02_product_recommendation.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "baking-sustainability",
   "metadata": {},
   "source": [
    "# Recommendation system for E-commerce\n",
    "\n",
    "- part 1: based on popularity\n",
    "- part 2: model-based collaborative filtering based on customer's purchase history and ratings provided by other users who bought similar items\n",
    "- part 3: cold-start, without product rating (?)\n",
    "\n",
    "\n",
    "References:\n",
    "- https://www.kaggle.com/shawamar/product-recommendation-system-for-e-commerce"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "varied-wealth",
   "metadata": {},
   "outputs": [],
   "source": [
    "import matplotlib.pyplot as plt\n",
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.decomposition import TruncatedSVD\n",
    "\n",
    "%matplotlib inline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "governmental-marsh",
   "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>UserId</th>\n",
       "      <th>ProductId</th>\n",
       "      <th>Rating</th>\n",
       "      <th>Timestamp</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>A39HTATAQ9V7YF</td>\n",
       "      <td>0205616461</td>\n",
       "      <td>5.0</td>\n",
       "      <td>1369699200</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>A3JM6GV9MNOF9X</td>\n",
       "      <td>0558925278</td>\n",
       "      <td>3.0</td>\n",
       "      <td>1355443200</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>A1Z513UWSAAO0F</td>\n",
       "      <td>0558925278</td>\n",
       "      <td>5.0</td>\n",
       "      <td>1404691200</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>A1WMRR494NWEWV</td>\n",
       "      <td>0733001998</td>\n",
       "      <td>4.0</td>\n",
       "      <td>1382572800</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>A3IAAVS479H7M7</td>\n",
       "      <td>0737104473</td>\n",
       "      <td>1.0</td>\n",
       "      <td>1274227200</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "           UserId   ProductId  Rating   Timestamp\n",
       "0  A39HTATAQ9V7YF  0205616461     5.0  1369699200\n",
       "1  A3JM6GV9MNOF9X  0558925278     3.0  1355443200\n",
       "2  A1Z513UWSAAO0F  0558925278     5.0  1404691200\n",
       "3  A1WMRR494NWEWV  0733001998     4.0  1382572800\n",
       "4  A3IAAVS479H7M7  0737104473     1.0  1274227200"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "amazon_ratings = pd.read_csv(\"./data/ratings_Beauty.csv\")\n",
    "amazon_ratings.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "particular-classic",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "UserId        object\n",
       "ProductId     object\n",
       "Rating       float64\n",
       "Timestamp      int64\n",
       "dtype: object"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Check for numerical and categorical data.\n",
    "amazon_ratings.dtypes"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "id": "directed-newcastle",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "UserId       0\n",
       "ProductId    0\n",
       "Rating       0\n",
       "Timestamp    0\n",
       "dtype: int64"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Check for null rows. None.\n",
    "amazon_ratings.isnull().sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "id": "guilty-property",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(2023070, 4)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Check dataset size.\n",
    "amazon_ratings.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "running-latex",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['UserId', 'ProductId', 'Rating', 'Timestamp'], dtype='object')"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "amazon_ratings.columns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "diverse-recipient",
