzbWpXWxu+puHFDDd/c3004EsW2xDSrLxjmaM8Ev3HH2orPVrnt7N1Yy3PnRrljW70WoFTA6grJnhv2MhMIs6u9esVjqz0ObttSz89OD/Oqzam/GSrVQrdjvmBSSxoStjSW860DPYwtEdgd7p6gtdqd9N+93Wqh2mPXAC0PZTOD9vvAiSw+X8oSiyTdGqBdob3WQ//kLMFw6j3bBib9NFU6k1qwmg6v39FEJGp48lQsg/TsuRFu2VSX1QK1C92woYbZUISTA9NLHnPo4gShiGHvxto1P9+d2xuxWy389IRm0dTqHemewGW3sL1p+fqFCXdua8DjsPLYsYEMj6w4RKKGSV8opQzarvYagCUz8ke6JpLOniVoqY38lJUATUTagLcAn8vG863WbDAWoHk0QLvC+loPURMrc5EKYwyDU/6sruu7vq2ahgonf/f4Kb78XCfdY7PcFq+Xlis3blj+ogqX1p/t6Vh7gFbutHFjRw3H+6e0EbJalWA4yvG+Ka5trVoy67uQy27lNdsbOTvs5blliiirmAlfEAPUpLBm77q2KhxWCwcujl1x3/B0gN6JWXYnkfGcr67cyYg3oLu/80y2Mmj/APwJsOgrhYjcLyIHROTA8PBwloZ0JV8ig2bXNWgLJXZEpjrNOe0P4wtGVrX+bLUsFuGf7tuN02blzx85BuRu/VnCUyeHqHTZ+O6hniXXr7xwfpRrW6vStgayo66MSNQwOKlTFyp1J/qnCEaiXJ/ii/3ejbVUue389WOn9AV/BeO+WGvBVDJoLruVa1orObjIjs+59WepBmhlDgLh6Nx4VH7IeIAmIm8FhowxB5c6xhjzgDFmjzFmT0NDQ6aHtKTZYKyLgGbQrlTutFFb5lixgvVCiQ0C2a7gf+vmOn70+3fw1790HR957Wa2LNF4PFtEhPV1ZVwc9S36ovWlZzs53DVBpcuetgXIbfGivN1aQFStwpHuCarcdjqSaEE0n91q4e6rGjnSPVG0GwbSZWwmNq2YSoAGsKejlpd6Jq+oXfalZy9gtQjH+qZSuo7UeGK9p3tXuRFMZUY2Mmi3AW8XkU7gG8BdIvLVLDxvynxBXYO2nPW1HrrHFg8wlpJosp7tAA1i1fZ/+aZ2/viNV11WXDNXrl5XweRsiDNDVxaZ7BrzETGGjfWpvRgup9ptp8xhpUcvuipFk74QZ4amky52utDu9TVsqi/jbx47RTjeNzJfWy3l0rgviFWESrc9pcft2VBDMBLlld7JuduiUcMrfVNsayxPud5joiyKdoPILxkP0Iwxf2qMaTPGdAD3AU8YY34108+7GrPxdyMeuwZoi2mvcTMdCDMxm3wafHDKT6XLhkdLl3BtWxUVLhvPLrI258zgNBZJrt1LskSEthqPXnRVyh4/MUjUxHq+robVIvzJPds5M+TlG/u70zy64jE2E6TaY085CE6saT0wb03r0Z4JJmdDXLOK/7OauQBN38zlE33VnCeRQXNpBm1R62tj2Z1U1qENZHmDwEKZeKe+2nPaLBZu2VTH48cHOTUwzfZ1sZ1xI94A+y6MsbOlClea3xy01bo5PThNQNu4qBT86JV+qt12WqtX37v2jTvXcfPGWj79+GnevqsljaMrHuO+4KqKSdeVO9lUXxbrPHBn7LZHX+7HKsJV61auF7mQy27BabPom7k8k9VCtcaYp/K1BhrEdnG67JaslYMoNOuqXNitkvQ6tEjUMDQdyOoGgXx3c0ctdqvwhWcuzN32r0+dIxSJ8rqrm9L+fG3VHgyp775VpWvaH+Lp0yPsbKlc09IAEeF/vWUHY74gn3nybBpHWDzGZoLUrrLrwo0bajh4cQxjDMYYHn15gC2N5ataoiMi1Hgcep3IM9pJYJ7ZUASPQ5OKS7FahJbq5AvWjs4EiESNBmjzlDlt7G6v4eEjvRzqGqdvYpYHX7jIDRtqaFihUvtqJDYK6NSFStYTJ4cIRqJzLd7W4tq2Kn5xdxtffKZTC6EuMO0P4QtGVt2O7aaOWsZ9Ic4Ne3m5d5LeidlVTW8m1Hjsep3IMxqNzOMLhnHr+rNltVS5OXhxnGjUYFmh8OvwdOyC3JjGwKMYFhffsbWe00PT/OL/e466MgcYuHuFHp2rVea0xS+8OnWhkvOjVwZoqHAm3Q92JX9yz3YePz7Adw/28Juv3qQzFHHdY7FgKNUdnAl7N9VitQjv+rfnaa1xY7MIVzcnV1B4MdVlDl7umcQYkxebqpRm0C7jC0a0xMYK1lW5CEaiSU1zJgK0hvL0Z4YKWV25k2f+x138r7fuoMpt58Ov2ZzR5tKxjQL6zlitbDYY4alTw7xxZ1PaAqmmShefePtOLo75tHjtPInyN4kSF6naUFfG13/zFl61uZ6T/dO89qrGNc0A1XgceANhJlPYBKYySzNo88wGI6t+N1NslspUNccX/J8cmKJjhZIQw9MBKl02nJqVvEK508aHbt/Ih27fCGQ2M9he4+bl3kmGpwMZmUZVxePxE4PMhiK8+ZpmOkfTl3X9hd2tfPbp8/z4+CDb1lVof1jg4mis4flaXnNu3ljLzRtrmfSFcNgsPHy4d9Xnqo6X+ugZn83oG0aVPM2gzeMLRnSKcwWNFS4EON6/dE/JhKHpgF6I80BLfCfeyYHUGsUHQhEOdY0T1WrwJeNrL1xkfa2HWzaltzWaiPCO3a04bBa+9GznXIHWUnZm0Eu505aWdc9VHvua63cm1sJptj1/aAYtLmoM/pBOca7EYbNQV+7kZP/yL/bGGIa9AW5cX5OlkZWeZLNuiXfoiTUvyfCHInzpuU66xnzUlzlYn2I1eVV4zgxOs+/CGB9701Urri9djQqXnQ/etpHPP3OBz/78PG+9rnnFLHwxOz04TVNl/mS0E1Otul41f2iAFhcIRTGguziTsK7KxYkVsjEDU36C4WjWp9SKYRNBulW67Vgl+fIo84MzgCl/OJPDU3nia/u6cFgtvOvGtow9R0u1m9+4IxakveHTT3Pbljpu21LPTCDCM2dHcFhjVfUXay+18G/7PXvXZ2ycSwlFovzgSB8Om4XDXeNsb6q4rAh3smOKRg1nhrwp9znNJLfdSrnTphm0PKLRSJwv3odT2zytrLnKxSu9k0z7Q1S4Fl/gejbezkjXPOWeRYRqjz3pnpyPnxikZ9zH269v4ZGjfUwHNEArdr5gmG/s7+Lq5goeO5bZ/pnNVW5++87NTPlD/NcrAzx5ahiAMoeVYCRKKGKwCFzVXMG9u1qTPm82ArifnhjiUNc4DquFF86P0ljh5Hfv2oo1xYxj78QsvmCEdXm0BCTWecSttdDyiAZocdrmKXmJvpqnBqbZ01G76DGJAC2dJTbU6tWWOZKuXzc8FaC12s3NG2v5j6N9eP26q6vY/eBIH/5QlJs3pnft2VLqyp387t1b+f+9+WqGpgNUe+x892AvxhhmghG+/mIXf/DNI3gDYd67d0NWxrSSnnEfz5wd5sYNNfzC7laOdE/wnYM9HO4aX/I6uJTTg7E1vI15NMUJ0Frt1gxaHtFNAnHaKD15iZ2cJwaW3ihwbtiLy26hXHtw5oUaT/IB2uRsiEp3rD9gudPGtE5xFrXOkRn+76MnaKtx05HGXrDJEBGaKl04bda5r8udNn7tVR28dnsjf/bwK+zvHMvqmBYTDEf53qFeypw23nxNMxYRdrdX01bj5qcnhwjFG8In61Q8QMu3It5tNe68WoPmL/EWdRqgxc1qgJa0KredSpeNE8tsFDg75KWh3KkFD/NEbZmDcV+I6RWyYcYYJmdDc1vuy102vDrFWbS8gTD3P3gAi0W476b1efP3ardaePXWBiqcNv742y/xtRcu5nQ8D+27yMCUn3fsap17jRAR3rhzHZOzIV68kFoQeWbQS0uVK+29d9eqrcbDtD8/aqH98KV+/vd/HufrL3aV7DVIA7Q4X2KKUzcJrEhEuKq5ctmdnGeHZrTERh6pSXInpz8UJRiJUhkP0CpcmkErRlFjeOH8KL/14AHODnn5l3ffkHc1IB02C6/Z3kDn6Axnh705G0c0avjSc52017i5uvnyRuSbG8rZ0lDOk6eGCIaTz6KdGphma9Pqq/5nSqI1XG+Opzm7x3x87HsvUe2xc7x/ik8/fppzOfwdyBUN0OJmE5sE8uwdTb66el0FJwemiUavrJE16Qsx4tWiqPkk0ZB5pY0CiXfOVYkMmtNesu9ei1HUGA5eHOdvf3yK+x54gcNdE3zy3mu4fWt9roe2qJs6aql223n8+CAmR/X4njw1ROeoj1dtWfxn9JrtDfiCEY73TyZ1vkjUcHbYy/Z1+Rigxaa4k91Q9NC+rss+0iEcifL73zgMBj50+yY++totuB1WHjs2kJbzFxJNF8XNBiM4bZaUd+OUqh0tlfiej3BhdIbNDeWX3Zd4t6sBWv6oKYsFXCutQ1sYoFW4bHj9YaLGaA/FAjc05ec7h3roGZ+lrcbNX7x9J6/f0ZSxWYN0vGDbrBbuuqqR7x3u5eTA9BUZrExKjP8Lz16g0mXjmiWax3fUl1HjsXO4ayKp814cnSEYjrK1sZxQJL+KQCf6rya6HOTCZ39+gUNdE/zjfbuYCcRmtm7ZWMujrwxwbth7xetNMdMMWpwvGNH1ZynYHS9Ae+ji+BX3nR2K71DSAG1JmXjnuRy33UqFy5ZygFbutBExBn+wtBfrFrpo1PCtA92MzQR5541tfPjOzdy7q7UglnTsXl9DtcfOs+fW3sfTGENkkaz/Ugan/Jwd8nLLprol37xbRNjVXsPZIS8Dk/4Vz3l6MPYGNh8zaFVuO7VlDi6M5GajQCAc4fPPXODObQ2XlVi5rr0aAR4+tPpWVoVIA7S42VBES2ykYEtDORUuG4cWedd46OIEVW773LonlXsiQnuNh+4V1pZMzoYQmKtvV+GKvYBrLbTC9sOX++mb9POWa5u5YX1NQWVDrRbhpo5azg/PMDIdWNU5DneN892DPfzNj0/xiUeO8ejL/UkthH/u3Ai2+PMvZ/f6agzw/SMrBxCJEhtbGvMzE9RR56FzZHUZtLW+8fyPo/2MeAP8xh0bL7u90mVnS2M5Dx/uXXRZTbHSAC1OM2ipsViEXe3VHO66MoP2YucYN3XUFtSLQClYX+tZsZvA5GyICpdtLltQngjQdKNAwQpFovzdj0+xrtKVV5XrU3HjhhosEru2pOpI9wTv+ew+jvVP0lrt5tq2Kp49O8Jr/uZJnjw5tOTjBqb8HLw4zp6OGspWKBdUX+5kfa2H7x7sWXGt3OnBadbXevI2e9lRX0ZnDqY4jTF84ZkLbGsq5/ZF1vvtXl9N78Tssr8D2Z6ZyDQN0OJmgxHcefoHk69u3FDDqcFppuaVbhia8nNhZIa9G1Mr3Kgyr73WTfeYb9kXkKnZ0Nz0JkCFM/a5N5D7bfdqdb65v5vOUR9v2NFUsG+aKl12rm6u5FDXeEo1x84Pe/ngl/ZTX+Hgv71uG+/du4Ff3tPOR167hZZqN/c/eICfnriyc4Ixhkdf6sdps/K6q5qSeq7d66s5M+Tl5d6lNwsYYzjQOc41rdlbS5eqjXVl9E/650pPZcsL58c43j/FzpYqvv5i9xUB1o7mKsoc1pKa5tQALc6nU5wpu2F9DcbA0e6Judv2xesB3awBWt5pr/UQCEcZXmaaaGJhgKYZtIJmjOH/PXmWPRtq8nLNUypu3liLLxjhWN/yfYATvIEwH/jiiwjw4Af3XtaWrqXazUO/eQtXN1fy4a8e5EevXL5D8Ccnhjg77OXuqxsv67W5nOtaq7FZhE/954klMzjH+qYYmPJzV5JBXy4kGthnO4v2+Wcu4HFY2bVEltdhs/DGa9bx6Cv9Ka0jLGQaoBG7iM0GwzrFmaJd66sRia05S3jxwhhlDis7W/L3HWKpaq9dfgu9MeaKDJrTZsFmEbwaoBWkE/3T9E36+eWb2vOmCO1qbW4op7bMwb4Lo0kd/+nHT9MzPsu/v+/GuaBjviq3nQc/tJcdLVV8+KsH+fMfvMKUP8SjL/fziUeO0VDhZG8Kra/cDivXtVVzpHtiyQr4jx8fRAReu70h6fNm28ZEgLbKdWir8UrvJD85Mcgtm+qwW5cOS+7c1sC0P8yxvuRKmhQ6DdCAmWCEqAGPBmgpqXTZ2dpYzqF569BevDDGjR212Jb5I1O50Z6ocbREsdqp2TDBSPSyAE1EYsVqdZNAQXr6TKwR+Z3b8jcgSJZFhFs31XFx1LfiWsr+yVm+9Fwn7755/bJ9Mqvcdr55/y188LaNfPn5i+z+5OP8ztcOYYzhF3e3plx2ae/GWoKRKEfmzSrM95MTg9y4voa68vzd4Z4IZi+sIoMWiRoeOzbAd+M9Ske8yW3q+Nsfn6LaY1907dl8t26KBczPn0suSC90+ioKTPiCgBapXY0bN9RwuGucaNQwPhPk1OC0rj/LU4kq4Uu9uPVPxQK3ynkBGsR2dJZ6Bk1ErCJyWET+M9djScXPTg1z1bqKvOv5uFp7Ompw2608fXp4yWOixvCDI31Uu+38yRu3r3hOl93Kx9+2gw/etpEbN9Tw/ls38Duv3cKGuiuzbitpq3HTUuVi34XRK9Z69k3McqxvitftyN/pTYiV1qkvd6acQYsaw/cO9fCz08Mc65/k2wd7uPOvn+SZM8uXR3nxwhhPnRrmw3duXrH1VWOli80NZTx/vjQCNF0VD0z4Ygug83VXTT7bvb6Gr7/YzblhL+fjf9C6/iw/uexWmiqdS9ZC65+I1XCqXhCglTttjM6srrxBEfl94ASQt3P3C9c83burhQMXx/jgbRuXeEThcdqs3LKplqdODS9ZtPTFC2N0jfn423ddT7Un+VI/WxrL11z6QkTYu7GOh4/0cnH08r+zn8Z3jL7u6vQFaKnuVEz2+I31HjpTqIVmjOGRo30c7p7gdVc38ZrtDfRNzPLEySF+/Usv8ne/vIu3X99yxfO/++Z2/uaxkzRUOPnArR08fHjlDQC3bKrj+4d7CUeiRT9TU9zfXZISAZquQUvdDfGCtR956BCfefIsDpuF69oWr7itci9WC22JAC1eZPPKDFpp9+MUkTbgLcDncj2WVLxwfpRQxPDqIpjenO/WzfVYLcJnnz5/xX2j3gA/emWALY3l/NINrYs8OvOub6/GabPw8zPDl2XRfnJ8kI31ZWxuSD0zl20ddWUpTXGeHpzmxQtjvHprA6/d3oBFhLYaD9/8rVvZvb6G3/v6Yf7hJ6eJzvt5GGP49E/OsL9znN+9a0vSr7+3bq5jJhhZdrdssdAADRiPT3HqGrTUbW4o43++5Wqq3Q6O901x+5Z6nDb9Oeartho3PUsUq+2fnL2sSG1CucuGLxgpmZ1Ti/gH4E+AJes7iMj9InJARA4MDy89/ZZNn/35eexW4dyQtyhqQiWUO23cuKGG7x3qvWyxeNQYvnuoBxH4xd2tOdsU4bBZeM22Bk4MTPNvP4sFkUe6J3j+3Ch3X9VYEJs1NjaUMTwdYNqfXHmd586NUumy8fodTZd9f1VuO1/54M384u5W/uEnZ/jyc530js8yORvikaN9/NNPz/CuG9t4794NSY/tlsQ6tBKY5tQ5PS6tQdMALXUiwm/csYnfuGMTvmBYe5nmufZaD//xUv+i0wP9k/7LitQmXKqFVnpZNBF5KzBkjDkoIq9Z6jhjzAPAAwB79uzJi0j2zKCXTfXlRTkN9JrtjXSP+Xjv5/bx1Q/tpb3Www9f6qdz1Mc7b2hLaWozE169rYH+KT9//dhJBiZn+fqL3TRWOnn/rR05HVeyNsbX310c9XFN6/IzIsPTAc4MeXnd1Y2LXv9ddit/98vXc2NHDR//wTE+89TZufs+fOdm/sc921MKWuvLnWxrKuf5c6P8zmu2JP24QpTxAE1EXMDTgDP+fN8xxvx5pp83FeM6xZkWuoYv/7XXeIhEDf2T/rmyGwn9k7OX7eBMuFQLrSSL1d4GvF1E3gy4gEoR+aox5ldzPK5ljc0EGZ0Jcuvm5MtEFJIqt51v3H8r7/7sC7z7sy9gTOwNxJ4NNexeX53r4SEi/NINbQjw5ecvcsfWev7pvt0F0/5ubifnyMyKAdoL50exyvLtsESE9+7dwPhMiP7JWWYCESpcNj72pqtWNb5bN9XxrQM9BMNRHLbiewOSkI1X1ABwlzHGKyJ24BkR+S9jzAtZeO6kjPuC8XpPxfsfrRRAW21sJ2f3mG+RAM2/aIBWHi/UWYo7OY0xfwr8KUA8g/ZH+R6cAXSNxdYPbVyk/lexWF/n4Rv338KHv3qQzQ3lbKjz0FzlzvWw5titFr7wazfxzNkR3npdS0HNLnTUJVcLLRCKcKhrnGvbqq5YGgFXbkqoLXNQm4Yg9dbNdXz5+Yu83DvBjRuKd1NaxiMSE+ONf2mPf+TFFEDC+ExQpzdVSUjUQlu4Ds0YQ//E4gGaNkwvPH0TfmwWobGiOMprLKW91sMPf+8O/undu/MqOEuoK3dy767U66nlmtthZV2la25n/lIOdU8QCEfn6pNlS6K23cGLV/aCLiZZSRnFawgdAYaAx40x+xbcn9MFtuO+kE7PqZLQXOXCapErdnIOTQeYDUWoXaSAZiKDVso7OQGMMU8ZY96a63Eko29ylnXx/2ulVuO6tioOXFy+Of3hrnGaq1xXZOMzrb7cyYY6T9EHaFmJSowxEWCXiFQDD4vINcaYV+bdn9MFthM+zaCp0mCzWmiucl1RC+1C/J1y/SLTDzarBafNgi9Y2gFaoUhkQ1daO5RrxbSztBjdurmOHx8fpGfcR1vNlQHYuWEvPeOzvOmadTkYXazE0zNnRzDGFMTO2NXI6qIrY8wE8CRwTzafdyWxDJoGaKo0tNW46V4wxTkXoC3RgsZlt+IPLVllQuWRCV+I2VCElurint5UmZXYYLJUW6UfHO5FgOvbqrM3qHlu2FDD8HRgybJBxSDjAZqINMQzZ4iIG3g9cDLTz5uKcV9QpzhVyWiv8dCzYIqzc2QGh9VClefKNWgQa5oeCC/eAFrll77J2AtWSx6uyVKFY1tjBbVljkXrjRlj+P6RPjY3lF9R2DpbbowXSZ/fC7rYZCOD1gw8KSIvAfuJrUHLm3524UiUaX9YM2iqZLTXehicCuAPXQq4LozMsL7Og2WJqYJYBk0DtELQN+HHIrCuSjNoavUsFuGWTbW8cO7KvqKHuiboGvOxq706N4MDtq+roMxhLep1aBlPGxljXgJ2Z/p5VmtiNtGHUwM0VRra46U2eidm53oZdo7OzG2tX4zLbmEmoAFaIeifnKW+3Im9CAvUpkLXuK3drZvqePTlAbrGfJc1j//+4V5cdgs7WnLXmtZqEXatry7qDFrJz+td6iJQ8j+KNdGLYeFILPjtHvOxuaGcaNRwcdTHncv0bHTarIx6g9kaolqDvolZNi3SRFypVCXaKr1wfnQuQBueDvDw4V7euHMdLntuExs3rK/h/z11jplAmDJn8b2Gl/ZbLGBsRjNoqrQsrIXWP+UnEI7OVQ9fjMtuJRDWTQL5zhsIM+UP06LTmyoNtjSWU1/uvGyjwN8/fhp/KMLv3701hyOLuWFDDZGo4WjPRK6HkhElH6CNawZNlZjGCicOq2WuFlqiWvjGFaY4dQ1a/uubiAXdzdW6QUCtnUhsHdrPz4xwYWSGkwNTfHN/F++7dUNeZGlvaI9vFCjSdWglH5XMTXE6NYOmSoPFIrTWuOkZi72YJ0psdNSX0TnqW/QxTpuVcNQUfe+7Qtc/oTs488nCpR/v2bs+RyNZvV97VQdPnx7mnn94mtZqNxUue9qyZ2tdGlPlsbOpvoyXeibTMp58U/JX2kSjdJ3iVKUkVgvtUgbNabOwrnLpaTGXPXapKNGG6QVjcDpAtduOW69nKk32dNTy+B/eyR1b6zk/MsMfvG4r1Z78afq+o6WSY31TuR5GRpR8Bm3cF8RhteAo8R1PqvjNf7dqDBzrm6J7zDe3g9OyTFugxGJgbyBM3RLFbFXujXgD1Ffo/49Kr6ZKF599/x7ODc+wuWHppRCZtlhGcmdLFf/5Uj+TvuJ781jyAdrETIhqj71oW0UotZjbttTzwvlR/v7x05wfmWFr4/LrSVy2WIBW6v0485kxhuHpALvXVyd1vO68VqkQEbascJ3Itof2dTE05Qfgn544M1c2qFiUfNpo3BekJo/StUplQ5Xbzqs21/P9I71cHPUtu4MTLk1xTukUZ97yBsIEwtEl23UpVYwSG2IS6y+LSckHaBO+WAZNqVJz57YGqtx2IlGz7A5OAKddM2j5biRep65BAzRVQsqdNipdNvom/bkeStqV/BTnmC/IliJLi2aDTo8UPrfDykdfu4VP/fAEW5tWmuKMvZfzaoCWt0amA8DSDe+VKlbNVe65EjPFpOQDtAlfkJoyzaCp0vTrt21kR0slN8QbDy/FNZdB0ynOfDXsDWCzyJIN75UqVi3VLs4MTROKRIuqxVnxfCerYIyJT3HqGjRVmqwW4VWb61fcJOOcK7OhGbR8NeINUFfuWLLhvVLFqrnKTdTA4FRxTXOWdAZtOhAmHDXUaoCm1LJsFgs2izAd0AAtX414AzQtU8tOFT5dWrK4lvhGgb4J/1yv4WJQ0hm0iXgfTt0koNTKXHarZtDyVCgSZWwmqOvPVEmq8dhx2S30TRbXOrSSDtASfTi1zIZSK3PZLboGLU91j/mIGt3BqUqTiNBc5S66UhslPcU5F6CV2RmK74BSSi1OM2j56/xwrJ+qdhFQmZTPU6zNVS4OdI5jjCmawvMlHaBN+BJTnJpBU2olTptm0PLV+REvAPXlei1LVj4HGyp19eVOgpEoU/4wVe7iWLZU0gGaTnEqlTyX3YpXNwnkpQsjM3gcVjyOkr6k5z0NCjMnsf5yxBvQAK0YjM8EEaFo/jOVyiSXzcrYTDDXw1Bx81/snz83phsEVElLZI9HvIGi6clZ4psEQlS67FgtxTFfrVQmxTYJaAYtH416AxqgqZJW6bZjtwqj3uJ5E1nSAdrglJ9GXVSrVFKc8SnOSNTkeihqnmA4ynQgTG2ZLtVQpcsiQl2ZkxFv8Wz4K+kpzsEpP+uqtLCjUsmY68cZKJ5FuMUgsZa2VlvWqTUq9DVy9RVOBoqoFlpJB2gDU362NlXkehhKFYREP04N0PJLqW52KvRgQqVffbmD432TRKKmKJYulewUZzgSZXg6wDptjaJUUpzaMD0vjc8k6jmWVoCm1EL15U6i5tLfRKEr2QBtxBskatApTqWS5NKG6Xlp3BfCZhEqnCU9IaLUZaU2ikHJBmgD8a73mkFTKjkum2bQ8tHYTJCaMkfRVE9XarXml9ooBhkP0ESkXUSeFJHjInJMRH4/08+ZjIHJeICmGTSlkuLUDFpeGvcFqS2x9WdKLcbjsOFxWBkpklIb2cighYH/bozZAdwCfEREdmTheZc1GM+gNWkGTamkuObWoGmAli+MMfEMmm7aUApi05yaQUuSMabfGHMo/vk0cAJozfTzrqR/0o/dKtTpwlqlknJpilMDtHwxG4oQCEdLbgenUkvRAG2VRKQD2A3sW3D7/SJyQEQODA8PZ2UssSK1LixFsBVXqWywWwWrRXQNWh4Zn4n9X2iAplRMfbmDKX+YQDiS66GsWdYCNBEpB74L/IExZmr+fcaYB4wxe4wxexoaGrIynoFJLVKrVCpEhAqXTRum55GxuSK1GqApBfN3chb+OrSsBGgiYicWnH3NGPO9bDznSgan/LqDU6kUVbhsOsWZR+ZqoGkGTSng0puVYqiFlo1dnAJ8HjhhjPn7TD9fMowxDEz5dYOAUimqcNp1ijOPjPuCuOwW3A5rroeiVF5IvFlJdNgoZNnIoN0GvA+4S0SOxD/enIXnXdJ0IIwvGGFdlTZKVyoV5S4bU5pByxvjvqBObyo1j9thxWW3MO4r/DeSGS89bYx5BsirlfiDk1piQ6nVqHTZ6Jvw53oYKm5sJkRTpb7RVGq+Go+DCc2gFab+Se0ioNRqVLjsTOkUZ16IGsOEFqlV6grVHodOcRaqRJun5ip3jkeiVGGpctuZmtUALR94/WHCUaNN0pVaoMZjZ9wXwhiT66GsSUl2101McT55agi7tSRjVKVWpdJtZzoQJho1WkMwxxIZAt3BqdTlajwOguEoE75QQb+BKcnoZGDKj8dh1eBMqRRVue0Yo90E8sFcgKZtnpS6TI0n9jfRMz6b45GsTUlGKINTfipdelFTKlWVrljSfVKnOXMusUut2l24GQKlMqE6nlXuGffleCRrU5IB2sCUn0p3Sc7uKrUmVe7YGxsN0HJvfCZImdOGw1aSl3GlllQzF6BpBq3gDExqBk2p1UgEaKWyk1NE2kXkSRE5LiLHROT3cz2mhInZ0NxUjlLqkkQtNM2gFZgpf4gRb5C6cq0dpFSqqjwll0ELA//dGLMDuAX4iIjsyPGYgFgGTTcIKLW4Go+j4DNoJTPP99C+LgC6x2IRdUO5XtiUSlWpTXEaY/qB/vjn0yJyAmgFjudyXNGoYWI2xM6WylwOQ6m8VV0EAVrJZdBGvAHgUsd7pVTySi1Am09EOoDdwL5F7rtfRA6IyIHh4eGMj2XYGyASNXOLoZVSl6vx2OkZ9xV0LbSSC9CGvQEsArWaQVMqZW67FZtFSi5AE5Fy4LvAHxhjphbeb4x5wBizxxizp6GhIePjSayt0SlOpRZX43EwE4wwUcA9OUsuQBuZDlDjcWCzlNy3rtSaiQhVbntJBWgiYicWnH3NGPO9XI8HLu1Oq9ZNAkotqhh2cpZclDLiDdJQodObSq1WKQVoIiLA54ETxpi/z/V4EhIvOppBU2pxiQLO3QW8k7OkArSoMYx4A7r+TKk1qCytfpy3Ae8D7hKRI/GPN+d6UD3js5Q5rFoDTaklJAo4F3KpjZLZxQkw6QsRjhoaNEBTatWq3HYm4m2Gip0x5hkg75qO9oz7CrrHoFKZ5nZYqXTZ6B7TKc6CMJzYwalTnEqtWilNcear3vFZ3cGp1Araajz0TmiAVhCGpxMlNvTCptRqaYCWW9GooWdiVrsIKLWCthr3XO3TQlRSAdqIN4DLbqHcWVIzu0qlVaXbxpQ/XND1hQrZiDdAMBzVDQJKraC91kPP+GzBXqtKKkAbjm8QiG3MUkqtRpXbTiRq8AbCuR5KSerWEhtKJaWtxs1sKMLoTGGumS2pAG1kOqAbBJRao1LuJpAPtEitUslpq/EAhVsLrWQCtEA4wpQ/rBsElFqjRIA2NasZtFzQIrVKJae91g0UbqmNkgnQRryxFKdm0JRam0rNoOVUz7iP2jIHTps110NRKq+1VscCtEIttVEyAdrcDk7NoCm1JjrFmVudIz466jy5HoZSea/CZac63jS9EJVUgCZAvRZ3VGpNLk1xaoCWC11jPjbUleV6GEoVhPYaz9zGmkJTQgGan9oyBzZryXzLSmWETnHmjj8UoW9ylg2aQVMqKW01bs2g5bthb0CbpCuVBuUOGxbRAC0XesZ9GAMdmkFTKinttR56C7QWWsYrtorIF4C3AkPGmGsy/XyLCUeijHiDbGuqyMXTK5W3HtrXlfJjLBaJNUz3a4CWbZ0jsUzAhjoPJ/qnczwapfJfW42bQDjK8HSAxkpXroeTkmxk0L4E3JOF51lSz/gskaihUTNoSqWFtnvKjc7RGUAzaEolq60mvpOzANehZTxAM8Y8DYxl+nmWc3bIC2iJDaXSRQO03Lg46qPSZdMaaEolqX2uWG3hrUPLizVoInK/iBwQkQPDw8NpP/+54XiAVlFY6U2l8pUGaLnROTpDR32ZtqtTKkmtNYlitZpBWxVjzAPGmD3GmD0NDQ1pP/+5YS/lThtuhxZ2VCodKjVAy4mLo1piQ6lUeBw26socdI9pBi0vnR3y6g5OpdKo0mXXOmhZFopE6Z2YZUOtlthQKhVttR7NoOUjYwznhmc0QFMqjRJTnIW4db1Q9cY3O2kNNKVS01bjplvXoF1JRL4OPA9sF5EeEflQpp9zvhFvkMnZkG4QUCqNqtx2QhGDPxTN9VBKxtwOznqd4lQqFRvryugZnyUYLqzrVcbroBlj3p3p51hOYoOAlthQKn3m9+PUtZ3ZcXH0Ug00pVTytjSWE4kaOkdnCqoeatFPcc6V2NAATam0SQRoE7PBHI+kdHSOzuBxWHU2QKkUbWksBy7FA4Wi6AO0c8Ne3HbrXP9ApdTaNVbGgoTBqUCOR1I6Ejs4tcSGUqnZ1BBbFqABWp45M+hla1M5Fr2oKZU2rdWx2kK9BbgzqlB1js7QodObSqXM47DRWu3mjAZo+eXU4DRbGwtnzlmpQtBU6cJmEXonCm9nVCHyhyJcHPXNTdUopVKzpbFcM2j5ZMIXZHg6wPZ1elFTKp2sFmFdlUszaFlyamCaSNSws6Uy10NRqiBtaSzn/LCXSLRwSgMVdYB2ejAWLW8toF0bShWK1mo3vRMaoGXD8f4pAHY0V+V4JEoVpi2N5QTC0YJ6U1nUAdqpwWkAtmuAplTatda4C+piV8iO9U1S4bTRXuvO9VCUKkhbEzs5h6dzPJLkFXWAdmZwmgqnjeYqbZKuVLq1VbsZmPITihRW8cdCdLxviqtbKnUHp1KrVIilNoo6QDs9OM2WpnK9qCmVAa01bqIGBib9uR5KUYtEDScHptnRrOvPlFqtao+D+nKHBmj54vSgV6c3lcqQ1upYyQddh5ZZnaMz+IIR3SCg1BptbiisnZxFG6CNeAOMzQR1g4BSGdJao7XQsuF4X3yDgAZoSq1JotSGMYWxk7NoA7TT8Q0C25q0xIZSmZBY26kZtMw61jeF3Spaz1GpNdrSWM6UP8ywtzA6oBRvgDagOziVyiSX3UpDhVMzaBl2vH+KrY0VOGxFe7lWKisS8cCxeFY63xXtX/zpIS9Vbrs2SVcqg7QWWuYd75vS6U2l0mD3+hrsVuGF86O5HkpSijZAOzM4zTbdwalURrXWaICWSUNTfka8Ad0goFQauB1WdrfX8Pw5DdByJhyJcqJ/mu3rdHpTqUxqi2fQogXUPqWQ7LswBsB1bdW5HYhSReLWzXW80jvJpC+U66GsqCgDtKM9E3gDYW7dVJ/roShV1Fpr3ATDUUZmCmPRbaH58fFB6ssd7GqvzvVQlCoKr9pcR9TAvgv5n0UrygDt52dGEIn9RyilMqe1WkttZEogHOHJk0O87uomrBZdqqFUOuxaX43LbuG5ApjmLMoA7ZkzI1zXWkVNmSPXQ1GqqLUkAjRdh5Z2L5wfwxsI84adTbkeilJFw2mzclNHbUGsQyu6AG3aH+Jw9wS3b9XpTaUyTYvVZs6Pjw3gcVh51Wa9limVTrduruPU4DTD0/m9NKPoArQXzo8RiRpu39KQ66EoVfQqXXZaq90Fs229UESjhsePD3LntgZcdmuuh6NUUUm86cn361bRBWjPnBnGbbdyw4bqXA9FqZLwtutbePrMCKMFUp27EBztmWBoOqDTm0plwDUtlVS57TxytC/XQ1lW0QVoPz8zwt5NtTht+q5TqWy4d1cLkajh0VcGcj2UovHtgz3YLMJd2zVAUyrdbFYLH7xtI48fH+Ro90Suh7OkogrQusd8nB+Z4fYtumZDqWy5al0F25rKeeRIb66HUhQOXhzn6y928au3bKDKY8/1cJQqSh+8vYMaj52//fGpXA9lSUUToBlj+MQjx3DaLLxx57pcD0epkiEi3Lurlf2d4/SM+3I9nIIWDEf50++9RHOliz964/ZcD0epolXhsvM7r9nCz8+MsC9P16JlJUATkXtE5JSInBWRj2XiOb65v5ufnhzif9xzFe21nkw8hVJqCW+/vgWA/zjan+ORpF82rl8Qe5P5L0+e5fSgl//9jmsod9oy9VRKKeB9t26gqdLJX/zHcYam/bkezhUyHqCJiBX4DPAmYAfwbhHZkc7nODUwzf/+z+O8anMdv/aqjnSeWimVhPZaDzesr+bzz1zgiZODuR5O2mTj+gXQM+7jN79ygH/66Rnefn0Ld1+ta8+UyjSX3cqn3nEt50e8vPkfn+G5syO5HtJlsvEW7WbgrDHmPICIfAO4Fzi+lpOGI1G++sJFHjnax6GuCSpdNv7mXddj0YrbSuXE/37HNfzBN47wwS8d4J6d6/iLe3fSVOnK9bDWKiPXL4gV1H7s2ACv9E1yrHcKq0X4szdfza/f1rHWUyulkvT6HU384CO38ztfO8h7PrePDXUe9myo5a3XN/Pa7Y05HZsYk9kmxyLyTuAeY8xvxL9+H7DXGPPRecfcD9wf/3I7kIlVe/VAfoXHySvksUNhj7+Qxw6FM/4Nxpi8K16YzPUrfns2rmELFcr/7XyFOGYozHHrmLNnuzGmIt0nzYtFDsaYB4AHMvkcInLAGLMnk8+RKYU8dijs8Rfy2KHwx18osnENW6gQ/28LccxQmOPWMWePiBzIxHmzsUmgF2if93Vb/DallMp3ev1SSuVENgK0/cBWEdkoIg7gPuCRLDyvUkqtlV6/lFI5kfEpTmNMWEQ+CjwGWIEvGGOOZfp5F5HV6Yc0K+SxQ2GPv5DHDoU//pzKo+vXYgrx/7YQxwyFOW4dc/ZkZNwZ3ySglFJKKaVSUzSdBJRSSimlioUGaEoppZRSeaYkArRstWpJlYh0isjLInIksU1XRGpF5HERORP/tyZ+u4jIP8W/h5dE5IZ55/lA/PgzIvKBDI31CyIyJCKvzLstbWMVkRvjP4uz8cemteLwEuP/hIj0xn/+R0TkzfPu+9P4WE6JyBvn3b7o71J8Efm++O3fjC8oT9fY20XkSRE5LiLHROT347cXzM9fpU++Xc8K5TpWiNewQr1uFeI1a5kx5+7nbYwp6g9iC3vPAZsAB3AU2JHrccXH1gnUL7jtr4GPxT//GPBX8c/fDPwXIMAtwL747bXA+fi/NfHPazIw1lcDNwCvZGKswIvxYyX+2DdlYfyfAP5okWN3xH9PnMDG+O+PdbnfJeBbwH3xz/8N+O00jr0ZuCH+eQVwOj7Ggvn560fafhfy7npWKNexQryGFep1qxCvWcuMOWc/71LIoM21ajHGBIFEq5Z8dS/w5fjnXwbeMe/2r5iYF4BqEWkG3gg8bowZM8aMA48D96R7UMaYp4GxTIw1fl+lMeYFE/vN/cq8c2Vy/Eu5F/iGMSZgjLkAnCX2e7To71L8ndtdwHfij5//s0jH2PuNMYfin08DJ4BWCujnr9KmUK5neXcdK8RrWKFetwrxmrXMmJeS8Z93KQRorUD3vK97WP6Hnk0G+LGIHJRYqxiAJmNMf/zzASDRNXmp7yOX31+6xtoa/3zh7dnw0XhK/QuJdDupj78OmDDGhBfcnnYi0gHsBvZRHD9/lZp8vJ4V8nWsUP+GCua6VYjXrAVjhhz9vEshQMtntxtjbgDeBHxERF49/874O4OCqINSSGOd51+BzcAuoB/4u5yOZgUiUg58F/gDY8zU/PsK9OevikNRXMcKZZwU0HWrEK9Zi4w5Zz/vUgjQ8rZVizGmN/7vEPAwsdToYDx9S/zfofjhS30fufz+0jXW3vjnC2/PKGPMoDEmYoyJAp8l9vNnhXEudvsosZS8bcHtaSMidmIXja8ZY74Xv7mgf/5qVfLuelbg17GC+xsqlOtWIV6zFhtzLn/epRCg5WWrFhEpE5GKxOfAG4BXiI0tsVPlA8AP4p8/Arw/vtvlFmAynip+DHiDiNTEU69viN+WDWkZa/y+KRG5JT5P//5558qYxIUi7heI/fwT479PRJwishHYSmxB6qK/S/F3gk8C74w/fv7PIh3jFODzwAljzN/Pu6ugf/5qVfLqelYE17GC+xsqhOtWIV6zlhpzTn/eJo27ZPL1g9gOkdPEdlb8Wa7HEx/TJmK7O44CxxLjIjZP/VPgDPAToDZ+uwCfiX8PLwN75p3rg8QWKJ4Ffj1D4/06sfRuiNjc+YfSOVZgT/wX/xzwL8S7XGR4/A/Gx/dS/I+ted7xfxYfyynm7Q5a6ncp/v/5Yvz7+jbgTOPYbyc2FfAScCT+8eZC+vnrR1p/l/PmelZI17FCvIYV6nWrEK9Zy4w5Zz9vbfWklFJKKZVnSmGKUymllFKqoGiAppRSSimVZzRAU0oppZTKMxqgKaWUUkrlGQ3QlFJKKaXyjAZoSimllFJ5RgM0lRYi8lz83w4ReU+Gn+vDIvL+RW7vEJFXFnuMUkotRa9fKh/ZVj5EqZUZY14V/7QDeA/wUAaf698ydW6lVOnR65fKR5pBU2khIt74p38J3CEiR0Tkv4mIVUT+RkT2i8hLIvJb8eNfIyI/E5EfiMh5EflLEXmviLwoIi+LyOZlnusTIvJH8c9vFJGjInIU+EjGv1GlVNHR65fKRxqgqXT7GPBzY8wuY8ynibUmmTTG3ATcBPxmvG8ZwPXAh4GrgfcB24wxNwOfA343yef7IvC7xpjr0/lNKKVKkl6/VN7QAE1l2huINcE9Auwj1otta/y+/caYfmNMgFjPsh/Hb3+Z2FTDskSkGqg2xjwdv+nB9A1bKaX0+qVyR9egqUwTYu8QH7vsRpHXAIF5N0XnfR1FfzeVUrmn1y+VM5pBU+k2DVTM+/ox4LdFxA4gIttEpCwdT2SMmQAmROT2+E3vTcd5lVIlS69fKm9olK/S7SUgEl/0+iXgH4ml+w+JiADDwDvS+Hy/DnxBRAyXphiUUmo19Pql8oYYY3I9BqWUUkopNY9OcSqllFJK5Rmd4lR5S0T+DHjXgpu/bYz5/3IxHqWUSpZev9Ra6RSnUkoppVSe0SlOpZRSSqk8owGaUkoppVSe0QBNKaWUUirPaICmlFJKKZVnNEBTSimllMozGqAppZRSSuUZDdCUUkoppfKMBmhKKaWUUnlGAzSllFJKqTyjAZpSSimlVJ7RAE0ppZRSKs9ogKaUUkoplWc0QFNKKaWUyjMaoCmllFJK5RkN0JRSSiml8owt1wNYqL6+3nR0dOR6GEqpLDp48OCIMaYh1+NIB72GKVVaMnX9yrsAraOjgwMHDuR6GEqpLBKRi7keQ7roNUyp0pKp65dOcSqllFJK5RkN0JRSSiml8owGaEoppZRSeUYDNKWUUkqpPKMBmlJKKaVUntEATSmllFIqz2iAppRSSimVZzRAU0oppZTKMxqgKaWUUkrlGQ3QlFJKKaXyTN61elIq4aF9XVfc9p6963MwEqXy08K/kWL7+yjm76+Yv7dcKMafp2bQlFJKKaXyjAZoSimllFJ5RgM0pZRSSqk8owGaUkoppVSe0QBNKaWUUirPaICmlCoKInKPiJwSkbMi8rFF7neKyDfj9+8TkY4F968XEa+I/FGy51RKqUxJKkBb7YVPRDpEZFZEjsQ//i3N41dKKUTECnwGeBOwA3i3iOxYcNiHgHFjzBbg08BfLbj/74H/SvGcSimVESsGaGm48J0zxuyKf3w4TeNWSqn5bgbOGmPOG2OCwDeAexcccy/w5fjn3wHuFhEBEJF3ABeAYymeUymlMiKZDNqaLnxKKZUFrUD3vK974rcteowxJgxMAnUiUg78D+AvVnFOpZTKiGQCtFVf+OL3bRSRwyLyMxG5Y7EnEJH7ReSAiBwYHh5O6RtQSqk1+gTwaWOMd7Un0GuYUirdMt3qqR9Yb4wZFZEbge+LyE5jzNT8g4wxDwAPAOzZs8dkeExKqeLTC7TP+7otfttix/SIiA2oAkaBvcA7ReSvgWogKiJ+4GAS5wT0GqaUSr9kArRVX/iMMQYIABhjDorIOWAbcGCtA1dKqXn2A1tFZCOx69F9wHsWHPMI8AHgeeCdwBPxa9RcZl9EPgF4jTH/Er+WrXROpZTKiGSmOOcufCLiIHaRemTBMYkLH8y78IlIQ3yTASKyCdgKnE/P0JVSKia+tOKjwGPACeBbxphjIvJJEXl7/LDPE1tzdhb4Q2DZshlLnTNT34NSSs23YgbNGBMWkcRFygp8IXHhAw4YYx4hduF7MH7hGyMWxAG8GvikiISAKPBhY8xYJr4RpVRpM8Y8Cjy64LaPz/vcD7xrhXN8YqVzKqVUNiS1Bm21Fz5jzHeB765xjEoppZRSJUU7CSillFJK5RkN0JRSSiml8owGaEoppZRSeUYDNKWUUkqpPKMBmlJKKaVUntEATSmllFIqz2iAppRSSimVZzRAU0oppZTKMxqgKaWUUkrlGQ3QlFJKKaXyTFKtnpRayUP7uq647T171+dgJEoppVTh0wyaUkoppVSe0QBNKaWUUirPaICmlFJKKZVnNEBTSimllMozGqAppZRSSuUZDdCUUkoppfKMBmhKKaWUUnlGAzSllFJKqTyjAZpSqiiIyD0ickpEzorIxxa53yki34zfv09EOuK33ywiR+IfR0XkF+Y9plNEXo7fdyCL345SqsRpJwGlVMETESvwGeD1QA+wX0QeMcYcn3fYh4BxY8wWEbkP+CvgV4BXgD3GmLCINANHReQ/jDHh+ONea4wZyd53o5RSmkFTShWHm4Gzxpjzxpgg8A3g3gXH3At8Of75d4C7RUSMMb55wZgLMFkZsVJKLUMDNKVUMWgFuud93RO/bdFj4gHZJFAHICJ7ReQY8DLw4XkBmwF+LCIHReT+pZ5cRO4XkQMicmB4eDgt35BSqrRpgKaUKnnGmH3GmJ3ATcCfiogrftftxpgbgDcBHxGRVy/x+AeMMXuMMXsaGhqyNGqlVDHTAE0pVQx6gfZ5X7fFb1v0GBGxAVXA6PwDjDEnAC9wTfzr3vi/Q8DDxKZSlVIq4zRAU0oVg/3AVhHZKCIO4D7gkQXHPAJ8IP75O4EnjDEm/hgbgIhsAK4COkWkTEQq4reXAW8gtqFAKaUyTndxKqUKXnwH5keBxwAr8AVjzDER+SRwwBjzCPB54EEROQuMEQviAG4HPiYiISAK/I4xZkRENgEPiwjErpUPGWN+lN3vTClVqjRAUxnz0L6uy75+z971ORqJKgXGmEeBRxfc9vF5n/uBdy3yuAeBBxe5/TxwffpHqpRSK9MpTqWUUkqpPKMZtDXQDJFSSimlMkEzaEoppZRSeUYDNKWUUkqpPKMBmlJKKaVUntEATSmllFIqz2iAppRSSimVZzRAU0oppZTKM0kFaCJyj4icEpGzIvKxRe53isg34/fvE5GOBfevFxGviPxRmsatlFJKKVW0VgzQRMQKfAZ4E7ADeLeI7Fhw2IeAcWPMFuDTwF8tuP/vgf9a+3CVUkoppYpfMhm0m4Gzxpjzxpgg8A3g3gXH3At8Of75d4C7Jd7ATkTeAVwAjqVlxEoppZRSRS6ZAK0V6J73dU/8tkWPMcaEgUmgTkTKgf8B/MVyTyAi94vIARE5MDw8nOzYlVJKKaWKUqY3CXwC+LQxxrvcQcaYB4wxe4wxexoaGjI8JKWUUkqp/JZML85eoH3e123x2xY7pkdEbEAVMArsBd4pIn8NVANREfEbY/5lrQNXSimllCpWyQRo+4GtIrKRWCB2H/CeBcc8AnwAeB54J/CEMcYAdyQOEJFPAF4NzpRSSimllrdigGaMCYvIR4HHACvwBWPMMRH5JHDAGPMI8HngQRE5C4wRC+KUUkoppdQqJJNBwxjzKPDogts+Pu9zP/CuFc7xiVWMTymllFKq5GgnAaWUUkqpPKMBmlJKKaVUntEATSlVFFbbkk5EbhaRI/GPoyLyC8meUymlMkUDNKVUwVtjS7pXgD3GmF3APcC/i4gtyXMqpVRGaICmlCoGq25JZ4zxxTugALgAk8I5lVIqIzRAU0oVg1W3pAMQkb0icgx4Gfhw/P5kzkn88dquTimVVhqgKaVKnjFmnzFmJ3AT8Kci4krx8dquTimVVknVQVNKFZ6H9nVd9vV79q7P0UiyYi0t6eYYY06IiBe4JslzKqVURmgGTSlVDOZa0omIg1g3k0cWHJNoSQfzWtLFH2MDEJENwFVAZ5LnVEqpjNAMmlKq4K2xJd3twMdEJAREgd8xxowALHbOrH5jSqmSpQGaUqoorLYlnTHmQeDBZM+plFLZoFOcSimllFJ5RgM0pZRSSqk8owGaUkoppVSe0QBNKaWUUirPaICmlFJKKZVnNEBTSimllMozGqAppZRSSuUZDdCUUkoppfKMBmgqbb5zsJtv7u9a+UCllFJKLUs7Cai0OTc8w2woQiRqsFok18NRSimlCpYGaCot/KEIk7MhAAam/LRWu6845qF9l2fX3rN3fVbGppRSShUaneJUaTE8HZj7/OLoTA5HopRSShU+DdBUWgxN+wGwW4XOUV+OR6OUUkoVNp3iVGkxNBXAZhGubq7kwsgMxhhEdB1aodNpaaWUyg3NoKm0GJoO0FDhZGN9GdP+MOO+UK6HpJRSShUsDdBUWgxN+2mocLKhrgyAzjWsQzPG8O0D3cwGI+kanlJKKVVQNEBTa+YLxjJmjRUuGiucuOyWNW0U6Bz18cffeYmfnBxM4yiVUkqpwqEBmlqzc0OxYKyxwolFhA21ZWvaKNA/MQvAwYvj+EOaRVNKKVV6NEBTa3ZmaBqAxkonAB11HoanA3gD4VWdr38ytiM0GI5y4OJ4egaplFJKFRAN0NSanRnyYhWhriwWoG1prIjdPji9qvMNTMUCtNZqN8+fGyFqTHoGqoqaiNwjIqdE5KyIfGyR+50i8s34/ftEpCN+++tF5KCIvBz/9655j3kqfs4j8Y/GLH5LSqkSpgGaWrMzg17qyh1z7Z2aq12UO22cWmWA1jcxS43Hzp3bGhj3hTjRP5XO4aoiJCJW4DPAm4AdwLtFZMeCwz4EjBtjtgCfBv4qfvsI8DZjzLXAB4AHFzzuvcaYXfGPoYx9E0opNY/WQVNrdnZomsZK19zXFhG2NZVzon+aqDFYUqyHNjDpZ12Vm6ubK6ly2zncNcHOlqp0DzurCqWe2MJxFpCbgbPGmPMAIvIN4F7g+Lxj7gU+Ef/8O8C/iIgYYw7PO+YY4BYRpzEmgFJK5Yhm0NSa+EMRusZ8NFY4L7t9W1MFs6EI3WOxzQKhSJRgOJrUOfsn/TRXubBahLpyBzOrXMumSkor0D3v6574bYseY4wJA5NA3YJjfgk4tCA4+2J8evN/yRLVl0XkfhE5ICIHhoeH1/J9KKUUoBk0tUajM0GiBqrc9stu39pYgQCnBqdpqXbzbz87x+CUn3VVLrY3VfC6q5uWPOfAlJ/d66sBcNmsjAWCGfwOlIoRkZ3Epj3fMO/m9xpjekWkAvgu8D7gKwsfa4x5AHgAYM+ePbpoUim1ZppBU2sy7Y91DHDZrZfd7nZYWV/n4fTgNP/5Uh/9k35u6qgF4MlTw4zOLB50+UMRxmaCNFfFpkydNgv+sJbaUCvqBdrnfd0Wv23RY0TEBlQBo/Gv24CHgfcbY84lHmCM6Y3/Ow08RGwqVSmlMi6pAG0Nu6Nunrf76aiI/EKax69ybNofm3502a/8VdreVEHfhJ/9nePcua2Be3e18o5dsVmngXgpjYUG4zs411W5AXDaLQRCyU2NlrqH9nVd9lFi9gNbRWSjiDiA+4BHFhzzCLFNAADvBJ4wxhgRqQZ+CHzMGPNs4mARsYlIffxzO/BW4JXMfhtKKRWzYoC2xt1RrwB7jDG7gHuAf4+/c1VFYmo2nkGzWa+4b1tTrNzGhjrP3JRmY4UL4VIgtlCiBtqlDJqVQDiC0VIbahnxNWUfBR4DTgDfMsYcE5FPisjb44d9HqgTkbPAHwKJN5sfBbYAH19QTsMJPCYiLwFHiGXgPpu1b0opVdKSCZbWsjtqfjl5F6CvskXmUgbtygCtucrFu25sY2tTxVwJDofNQm2ZY67W2UKJzNq6KhcXR324bBaiBsJRg92a2m5QVVqMMY8Cjy647ePzPvcD71rkcZ8CPrXEaW9M5xiVUipZyQRoi+2O2rvUMcaYsIgkdkeNiMhe4AvABuB98Xe6lxGR+4H7Adavz8/yA2pxl9agXZmMFRF2r6+54vZ1Va4lpzj7JmNtntbFy3Y444GfPxTBbs2PJZOFUjJDKaVU4cr4K54xZp8xZidwE/CnIuJa5JgHjDF7jDF7GhoaMj0klUZTy2TQlrKu0sXYTBBf8MryGQOTfipdNsqcsfcOTlvsVzTZEh1KKaVUMUgmQFvT7qgEY8wJwAtcs9rBqvwz5Q/hsFpSym41VbowxDoQLBSrgeae+9oZX9vm1wBNKaVy5qWeCd7/hRf5X99/hSdPDumb5ixI5lV1LbujNiY2BYjIBuAqoDMtI1d5YdofptKd2r6PxAaAkwNXtnCKdRG4lGR1xqdOAyEttaGUUrnw/cO9vOvfnudY7yTfOdjDr39pP7/5lQO6eSvDVgzQ1rg76nbgqIgcIVZj6HeMMSNp/h5UDk3Nhqhw2Vc+cJ6aMgd2q3By4MpenYkuAgmJ3aEBfbemlFJZ96393fzBN4+wq72aH/+3V3P446/nT+7Zzs9OD/O10ivnk1VJpT7WsDvqQa5sPKyKyLQ/TIUrtQyaRYSmShcn+y8P0ILhKCPewOVTnIkMmharzbj/+f2X2Xd+jAqXna1N5VzfVp3rISlVskKRKCf6pzjRP8Wbrm2mMsU3wukQDEf5+8dPc+OGGr76G3vnlrL89p2bee7sKP/n0RO8emsD6+s8WR9bKdCaZGpNpv2hVV041lW6ODkwhTGGRHvDRG20+Rm0xCYBvxarzShvIMxXX+ii0mWje9zHyYEpDdAKWCAc4fPPXKB7zMfwdID7blrP63Ys3V5N5ZefnR7md756kJlg7I3pgc5x/uZd12d9HA8f7mFgys9fv/O6y9YZiwh/9c7ruOfTT/Mn3z3KN+6/NetjKwX5UbdAFaypVWTQIFZqY9wXYnj6Uk/qgalLNdASnDrFmRXnh2MbNt52fQt3bmvEF4zour8C9v3Dvfz1j07x+PFB9l0Y41M/PE40quuFCsU//OQ0NWUO/uU9u3nfLRv49sEejnRPZHUMkajhX586x7WtVdyxtf6K+1ur3fzB67fxwvmxRdcTq7XTAE2tybQ/tKoArakysVHg0jRn30SsBtr8DJrdKlhENwlk2rl4gNZQ7qTGE8uIjvm0SX2h+ub+brY2lrP/z17H//cL19I56uOp00O5HpZKwrG+SQ53TfDrt23krde18Cf3bKehwsknHjmW1SD70Zf76Rz18ZHXbp6b5VjoHbtasFqEHxzpy9q4SokGaGpNpv3hVU1xJoKwV/om527rGo01nmiuvrQGTURw2qxaZiPDzg3NYLMIdeVOasscAIzPhHI8KrUaZ4emOdQ1wS/vaUdEeNM162iqdPLFZztzPTSVhIf2deG0WfilG2J9iytcdv7HPVdxpHuC7x1eWOEqcz738/NsbijjDTvWLXlMXbmTO7bW88iRPs3QZoAGaGrVQpEovmAk5V2cAB6HjY46D0e6JuZuO9ozweaGMsqdl2fknDYLQd0kkFHnhr2sr/NgtQg1nniAphm0gvTtAz3YLMI7dsde4O1WC++7ZQM/PzPC2aErd06r/OENhPn+4V7edn0L1fG/Q4Bf3N3KzpZKvvDMhayMY3DKz9GeSd55YzsWy/It9u7d1ULvxCyHusazMraEUKT437RrgKZWzRvvIrCaKU6A3etrONw9gTEGYwyHuiYWbQ3ltFt0k0CGnRv2srmhHACPw4rDZtEpzgIUikT57qFe7rqqkYYK59zt7755PQ6bhS8915m7wakVff9wLzPBCO9d0D7OYhHedn0Lx/un6I+3w8ukp07FpsNfe9XKnX1ev2MdLrslq9Ocjx8f5Pq/+DE/PTGYtefMBQ3Q1KolGqVXule3/XtXezXD0wH6Jv10jfkYmwlyw2IBms2qZTbWwB+K8MTJIfxLrOMLR6JcGJmZC9BEhFqPg/EZDdAKzVOnhhnxBvjlPe2X3V5X7uSt1zbzg8M6FZXPHj7cy9XNlexqr77ivruvagTgiZOZX0v45MlhmqtcbG+qWPHYcqeN113dxA9f7s9KVmtoys+ffOcovmCE/9/DLzM5W7xLMTRAU6s2FW+UvtoMWuIidKRrgsPxqc7d66uvOM5lt+guzjX4+ZkRfnJikKdODS96f/f4LKGIYXND2dxtNR67TnEWoMePD1DtsfOa7VdmPm7ZXMd0IMz5kZkcjEytJBSJ8nLvJLdvqVt0Uf6WxnLaa908cSKzAVowHOWZsyO8ZnvjkpsDFnr79S2MzQR5/tzoygevgTGGP/7OS8yGInz6V65neDrA/330REafM5c0QFOrttYA7ermShw2C0e6xzncNY7HYWXbIu/YnDYrAZ3iXJVAKMIL52MXzVd6Jxc95txQbAfnlsbyudtqyhyMz4S0lUuBOd4/xTUtVdgW6Y2bqGv3Us9EdgelknJ6cJpgOMq1S9QfFBHuvqqJZ86OMBvM3IzCgYtjeANhXrtIkL+U27fWY7cKz2U4QPvm/m5+dnqYP3vz1fzC7jZ+89Wb+Mb+bp47W5wNijRAU6s2N8W5ygrXDpuFa1oqOdw1waGuCa5vq8a6yIJUp82iU5yr9GLnGLOhCG67lZeXCtDiJTY2NcwL0DwOgpHoXKFMlf8iUcPpQS9XNy8+LbWlsRy33cpLPYv/Hqjcejn+/3Jda9WSx9x9dSOBcJTnzmUuIHnq1DB2q3Dblitrny3F47BxfVv13JvBTPnuoR6uWlfBr96yAYD/9rptNFQ4eejF4mw5pQGaWrWp+Nz/WlqQ7F5fw8u9k5zon+KGDdWLHuO0WbTMxiqEI1GePTvCpvoydjRX8krv5KIZsXPDXhoqnFTNW0t4qdSGTnMWihFvgGA4yo6WykXvt1qEa1orNYOWp17qnaTCZWPDMm2Tbt5YS5nDyk8zuA7tyZND7N1YR5kztZmRWzbV8XLvJN5AOCPjmvSFOHhxnNfvaJqbenXZrezZULPkm89Cp62e1Kpd2iSw+l+jXe3VfD6+dXx3+5UbBACcdivBcJSoTrel5Ej3BFP+ML90QxsjM0EOdo3TP+mnZV6dOYBzwzOXrT8DLiu10V6rffYKQf9krBPHhWEfDy3RxNphtbCva4wHn7+4aLb6PQt2Dy601HmXevxKx6f6/Nkez0rnW248qX4vPzs1TGOFk6+/2L3scR31Zfzn0T52NldetkYsHd/rhC/ImSEvWxvLU378TDBMJGr428dOLbpUZSUrjf+lngmiBkLh6GX3RaOGi6M+ZoMR3A5rys+bzzSDplYtEaAtrFuWivm7lXYtskEAwBXvxxnULFpKzgx5qfbY2dJYTms8KFv4TtMYw9kh72XrzwBqyuLdBDSDVjAGJmexWuSy8hoLtdV4CEcNQ9P+LI5MrSQciTIw6Z/7O13OtqYKpvxhhr2BFY9N1YX4BpLNC64HydhQW4ZFLp0j3U4NTONxWGlb8IaxtSb2de9E5suPZJsGaGrVpvwhPA7roguSk9VW46a+3Mn6Wg/15Yu/sDjt2o9zNcZmgjSUOxERmqtcWC1yxUaB0Zkgk7OhuRIbCU6bFY/DWlA7OUXkHhE5JSJnReRji9zvFJFvxu/fJyId8dtfLyIHReTl+L93zXvMjfHbz4rIP0my29pyoH/ST2OFc9HMWEJbTSwA6BkvvhezQjYw5SdizFywsZyOuli2+2K880o6dY76cNktc634UuGwWWir8cz19U2nqDGcHpxmW1MFlgV/gomgtnc8/T+PXNMATa3atD+0pvVnENuZ9Duv2cxv3blpyWOc8QzaUnW81OJGZwJza8nsVgtbG8uvWCB+dujKDQIJtfGdnIVARKzAZ4A3ATuAd4vIjgWHfQgYN8ZsAT4N/FX89hHgbcaYa4EPAA/Oe8y/Ar8JbI1/3JOxb2KNBib9l/WxXUxtmQOX3aIBWp5JZH+SyaDVlzsoc1jpzECm6uLoDOtrPVcEQcnaVF9G78Rs2jd19Y7PMhOMLDp16nZYqS1z0KMZNKUumfaHV11iY74P3r6R9+7dsOT9Tptm0FLlC4bxh6LUlV1qF3NNa9UVGwUSGbUdzVcuLK/xOAqpm8DNwFljzHljTBD4BnDvgmPuBb4c//w7wN0iIsaYw8aYRBn0Y4A7nm1rBiqNMS+Y2A/tK8A7Mv6drII3EGY6EGZd1fIv8CJCW7WnKLMNhax3fBa33UqNZ+U3vCLChroyLo6l9//QFwgzNB2Yy9CtxsaGMqIm/dm9U4PTCLBtianX1mq3TnEqNV+6ArSVuOyxX9OAZtCSNuqNBVa1ZZemja9rq2J0Jji3mBzgaM8krdXuRdct1ZY5mPSFCmVzRiswf3V1T/y2RY8xxoSBSaBuwTG/BBwyxgTix/escE4AROR+ETkgIgeGhxcvCJxJifY/K2XQAFpr3AxM+Uuil2Gh6J2Ypa3GnXRh2I46D2Mzwbmd9OmQCPg2rCFAy9Q6tNOD07TXevAssd65rcbNhC+UsR2kuaIBmlq1KX9o1W2eUqEZtNQlFvfXll+eQYPLNwoc7Z7g+vbF6y7VeBxEjEnri0A+E5GdxKY9fyvVxxpjHjDG7DHG7GloSL7AZ7oMxIPu5iTWDrXVuImaS49RuRWKRBmcSm6DQEJHfSyI6hxNXyB0cXQGq8jcOsXVcNgstFa70xqgzQYj9I7PXrGRab7W+Jj7iiyLpgGaWrVYBi0LAVoig6bFapM2mgjQPJcCtB3NlVgtMlcHa2wmSNeYb67C/ELV8emWAul11wvMb0DZFr9t0WNExAZUAaPxr9uAh4H3G2POzTu+bYVz5oX+ST+VLtuSGYb5GitiQdxIBnYBqtSNeANEDaxLIvuZ0Fzlxm4VOtM4ldg56qO1xo19DZu+IBY89o7Ppi1De3FsBgNsrF86s9dS5UaAniKbutcATa3atD+UlSnOS5sENIOWrLGZIBUuGw7bpT9xl93Kjetr+Gm8l9/ReKB2/SKNmYG5QpUzhTFtsB/YKiIbRcQB3Ac8suCYR4htAgB4J/CEMcaISDXwQ+BjxphnEwcbY/qBKRG5Jb578/3ADzL8faxKbINAcpmPmjI7wqUgXuVWYjlC3RK72BdjtQjraz1cTFMGLRSJ0js+u2yR3GR11JURMYbuNAVLF0Zm5r7fpbjsVurLnfQW2eYXDdDUqk3Nhte8izMZl6Y4NYOWrLGZwGUbBBLefO06Tg5Mc27Yy9HuCUQuTX0ulKhv5w3k/889vqbso8BjwAngW8aYYyLySRF5e/ywzwN1InIW+EMgUYrjo8AW4OMiciT+0Ri/73eAzwFngXPAf2XnO0qeMYYRb2DZ+mfz2SwWqj12zaDlicRyhMX+Xpezoa6MgUl/Wna394zPEjFmTRsELo0rFkh1jqQvQGtLIrPXWuOmr8im7bWTgFoVfyhCMBLNSgbNahHsVtE1aCkYmwmypfHKLen3XNPMJ/7jOP/1cj8v9UyytbF8yULDZfGq3DPBgsigYYx5FHh0wW0fn/e5H3jXIo/7FPCpJc55ALgmvSNNL28gTDhq5qakk1FX7tQixHlidCZAmcOKy55aFfyOujIM0DXmW1Xl/vkSmbgNaega4nHYaKp0piW7FwhF6JuY5dXbVl7XWVfu4Ej3BOFIdE21OfNJcXwXKusuNUrPTozvtFkJ6BRnUoLhKFP+8FwNtPnWVbm4cUMNP3x5ILZBYIn1ZwA2qwWnzVIoU5wla8IXWyNY40k+A1NX5mDEG1i0N6vKrlFvMKXpzYT1tR6sImkpDHtu2Mu6SldSaxiT0REvAxKJru336+KYj6iBjUlk9hK//xOFsWY2KRqgqVWZ8scbpWdhFyckGqbn/1RbPkjULltqyuRN16zjRP8UozNBrlti/VlCmdNWdFvXi03iBSnVDJo/FGU2qH9TuTY6E0x5ehNiOybbaz2cXWOAFopEuTjqu6If71p01JURDEcZmFrblOOFkRksAuuTWBs3v39wsdAATa1KIoOWjSlOiO3k1AxacsbmaqAtEaBd2zz3+a5lMmgQm+b0FcAatFI2EX9BqnYn/yJfH//dGCnhac6BydzXggtFokzOhqgrTz1AA9jcWEb/hB/fGpYhdI35CEfNFe3e1mKuDMgay210jszQWu2eW4e8nESR34kC6X6SDF2DplZlOp5By0aZDYhPcebhGrS1XBgzZWwmtvh7qXflrdVudrVXc7x/isPd41c0UJ+v3Glj3Fc8F7xiNO4L4bRZcDuSX8OUqI836g0suzsunaLG8PUXu/D6w1S67exorlxyB3GmvdQzwTf2d7OlsZwP3NqxbP/STLq0QSD1KU6ALQ3l/PTEEOeHVx8InRv2YpFLQVU6VLnt1HjsdI7OcNuW+lWdIxiO0jM+y21bFtaSXlyFy45FNIOmFFOziTVo2QnQXDZL3u3i7B2f5f88eoLnzo7keiiXGZ0J4rIv/4L9p2+6ij9/2w5sluUvAWVOW8FsEihVE75gSuvPIFYfL9ulNk4PTHOsb4pQNMrF0Rm+fbCbkens7yTtHvPxnYM91HjsnB3y8p8v9eVsLd6lEhury6C11Xhw2CxrmuY8N+Sltdqd8iaFlXTUldE56lt1J5LO0Rkixixb/2w+q0Wocts1QFPqUgYtW1Oc+ZdB2985RtTA8+dHcz2Uy4zNBKkrcy7bNmbvprpl+58mlDltzATCupg8j034QimtP4PYBpBqj53RLJbaeO78KFVuO7995xY+etdWbBYLj58YzNrzQ6zo8ldeuEil285vv2YLd2ytZ9+FMfZdGMvqOBJG57Ldq8ugWS3Cxroyzg2tLkDzhyL0TsyyeZkq/au1tamcmUCYnlX2DD3eP4XDamFTClOv1R7H3KaZYqABmlqVRHX5rG4SyKNenMFwdK7Q69GepacIc2FsJrjk+rNUlTltRI0WCc5nE7NBqlPMoEFso0C2MmiDU37ODnnZu7EWq0Uod9q4bUs9L/dOZrW46JMnhwiEIrz/lg2UO228cec6NtWX8eSpoZy8CRmdCeK2W1Oanl5oc2M5ozPBVTUL7xyZIWpI6/qzhO1NlVhFONY3lfJjo1HDib4ptjWVp9TZoMbj0AyaUpOzIawWmauVlWlOm5VgHmXQjvdPEghHaSh38lLPRN5kmILhKGMzQepXsW1/MXO10HQnZ17yhyL4Q9G5BdKpqCtzzE2xZdrz50exWYSbOmrnbrtjaz1uu5UfHx/IyhiGpwMc6hpn9/oaGuM9Sy0i7GqvZtofZnAq+9OtY94g9auc3kxI7L58dhVLLc4Ne7GtUKV/tdwOK5sbyzjWP5Xy9fFw9zjTgTA7WhYvor2UGo+daX+YcI43f6SLBmhqVSZnQ1S57ctOo6WTy24hHDV5sw7t4MVxajx2XrWljglfiO6x/GgxMjjlxwAt1cn39VvOpW4CGqDlo0S2YLUZtNlQBF+G/29ngxEOd41zfXv1XPswiLXnec32Bs4MedPaXHspX36uk0jUcMeCRetb40VezwxNZ3wMC43OBFZVA22+pkoXZU4bT58eTulxxhhODU6zoc6z5v6bS9nZUsXYTJD+FCv8P3ZsEKsIV61LrQBvjceBoWD6B69IAzS1KokALVsSPSXzoeTD+EyQ88Mz3LChhvaa2DvPxHRnrvVNxgLFliT7Mq5krh+nbhTIS4n1NtWr+FtM7PLN9DTn8f5JQhHD3o21V9y3d2MdZU4bT54cyugYZgJhvvJ8JztaKqlf0BKrym2nscLJmcG1F3xNRTgSZcIXWvNyBIsI17ZW8eNjgwxNJx8InRueYcQbZHd7zZqefzlXN1cikNI0pzGGH70ywObGspQ3LlSXxf4OimXnuQZoqxSNGsZ9Qfon8yNzkm2Ts6GsrT+DSw3T8yGT83LvJAa4YX0NTZUuHDYLL+VJgNY/4cdlt6S8aHwpZZpBy2tzNdBWM8UZn1rLdE/Oc8MzlDmstFZf+abBYbNwx5Z6zg576VrlYvJkfGN/N1P+MK/eunjLoG1NFVwYncnqMooxXxBD6j04F/OqzXUEI1G++kJX0o957twIZQ4r17alNo2YinKnjY76Mo71Jb9O9+TANF1jPnY2pz6uYitWqwHaKnzl+U52/PmP+JvHTvHPT5wtySBtKusZtNg7KV8eVD4fnQlS5rRR43FgtQg7mivzZqNA3+QszVXutE09X1qDlvufu7rShC+ELb7oPlXZKLVhjOH8sJdNDeVL/k7u3VSLx2HNWBZt1Bvgn584wy2bamlfYq3V1sZyIlHDhZHsZdHG5kpsrH29aH25k7uvauRrL1xMajPVqDfAqYFpbt5Yl7HpzYSdLZUMTQcYTLKrwMOHexGBq5pT7y9aWWS10DRAW4X/fKmfxgoXd1/VCMBQDhaX5tqUP5zVAC2RQcuHqTavP0TFvBfE69uqeKV3cs1959YqEjUMTPppqUrP+jOY148zD37u6krjs7ESG6sJyG1WC1Uee0abpo96g0z5w8vuEnTarNy+pZ5Tg9MZyUT/fz88wUwgzCfvXbrnfUd9GTaLcHqV5SpWIxEYp2vH9Qdv38joTJBHjvateOwL50cRYdFp53S7trUKp83C94/0rlgT7ZXeSb7wzAV+YVfrqoqgWy1CpdteNKU2NEBLkTGx7b93bmvgjni6vFii9VTE1qBlrxGFI/4uLx92E04HwpfVf7uurRpfMMK5NDQtXosRb4Bw1NC8yFTSWpTHa6Gp/DPhW12JjYQaj4PxDAZo5+IZqZX6PN6yqQ6Pw8pHHjrEQIoLypfz8zPDfO9wLx++czPbmpbOyNitFjbWl2V1HdqYL4jDZknbTvhXba5je1MFn/v5+WX/XsdnghzsGuea1qqsLFOpcNl5+/UtXBz1LbuRIRCO8EffPkptmYM/f9vOVT9fMZXaSCpAE5F7ROSUiJwVkY8tcr9TRL4Zv3+fiHTEb3+9iBwUkZfj/96V5vFnXc/4bHz7b2Xsj8tpy+g70HxkjMn6JgGnPRGg5X6qzesPXzaldH17bK3E0e6JHI0opj/NGwQSyjRAy1sTvtCqNggk1Gb4xezc8AxVbvuKWSKX3coHbu1gzBvkvZ97Yc0FdI0xnBma5mPffZmN9WV85LVbVnzM1qYKRryBrO0AnJgJxqaZ07QcQUT4b6/fytkhL2/752cWXXpzamCKf3nyLAB3blt8PV4m7Gqv5trWKn5yYpAT/VNXzDZM+0N88j+Oc3Jgmr/8pWupWsMa2hpP8WTQVkyBiIgV+AzweqAH2C8ijxhjjs877EPAuDFmi4jcB/wV8CvACPA2Y0yfiFwDPAa0pvubyKbEYscdzZUc65ui1lNcrSWSMROMEIma7K5Bi2fQct370hhzRQZtU3055U4bL/VM8q497TkbW/+EH5tFaKhITw20hDKnLaNZFrU6/lAEbyC8tgxamZ0pf5hQJJr2tUjRaGz92famiqSCkPZaD5//tZv4wBde5PWffppbNtUiCG01blqq3SuOb8IXpGd8lrGZIKcHpzkfb7T96V/ZldRuwA3x9Wk94z6q3JlbOJ8w7gutqn7dcu65ppmv/cYt/P43DvP/njpHfbkDt92GMYaZYIQRb4DmKhfvuXl9Wta+JUtEeMeuVnrGfTz4wkWcNgst1W5EIBCK0jcxiwHeu3c9d13VtKbnqvY4mJqdIBiOzu3+L1TJzFHdDJw1xpwHEJFvAPcC8wO0e4FPxD//DvAvIiLGmMPzjjkGuEXEaYwp2EVbx/umsAhsX1fBsb4pasoc9GSxEnY+SLzDzEWZjVxncqZmw0SihvJ56yMs8Y0Cx/tTr5idTn2TszRVutLe+LnMYaVnTDNo+SZROX4tL/Lzd701VqRv7SLAqcFpfMFISlXqb9lUx9d+Yy9f29fF/s6xuWurRWBLYzm1ZQ7uvrpxLlgb9wU53jfFy72Tl+0CrXLbecu1zfz9r1yP05bcFOK6KhcWifXY3ZligdRUGWMY8wXZtMLU72rcurmOR3//Dn7v64eZ8IXwBSNYLcK6KhfXtVVx57aGjG8MWIzbYeX37trK2WEvpwe9sZIgJra++K6rG/nvr9/O1avYGLBQohZa/+QsG+rS//PNpmQCtFage97XPcDepY4xxoRFZBKoI5ZBS/gl4NBiwZmI3A/cD7B+/fqkB58Lx/un2NxQPveOrMbj4JXeyVU3hC1Ek77sB2iJi+xMjndxDntj62MqFuya29FSybcOdBONGixpDpCSYYyhb8LPzpbKtJ870TC9lH7HC0FirdZapoMSU4/jM6G0B2jPnYv1qE01CNnTUcueeMeBf3vqHD3js1wcm+Fo9wQf/upB7FahtdqN02bl1GCsuOy6Shdv2NHE1sYK6sodc9fnZIMziK1Da6p0raplUqrGfSGC4WjKTe6TVV/u5K3XtWTk3GvhtFvZ2VK1aAC8I03XrkTJmd7x0gjQ1kxEdhKb9nzDYvcbYx4AHgDYs2dPXr8KHO+b4qZ5O19qPQ6iprAqFz+07/JaOe/Zm1pQnO0+nAB2qyDkPoM2PB2b6itf0CR+R3MlvmCEi2M+NtZn/6LQP+lnNhShJc0bBCC2SSDWjzP36//UJYmyBZWr2O2WUFOWubpRz58bpbbMsaYp2Eq3nR1uOztaKnnDjnW0VLvY3zlO97iPqdkQG+vLYsVn0zRd11rt5lhfrDVRJruk9IzHsn2ZCtBKWeLvYWi6YCfq5iQToPUC8xfWtMVvW+yYHhGxAVXAKICItAEPA+83xpxb84hzaHwmSN/k5VmKuQtcCa3RycUUp4jgsFlyvklgOL54ebEMGsQC+FwEaP/2s9if1oa69PfUK3MWRi00EbkH+EfACnzOGPOXC+53Al8BbiR2ffoVY0yniNQRW5pxE/AlY8xH5z3mKaAZSKRV3mCMyWzZ+yQlXoAqXKt/n13htGGzSNo3OkWihn0XRlNu1bMcq0W4++om7r760hqlhW8216q1xs2Bi+OMp6HC/3ISreFqyrJ3DS0Vib+HZOuu5bNk/rL3A1tFZCOxQOw+4D0LjnkE+ADwPPBO4AljjBGRauCHwMeMMc+mbdQ5ciK+xmjHvArHifUfhbxRYLGL3HJZtalEBm0N79xXw2mz5HyTwPDci+Ll3/uWxnJsFuF4/yRvua552XOMegP85lcO8L/esoP1aQiofvTKAF95/iK3b6mnOc07OKEwugmscTOTH/hfwDXxj4Xea4w5kNFvYBUGp/w4bZaUpvEWEpGMlCU41jfJtD/Mpvrk15/lg7bq2N9j78RsZgM0zaBljNNmwWG1MFgE9UlXDNDia8o+SmwHphX4gjHmmIh8EjhgjHkE+DzwoIicBcaIBXEAHwW2AB8XkY/Hb8ubd6CpSiwCn7+QsTpejXtspnCmONdqLoOW5h1IK3HYLDkPEoanA1gtgst++SJbl93K5oZyjq/Qcy4SNXzrQDfd47O01bhTqvezMJB+983tnByY5k++c5RrW6t4w8617X5aSpkj3o8zjwM01raZaQZ4RkRWrsWQR4amA6sq5rlQTZk97TMAz8fXn23MwCL4TGqqdGK1CD3jPq5tzdxGge4xH267NeVek2plIkKFy5ZSX9J8lVRu3BjzKPDogts+Pu9zP/CuRR73KeBTaxxj3jjeN8W6Stdl25OtFqGqxEptTM6GsAiUO7JXqBZiC35z3eppeDpAudO26PqUHS2VPHduZJFHXfLUqSG6x2dpr3Xz8OFePvamq1aVAfnJiUH+4SenGYqP55/fvXtuUXa6FUjD9HRtZlrMF0UkAnwX+JQxV+6WyMVGp6Ep/5qmNxNqPI6098F8/vwomxvKsp5lXyub1cK6She9Gd6Z3zOe2Qxdqatw2Yuiw09hFwnJsuP9U4vuNKnxOEqqWG2iUXq2dyvmRQbNG1jyRXFHcyWDU4Eli2x2j/l48tQQu9qr+dQ7rmXCF+Lx44Mpj2HUG+CJk0Nsbijnf7/jGn70B3fQkcF1b5fWoOV1gJYp7zXGXAvcEf9432IHGWMeMMbsMcbsaWjITgHQoekAlWkI0GrLHPhDUWbT9OYnFImy/8IYt26uS8v5sq21xk3vxGxGdy13j/vSXgNNXVLpLo4MmgZoSQqEI5wd8i5apyWT1bgf2td12Uc+yHYXgYR8WYO2VGPqRPB+on960fufPDWEx2Hjbde1cPuWelqqXHxzf/eixy7ncPcEAvz9r1zP+27ZQFtN+jcGzGezWHDZc79BYwWpbGZi4WampRhjeuP/TgMPEZtKzTljDINT/vRMccbXQY2l6Rr2cu8kM8EIt26qT8v5sq2t2k0gHJ1rZp5u0aihZ3xW159lUKXLzuBUgEWS3QVFA7QknRuaIRw1XLVukQxamZ1pf7hkyhDkKkBz2Cz4cr2Lc3rpDNrVzfGdnP2TV9w37Q9xenCaG9bX4HZYsVqEd+5p55mzI3Nb7pMRNYbDXeNsbijPyIaApXgctpwHxyuY28wkIg5i62AfWXBMYjMTzNvMtNQJRcQmIvXxz+3AW4FX0j7yVZgOhPGHomnLoEH6dqIn1p/dsinzjbgzobUm9nfVM5Head+EYW8gVgNNpzgzpsJlYzYUYbrAs/4aoCXp1GBs8fdi28YT74Ry1VEg2+8SchagWVee4sxkxjESNYzNBCh3Lv6915Y5aK5yLbpR4HDXBFEDN26ombvtXTe2AfC9QwsTPUu7OOpj3Bdi9/rq1Aa/Rh5H7tf/LccYEya2Kekx4ATwrcRmJhF5e/ywzwN18c1MfwjM9RUWkU7g74FfE5EeEdkBOIHHROQl4AixDNxns/QtLWsoXkLg/9/emYc3dpf3/vtKsiRr8yJb3mfs2bdMkkkyScgkhMAkQwKkhVDCcpu2gbS3wKVwSy+UW5py21JKCxQItNxCWdMAaSHTEEIymUlCYJjMTDKZxcuMZ/G+W5Zk7cuvf5xzPBpbtmT7SOdIej/P48fy0dE5r7X89J53+6oZQVMrC3D4/BQ2NzgLKiWkJh6nFSYDYXgmPykynoGWf5TPRbHXoRW2yruI6R4NwGw0ZKz1Ua5AB7whbPAUrq1cCIGPP34SU7NRfOv3bsjrYMV0/OH43FVmIZFSnNo5CVPBKFJi6blTWzNIPgkh8Eq/F2tqbVfoZLbV2rCl0YXjfd6cbXi13wuz0ZB3KZr52M0mzev/srHSZib5vvZFDnudWvapifLF46xc/RJeaTbCWmFQpY42FEvgWN807r9B34owS6HIIuVLUWBuBhrXoOUNJbI87o8U9DtZbTiCliPdIwFs8DgyapjNRdBU7oTKxveP9OPx44M41DOBA12Fm1yiXYrTiGAsoVldgTIDbbEaNAC4tq0a58ZncW7sch3aa4M+jAei2LWmZsH+mxoc6B2fzen88WQKp4Z82NHiKrgIsM1s1HsXZ1kxJhdAuxaJ5i6XWrs6dbSHuicQiafyNvKlULRUV2I4T40CA/L3BKc484cSQRsr8kYBdtBypGc0sOhUbIfVJM3OKYCGW7o9f/1kJ27bVI92tw3/+EwPUqn8Oy5CCPgj2jUJCAGENar1m5SLhpeKoL33prWwVRjxpQPn5rb9+NgAKoyEna0Lo14bPQ4MzYRzik5dmgwimkhhZ2v18o1fJTazUfP6P+YyyhBONcZsAEon+upnOT51agR1DjNu7CjODk6Fljw2Cgx4Q6h3WjQRLC8XLkfQijvFye+QHPCF4hj1R7B5EQfNQASHxYSpPHX9zCeVEvjIY6/Caa3AP77zanx07yZ0jwbw5KmRjPsf75vGvi+9iL99qivrINVshONJxJNCsyYBQDvJocVUBNKptZvx+7d04GenRtA14scLZyfwo2MDuKqlKuNQyg0e6T11Poco2qhcd9SSB73NbNgtJsSSqbJphNE74/4o7GYjLCoNOnXbLfAGY0gkUys+RiyRwsHucdy1vRHGAo/gURtF03bIp/5F98C0NKSayR+WCiNsZmPRqwmwg5YD3aOSU7OYgwYAdrNx0flXanNhchbdowF8bO8m1DsteOvOZmxucOKLz57NuMB+/fkLuDgZxLdeuoi7v/xLvDYws+Jza6HDqWCZc9C0SbXlkuIEgA/cug5Oqwn/5z9O4g+/dwwbPU7cc1XzFfsoTQw9o1Iq9FwODtq4XxrxYc9y/nxQaZYcgZlQ+Shm6JmxQAQNLqtqx/M4LUgKgb5VlGn0jAUQjidxz1VLS50VAx5ZUWA4D41fFyeDmuj1lhsNLiunOMuBbvlLVBmjkAm7xVSQYbWPHunHIwclYezJQBSPHumHwUD46N6NuDgZxIGuKwefjvjCONg9hj/Y04GXP/UmqctwZOVRNF04aBrVQikz0LLVf1XZKvD+PetwctCH5qpKfPfB3XMOznxq7WYYDYRz45lnp6UzFojA49KmM06Reyqngcx6ZsIfvaLhZLUox8q1HjITp4d8cNvN2N1RnOM10jEZZEUBlctWgtEERv0RrK8v3sL1YsHjtGCiyCNo3MWZA92jAVTbKuBZYkG0W0yYLFAErW86CLvZCLdDKjJ99Eg/UkKgurICn/9FD/btuHwF+8OjA0gJ4N03rEGt3Yzr1tbgl+eyKdssji+UPwct21gMsyyJpFUn58Rs7l+K77+1Aykh8K4b2lC3xLgBo4FQ77Cgd2zpL8ZUSmDcH71iTEchsVmUCBo7aHpgLBDB1SrWIqY7aHflLg87RyyRQveoH++8vg2mEqmtaqmuxMmhGQghVOuQvzgZBACsq7PDy9HovOJxWXFycEZrM1ZFaXyS8kzPqB+bG5xLfkjtZmPBogt9UyGscduvsMdAhN0dtTg/EZy7Ck4kU/jh0QFs9DjwUu8kHj3SDyGkKJgSCVsuWkbQlMiVVuMeJgIR1Oc428luMeGjezfN1bIsRb3TkjXFOTQTRiyZUjWttRxsSgSNHTTNEUJy1pe6YFwu1gojqiorcqqFzMSZYR/iSVES6U2FlupKROIpVdf18xPS87uOI2h5p8FpwXiRqwmwg5aFVErg7Njsoh2cCnaLCaFYUjU9u8UIROKYCsbQ7l4o73Pd2hoYifCDI30AgOd7JjDii+CG9ssphzW10uNWKo6shxSnVt2EE4Eo6pzqt8Z7XBYMeENLvnfOymM7GjRLcUoRNLWmzTMrJxBNIBxPqp7uzuVCIRNCCLx4bgIepwU3ryvu7s10muVCfjXTnOcngjAQsDbD+s2oS4PLWvRqApzizIIyAmHLEvVnAOYKt6eCUbSa8/fh65uSHKu1tQvP4bRWYHuLC48fH8S6Oju+crAXHqflitq5pmppSvbAdAhXtSx/2KnioLk07eLU5gM3ORvDLXmYju5xWiGEdHW9Y5HX5KycAvU4L0fQCqnNqkTQOC2jPcrogAaXVdWO5nqnBa8NzCCVEjAsowuzZzSAMX8U913XuqzH6Z0GpwVGIgzPhFUbbXNhYhatNbaMHd2MuigXMOP+CFwqKG5oAUfQsnBmeHGJp3QKVUTdPx2CyUCLps5u7HAjEEngL544g9aaSvz/373+ipZ3k8GAlurKFUfQ/OE4iACnBp2EFqN2TQKxRAq+cHzJerKV4smhQPvsWAAuq2nRZoN8YzQQLCZ1ps0zq0OReVKzSQCQ3oehWBIj/uV1vr1wdgLVlRWq1sTpAZPRgIYqi6oRtAsTQayr5w7OQqBczBbzLDSOoGXheN80zCYDtjUvHUFzyEXUU3n+Ars0FURLTeWihbjtbhs+dfdWrKu3444tHhDRnJOpsKbWhl9fmEIimcp4nJODMzCbDLhP1opMxxeOw2kx4bGjA+r8Q8vAXCGnODVoElAcE6UxQ03cDjNMWTo5z44FNKs/U7BbTKrpNTIrZzxwOYJ2aVI99ZL0RoFcZ+1dmgyibzqEt+xsKvrZZ5loq7HhxMAMUkLAsMpGgVRK4OJkEDeVUBpYzyjlIMU8aoMjaFk41ufF1a1VsJiWjlwoKc58TJ5WiCVSGJ4Jo929+BUYEeEDt63DG7c2LNrU0FZrQzIlMOxb+MaNxpP46YkhfPqJ0whEFqazfOE4qjTSkDMZDKgwkiZNAkqHrtuufgTNZJA0Xs8t0smZTAn0js9q7qDZzEZOceqAMTnCpWaTgHQ86f21nFEbB7vHYTMbcf3a4h+tkYk1tTZEEylVojAj/gjC8SRH0AqER14vi3lYLTtoSxCJJ3F6yIfrclh85hy0PEbQBmdCSInM9WfLYY178UaBo31eROIphGJJ7H9teMH9WulwKtgtJoQ0dNDq89AkAEiST4sVaPdPhxBNpDRrEFCwmY3cJKADxgNRVFYYsw5MXi52sxE1toqcHbRzYwH0TsziDZs9BdeGLRRKU1XfdHDVx7ogd3DyDLTC4LCYYDcb5y5oipHS/FSpxMlBqXX8+hxmT1lMUnQnnynOwWmpFqJtlQ6ay1qBalvFAgctmRL4Ve8k2t12bGl04t9fXliEfn4iiLYa7TqQ7GYTZjXo4lRkvPIRQQOAjQ1O9E0FM3ZyKh2c6Q0CWmA3c4pTD4z5I2issqo2m0uBiLDB40BvDkOTU0Lg6TOjqLFV4MYSGEy7GLV2M+xm45zA+Wq4MCE5ees5glYwGlzWuZKAYoRr0JbgWN80AOQ0HJSIUGs351XuadAbQo2tQhWpn3V1dnSNBJBMibnakVNDPvjCcdx7dTM2NDjw6SfO4NSgD1fJIt/TwRj6p0N4741rVn1+YGVdiDazESENmgSmgtLrWqdyWklhW5MTKSHJ5VzTVn3FfWdlJQutVAQUOIKmD8YD6qoIpLPB48DTp0ez7ndycAYjvgh+p4QG02aCiLDGbZ/rnl8NFyZm4bCY8vbaMQupd1rmmmqKkdL9ZKnA8UterK+3o8aeW1qr1m7Ja4pzaCaMFpWiV9ubqxCOJ+cGJwoh8NK5CdQ7LNjU6MS917TAWmHAo2lRNGUqs1ot5yvBbjEhqEGTwORsDBaTYW4eWDYUrU3lJxvKKJSuDDJcnSN+rKm1Za2DzDc2+bmPJlgwXUvG/erqcKazvt4Bbyi+5IVmIpXCs51jaK6yYmfr8kf1FBtram2YCsZWXft6YVLq4FQ78sksTrFH0NhBW4RUSuB4v3dZxa91DnPeUpyz0QS8oThac+yuysYGjwMWkwGnh3wAgN6JWQz7ItizsQ4GIlRVVuAtO5ux/8TQ3Nyx1wZ8IMJcRE0L7BajJnPQJmejqHNY8ra4ttXYYDcbMzpop4Z8K5pZpzY2FkzXHCEExlRWEUhnszxO6PTwwvehwvE+L7yhOO7c3rjqzsZiQKlDW22a8/z4LNefFZgGlwVj/kjRqgmwg7YIFyZnMROK47r23LUPa+3mvEXQhrxS/VlrjToOWoXRgK1NLpwZ9iOZEnjh7ARcVhOuTUuv3XddK4KxJA71jAMAXhucwUaPQ/Xi5OVgM5tUd9ByiXZNzsZQl4cRGwoGA2FLk2uBg+YNxjDoDWvqFCvYWDBdc2ZlFYF8NYzc0F4Lm9mIZzszpznjyRQOdY9jTa0NGz3l4Wy01lTCQCtXXwGAUCyBYV8E6+q4/qyQeJxWROIp+CPFqSbADtoiHLvkBYCcGgQU8uqgzYRAQE7ajrmyQ05z/vLcBC5MBHHLhror6kluaK9FncOMn58ehRACJwdnNE1vAlJnzmrmoM13xnKtg5uSI2j5ZGuTE90jgSuu9k7JEU49RNBY7kl7xtJUBPKBtcKI12+qxzNnxpBKLYw6vHxxGv5IAnu3LT7Gp9SoMBrQvIrh3gBwon8GALC9Zel5moy6KHW7E0U6C40dtEV4+dI0au1mdCzjisdtN2M2mkAkrn6NzqA3jDqHRVWJkI0NUprz2c4xWCsMV2h2AtL0+Du3N+JQ9zjOTwQxORvD1fMK2AuNzVz4FOejR/oxMB3CTDieV3mlrU0uBKIJDHovTy5XHLQdzdo7aDaLvuWeiGgfEfUQUS8RfSLD/RYi+qF8/xEiape3u4noEBHNEtFX5z3mOiI6JT/my6SxV5IvFYF09u1oxHggilcHZq7YHo4l8cLZCXTU2csuVddWa8OgN4R4MrWixx++MAWjgRassUx+aSjyWWjsoGVACGncxM3r3cu6SqyVRzDkI4omNQioFz0DpCvDLY1OCAA3rXNf4fwp0SWryYhQLImvHDwHALhGBxE0taSeBqZDeKZzNGt9ghACs9FE3lO7SqNAZ1qa8/SQD2vdNs2GA6ej1KBN63DUBhEZATwC4M0AtgF4NxFtm7fbgwC8QogNAL4I4HPy9giAvwDwpxkO/XUAHwCwUf7Zp771uZOuIpAv3rDFgwoj4ZkzV6Y5HznUi9loAnu3NuTt3Hqlw21HPCnwqhwJWy6Hz09hR0sVnEWqCVmsXHbQOIJWMvSOz2LMH8WtG+qW9bhaudtTbQfNF44jEEmoVn+Wzu4ON5qqrHjd+sz/a0edHZUVRjxxYhhmo2GuiFgrbGYTIvEUEiu8klVIJFP40bEBPN8zgdEsH95wPImUQN4dtC2NThBd2cl5clAfDQLAZQdNpynO3QB6hRAXhBAxAI8BuHfePvcC+I58+3EAbyQiEkIEhRAvQXLU5iCiJgAuIcRvhOTFfxfAb+Xzn8iG8kWTTwfNZa3Azevr8PSZyxcv58YC+JcXz+Patmq0l2Ed1QaPAwYCnuseW/ZjQ7EEXhucwc0s8VRwlGaaYu3kZActA788NwkA2LPxSqclW+2SUkSudifnkFeqfVCrgzOdjjo7PnzHxkWdD6OB5nRItzW7NJ8Ybpc1T0OrTCP/6vzU3Os0X6t0Pkp7vRrz55bCZjahw22fc9C8wRiGZsK6cdBMBgMc+tXjbAGQLhA7KG/LuI8QIgHAB2Cpb80W+ThLHbOgjPmjsJnVVxGYz13bG9A3FULPWACplMCf/+QU7BYT3nxVU17Pq1esFUZ01NlxsGt82Y89dsmLeFLg5vXsoBUau8UEh8VUtBE0HlSbgZd6J9FRZ8eLZyeX9bjLETR1vfXBmTAMBDRWqe+g5cKOZheO93kXDFDVAsVJCkWTcK0wXeALx3GoexxbG52IJFI4PeTDm5ZI2ygOWiG6V7c2uebqzvTUIKBQY6/QawRNU4joIQAPAcCaNeoMcs7EeCB/M9DS2butAf/3p6fxwLdehsdpxakhH/7+HTuRyNA4UC5saXThZ6dG0D8VmpPLy4XDF6ZgMtCyGs4Y9fC4LKpoqWoBR9DmEUuk8JsLU9izzPQmcFkGaEplwfS+qRAaXVbNolfrPQ7cfVUj3nZNsybnT0dJs61maOQzZ0aREgL37GzG9mYXxgNRTCwRAg/K0lKFcdCc6J8OIRCJzzlo23XkoNXazJjWZ5PAEIC2tL9b5W0Z9yEiE4AqAFNZjtma5ZgAACHEN4QQ1wshrq+vr1+m6bkz7s+fikA6HqcVn7l3B25or4W1woB37GrFfde1Zn9gCbNFLu84uMw05+HzU7i6rTrvEXgmMx6nBePcxVkavNrvRSiWxC0rcNBclSaYDOrqcUbiSfRNBbGxQbvaL5PBgK+99zrsWqP9FaBdnsW1UrmnWCKFU0M+XLe2BrV2M7bL3ZFnhn2LPmY2IjkkDmv+F1glnfxnj5/Ega4xtLttmorTz6faZsaMPlOcRwFsJKIOIjIDuB/A/nn77AfwgHz7PgAHxRIdIkKIEQB+IrpJ7t78XQBPqG967hQqggYA/+Omtfjqe3bhx3/0Ovzj71wNg6E8xmoshtthwfp6O57rzj3NORtN4NSQj+vPNKTBZeUuzlLhV72TMBBWVC9ARKixmzGtYgStd3wWKQFs0tBB0xPKVehKI2jnJ2aRSIk5x6yqsgJtNZU4vZSDFk2CcDl6l09u21iPB/d04KXeSbzar/3cuflIerP6c9DkmrIPAfgFgC4APxJCnCGizxDR2+TdvgnATUS9AD4GYG4UBxFdAvAFAL9HRINpHaB/DOBfAfQCOA/g54X4fzKhqAg0sJajZrxxawOOXJjOef15+eIUkimuP9MSyUErTjUBjrnO45e9k7i6rXrFUQu3XV25p7NjAVgrDHNyI+XOXJNAdGVNAl0jflhMBrTXXX4+tzdX4ekzo5gOxubqCNOZjSZgs5gWlbVRczaayWjAX7xlGz5+12Y83zOOHTpKbwLK+zsKIYTuBpUKIZ4C8NS8bZ9Oux0B8M5FHtu+yPZjAHaoZ+XKCcgqAp48qQgw2bljiwffePECXuiZwD07l26YSKUEvnboPGrtZlzH9Wea4XFaEE1IagJ6ykbkAkfQ0vBH4jg56FtR/ZmCpCagTjhVCIGzYwFs8DhhLPP0goISQVvJLLRUSqBnNIBNDU6YDJff+luapOikIhw/n2A0AYelsELl1goj9u1oQmuNvhzzOqcFkXhqVWoOzMoYz7OKAJOd69fWoK22Eo8c6s2otJDOf746hGN9Xnxi3xZVB4wzy8Mjf17Gi7CTkx20NI5enEYyJRadCZYLbodFtQha10gA/kgCmxvKa2r3UlTLV0Ar6SQ8NeRDIJqYK/ZVqHNYYDYZMOLL/AEuxJDaYkGRu5qcLc6ajmJG+YLxONlB0wqT0YA/vXMzOkf82P/a8KL7hWNJfPapLly7prrsmyu0RikJKMY6NP7WSePX56dgMRlw7ZrqFR/D45RaeleaAkpPlz0vi5Rr2SCgN2psZpgMlHXwYDiWXBB1fK5rDARg87zn00CERpcVI74wMjEbzc+Q4GLELc/6m5yNYa27/AaWasmY3InGKU5teevOZvzLCxfwD8/04M1XNcJiujI6Fk0k8dMTQ5gOxfCdP9hd9s0VWqNEnIuxkzOnCFo+NO70yK/PT+H69ppVhaMbXBaE40kEVNCLPDsWQHO1dcXzvkoRg4FQ77RkvRr65ksX8MUDZ3F66HLx/4Gucax12+Y0JdNpqrJi1Je5kJQjaJep5wiaZnCKUx8YDIRPvHkLBr1hPHKwd07VRCmh+PJz53BqyIePvWmT7mpIyxHlgqYkI2hpGnd7IU3SPkpE+4UQnWm7zWncEdH9kDTu3oXLGnc7oJNC28WYDsbQNeLHx+/avKrjzGl/+SKrcqxmown0T4fw+k35m6lUrHhc1iWvhsb9EQz7IjAS4Z3/fBh/+Pp1OD3kQ+eIH/u2N2Z8TFNVJY5cnF4gBB6OJRFLpNhBk7kcQSu+xa7YKZSKAJOd2zbV401bPfjywV48fnwQr9tQh5fOTWLUH0Gdw4KHbl2HD79xo9ZmMpAUWpxFqiaQyyd9TuMOAIhI0bhLd9DuBfCwfPtxAF9VNO4AvEREG9QzOT/85oI0r/KmVc6ruSzOGl1VavLk4AxSArobs6AHPE4LBqZDi96viI3/0evX41fnJ/GlA+fQUl2JP7ilY9Fu2KYq6XWbn+ZUHBH+UpTI1zBmJjtjBZyBxmTnn993HQ52j+O7h/vw81MjuGVDHW6trMOOlipUGLm8W094XJYlh5HrlVy+dTJp3N242D5CiAQRKRp3OWklFUomZSkOn5+C3WzEztbVhaQb5QU0mwB3Nk4MzKC5ysoLcgY8TguOXZpe9P4zw3601VSipaYSj//RzRj1R9BSXQkiWnQkRoPLCgIWNAoMyDqoxdaenS/MJgNcVhNH0DRgwh+dE39mtMdkNODO7Y24My0qr+bIHUY9PE5rUUbQdOHmF0omZSl+fX4SuztqV33lczmCtvI3w0QgikFvWBfal3qkwWWFNxRHLJFacN9MSBIYVwbRmowGtNbYsjZsmE0GuB0WjM5z0JQatqY8CNUXK3VOC0fQNGAsEJkbGcAwTO40uCyrDppoQS7eSD407nTFmD+C8xPBVY3XUKg0G+Gyri7ffWLACwKwkx20jChRhIkMUZwzw1J6U5FMWg5NVQs7OU8P+VFVWcEpzjTq7JaMzz2TP1IpgZGZCJqr2EFjmOXSVF2JMX8k6+w6vZGLg6a6xp3eOHJRSpettv5MQZGWWAkpIXBiYAYbPA7u3lyEy105C5/jM8N+NLqsc/O6lkNTlRSZ80cuNwqcHvahmaNnV1DnNGOKHbSCMjkbRSyZ4nEvDLMCWqorEU+KrOOZ9EZWBy2PGne64ZU+LyorjNjapM68scaqlYuzDkyH4A3FOb25BMqgzvF5z7E3GEPfVHBF0TPgcqNA90gAgNRJe3EyiJZqjlqk47ZbMMkpzoIy4JUiuy3soDHMslEubAa9izeX6ZGc8jb50LjTE6/2e9FYZcWPjg2qcjyP04re8Zz6IxbQOeKHkQhbm1bmZJQDSgRtYt6ojZNDPggA6+pWNkC1sUr6EHcO+7C7oxadw34IAY6gzaPOYYEvLNUAmk26KGMteYZmJAdNb9JfDFMMKJ+boZkwrtfYluVQ9qtrJJ7EmWG/qmLkjVUWjAeiK8p394wG0F5nY+22JXDbLTDQwsGDnXL9WVPVyhwql9UEm9mIk3JjgNIgwA7aldQ5pVlo0ypJmjHZUa78W/i9yDDLRvncDHozq8XolbJ30E4O+pBICVUdtAaXFcmUwOQyRdMHpkMYD0SxuZGjZ0thNBDqHJYFw2rPDPtQY6tApXllzi0RYUujCz8/NQpfKI7Twz7UOy1cCzgPZRYaj9ooHEPeMGpsFbBzswrDLJtKsxF1DnNppjhLEWVezYtnJwAAbSo7aIBUI7UcYeNDsvbmFtbezEqDy7qg4LNzxL/i6JnCLRvceKXfix+83IczQ37sWGE9WylT72Q1gUIz6A1z/RnDrIKW6sqii6CVrYOm0DcdgttuVnWMguKgjfoiy9JiO9g9DrfdjLpVDqMsh2GJHqfliqGyoZhU0H/HZs+qjttUVYk9G+rwb7+6hKnZKO7a3rBaU0uOyxE0TnEWiqGZMDbUO7Q2g2GKltYaG7pklZlioaxTnEII9E+HVE1vAtJQPEAaLDmfR4/0X/GjEI4lcfj8FDY1cvQsFzyuK1Oc3aMBCLHy+rN03n9rByYCUaQEsJ3FjhegXEDwqI3CIITAoDfEETSGWQUtNZUYmgmjiCaAlbeD5g3FEYwmsMatroNW77CAMhSxL8Wvz08imkhxejNHPE4rpoIxJJKSmsBcg4AKIzFev6keGz1StGI5EdBywW42wmIycIqzQEwHY4jEeQYaw6yG1ppKRBOpohqyXdYOWv90EABUj6CZjAbUOSwY8+U+rPa57nHYzEZ0rHBERLnhcVkgxOU0W+eIHy6rCdUqaGYSET51z1a8fVcLT27PAJHUpMFyT4VBqZvhDk6GWTnK52eoiOrQyroGrW8qBLPJkBdB8kaXNWOKMxOxRAo/PzWCN2zxwLRKLdByQWm+GPNH0FhlReewH9uaXVk1N3Pl9s0e3L7KerblML9u8D03rinYuVdCncNcVFeixQzPQGOY1aN8fga9YVy7pkZja3KjrL2B/ukQ2moqYVDpSz2dBtdC4e3FePHsBLyhON5+bYvqdpQqSp3feCCKZEqge9SPbU2cjiwUeoygEdE+Iuohol4i+kSG+y1E9EP5/iNE1J523yfl7T1EdFfa9ktEdIqIThDRsQL9K1cwNwONU5wMs2KUz49ywVMMlG0ELRpPYtQXyVuUpMFlxSv9Mznt+5MTQ6i1m3Hbpnr8WCU1g1JnTu4pEMHFySAi8RS2NbsQS6Q0tky/qNnd63aYcUoe5KsHiMgI4BEAewEMAjhKRPuFEJ1puz0IwCuE2EBE9wP4HIB3yfJz9wPYDqAZwAEi2iSESMqPe4MQYmXSICow5A3DaTWhSoX0PcOUKw6LCdW2iqKahVa2DtqANwwBYK3KDQIKDS4rpoMxRBNJWEyLD071R+I40DmG+29oQwWnN3OmzmGea8Q4dkkSu9/W5MKJgZlVH7scxpSsljqHBdPBGFIpAYNB/Qj0CtgNoFcIcQEAiOgxAPcCSHfQ7gXwsHz7cQBfJSknfi+Ax4QQUQAXZU3h3QAOF8j2JRn0hrn+jGFUoNhmoZWtR9A/HQQBaMtTXUejK7Og93yePj2KaCKF3+L05rIwGQ1w28149Eg/PvmTU2irrcQGD8+JKhR1DgsSKQFfOK61KQotAAbS/h6Ut2XcRwiRAOAD4M7yWAHgGSI6TkQPLXZyInqIiI4R0bGJiYlV/SPzGZoJc/0Zw6hAa01lUTUJlLGDFoLHZVmxLFA2lNEd3aOBJff7yStD6Kiz45q26rzYUcqsddvhC8fwgVvX4ckP38rC3QXE7SgbNYE9QohdAN4M4INEdFumnYQQ3xBCXC+EuL6+vl61k0sz0MI8YoNhVKCl2oZBb/HMQivLFGcqJQ2ovaqlOm/nuHZNNWxmI57vGcfebZmn0R+5OIXDF6bwZ/s2q9Z9WE488p5dEBCqDKfVG3pPsyoC8oMzYWzUx+y+IQBtaX+3ytsy7TNIRCYAVQCmlnqsEEL5PU5EP4GU+nwxH/9AJvzhBGajCXbQGEYFWmsqEY4n4Q3FUWs3a21OVsoy5NA7MYtIPIW1Ks8/S8diMuJ16+vwfM9ERm+9e8SP/SeGcccWDx66dV3e7ChlGqusJemcFQNK7Wb/lG4Kbo8C2EhEHURkhlT0v3/ePvsBPCDfvg/AQSF9OPcDuF/u8uwAsBHAy0RkJyInABCRHcCdAE4X4H+ZY0Dp4OQaNIZZNUon58C0btatJSlLB+3YJS8AqK4gMJ/bN9djaCaM8xOzV2wf9Ibw70f70Vxdia+8+1qefcYUHfUOC2xmIy5NBbU2BcBcTdmHAPwCQBeAHwkhzhDRZ4jobfJu3wTglpsAPgbgE/JjzwD4EaSGgqcBfFDu4GwA8BIRvQbgZQA/E0I8Xcj/q0cukeD6SoZZPcrnqGds6dIjvVCWKc7jfV7YzEa48xzivH2zVIvyfM8ENnikNJA3GMN3DvfBYTHhd29eC7uKIu0MUyiICGtqbXqKoEEI8RSAp+Zt+3Ta7QiAdy7y2L8B8Dfztl0AcLX6luZO14gfFpOBFUYYRgXa3XZUVhiLRjS9LL2D433TWFtry3vdV2uNDRs8DjzfM4H337oOvnAc3z58CclUCg/sWQ+nlecaLZdim7hfyrS77Tg3XhxXosVK16gfmxudHGVnGBUwGgibG51F46CV3af+4mQQl6ZCWFdfmJTB7Zvq8fLFaXQO+/Hub/wG07MxvPfGtfDkQV6KYQrJWrcNA9NhJFPF0RFVbAgh0Dnsx9ZGl9amMEzJsLXJha6RQFF0cpadg/Zs5ygAaahpIbh9swexZApv/epLGJoJ4303rcH6AjmHDJNP1rrtiCVTGPXnJmnGLI8xfxTeUBzbmtlBYxi12NbkhC8cx3COUoxaUjYpTiU19oMj/Wh0WVFToBbbGzpqUO+0oLWmEl99zy680KPuEEuG0Qqlk7NvMshdhnlAScNsLdDFJMOUA8oFT9ewX/frVtk4aAAwG02gfyqEN2zJj/5mJiwmI174+O2wmox6kcRhGFWYc9CmQ3idxraUIp2yg7alSRdz5himJNgslwx0jfjxpkVmlOqFsnLQekb9ECj8FanNXFZPM1MmNFVVwmw06GbURqnROeJHW20lXNxMxDCq4bCYsNZtQ9eo/hsFyqoGrXMkgKrKCjRXcYE+w6wWo4HQWlupq1EbpUTXCDcIMEw+2NroQucwO2i6IZZIoXc8gK1NTpZVYhiVaHfbcYkdNNUJxRK4OBnkBgGGyQPbml3omw4hGE1obcqSlI2D1jniQzwpsK2pSmtTGKZkWFNrQ99UsCha1ouJntEAhOAGAYbJB1ubXBAC6B7V9xzHsnDQhBD45blJ1DstWFfPE7kZRi3a3TaEYklMzsa0NqWk6BqRvjgKNQ6IYcqJrXLjjd4H1paFg/ZS7yRGfBHcuqEOBk5vMoxqrJUliPq4UUBVjl6aRrWtAq01+h4DwDDFSEt1JWrtZhy9NK21KUtSFu2F//LCBTitJlzTVq21KUXLfIklhgGAtbXyqI2pEK5vr9XYmtIgEk/i2c4x3H1VI9fLMkweICLcua0B//XaMCLxJKwVRq1NykjJO2inh3x4qXcS+7Y3sp5dCcKOo7a01thgNBDOsianarx4dgKz0QTu2dmstSkMU7Lcs7MJjx0dwPM9E9i3o1FrczJS8h7Llw6cg8Niwu4OvrpnGLUxmwzY3V6LA51j3CigEk+eHEGNrQKvW+/W2hSGKVluXueG227GkyeHtTZlUUraQXvh7AQOdI3hQ3ds0G0Ik2GKnbt3NuH8RBBnx2a1NqXoCceSONA1hn07mlDBEX+GyRsmowH7djTiua5xhGL6HLdRsitAPJnCZ/7rDNrdNvz+Le1am8MwJctd2xtABPzs1IjWphQ9h3rGEYol8ZadTVqbwjAlzz07mxCOJ3Gwe1xrUzJSsg7ad359Cecngvj0W7fBYipM9OzRI/1X/DBMOeBxWrG7vRZPsYO2ap48OYw6hxk3ckkGw+SdGzvcqHda8ORr+ly7StJBOzPswxeePYvbN9fjji36FkNlmFLgnp1N6B2fxdkxbhZYKS+dm8TPT4/it65p4YYmhikARgPh7de24Bedo3jx7ITW5iyg5Lo4x/wRPPjtY6iqrMDfv2OnprZwFI0pF/btaMRf7j+Dn50cwaa9Tq3NKTrG/BF85LFXsaHegY/duUlrcximbPiTN23C8z0T+JMfnsDP/tceNFXpZ/ZgTpdpRLSPiHqIqJeIPpHhfgsR/VC+/wgRtafd90l5ew8R3aWi7QsY9UXw4HeOIhCJ45sP3ACPi0XRGaYQeJxW3NBei0df7sfpIZ8mNuRjncp2TDXwheP48KOvIhRL4uvv2wWbueSumxlGt1Sajfja+3YhGk/iw4++Cl8orrVJc2R10IjICOARAG8GsA3Au4lo27zdHgTgFUJsAPBFAJ+TH7sNwP0AtgPYB+Br8vFUIxJPomc0gM/+vAuv//whnB2dxVfecy2LDDNMgXn4rdthMhDu++df44kTQwU9dz7WqRyPuWKmZqP4/C+6sefvDuLlS9P47NuvwgYPRx8ZptCsr3fg796xE8f6vLjlcwfx9093Y2o2qrVZOaU4dwPoFUJcAAAiegzAvQA60/a5F8DD8u3HAXyVpBHY9wJ4TAgRBXCRiHrl4x1Ww/i9X3gBvROzEAIgAn77mhZ8dO8mtMnTzYsBToPmF35+C8e2Zhf+68N78Mc/eAUfeewEGlxW3LSuYLO88rFOIYdjrpgvHTiH7x/pw907mvDHb1iP7c1VahyWYZgV8Narm7G+3oFHnu/F1184j5lwHH/721dpalMuDloLgIG0vwcB3LjYPkKIBBH5ALjl7b+Z99iW+ScgoocAPCT/OUtEPTlZP48vyj+LUAdgciXHzRN6skdPtgBsz1Joast7F25a1J6bP7esQ69diT1p5GudynZMAKtbw74m/yxBwV7zDK/vfJa0JYfHr+r88+7P+rys1p5lHG+BLWqfOxtp5yvK9SrPr9Wyn5PPyj85str1KyO6KHYQQnwDwDfyeQ4iOiaEuD6f51gOerJHT7YAbM9S6MkWQH/2aEU+1zA9PcdsS2bYlszoxRa92LFccmkSGALQlvZ3q7wt4z5EZAJQBWAqx8cyDMOslnysU7x+MQyjGbk4aEcBbCSiDiIyQyqm3T9vn/0AHpBv3wfgoJCE+fYDuF/unuoAsBHAy+qYzjAMM0c+1qlcjskwDJMXsqY45VqNDwH4BQAjgG8JIc4Q0WcAHBNC7AfwTQDfk4trpyEtZJD3+xGkotoEgA8KIZJ5+l+ykdcU6grQkz16sgVge5ZCT7YAOrEnX+tUpmMW+n+DTp5jGbYlM2xLZvRii17sWBYkXUAyDMMwDMMweoH1RBiGYRiGYXQGO2gMwzAMwzA6o+QdtEJItWQ5/7eIaJyITqdtqyWiZ4nonPy7poD2tBHRISLqJKIzRPQRrWwiIisRvUxEr8m2/JW8vUOW4umVpXnM+bZlnl1GInqViJ7U2h4iukREp4joBBEdk7dp8v4homoiepyIuomoi4hu1vK9XA5ouX7pZe3iNSurTbpYr3itUp+SdtAoz1ItOfJtSPIx6XwCwHNCiI0AnpP/LhQJAP9bCLENwE0APig/J1rYFAVwhxDiagDXANhHRDdBkuD5oizJ44Uk0VNIPgKgK+1vre15gxDimrQ5Plq9f/4JwNNCiC0Arob0HGn5Xi5pdLB+fRv6WLt4zVoaPa1XvFapiRCiZH8A3AzgF2l/fxLAJzWwox3A6bS/ewA0ybebAPRo+Bw9AWCv1jYBsAF4BdKk9kkApkyvYQHsaIX04b0DwJMASGN7LgGom7et4K8VpJlhFyE3FmlpS7n86GH90uPaxWvWFTboZr3itUr9n5KOoCGz/MsCqSkNaBBCjMi3RwE0aGEEEbUDuBbAEa1sksPzJwCMA3gWwHkAM0KIhLxLoV+zLwH4MwAp+W+3xvYIAM8Q0XGS5IQAbV6rDgATAP5NTqf8KxHZNbKlXNDj+qXp681r1gK+BP2sV7xWqUypO2i6R0jufMFnnRCRA8B/APgTIYRfK5uEEEkhxDWQrgR3A9hSiPNmgojeAmBcCHFcKxsysEcIsQtSmuuDRHRb+p0FfK1MAHYB+LoQ4loAQcxLEWj1Xma0odCvN69ZV6LD9YrXKpUpdQdNr1ItY0TUBADy7/FCnpyIKiAtdD8QQvynHmwSQswAOAQpJF9NkhQPUNjX7BYAbyOiSwAeg5Q2+CcN7YEQYkj+PQ7gJ5C+ELR4rQYBDAohjsh/Pw5pEdT0fVPi6HH90uT15jUrI7par3itUp9Sd9D0KtWSLjnzAKSaioJARARponqXEOILWtpERPVEVC3froRUV9IFadG7r5C2AIAQ4pNCiFYhRDuk98pBIcR7tbKHiOxE5FRuA7gTwGlo8FoJIUYBDBDRZnnTGyFN3tfsvVwG6HH90mKd4DUrA3par3ityhNaF8Hl+wfA3QDOQqoT+JQG5/93ACMA4pA8+wch1Qk8B+AcgAMAagtozx5Iod2TAE7IP3drYROAnQBelW05DeDT8vZ1kLQQewH8GIBFg9ftdgBPammPfN7X5J8zyvtXq/cPpK61Y/Lr9VMANVq+l8vhR8v1Sy9rF69ZOdml6XrFa1V+fljqiWEYhmEYRmeUeoqTYRiGYRim6GAHjWEYhmEYRmewg8YwDMMwDKMz2EFjGIZhGIbRGeygMQzDMAzD6Ax20BiGYRiGYXQGO2hMXiCiS0RUl6djNxPR44vc9zwRXZ+P8zIMUx7w+sXoAVP2XRhGXwghhnF5UjbDMEzRwOsXkyscQWNWjSzz8TMieo2IThPRu+S7PkxErxDRKSLaIu9bS0Q/JaKTRPQbItopb3+YiL5HRIeJ6BwRfWCJ87UT0Wn5diURPUZEXUT0EwCV+f5/GYYpHXj9YvQKO2iMGuwDMCyEuFoIsQPA0/L2SSHELgBfB/Cn8ra/AvCqEGIngD8H8N204+yEJPh7M4BPE1FzDuf+nwBCQoitAP4SwHWr/m8YhikneP1idAk7aIwanAKwl4g+R0S3CiF88vb/lH8fB9Au394D4HsAIIQ4CMBNRC75vieEEGEhxCQkwd/dOZz7NgDfl493EpL2GsMwTK7w+sXoEq5BY1aNEOIsEe2CJGD810T0nHxXVP6dRG7vtfnCsCwUyzBMXuH1i9ErHEFjVo0cyg8JIb4P4PMAdi2x+y8BvFd+3O2Q0gh++b57ichKRG4AtwM4msPpXwTwHvl4OyClGRiGYXKC1y9Gr3AEjVGDqwB8nohSAOKQ6ioytpEDeBjAt4joJIAQgAfS7jsJKTVQB+D/yd1O2fg6gH8joi4AXZDSEQzDMLnC6xejS0gIjsIy2kNEDwOYFUL8g9a2MAzDLAdev5h8wClOhmEYhmEYncERNEa3ENFVkDum0ogKIW7Uwh6GYZhc4fWLWS3soDEMwzAMw+gMTnEyDMMwDMPoDHbQGIZhGIZhdAY7aAzDMAzDMDqDHTSGYRiGYRid8d8xb2ExrWa1mgAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 720x720 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "fig, axes = plt.subplots(2, 2, figsize=(10, 10))\n",
    "sns.distplot(sales_train_by_month_clipped['item_id'], ax=axes[0][0])\n",
    "sns.distplot(test_set['item_id'], ax=axes[0][1])\n",
    "sns.distplot(sales_train_by_month_clipped['shop_id'], ax=axes[1][0])\n",
    "sns.distplot(test_set['shop_id'], ax=axes[1][1])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Pretty different, specially the shops. Let's look at the last month of the train data, since time might explain this effect."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:xlabel='shop_id', ylabel='count'>"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 720x720 with 4 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "sales_train_last_month = sales_train_by_month_clipped[sales_train_by_month_clipped['date_block_num'] == sales_train_by_month_clipped['date_block_num'].max()]\n",
    "\n",
    "fig, axes = plt.subplots(2, 2, figsize=(10, 10))\n",
    "sns.countplot(data=sales_train_last_month, x='item_id', ax=axes[0][0])\n",
    "sns.countplot(data=test_set, x='item_id', ax=axes[0][1])\n",
    "sns.countplot(data=sales_train_last_month, x='shop_id', ax=axes[1][0])\n",
    "sns.countplot(data=test_set, x='shop_id', ax=axes[1][1])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Yup, big random dataset on test set. Also there's some shops and items that are not present. Let's check the intersections"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 61,
   "metadata": {},
   "outputs": [],
   "source": [
    "def intersection_cnt(a, b, idxs):\n",
    "    a_uniques = a[idxs].drop_duplicates()\n",
    "    b_uniques = b[idxs].drop_duplicates()\n",
    "    intersection = pd.merge(a_uniques, b_uniques, on=idxs)[idxs].drop_duplicates().shape[0]\n",
    "    return {'intersection_cnt': intersection,\n",
    "            'on A and not on B': intersection / a_uniques.shape[0],\n",
    "            'on B and not on A': intersection / b_uniques.shape[0]}"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 62,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'intersection_cnt': 111404,\n",
       " 'on A and not on B': 0.2626684648829116,\n",
       " 'on B and not on A': 0.520093370681606}"
      ]
     },
     "execution_count": 62,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "intersection_cnt(sales_train, test_set, ['item_id', 'shop_id'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 63,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'intersection_cnt': 28680,\n",
       " 'on A and not on B': 0.9095810472233675,\n",
       " 'on B and not on A': 0.13389355742296918}"
      ]
     },
     "execution_count": 63,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "intersection_cnt(sales_train_last_month, test_set, ['item_id', 'shop_id'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 64,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'intersection_cnt': 4737,\n",