   "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>Rating</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ProductId</th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>B001MA0QY2</th>\n",
       "      <td>7533</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B0009V1YR8</th>\n",
       "      <td>2869</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B0043OYFKU</th>\n",
       "      <td>2477</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B0000YUXI0</th>\n",
       "      <td>2143</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B003V265QW</th>\n",
       "      <td>2088</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B000ZMBSPE</th>\n",
       "      <td>2041</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B003BQ6QXK</th>\n",
       "      <td>1918</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B004OHQR1Q</th>\n",
       "      <td>1885</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B00121UVU0</th>\n",
       "      <td>1838</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>B000FS05VG</th>\n",
       "      <td>1589</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            Rating\n",
       "ProductId         \n",
       "B001MA0QY2    7533\n",
       "B0009V1YR8    2869\n",
       "B0043OYFKU    2477\n",
       "B0000YUXI0    2143\n",
       "B003V265QW    2088\n",
       "B000ZMBSPE    2041\n",
       "B003BQ6QXK    1918\n",
       "B004OHQR1Q    1885\n",
       "B00121UVU0    1838\n",
       "B000FS05VG    1589"
      ]
     },
     "execution_count": 7,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Why not using the mean of ratings? Because one product with 5-star rating can rank higher than one product with 10 5-star but 1 1-star rating.\n",
    "popular_products = pd.DataFrame(amazon_ratings.groupby(\"ProductId\")[\"Rating\"].count())\n",
    "most_popular = popular_products.sort_values(\"Rating\", ascending=False)\n",
    "most_popular.head(10)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "id": "fallen-notification",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<AxesSubplot:xlabel='ProductId'>"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAX0AAAFDCAYAAADWCd6MAAAAOXRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjQuMywgaHR0cHM6Ly9tYXRwbG90bGliLm9yZy/MnkTPAAAACXBIWXMAAAsTAAALEwEAmpwYAABVhElEQVR4nO2de9xdw9X4vysJgiARcWkS4lq3khB3bYW6q6RaSt+XuLS0pUW1dSmlqB9t1aVFX29F8LbulGpRgrbULTdJJEgQJHWJuFeVxPr9sebkmTNn73P2fp7nPHnirO/nsz/n7Nlrz549e/baM2tm1oiq4jiO47QGPRZ3AhzHcZyuw5W+4zhOC+FK33Ecp4Vwpe84jtNCuNJ3HMdpIXot7gTUY5VVVtEhQ4Ys7mQ4juMsUUyYMOF1VR2QdaxbK/0hQ4Ywfvz4xZ0Mx3GcJQoReSHvmJt3HMdxWghX+o7jOC2EK33HcZwWolvb9B3HcVI++ugj5syZwwcffLC4k7LY6d27N4MGDWKppZYqfI4rfcdxlijmzJnDCiuswJAhQxCRxZ2cxYaqMn/+fObMmcPaa69d+Dw37ziOs0TxwQcf0L9//5ZW+AAiQv/+/Uu3eFzpO46zxNHqCr9Ce/LBlb7jOE4L4TZ9x3GWaIac9KdOjW/2uXs3lOnZsyef+cxnWLBgAWuvvTbXXHMNffv2zZWfPHky//znP9lrr70AuP3225k+fTonnXRSZyW7MEuM0s97sEUekOM4Tmey7LLLMnnyZABGjx7NJZdcwo9+9KNc+cmTJzN+/PhFSn/fffdl33337Yqk1uDmHcdxnA6w3XbbMXfuXAAee+wxtttuO4YNG8b222/P008/zYcffsiPf/xjrr/+eoYOHcr111/P2LFjOeaYYwA49NBD+e53v8v222/POuusw0033QTAxx9/zLe//W023HBDdt11V/baa69FxzqCK33HcZx2snDhQsaNG7eo1r7hhhvy97//nUmTJnHmmWdyyimnsPTSS3PmmWfy1a9+lcmTJ/PVr361Jp6XX36ZBx98kDvuuGORyeeWW25h9uzZTJ8+nWuuuYaHH364U9K8xJh3HMdxugv//ve/GTp0KHPnzmWjjTZi1113BeDtt99m9OjRzJw5ExHho48+KhTfqFGj6NGjBxtvvDGvvvoqAA8++CD7778/PXr0YPXVV2fEiBGdknav6TuO45SkYtN/4YUXUFUuueQSAE477TRGjBjBtGnT+OMf/1h4DP0yyyyz6L+qNiXNFVzpO47jtJPllluOiy++mPPPP58FCxbw9ttvM3DgQADGjh27SG6FFVbg3XffLRX3DjvswM0338zHH3/Mq6++ygMPPNApaXbzjuM4SzSLewTfsGHD2Gyzzbj22mv54Q9/yOjRozn77LPZe++2dI0YMYJzzz2XoUOHcvLJJxeK98tf/jLjxo1j4403ZvDgwWyxxRastNJKHU6vNLsp0RGGDx+ulUVUfMim4zgAM2bMYKONNlrcyegS3nvvPfr