       " 'on A and not on B': 0.21722382721144587,\n",
       " 'on B and not on A': 0.9288235294117647}"
      ]
     },
     "execution_count": 64,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "intersection_cnt(sales_train, test_set, ['item_id'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 65,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'intersection_cnt': 3991,\n",
       " 'on A and not on B': 0.7372990947718455,\n",
       " 'on B and not on A': 0.7825490196078432}"
      ]
     },
     "execution_count": 65,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "intersection_cnt(sales_train_last_month, test_set, ['item_id'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 66,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'intersection_cnt': 42, 'on A and not on B': 0.7, 'on B and not on A': 1.0}"
      ]
     },
     "execution_count": 66,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "intersection_cnt(sales_train, test_set, ['shop_id'])"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 68,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'intersection_cnt': 42,\n",
       " 'on A and not on B': 0.9545454545454546,\n",
       " 'on B and not on A': 1.0}"
      ]
     },
     "execution_count": 68,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "intersection_cnt(sales_train_last_month, test_set, ['shop_id'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Benchmark submission\n",
    "\n",
    "The sample submission uses the counts for the last month, and imputs the missing values with zero."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "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>date_block_num</th>\n",
       "      <th>shop_id</th>\n",
       "      <th>item_id</th>\n",
       "      <th>item_cnt</th>\n",
       "      <th>ID</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>4.418487</td>\n",
       "      <td>31.642857</td>\n",
       "      <td>11019.398627</td>\n",
       "      <td>0.255649</td>\n",
       "      <td>107099.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>11.237777</td>\n",
       "      <td>17.561933</td>\n",
       "      <td>6252.644590</td>\n",
       "      <td>1.089856</td>\n",
       "      <td>61834.358168</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>30.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>16.000000</td>\n",
       "      <td>5381.500000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>53549.750000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>34.500000</td>\n",
       "      <td>11203.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>107099.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>0.000000</td>\n",
       "      <td>47.000000</td>\n",
       "      <td>16071.500000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>160649.250000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>33.000000</td>\n",
       "      <td>59.000000</td>\n",
       "      <td>22167.000000</td>\n",
       "      <td>20.000000</td>\n",
       "      <td>214199.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       date_block_num        shop_id        item_id       item_cnt  \\\n",
       "count   214200.000000  214200.000000  214200.000000  214200.000000   \n",
       "mean         4.418487      31.642857   11019.398627       0.255649   \n",
       "std         11.237777      17.561933    6252.644590       1.089856   \n",
       "min          0.000000       2.000000      30.000000       0.000000   \n",
       "25%          0.000000      16.000000    5381.500000       0.000000   \n",
       "50%          0.000000      34.500000   11203.000000       0.000000   \n",
       "75%          0.000000      47.000000   16071.500000       0.000000   \n",
       "max         33.000000      59.000000   22167.000000      20.000000   \n",
       "\n",
       "                  ID  \n",
       "count  214200.000000  \n",
       "mean   107099.500000  \n",
       "std     61834.358168  \n",
       "min         0.000000  \n",
       "25%     53549.750000  \n",
       "50%    107099.500000  \n",
       "75%    160649.250000  \n",
       "max    214199.000000  "
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "solution_benchmark = pd.merge(sales_train_last_month, test_set, how='right', on=['item_id', 'shop_id']).fillna(0)\n",
    "solution_benchmark.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [],
   "source": [
    "solution_benchmark[[\"ID\", \"item_cnt\"]].rename(columns={\"item_cnt\": \"item_cnt_month\"}).to_csv(os.path.join(TMP_DIR, 'benchmark-submission.csv'), index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Successfully submitted to Predict Future Sales"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 2.14M/2.14M [00:06<00:00, 336kB/s] \n"
     ]
    }
   ],
   "source": [
    "%%bash\n",
    "kaggle c submit -f ${TMP_DIR}/benchmark-submission.csv -m 'benchmark reproduction' competitive-data-science-predict-future-sales"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The score is as expected: 1.16777. Let's try and beat that with some simple heuristics. Instead of replacing NaNs with zeros, let's replace them with the mean of the other items counts."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "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>date_block_num</th>\n",
       "      <th>shop_id</th>\n",
       "      <th>item_id</th>\n",
       "      <th>item_cnt</th>\n",
       "      <th>ID</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>6.072183</td>\n",
       "      <td>31.642857</td>\n",
       "      <td>11019.398627</td>\n",
       "      <td>1.909344</td>\n",
       "      <td>107099.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>10.587572</td>\n",
       "      <td>17.561933</td>\n",
       "      <td>6252.644590</td>\n",
       "      <td>0.874654</td>\n",
       "      <td>61834.358168</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>1.909344</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>30.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>1.909344</td>\n",
       "      <td>16.000000</td>\n",
       "      <td>5381.500000</td>\n",
       "      <td>1.909344</td>\n",
       "      <td>53549.750000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>1.909344</td>\n",
       "      <td>34.500000</td>\n",
       "      <td>11203.000000</td>\n",
       "      <td>1.909344</td>\n",
       "      <td>107099.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>1.909344</td>\n",
       "      <td>47.000000</td>\n",
       "      <td>16071.500000</td>\n",
       "      <td>1.909344</td>\n",
       "      <td>160649.250000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>33.000000</td>\n",
       "      <td>59.000000</td>\n",
       "      <td>22167.000000</td>\n",
       "      <td>20.000000</td>\n",
       "      <td>214199.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       date_block_num        shop_id        item_id       item_cnt  \\\n",
       "count   214200.000000  214200.000000  214200.000000  214200.000000   \n",
       "mean         6.072183      31.642857   11019.398627       1.909344   \n",
       "std         10.587572      17.561933    6252.644590       0.874654   \n",
       "min          1.909344       2.000000      30.000000       0.000000   \n",
       "25%          1.909344      16.000000    5381.500000       1.909344   \n",
       "50%          1.909344      34.500000   11203.000000       1.909344   \n",
       "75%          1.909344      47.000000   16071.500000       1.909344   \n",
       "max         33.000000      59.000000   22167.000000      20.000000   \n",
       "\n",
       "                  ID  \n",
       "count  214200.000000  \n",
       "mean   107099.500000  \n",
       "std     61834.358168  \n",
       "min         0.000000  \n",
       "25%     53549.750000  \n",
       "50%    107099.500000  \n",
       "75%    160649.250000  \n",
       "max    214199.000000  "
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "benchmark_mean_imputation = pd.merge(sales_train_last_month, test_set, how='right', on=['item_id', 'shop_id'])\n",
    "benchmark_mean_imputation.fillna(benchmark_mean_imputation['item_cnt'].mean(), inplace=True)\n",
    "benchmark_mean_imputation.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "benchmark_mean_imputation[[\"ID\", \"item_cnt\"]].rename(columns={\"item_cnt\": \"item_cnt_month\"}).to_csv(os.path.join(TMP_DIR, 'benchmark-mean-imputation.csv'), index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Successfully submitted to Predict Future Sales"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 4.62M/4.62M [00:13<00:00, 348kB/s] \n"
     ]
    }
   ],
   "source": [
    "%%bash\n",
    "kaggle c submit -f ${TMP_DIR}/benchmark-mean-imputation.csv -m 'benchmark reproduction replacing zeros with mean' competitive-data-science-predict-future-sales"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Surprise surprise: `1.99060`. Even though the mean was relatively low (~1.909). Let's try with the median."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "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>date_block_num</th>\n",
       "      <th>shop_id</th>\n",
       "      <th>item_id</th>\n",
       "      <th>item_cnt</th>\n",
       "      <th>ID</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "      <td>214200.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>5.284594</td>\n",
       "      <td>31.642857</td>\n",
       "      <td>11019.398627</td>\n",
       "      <td>1.121755</td>\n",
       "      <td>107099.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>10.897239</td>\n",
       "      <td>17.561933</td>\n",
       "      <td>6252.644590</td>\n",
       "      <td>0.927854</td>\n",
       "      <td>61834.358168</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>2.000000</td>\n",
       "      <td>30.000000</td>\n",
       "      <td>0.000000</td>\n",
       "      <td>0.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>16.000000</td>\n",
       "      <td>5381.500000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>53549.750000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>34.500000</td>\n",
       "      <td>11203.000000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>107099.500000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>1.000000</td>\n",
       "      <td>47.000000</td>\n",
       "      <td>16071.500000</td>\n",
       "      <td>1.000000</td>\n",
       "      <td>160649.250000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>33.000000</td>\n",
       "      <td>59.000000</td>\n",
       "      <td>22167.000000</td>\n",
       "      <td>20.000000</td>\n",
       "      <td>214199.000000</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "       date_block_num        shop_id        item_id       item_cnt  \\\n",
       "count   214200.000000  214200.000000  214200.000000  214200.000000   \n",
       "mean         5.284594      31.642857   11019.398627       1.121755   \n",
       "std         10.897239      17.561933    6252.644590       0.927854   \n",
       "min          1.000000       2.000000      30.000000       0.000000   \n",
       "25%          1.000000      16.000000    5381.500000       1.000000   \n",
       "50%          1.000000      34.500000   11203.000000       1.000000   \n",
       "75%          1.000000      47.000000   16071.500000       1.000000   \n",
       "max         33.000000      59.000000   22167.000000      20.000000   \n",
       "\n",
       "                  ID  \n",
       "count  214200.000000  \n",
       "mean   107099.500000  \n",
       "std     61834.358168  \n",
       "min         0.000000  \n",
       "25%     53549.750000  \n",
       "50%    107099.500000  \n",
       "75%    160649.250000  \n",
       "max    214199.000000  "
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "benchmark_median_imputation = pd.merge(sales_train_last_month, test_set, how='right', on=['item_id', 'shop_id'])\n",
    "benchmark_median_imputation.fillna(benchmark_median_imputation['item_cnt'].median(), inplace=True)\n",
    "benchmark_median_imputation.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [],
   "source": [
    "benchmark_median_imputation[[\"ID\", \"item_cnt\"]].rename(columns={\"item_cnt\": \"item_cnt_month\"}).to_csv(os.path.join(TMP_DIR, 'benchmark-median-imputation.csv'), index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Successfully submitted to Predict Future Sales"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 2.14M/2.14M [00:06<00:00, 335kB/s] \n"
     ]
    }
   ],
   "source": [
    "%%bash\n",
    "kaggle c submit -f ${TMP_DIR}/benchmark-median-imputation.csv -m 'benchmark reproduction replacing zeros with median' competitive-data-science-predict-future-sales"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Better than before: `1.39120`. Still worse than zero-imputation. Now let's try setting everything to zero."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [],
   "source": [
    "zero_submission = pd.DataFrame({'ID': test_set['ID'], 'item_cnt_month': 0})\n",
    "zero_submission.to_csv(os.path.join(TMP_DIR, 'benchmark-zeros-only.csv'), index=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Successfully submitted to Predict Future Sales"
     ]
    },
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "100%|██████████| 1.73M/1.73M [00:25<00:00, 70.2kB/s]\n"
     ]
    }
   ],
   "source": [
    "%%bash\n",
    "kaggle c submit -f ${TMP_DIR}/benchmark-zeros-only.csv -m 'zeros only' competitive-data-science-predict-future-sales"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "1.25011. Not much insight from this other than knowing the dataset is also very skewed to zero on the test set.\n",
    "\n",
    "# Choosing our train set\n",
    "\n",
    "We now know the ids on the test set are probably randomly generated. We also know the test set is skewed to zero. This makes me want to use the ids from the test set to train.\n",
    "\n",
    "The problem with this approach is that the test set is pretty big (~214k rows). We won't be able to use the test set ids for each month or we'll have a train set of $33 x 214,000 = 7,062,000$ rows, which is too big for us right now. On the other hand, I want (further down the line) to have at least a two-level ensemble, so I need enough data to generate oof predictions and be able to test the 2nd layer on."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "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>item_id</th>\n",
       "      <th>shop_id</th>\n",
       "      <th>date_block_num</th>\n",
       "      <th>item_cnt</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>7.068600e+06</td>\n",
       "      <td>7.068600e+06</td>\n",
       "      <td>7.068600e+06</td>\n",
       "      <td>7.068600e+06</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>1.101940e+04</td>\n",
       "      <td>3.164286e+01</td>\n",
       "      <td>1.700000e+01</td>\n",
       "      <td>2.259512e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>6.252630e+03</td>\n",
       "      <td>1.756189e+01</td>\n",
       "      <td>9.521905e+00</td>\n",
       "      <td>3.369709e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>3.000000e+01</td>\n",
       "      <td>2.000000e+00</td>\n",
       "      <td>1.000000e+00</td>\n",
       "      <td>-4.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>5.381500e+03</td>\n",
       "      <td>1.600000e+01</td>\n",
       "      <td>9.000000e+00</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>1.120300e+04</td>\n",
       "      <td>3.450000e+01</td>\n",
       "      <td>1.700000e+01</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>1.607150e+04</td>\n",
       "      <td>4.700000e+01</td>\n",
       "      <td>2.500000e+01</td>\n",
       "      <td>0.000000e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>2.216700e+04</td>\n",
       "      <td>5.900000e+01</td>\n",
       "      <td>3.300000e+01</td>\n",
       "      <td>2.253000e+03</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            item_id       shop_id  date_block_num      item_cnt\n",
       "count  7.068600e+06  7.068600e+06    7.068600e+06  7.068600e+06\n",
       "mean   1.101940e+04  3.164286e+01    1.700000e+01  2.259512e-01\n",
       "std    6.252630e+03  1.756189e+01    9.521905e+00  3.369709e+00\n",
       "min    3.000000e+01  2.000000e+00    1.000000e+00 -4.000000e+00\n",
       "25%    5.381500e+03  1.600000e+01    9.000000e+00  0.000000e+00\n",
       "50%    1.120300e+04  3.450000e+01    1.700000e+01  0.000000e+00\n",
       "75%    1.607150e+04  4.700000e+01    2.500000e+01  0.000000e+00\n",
       "max    2.216700e+04  5.900000e+01    3.300000e+01  2.253000e+03"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def train_set_month(m):\n",
    "    df = test_set[['item_id', 'shop_id']].copy()\n",
    "    df['date_block_num'] = m\n",
    "    return df\n",
    "train_set = pd.concat([train_set_month(m) for m in range(1, 34)]).merge(sales_train_by_month, on=['item_id', 'shop_id', 'date_block_num'], how='left').fillna(0)\n",
    "train_set.describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 7068600 entries, 0 to 7068599\n",
      "Data columns (total 4 columns):\n",
      " #   Column          Dtype  \n",
      "---  ------          -----  \n",
      " 0   item_id         int64  \n",
      " 1   shop_id         int64  \n",
      " 2   date_block_num  int64  \n",
      " 3   item_cnt        float64\n",
      "dtypes: float64(1), int64(3)\n",
      "memory usage: 269.6 MB\n"
     ]
    }
   ],
   "source": [
    "train_set.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Yikes. 269MB for just the ids and labels. Let's see what we can do later."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Mean from MSE\n",
    "\n",
    "We can extract the mean of a dataset from the MSE.\n",
    "\n",
    "$\\bar{m} = \\dfrac{1}{N} \\sum_{i=1}^{N} y_i$\n",
    "\n",
    "$\\text{MSE}_0 = \\dfrac{1}{N} \\sum_{i=1}^{N} (y_i - 0) ^ 2 = \\dfrac{1}{N} \\sum_{i=1}^{N} y_i ^ 2$\n",
    "\n",
    "$\\text{MSE}_1 = \\dfrac{1}{N} \\sum_{i=1}^{N} (y_i - 1) ^ 2 = \\dfrac{1}{N} \\sum_{i=1}^{N} (y_i ^ 2 - 2y_i + 1) = \\dfrac{1}{N} \\left( \\sum_{i=1}^{N} y_i ^ 2 - 2 \\sum_{i=1}^{N} y_i + N \\right) = \\text{MSE}_0 - 2 \\bar{m} + 1$\n",
    "\n",
    "$\\therefore \\bar{m} = \\dfrac{\\text{MSE}_0 - \\text{MSE}_1 + 1}{2}$\n",
    "\n",
    "If we run this on the public LB we can get the mean of the test set data. "
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.8.2"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}