06cP8+fPZeuuteeihh1h99dWrZLLyQ0QmqOrwrDi9pu84jtNN2WeffXjrrbf48MMPOe2002oUfntwpe84jtNN6Sw7fox35DqOs8TRnc3SXUl78qGh0heRT4vI5Gh7R0SOE5GVReQeEZkZfvsFeRGRi0VklohMEZEtorhGB/mZIjK6dGodx2l5evfuzfz581te8Vf86ffu3bvUeQ3NO6r6NDAUQER6AnOBW4GTgHGqeq6InBT2TwT2BNYP2zbAZcA2IrIycDowHFBggojcrqpvlkqx4zgtzaBBg5gzZw7z5s1b3ElZ7FRWzipDWZv+LsCzqvqCiIwEdgrhVwEPYEp/JHC12mf4ERHpKyJrBNl7VPUNABG5B9gDuLZkGhzHaWGWWmqpUitFOdWUtekfSJuSXk1VXw7/XwFWC/8HAi9F58wJYXnhVYjIkSIyXkTG+5fccRyncyms9EVkaWBf4Mb0WKjVd4qBTVUvV9Xhqjp8wIABnRGl4ziOEyhT098TmKiqr4b9V4PZhvD7WgifCwyOzhsUwvLCHcdxnC6ijNI/iGr7++1AZQTOaOC2KPyQMIpnW+DtYAa6G9hNRPqFkT67hTDHcRyniyjUkSsiywO7AkdFwecCN4jIEcALwAEh/M/AXsAs4H3gMABVfUNEzgIeD3JnVjp1HcdxnK6hkNJX1X8B/ZOw+dhonlRWgaNz4hkDjCmfTMdxHKcz8Bm5juM4LYQrfcdxnBbClb7jOE4L4UrfcRynhXCl7ziO00K40nccx2khXOk7juO0EK70HcdxWghX+o7jOC2EK33HcZwWwpW+4zhOC+FK33Ecp4Vwpe84jtNCuNJ3HMdpIVzpO47jtBCu9B3HcVoIV/qO4zgthCt9x3GcFsKVvuM4TgtRSOmLSF8RuUlEnhKRGSKynYisLCL3iMjM8NsvyIqIXCwis0RkiohsEcUzOsjPFJHRzbopx3EcJ5uiNf2LgLtUdUNgc2AGcBIwTlXXB8aFfYA9gfXDdiRwGYCIrAycDmwDbA2cXvlQOI7jOF1DQ6UvIisBnwOuAFDVD1X1LWAkcFUQuwoYFf6PBK5W4xGgr4isAewO3KOqb6jqm8A9wB6deC+O4zhOA4rU9NcG5gFXisgkEfmtiCwPrKaqLweZV4DVwv+BwEvR+XNCWF54FSJypIiMF5Hx8+bNK3c3juM4Tl2KKP1ewBbAZao6DPgXbaYcAFRVAe2MBKnq5ao6XFWHDxgwoDOidBzHcQJFlP4cYI6qPhr2b8I+Aq8Gsw3h97VwfC4wODp/UAjLC3ccx3G6iIZKX1VfAV4SkU+HoF2A6cDtQGUEzmjgtvD/duCQMIpnW+DtYAa6G9hNRPqFDtzdQpjjOI7TRfQqKPcd4HcisjTwHHAY9sG4QUSOAF4ADgiyfwb2AmYB7wdZVPUNETkLeDzInamqb3TKXTiO4ziFKKT0VXUyMDzj0C4ZsgocnRPPGGBMifQ5juM4nYjPyHUcx2khXOk7juO0EK70HcdxWghX+o7jOC2EK33HcZwWwpW+4zhOC+FK33Ecp4Vwpe84jtNCuNJ3HMdpIVzpO47jtBCu9B3HcVoIV/qO4zgthCt9x3GcFsKVvuM4TgvhSt9xHKeFcKXvOI7TQrjSdxzHaSFc6TuO47QQhZS+iMwWkakiMllExoewlUXkHhGZGX77hXARkYtFZJaITBGRLaJ4Rgf5mSIyOu96juM4TnMoU9MfoapDVbWyVu5JwDhVXR8YF/YB9gTWD9uRwGVgHwngdGAbYGvg9MqHwnEcx+kaOmLeGQlcFf5fBYyKwq9W4xGgr4isAewO3KOqb6jqm8A9wB4duL7jOI5TkqJKX4G/iMgEETkyhK2mqi+H/68Aq4X/A4GXonPnhLC88CpE5EgRGS8i4+fNm1cweY7jOE4RehWU21FV54rIqsA9IvJUfFBVVUS0MxKkqpcDlwMMHz68U+J0HMdxjEI1fVWdG35fA27FbPKvBrMN4fe1ID4XGBydPiiE5YU7juM4XURDpS8iy4vICpX/wG7ANOB2oDICZzRwW/h/O3BIGMWzLfB2MAPdDewmIv1CB+5uIcxxHMfpIoqYd1YDbhWRivzvVfUuEXkcuEFEjgBeAA4I8n8G9gJmAe8DhwGo6hsichbweJA7U1Xf6LQ7cRzHcRrSUOmr6nPA5hnh84FdMsIVODonrjHAmPLJdBzHcToDn5HrOI7TQrjSdxzHaSFc6TuO47QQrvQdx3FaCFf6juM4LYQrfcdxnBbClb7jOE4L4UrfcRynhXCl7ziO00K40nccx2khXOk7juO0EK70HcdxWghX+o7jOC2EK33HcZwWwpW+4zhOC+FK33Ecp4Vwpe84jtNCuNJ3HMdpIVzpO47jtBCFlb6I9BSRSSJyR9hfW0QeFZFZInK9iCwdwpcJ+7PC8SFRHCeH8KdFZPdOvxvHcRynLmVq+scCM6L984ALVHU94E3giBB+BPBmCL8gyCEiGwMHApsAewCXikjPjiXfcRzHKUMhpS8ig4C9gd+GfQF2Bm4KIlcBo8L/kWGfcHyXID8SuE5V/6OqzwOzgK074R4cx3GcghSt6V8I/BD4OOz3B95S1QVhfw4wMPwfCLwEEI6/HeQXhWecswgROVJExovI+Hnz5hW/E8dxHKchDZW+iOwDvKaqE7ogPajq5ao6XFWHDxgwoCsu6TiO0zL0KiCzA7CviOwF9AZWBC4C+opIr1CbHwTMDfJzgcHAHBHpBawEzI/CK8TnOI7jOF1Aw5q+qp6sqoNUdQjWEXufqv4XcD/wlSA2Grgt/L897BOO36eqGsIPDKN71gbWBx7rtDtxHMdxGlKkpp/HicB1InI2MAm4IoRfAVwjIrOAN7APBar6pIjcAEwHFgBHq+rCDlzfcRzHKUkppa+qDwAPhP/PkTH6RlU/APbPOf+nwE/LJtJxHMfpHHxGruM4TgvhSt9xHKeFcKXvOI7TQrjSdxzHaSFc6TuO47QQrvQdx3FaCFf6juM4LYQrfcdxnBbClb7jOE4L4UrfcRynhXCl7ziO00K40nccx2khXOk7juO0EK70HcdxWghX+o7jOC2EK33HcZwWwpW+4zhOC+FK33Ecp4VoqPRFpLeIPCYiT4jIkyLykxC+tog8KiKzROR6EVk6hC8T9meF40OiuE4O4U+LyO5NuyvHcRwnkyI1/f8AO6vq5sBQYA8R2RY4D7hAVdcD3gSOCPJHAG+G8AuCHCKyMbZI+ibAHsClItKzE+/FcRzHaUBDpa/Ge2F3qbApsDNwUwi/ChgV/o8M+4Tju4iIhPDrVPU/qvo8MIuMhdUdx3Gc5lHIpi8iPUVkMvAacA/wLPCWqi4IInOAgeH/QOAlgHD8baB/HJ5xTnytI0VkvIiMnzdvXukbchzHcfIppPRVdaGqDgUGYbXzDZuVIFW9XFWHq+rwAQMGNOsyjuM4LUmp0Tuq+hZwP7Ad0FdEeoVDg4C54f9cYDBAOL4SMD8OzzjHcRzH6QKKjN4ZICJ9w/9lgV2BGZjy/0oQGw3cFv7fHvYJx+9TVQ3hB4bRPWsD6wOPddJ9OI7jOAXo1ViENYCrwkibHsANqnqHiEwHrhORs4FJwBVB/grgGhGZBbyBjdhBVZ8UkRuA6cAC4GhVXdi5t2MMOelPmeGzz927GZdzHMdZYmio9FV1CjAsI/w5MkbfqOoHwP45cf0U+Gn5ZDqO4zidgc/IdRzHaSFc6TuO47QQrvQdx3FaCFf6juM4LYQrfcdxnBbClb7jOE4L4UrfcRynhXCl7ziO00K40nccx2khirhh+ETjLhscx2klvKbvOI7TQrjSdxzHaSFc6TuO47QQrvQdx3FaCFf6juM4LYQrfcdxnBbClb7jOE4L4UrfcRynhWj5yVllyZrM5RO5HMdZUmio9EVkMHA1sBqgwOWqepGIrAxcDwwBZgMHqOqbIiLARcBewPvAoao6McQ1Gjg1RH22ql7VubfTvfAPhOM43Y0i5p0FwAmqujGwLXC0iGwMnASMU9X1gXFhH2BPYP2wHQlcBhA+EqcD22ALqp8uIv068V4cx3GcBjSs6avqy8DL4f+7IjIDGAiMBHYKYlcBDwAnhvCrVVWBR0Skr4isEWTvUdU3AETkHmAP4NpOvJ8lFm8VOI7TFZTqyBWRIcAw4FFgtfBBAHgFM/+AfRBeik6bE8LywtNrHCki40Vk/Lx588okz3Ecx2lAYaUvIn2Am4HjVPWd+Fio1WtnJEhVL1fV4ao6fMCAAZ0RpeM4jhMoNHpHRJbCFP7vVPWWEPyqiKyhqi8H881rIXwuMDg6fVAIm0ubOagS/kD7k966uDtox3HaS5HROwJcAcxQ1V9Gh24HRgPnht/bovBjROQ6rNP27fBhuBs4J+q83Q04uXNuw8nDPxCO48QUqenvABwMTBWRySHsFEzZ3yAiRwAvAAeEY3/GhmvOwoZsHgagqm+IyFnA40HuzEqnrtM98A+E43zyKTJ650FAcg7vkiGvwNE5cY0BxpRJoOM4jtN5uBsGx3GcFsKVvuM4TgvhSt9xHKeFcKXvOI7TQrjSdxzHaSHctbLTbtxfkOMsebjSd7oE/0A4TvfAzTuO4zgthNf0nW5HmVaBzyJ2nHK40ndaBv9AOI6bdxzHcVoKr+k7TgbeKnA+qbjSd5xOoKP9EM2SzZPvDrLO4sHNO47jOC2EK33HcZwWwpW+4zhOC+FK33Ecp4XwjlzHcRYb3aFTu9UosjD6GGAf4DVV3TSErQxcDwwBZgMHqOqbYRH1i7A1ct8HDlXVieGc0cCpIdqzVfWqzr0Vx3Gc8nSXkUxd5Z+qSE1/LPBr4Ooo7CRgnKqeKyInhf0TgT2B9cO2DXAZsE34SJwODAcUmCAit6vqm511I47jOK1CRz4QDW36qvo34I0keCRQqalfBYyKwq9W4xGgr4isAewO3KOqbwRFfw+wR6EUOo7jOJ1GeztyV1PVl8P/V4DVwv+BwEuR3JwQlhdeg4gcKSLjRWT8vHnz2pk8x3EcJ4sOj95RVcVMNp2Cql6uqsNVdfiAAQM6K1rHcRyH9iv9V4PZhvD7WgifCwyO5AaFsLxwx3Ecpwtpr9K/HRgd/o8GbovCDxFjW+DtYAa6G9hNRPqJSD9gtxDmOI7jdCFFhmxeC+wErCIic7BROOcCN4jIEcALwAFB/M/YcM1Z2JDNwwBU9Q0ROQt4PMidqapp57DjOI7TZBoqfVU9KOfQLhmyChydE88YYEyp1DmO4zidirthcBzHaSFc6TuO47QQrvQdx3FaCFf6juM4LYQrfcdxnBbClb7jOE4L4UrfcRynhXCl7ziO00K40nccx2khXOk7juO0EK70HcdxWghX+o7jOC2EK33HcZwWwpW+4zhOC+FK33Ecp4Vwpe84jtNCuNJ3HMdpIVzpO47jtBCu9B3HcVqILlf6IrKHiDwtIrNE5KSuvr7jOE4r06VKX0R6ApcAewIbAweJyMZdmQbHcZxWpqtr+lsDs1T1OVX9ELgOGNnFaXAcx2lZRFW77mIiXwH2UNWvh/2DgW1U9ZhI5kjgyLD7aeDpjKhWAV4veNnuINtd0tEdZLtLOpY02e6Sju4g213S0R1k8+TXUtUBmdKq2mUb8BXgt9H+wcCv2xHP+CVJtrukozvIdpd0LGmy3SUd3UG2u6SjO8i2R76rzTtzgcHR/qAQ5jiO43QBXa30HwfWF5G1RWRp4EDg9i5Og+M4TsvSqysvpqoLROQY4G6gJzBGVZ9sR1SXL2Gy3SUd3UG2u6RjSZPtLunoDrLdJR3dQba0fJd25DqO4ziLF5+R6ziO00K40nccx2khXOk7juO0EK70lyBEZNc6x86rc2xFEVmxOalyuiMi0k9EZHGnIwsRWanOseEdjPvE4O6lWyEiqxSUW0pEhonIqs1KS5eO3mkP4QF+HRvTf5eqPhQdO1VVz472VwaOAf4JXAGcAmwHzADOUdU3I9lfAXEvtmKz2u5X1Qcz0rEh5jJiYAiaC9yuqjMyZJ8AHgrbP1T1+Tr3d6GqHhf+H6uqF0XHxqrqoZH4JSJyvKr+KZLpAYwBVs+I+zjgB0Bv25XXgR+r6nUiMlhVX4pkdwdGJfd3m6relcQ5EhikqpeE/UeBysy/H6rqTYn8bVFePK7mfqMuIrI+sFHYnaiqczoqW/T+IvmlVPWjJGwVVX092r9BVQ8I/89T1ROjY39R1d2S8wVzRRKn4TGtM5pCREYAm4TdJ1X1/gyZHwM3qOpTIrIMcBewObBARL6mqve2Jx1l3hERWRN4TVU/CPEfCmwBTAf+V1UXROL3ishu8fsY4tgNe2/juTyVY0sD/xXnBfB7Vf1PIjoYmCAiR8e6oggiMhAbVQjwzzjNIjJaVa/KOGcp4GpVPSgK2xO4FMvX7wD/B/QOz2a0qo6LZH8D/EpVnwwfw4eBhcDKIvJ9Vb0245qbAj/E/JdV8uJ8VZ1S6EbLzORaHBvwW+D3wHHABOCX0bGJieyfgfOAy4AHgF8BnwXOxF7wWHZ0xnZ8yPTjEtkTgcnAScB/h+2kSlhGmjfFXEmMBZ4JD/+mEP82iezEOveT7q8NPAV8Kez3Bu4AfgcslcieHvJjnShsHeCP4X5mReEXBtkDgR3DdmAIuyiJ9yFgcLQ/GegPrAmMy8iLfYBzwvN4E/gH8AvgS8BqiWxf4A/A88AtYXsW+B9AMBce7ZEtc38jgDmYcvsLMKTO85hU5FjY3w2YBdyJlenfYsp5FrBbRr4NBB4F/gr8Mmx/BR4DBiayT9I2Eu9I4H5MeW2EKfN2pYNy78g0YLnw/zysvP83ViEZk8h+I5SbAVHY18Kz3CwjLzYO6bsK+G7YrgphG2fIbxHK2RXA8LC/BbBFIncyVgmq7L8ITMHesZPTdxE4MglbPpSRK5LwySHvtwPmA9uG8I0yysmT0f/jgD+E/6unZSiEjwRmAocDm4Xt8BA2MpXP2ha7Um+YQJgS/e+FjUm9BVgmzRRgcvgVYG7WsQLXWzYj3mdIlGoIXxqYWSDOVbAWyCxgYXJsUtb/SkHLiGsQ9pJ/E3gQuCDnmjOB3jn39x6wb3x/OXFIen9YbT3e/3X0/5EG+dAzvITfz8mLa4AzgB5JGk7DPm7PtFO21P0Bm4T/Xwn5WHlpc59Pxsuc7s8g+oBE4WsDMzLCbwUOzQg/hNoKTFyGbgaO6qx05ORb1jsyPfo/IXkuT2TEcTAwFVgDU3ZPZaUryI4Dds0I/wLW6sg6ZydM4T6AfQTvB+5LnxGwfJqPoZw+mMiujH1wvxv2B4Sycm7GteNy8VJybHKdZ/en+JmneVzJy5znNyQrn7O2bm/ewRQrYJO7gCNDc/Y+oE8i20NE+gErAH1EZIiqzhaR/nE89VDVf2eYQj8GPgW8kISvEY5VEUxSw4DtgR2AdbHa/m+xWlJWmntE/ysJqLJNisgW4e+JWE3nHuCaSriqTozEF6rqBzn3N1dV45nQH4jIVqr6eCK+FZDG0S+J75hoN9PBU7Bnbh+2bbEWyr3U5sW2qnpwEr8CZ4nIa1hetke2zP0trWHCoKreJCIzgFtE5ESqTR0Ay4nIMOzZLRv+S9iWTWR7YS2IlLnAUhnhG6vql9JAVb1aRH6UBP8nNPlfxVoq34/T2MF01JDzjrwkIjur6n3AbMzM8kJ497LiuEZEPgAmYTXsHTUynSUMVNV7MuK4N5igFhFs4edjrdqdVfWJBvfyr2j3ohC2UESWTeTeEJEvAHeKyKewGvdvNDLHRrwlIkcBKwJvisjxwA3YR+q9DNl9sPzfATgi3EcvassQQC9VnZ1xH7ODqakhS4LSHy8ie2hke1XVM0Xkn5gZJ+b/YTUGsCbPb0VEsebhmY0uFDL6YGpfiuOAcSIyE6jYwdcE1sNq8CnvYrbMSzDzz/N1LrsSVjOqvEWx4k6VzPnR/ynAalGYAjtHx+eKyC4a2Q8BRGRnav0dHQZcKiIr0Hbvg4G3MdtszKMi8g1V/d8k3qOwmhBJ+MwQz83YTOyzVTUt+EV4R1VntlO2zP19JCKrq+orAGq21l2w1sO6iezLmNkF4JXof2U/ZgzwuIhcR3UZ+ipmhkjJHGQR+nDSjsrjMHPKAKzl93yQ3QtTqo3SMRgzd2WlI71+3jvydeBqETkDy9fJIjIZM8N9L4ljKlZeBfso9QfuC30BqqqbJXH3EJFlNLHfi0hvanXYo5geOCRUAurRJ+67UdWxId5lMIUdX2u/8Pdy7DmPwz50+4Vzb4nERwOnYhXC3YCDsLL/AmbaijkKuBgz5xxXKXfALljNP2WBiKypqi8m6VsLWJAhX0O3n5ErIiuq6jsl5Hti97UgFNChmKnn5UTuXWqV6r8xu+lxqvrPRL4HtZ1fj6vqwow0HITZ87bEOmUex2q1D6tqlziYE5FNgNswE9CEEDwcq03sq6rTM85Znej+ogIYy6yK2dL/Q9sHakvM3DZKVV9N5E/GavcDMTPZw2GblOadiFyF2eXPil9YETkV2EBVD2mPbMn7+wIwL60hikhf4GhV/Wl6TlFEZCOyBwNkPYsLsJbscZXaqIgsD1wAfKCq3+1AOjYG9m2UjrLvSDhnI2AD2loUj6vqx4nMWvXSp6pVLerwTLfF8v+FEDYEU5bjVfXMSHZ/4K+q+lq9awTZczBle4yqvh/Clgd+DbyiqidHslfWT7Ie3uh6nYGIjAJ+hvWTxe/1ScCJqvqHhnEsAUr/WeBHqnpdB+LYAPiBqn4jCuurqm/lyA9X1fHR/sr14lfVN+pceznsY7E9VuNcWlXXio6vio0yWg+rvZ+b95ETkROwjmxNwvsDP1PVI5Lw3lgHWWXEw3Tgd6nZJzIbLbol4HWNRvdkpGVnqkeV3JcnG52zAZYP22Gdqa+r6uej4ytitc0tsM4wsI/2JOAIVX27oOzhcR6KyJdV9eaM9CyNvShnNUp7WURk1yyTRCKzRWKSi48thdVYD6XNrLgmZtY7WaORRSLyvZoIIlT1l/WOdzbBRLmwTGWtQJzHYCNWlsNaCO8Bv1DV1LxzE1a+3sc6cysj6KZlxNkT+CnWSonz+ArgVK0ecVQ2vbtj/W/jYnOMiByuqmMKxvHj+IMWhW8OnED1e/2LRqasRecvAUp/LWz0RR/gW6o6q47sZtjIkE9htdFLsK/2NtiQpgsi2cex0QpvJnHsio02GByFPU9bczTOsEpzdJ2MtCwfrlux62+FNacf0upFY+7Cvth/w0a6rKDVwzTjOC/HPiCLhqOJyLcxG/8FqnphXt7UQ0TuzwheGesHOUhVJ0eyf8ZGC91WxkwjIuvQlhfbY8/oUVXdJ0N2XdqGo01X1WfrxNtQVkTuxlpcR0emjz2xWvNdGobMhvCRlBiSWiddL6rqmtF++mEF8zD7Rew9rFL+lSZ8sC2vF4KfVdX3ReSzqvr3SPb06NSjsBFMi1DVn0SyfTDl+WVMKX2ItZh+UzFvNEhzHO/ESPZTwLlYS6YPbSbEMcBPNRn+moeITFXVz9Q5vkK49rsN4hlCWz/Sdpgyf1xV98qQjfN4lqr+O0Pmu8AtWmf4cCR7DlapmYg93wsrHycRmaiqdfM1iqeqDDWQHQwcqKo/byjb3ZV+hfCSjsVMJYuai6q6byTzKGbnfxjYA6tBX4UNy0prt98AjsZGBcwLYV/DvvwjteiY1+y0TiKMFybUNLCRLTVKUkSeUNXNo/26hUJEtsc+ZtOADbHRJSek5qsG6av7YkVyw7GWxeeisJGY/XcXbGTEtcCfNGf8vYjcijXN38by4R/Yhy9rfkMZJTMd+/hcV++jEMkfBJyNDf/dFFgV+whMTuQewl6el8L+5HCvywNXquoukWyeW3DBOhGXj2Q/Bh7BzGIVtg1hqqo7V0Ug8hzwG6yysjCEVfpwNlTVzElMIjJJVYfVyYfbsJFB9wIHhPu6DrNBz1XVU5I0T6NtVaa497YqzSJyH3Cmqj4Q7NyfDXGeDKyqqkdGsvuRjWAfn6oBAdI2/2Yu9hE5GVPmNfNvkvM2pK2SsS02j2BEdPy/MR14TXLewVhL5fdR2NvAv7AP5LXAjRW9kXHdqcAwNRNzX6zMPa2qx6fPR0TyWkMCLKuquf2uIjIA2B/rM/gUcKuqfj9PfhFaYIjP4t6wZRPHYZ1VI4DPV7ZELh0O9VyDeAsNGwO+nHP+0sBpGeGbET6oBe7tCWxEzMphq9rPkF8Rq8m9gLUcPpcT734525cxm3XRvK8ZNhrCl8M6IW/FOi2vJHtY3ReBVQpe6/5oewcboZU33G5zzPzxLNaBfDzwqTpx98SU/nuYrXmDHLnCQ1KxeQd7x+UxbDsBr6ZlCLOF7xmFPV8nvf3Cc56KddAfG5750UTDIYs+r7i8Zd0v1nH8VHLsOKxP6E/hXelTIt4J0f803o+wCtyVGdu7GXGXmX9zCjYX5RGswvctbCRdz4x4H826J+xDOCEJmxTyqDKBbB42v2E01jqPZWck+z3DOTcSjcsPx14kma8SHXspI2yFcM27sXkN5wNzirxfi+IoI7w4NqzJOCN+WerIPhUecGUyxox4P+ec/TGl9Rg5yilkcDrRac9wvQtzzhmGzcSbGLbLgfXDsV6R3GzgufAA0+25JM6Dg+yJWEfZlqFwX43Vptr9YuXcw2pp4c+R2yy8FAszjr2GDVXdmYIfwnDepBKy22KmmhexD8Q3kuM7Yh/TyzBlOgpbe/lMYJlEdlad6zyb7N8JjMiR/VtGWJ+Qzhsxc0PdSkk451isZTsHMzs1km+k9P+BDY8E68y9Ozr2dM4562DK9FFs6OHQDJl7sclYA7FZqDeHcCGZJ4G1gDfNuVaWopscxVV3/k14Jydgczh2A1ZqT14RzQ/KksWGt+6L1frnJcfuIKmQhvCzgY8zwrbOScN5GWGVjvTPVt6nIuWoKo4ywotjC5lSM8koR/YBqmuL8ZbWFKdiHadTsVrls9H+lIy4DwoyZ2G124eyCn+Q/TI2+SidNTcZsy/WzFwteH+3YQsex2GC1WbSD0ThFwurOV2cbP+HfWC+mBPHauHlfijInQtsniHXH7Mz348prosIk50a3Gtd5ZVzzk7Yx+c/Sfj49MXCWirnUVsL/R3JRyOEHwVc20lleljIj9fqyPTFavqTMeV1YSibO2fIVsryFKwDs/K/piyHsvgY1kp5EPh0CB9AmHiUk55NQtmfDRyQcXxN7IMwLZSdNaLn/+VE9rPAmjnXGZ4RNgX7WK+JmQmHRHFPz5BfmbaZ4PeF+/1f4LBEbgbR5KwofIWMcjGpTt4sl+wvi5lmsmQH5sVTsOwch1X0pmIf4nX5pCn9cKNrYMq/Mt3+FKB/htyKJeJcq96WIV/IPBAV0iEZ4UOwyUDnJOlYKdofgSnG47GRPkXvZ0CyX/jFonaq/SGY2WLVjHO/EV6kudjHYfsSafwUVnN9GPuA/rSObCGlj3WQ/xIzfTyAzVTun8jUM4dsnOyvitWG78eazueHeB8mpxmeEWcP4L+SsGOSfalXXrEP6fepbhUODWm7NpEtVZZLPK+4hn8TNkM5U5k1iKdGsZY8/yBs4tmrWIXqXmxi4lwS1wjJeb2wwRQ/IHsG+Pex1tpaUdgQzJz1g0Q2930vkP4+mLWhb8axLZJtGJGbkwLPZiqmU04smsZ2P4iu2jAb6UvAT7Dm1L7h/xPY1PFrItlnsU64IvE2rG1GsoXNA0G+pvYRHXs62X+UYIsOL/Xr2HCsq4DfZpw/DKuNxmaj9SqFvOD9LJ/s/6BIQQuyY4BdqaNIG5zfB/uoTKbW7l1pcfwKMwtVtT4S2XPC83485Feu6SM8u0sxh11rF0znzlhL5jtk1K6DzIpYp+Kvsdq4BPnZ1NqaS7VcGtxPar4qU5YHEcw7Yf97wI/Dtl4i+3F4Tj8Jefy9eMuIeyA2ZnzpsL9qeE7/zJDdHZt9OiQJPzwn3T0r5RtT5sMJrYlEbl+s1fl34I3wWxlVNCBD/ptYhWF+2F7ARgmWKtPJ/qXR/x1pMzu+BOyVyN6fsT2BtUKGFrz+piGfc02TVfJlbm5xbFjTbFhG+FDMLHNVFLYWZnq5Jy3AGefH/jEebiBb2DwQjj1BRi07pC9tbse+hX6BjbcHqy2msqXMRkVfQszO/FJ4Qb6d9XLUK/Dk1GIimd5Y38ktWP/JWGx0Vc9ELm1xVG2J7I8JfSQF0ljYAV7GueuFfM9y7HVbiPMozLTxAGZvrXlZaYe5qkQ6ypTla4F9ov2nMYV+GjaHI5Y9A3Pcl7klssdhnZsPY5WRr2NK9AIS5RzK4N8wk9WzwHca5RM2iWr18H8ANihhkwy5W8L9bEe5lvIKJB2yJc59sc7zuJ/Qn4jVzscXjHM4Gf1CnbF1eoSdnsD6teaZZNQ4sU7WV7EOldsrWyIzKet/znUKmwdC2KigXA4FPhO2w8ILNiqRnRoXFmD3aL/mA0Fxs1HhlzDIC9aqugxzL5A3MqFwLSbI/B6rtd+IKa26/TPhhR5OnY9IJa86UKbqOcC7n9Chj3WcP4N1RE+NlVPGs+sZ7jPz/rAp8u9kbO9ibiNS+TLpKFOW63kK/XsH8nQ6YbQZZnv/ANgyR3YqbbX2vtggiQvy0o99VJ/HWlDfwlrHV4T36YgOpHl+uPaPMLPqcnVkv5eznQC8kZfH1I4CKlxus2SJWkKEyV+0ea8tVglqb4Z11YY1c/plhK9MtnfCosM7K0Mj+1M7bHLlRDYd9vglzGaeWzPAhhRejXWoTsC8Qm6eIXcRVku8KBTspUL4GiS1AsqZjQq/hBlx9cSa35OA9+sU6Ia1GMyUU6gGhX2YXsM+VK8QeQLNkC3z8lS8e34XuB5rud2GTV1Py8W06P/jhP4BrGXXaERHvdEgk0qW+zLpKFOWpyf7K0f/06GG36BtxJlgpr23scrHsEQ2zYtcj48Z18kd0hiOT6XNR897tNX4+1E7eud5rD+kssX76eirFTGz3BmYi+T5oWxcRNJZHd6fs8hu9byVyFY606diH/V+IbxH/FwbPP/MkXNUv383YC3YHphOKjRAZElwuHYB8BcR+T7Vvl7OC8cWISIV293xmrM4RsRK1Hd0Fs+y/WLG+SsDm4nIEZrhgkBtSnSN/5cMjsPGu6+B2VorMxdXx2ogMR9JvrOldDGJDzS4h1Cb2fm0qk6gASLyGWzy1Vex/oWT64ivqGHClKo+J+afKGVnVb06xD1aMxaiiDgOa7LPCzN4f4e10rJYp87kKDSatEc5B3gfichANR9J72ETcsDyN3V0tnk0uUYwT5vv0DZTe0UKICIPqeoOSXCZdJQpy++KyAaq+gy0uRAJE5nSWa7HYuYrsM7UzUNcw7B+ls9GsoNE5OJof414X6t9BT0rIp9X1b+GYwuBI0TkbKw1mPKRmm+c90XkWW1zhvemmEPFmHTSWg9sEtr3SZzPqbmJ+EvYKrPoD8PK4TGYUq0wEfN1X/MOicjXk6CNkv3Ks1sZM0vG5/6KWv9GK2MTyo5Nr5WwgYZFfIBbxbwPN6TbK31VvVzMo+ZZVK+ac7aq/jERX4DVQGpcCmewrmY4S8tJw2FZ4UHZ3oCNEIjDN9Mwozf4UDkRc58wLaT7/ShuxWZEVs5dDnMr8IKqVhVSrFZxb5jmXeNsKZEt/BKKrT51YNgWhvTspqrPZdz2hiIyBVMwQ0SkX3j5epDtvjr2mHgs1kGdx4caZjmGj8gydWTnUe11tB5HYDberwOHBRcceQ7wjscqGTdj5ey+4MZhR2yOwyJUNVW+9bixzrGsqfZl0jGkRDpOB+4QkZ9SXYk6hVolsyCqhOyDrRA1HyuDP0tkf5Ds16tg7J8VqKqnikjqORdApc0b5t6VQDHfUlUVjZA+Qnk8OKRrMrC31jqU+xRtrhq2itJ9KrVuvw/DOoazqPrQaOQwLswm7oOZgF7H+hxixif7irU4vqfZTuMq77UAA6R6hbdCrpWXGDcMHUUSB1hhev23VDV9uGXjrXGbEIeJyPlYs/RKzNbfX6u9Re6L1ZrewArbJVh/xBDMGdhVSdybU8DZkoiMrpfuOF4xp3bXYi4NahxTJfGulQS9rKofivnM/5xWu5hN86KRi4nXiD6A2Edo0X7yoSrswyS5Rl0HeEFmJcxRXewt8jZVfYomkOdjpWg6yrivCPKV5fbiStTP0mcvIhMxJfsmNqplZw1rDYjIDFVNa7SFaEd61yRZvjCEDwQ20mg5yFDJOhz7aD6IOTDM9NcV3ExMJEyY0wJLeRYhpPdnmPuOtwjDc7GhzidptQO2T6vq0znx7KDJko8Z7/XtodK1OjbP4hQa0O2Vvph75CPIWN8UW6asqCOn1AHWNtjQwCcwR1pvtiNtnwbGqup2SfgkDf41wsdlK1X9SEQEs3VuFsk+gdV8VsJs5JuFWu6qmI2uoY+cAulcERY1ZzuNUItZZCLIkakocsFMRlXeUhNFXuZDtbBMTVsKOsArg5iDv8uxcnkn9pF+Mxx7TFW3jmT3y4uGDH8zJdNxf53DqolfnxLx7o3dX0/gjxq81IrI57F3Zu9EfjTWWvh0CJqBDbW9OpEr7NMnyK+HzZFIFeCOWKXj2ShsDtbivxAbZFBFXCkRke2wFuD22PDv2bS5/h6vkf9+MdfKecpSNfJwKyIPh+vfpG1+k3pi7/lxqrptkhfXAN/W6gVd2l2xaUS3N+9gGfIWNla44uFuEDay5P8wRQI0dIDVPw5Q1UeD4v8mtlDLnVQ7couV0R/JtrutgU09T1lJRL6ENT2X0bZFGjTDBvlxxb4qIs9XTCqq+pqIpDWbY7Da+Oti3iWvxEYGPQN8XVWnJvLHYjW63mF/PhkLo0ut33QlLICNKbL5UZw1tZjwUampxQTiZn/alK0ibdU0YEpRQal1gHc++Q7wsp51nMa4r+BSrBPwEcx09KCI7BuUUNrUzuoXqnBHgduI03i5Rg7MNHIiVvD80Vin9oYhKFM5Y2aGtbCO+LhSNJ7ovYviPA4b0TIRe+e2AH4uIqrVTs2+h030+jdWCbg161lEXEh239Lb4Vict/diz2/zsMUokXkltPIfJix+I+aZ84uYCXIQ4b0JZD2jwViLIq18rKKq11dd2JT/dSKSuvF+EtNrk0TkEFV9JDomiSwi8kvMxUWpRd+r0BIjChbHRs76plnHKOEAK8j3x0YkPII19UeTPSY8je9zWNM4cxwwtb5uVgvhq1M7lr7eyIvUiVW8iPKfaFsgfSesxhrLnkHBhdFz7qEfVqBvTMIfxl74nlFYT8wUU3eN3ALPekdsxaPK/k3Yx+Q+kglSlBu9U8YBXlxmZqbPPn12yf6IcM62ZdKXk46Vc7b+JA62sL6UQ4Bdwv7XsAljR5Os7RzK96SQ1pWwIZM7Yx/EgzuQx4+QP5w4s1xQwKdPkHu8znWnFk1jzvkbElbZwz5+L2Nzfb5f55x1gvwz2BDSpZPj12EVgm2wWeifCv8vBW7IymNMpzyLdfT2yMt/rC9rPGZu+xkZc5ga3nNHMqwrtlCY9qd6oeUeQfE8msgWdoCF1fCfDb91FQI2dr2wi4eS9zeb4g7Xno7+p94g02F8hRdGb5C+dChe7kLwWcewD8JRWEf8DsmxU5P9cUTzHrAhb1uGF+KuRDb2MRNvWf5m0rHVx2OdfHVn59J4zPsTJA69sA/MTGB+hvzuoSxV5o5cRjQvI5FdmFEuKvsfJrK/w4ai/hFrGd8a7m8s0eTF6H0aknG9IdR6ES2j9OsNJ653rK5PnwJlrqbygpmXzscqRn/CJj3WuCjAWrNTMB9Hh9B4QueGmHXhSWwOTuYMeOwj/C1srsvUsN2FTXxMHfzFQzD7Yn1rD4XnkaX0J4XfDbAJdU9iTuZOz7rHrG1JMO8ciA3PvFREKk3Mvpjp4cBYUFX3zItEI5/wgR2x6evzRKSPiKD5TczngAkicrpGPrbzCCaQ11T1g2DHPxRr6k4H/lejDiktN/LiJhEZi7l/uFVEjsNe8J2ptV+WWRg97z6WotYEOEFELsWawC+FsMG01SBT/gcbY/0YcLGI/FVVvxeO7Yf5M6qwolaPsJipYYiciPy/JN7nqW8yiVkhI2wI8CMROUPzV2XTBvGehw3PW9QkV9UpYmvqnhYLisiF2It6NdVmymNFZC9VPTaJ+zms5l5jlxaRl5Kgz6jqZqH/ay7m1mOhiPwf9mGKWVHzF9ZOh5iWGRZbs/BI3jGx4bgHYsOrX8JqxudoxuIlgfGSvS7z10lGCQU7/S1YX8TlmIlkGPCAiOyn1eaTdTVaja0eInIjVgE5H6s0LARWlLBAvEb9WmodwpdRu4Z3ZtTReW8BBwVT2YNkL4yuQfYZ7GN5VuhbOghr2a+XcU71BcNXY4lAbFlAtG1YVjxcCRG5BPi9FrR3ici3MFthnxD0LubO9NIM2YGY7W8V7GHG9v90xMo0zG3D+yJyHuYJ7w+Ehcs1Wk+zHSMZDsVqEeti69K+FOI+T6uXExyHvUjjkvN3xmrYO0dhWZ2M/bDW1INavQbp0ljH+kiSNVaxjvV08eopGjqug1K6FMvDg7Ca5bBIdqaqrp+VDyIyS1XXi/Ynq+rQHNl+WqBjPnRE36tRZ5lUL415P2bmiV/MzE5rsRWpcisOIvKMqm6QEV5xPbx+En40lvep0kZEvqPRMoGhvG2B+YF/EXMg9obYkMZJGo2yEZEJqrplThqrjoktap+OQV+EhnH2QfZ9bIZzTbSYiTFdUGYKNhjjHZKPqybLO4otHnMrtspXPFR5aczE+Uokeyf2LjyQxPF5rM9pzyis8CAREZkdpbPyWykXqtHqeZI99j6+v7i/8Ns5+mYdrLP8m0n4JK2zSE4RloSa/iJUdb4Yu2B2y32wmWsVngF+ISJrYDbCa7V2rDsAIvIjbBTHTho6T0NGXyQiK6tqXANFVeeKyJ+wlbW+SJvSV2rH3vbQtrH4X8BG73wM/J/YaJ2Y8dQZyUD4UETpGEvbhJl6fBe4TUQyF0ZPZNMas2KdeBep6p+S65epxUA0dj+0cI4Um0RyH20f2wpPicje6TVFZB9syn3MRDIQWzbuTszfTl2CYkw7yybQtjRmeh2leqJTXHFY3nYlr+LwgYhspaqPJ+FbYbM907RdUifdv0qCrsCa+D2xCX03iq28tS3JaClgI7F5FimS3hu27sJfM2SzKDN880zalGJaBrI4VlW3F5ERtD3XP2n2uszrpgof7AMlttxoTOFBIvVa5KFCGBMPWPgJZnrJY1PJGJoZdNI3M+Q/mxFWjiI2oO6wYQX4Yqwm8x72YGrcMwTZtbDOyknk2LswJZJn8047iDfBHERdR4bfmow47iZ0PAI3E1y3EjprE9njKLg6UZAv5GY6yPbGOqkqLoKPyLrnks+h3tT8moVqsJdnj4zwr2MzLeOw9cNzuZI2D5djsY95+vzGhrjjvp6NMLPIoQXvZQTJOgsN5FOXBqeS31me9ldsgXVYTqdtFugMzDTU0D0GNqRwP2ypxKzjn6LNW2tfbHRMzeIclHDDjK0JW/j97Ei5ahB3mb6F3EV/0njS97zosQzZF+scm9Tg3GOxwRGzKdgxi1XWK1aaweFZF+7QbcpD6uQHfg7WMTYuKIr+1FlmLuP8YWSs6kSGd8y8Y+HlzOxwyzl/MGYa+FtQAG+G/UmEERYZ5zQcyUAJN9OdmP8/Tvan0eYf6GtYzbg/1qJpt8OuKP5lqP5QHU72x1kwm+2NWA13+5A3+2TITqW2w3cO1s+wYSJb4846hA8i8ZtCiYpDdGx1zDa8JcGHTI7cH6L/I7E+jCvDNQ/NkF+T4KQO66/4CjmL6JR8FocAX4ied96ooDKePntjlbZ9w3P8ITYk8iIyVq8jw58Q+b6Falxy0+auO3XlXXiQSIP7qVntKytfGsTRsKIa5L6BTeR8Mfx/BquMPo0Nr258rY4Uiq7YwkN8MBTiZUJYo7Vve2Emi99hjruuwxY7j2XGkaGAMXPK/WnhD7+r0bbYQcMFNbCa50jMn8g2NPBBT+PVicq4mX6XEl4d66QpdRs7Ofr/e6zpXbaAX50TfgnJCJ8CcV2MuYR+kZzaJrU12jXJWdgD66BOWxAbk9GCoETFIeN4XZfUVHu//AdhpBHWH5K2Fk/CPgpPYRWjpzCTz5Mkfu+x1t4Pov25Ubn4ZiJbZlTQpKz/Ofd2Q4j7D5gr6kswV9tnA3dkyP+H4iPcRtfbEtkh4f7mYcpzZvh/PQXXXch6R5Jj7VkBLrOiGo49SdsqYv+izRPrcmQ4q8uMv2yCunrDanF7hJdxTih8L5MxXApb3GMM5sbgdqxmkvdyb4J1PI2lzZRwVQjbJJEditUKZmCTP+4NL9Yj5Ky9W+L+4hp+3dWJaIeb6XBsUoM0ZH0cKopgQVqIMRNT75DPm0THsrye3p5sf8TMc7dT6+76WAo2dalecGU+tobCoppdB55HmRZEmYpDWZfUcc35sXrPMyiCZbEW17uE9RCwfoa0dbLIY2ccV3ief01kp4TfXuFZ94zyqCOePqdF8b6SxlO2/HbGFtKdaSZNyltWC+KdRLZS4XqXapfauRUuClRU07xI86poPnX7jly1mWx3AXeJOeDaByvgc0VknKp+LRI/GRvneoI2GL2hqk+K+SD5Gm0+SP4GHKW1Qx3HhvBH40AR2RZrcm+ehA8Gfk7b9PyfaxgJICJ/UNVRkfgsqkcyrAl8KxoKFo9kkKyRKWHEyQK1zuLM280Jr/AW1tn8anogY3jgj7GOqp6Y0q74Yvk8VhtLGYTZsX9LWwfpcDKcpanqRVhH+lrYkL4xIrIs9kyv1TBzOTA+538NUj3jeNGIC+xFW1pVF70Ham/PkWJOrR7AWgb7a/VQvwr1OstHJrLbRv/PwtZVmBgGD9yA9Q3EVDx4CrCMiKyhqi+H0VPpDNCFakNxP8SGR84P9/Kv2n5qRKMZ1gRHcGrDi9Mhgj3C9ZbHapIrYaaFZaidcbwSxT19fhiuuUDMmWLVvaQJ7iwkmckcwlbFzFWbhP0ngUu01tlZvTJWdUxVs4YI56VpV2wk295Yxe86bAnIf+WcsqyIDMPMUEuH/xK23jnnVF8zfCGWOERkBWy41tUZxwZgymYh1vyrN8U7Pm9lzRiSV2YoYQi7B+vAfQRrTm+JLTA+Px1yJSJnUH94108i2SMxO16Wm+krVDUdnVA5r5Gjs7MxBf5YxrHzVPXEJKwXydR8Md82kua1mLfDY4G9MLPCZBF5TqMhbvUIhXoM5pOojFfLenH2wV70ozAXACdExyrD7QSrEEzEWnhAjYtgwrDIuOIwHVuB6oNELnY8lw6NLOxjRUT6Yk7GHo7CxmKjpJbHJq0twCpKO2PP6YBItqa8hvAe2ESneOjh8VgLuCf2kR6Jfdi3xfzK/CSNp+A95PljEsysuVoifzxwoRZQVsmQ26pDWM14UCS7A2aiHEvbR3tLzBT0X9pOVwcZaVDM535N+kXkPqxSc1OjimqQv7/OYdUCfpa6vdIXke/VOx7XhEVkY6zJNQSrMU/Clgj8K2Z7jsex74DVPj/GOgvPxmojS2MFL36pLsbGxV9N9YSkQ7BO5SqHXZKMIReR/8ZaIftibg2KvuDLaO24932o9ZD4c03cTEv12PtfYB+KRWgyt6AoIjJWVQ8N/0drQX85IjII82b4KjYbOMudcEW2F7b62YGYj58HsJr+bZFMPR85/8FmW1+iwb9QOKcvNlrqEOxlvyCp9VZ8yOSS3q+I9NIw2S58TDbEKhpvJHKVcexCKJ/a5pJ6iqrWHWIqNkflc5j9OJ2Q1AvrkFTMRLg19iF6MeTBvyLZSzE3v6cmcZyN2YfTceGfCvf9z5B/XwhpqKkgBPnh2LuxEOvMrvFM2o48Ho+9mxOw/o2HsM7idzPiXoi5KEiHPgswUFWXjmQfwTztTkriGAr8j6puE4WtglUU3sQqIT/Hhk8+i1kWZkWyz0fXrLAC5uL561rtZbPwB6IRIrJNao3IlFsClP7HWGbdib3MqUe+eOLQI1hnzdMisjVwtKqOFpFvYKNvvhLJPobVwvtgduZRqvqg2GSpX2myqIWI7EnGhCRVTZvllSbilnFtT0S+APwG62NYIwp/GThFVa/MiKfdXvbEvALmoRpNEIvOqZrsFsJWUfMDXtlf1FJpT/rEPDfuoBkuYKOm7l5Yp/V1mCvhmqZuMCfl0Qv7KB6kqtuFF/YErFY5Bnu+mTMxReScrLTlyB6K1YDnY62ZS7DOxQ2wiTXXRrJrJaf/U83zap5L6juwyUTTxOadTMTMCOsCl6vqhUXSmJHm5bHKzla0zdYditn6v57TUkNVPw6mnk2B2Rkftc+HvHgLqy0/hNn1P8J8+qRmwrLpjl1iV/zfv4L5nPp2JDeTOjOZVXVwtD9dVTfOuV7VMRH5C5b/K2AVkSsxvfFZrFWwU4F72A8z3ewRhWV9IPpgz6bqA1Eg/kwX3TVokztIOrph9vJzMcV/BVbTyPSVQ23HRtwZli7RNqnOsY46yzqexDlXCB8G3JOEPY+1RG6k1o/LpGS/MtTti6GQ1B3qViK9I7BO8tex8eND8vIiydOO5lOfZP8+bPRJ5vyLdsT/2/D7L6xz+HQy1jlt77PHhoKugg2XfQebGAQ2ymtKB9MeO9c7hTDiCVM6heMG7swJXyeUoy9W0p0hMwprmb2MVXgexTqv52DmyqqySlsH8tqY2QxscMVfEtk/UtvBv2hrcD/LY0r3x1jLKR29czQZy5KGY+nawvWWYk2HbT8RfoU6I9oKPI+iI9z2I/E3VeCc3KGj8bYkdOQ+gX31ThKR7bGa4K9E5ESt9R/zrIichimP/bAPBWI+ZHoksvF+6ra1agUoscUsTsYK/mrYl/k1rPP1XDWfGTEPau3MS9SakbsmwW9iU/1/hLlXPVzbZhSmzbCrsZrT8pi5Zho2bnpHzC65T5LuTTHXxrEp6BeauGDGRsrsrta5/RXgHhE5WK3zMu0JjFfuSVfnQhObdwOmE60apdl+1DfHPsrTKYiI3Kmqe6pqxYXAz2nLy0adbD1FpB+1911JY1zDXajWCnpdRN7T4NddVV+V2g7UhulNguMW1y7A/4a43w2t3/j8vNaWYLX4WHZpbFLcc8BzYrNcR4Wa7Z3J+adj+b8s9g5updaKXgvrs4pNij01rHpGcAUR0nuPmN+hmF/kpDf7JkS+htXuh2Kt/cexD9COGrlgCNfLncmMtR5jLiB/KdY0zQtD/CoiryfH8gZQpPfRh1o9lImq3iIipzaWrD6tiFC3V/oVxDpnh2H+4+dgSjflcKxWdDJWSI8N4cthNeSY00RkOVV9X1X/EF2nYruPuQH7kIyoFDKxlWoODcd2S+QvDw/4OswWXVdhqX2mzw5NyGtE5FZstmfKxqq6abDhzlHVionjLkncO4jISOzl+n+0jZQZDtwiIt/XyD6OjWB5MqTlJhGZEeROpLYgFfaPH9KR1ycjJFPwQyfV/mrrBRyMOS37G3CG2MiL2N9MYUWnqmc0SmfEhlSPQqmKiupRKC+KOYJbAXMhcT42S/oLWO24LVEl0ht4SUS+g5X1LbCOWcRG2KQjZx7HWotZae6bIbsT8GZQdvthI4e+JyKfU9WqClBU3l/UsMKTqr4gteshjxeRK7D3ZF+sH6ZilqnqgNcM1w4i0l+T/pWI/8EmH/0G85b7TI5cI26kupKRLsWqWEUkaynWivM5odoRnWAtm/hessp8Pyxffl0koXkfiDp9WUKyZkhu3KFZ0G0RkcOxhY17Y51UN2j22pGdca1Vs+IWW1T80znnZB4TW1WrssD4R7QtRzg7kZuk1aN5+mCd0ZthzeW1omO5Sw9m7D+BjfNNrzcEs5NvHoWNx8ahx46rBmGmo3W1xBC0jHz4AKtpL8g4fLyq9o1kp2no0BRbx3YPtRFPy2HO2TaLZBeSr+i2VdVlI9mLM2QWodUOsKqeR4N7WxEzJyj2Mu+OrcvwAqY4Xo5kC6c3yK+K+ahZA+uMrSzePQLrL/pFJDsNG8k2MyONqR07zuPxwGfVhnv2wkwPcR5PCtf6WES21tB5K7YK1BMadT6H1vQ3sIlsTwBj1Dx9LgusqtXrxp6LtThfF+v4vQGrLS+FradQ9VEI19ucNnv+p7GPamWd4ywfPDWkeVFHrjdmvroxCqvXh1T1IROR09PDWL/P39JWdqMPhNZ6Fi2cjnpC3XrDCsMUrClZYwtMZFfC7P8zsPHE88P/c0lmPpK9OMXskOHpZJK/YPbz1aKw1bBp0/cWuIfNsRr3s9QudjIm55yvUGtXrEwx/xXV082zppjnzs4jmeSF1Uw3z5BbCfhRTh4/1SiPg/w/yPEtQ2KDxOzCA8P/+wkuDrCa4pOJ7DSCD6AC8Y6ut6Vp6GB5XTUnvHB623HNrwCfzjk2KuN5bBr+30WwaWOVqnQi11Zku5kYAvx3gXTl+YSaGv2/HzMbgXWAjy8Q72rAMZhNv2bWap3z6s2c7YkNILgG68e4qSPPpESaTk+2H2MVic8065pLgnlnRAnZPDPMaGrNMK9jNbKYgZhtL23GfxWb6v5XMTevYCMHbsdaIbmEZvCqWEFdnsQspRmjaEL4TVjLJqaeaSXdXyAia2oyiiHYY6tq3RotLJ2Ev415FY2p5PFOWszUdRj2cchieLJ/PGZjvRnrf7hPRO7G+iyuTGTPIN8++p3kPmqGlea16rBO8UJI9pjwxyRMmNFq+/8ZFExviHt1TAFo+P0O5s5jBjb8eFErIpSVTDQyXQa+CfwutARfw8wyf8PMpuck59b0S4Xw2VgFKU5vZu09tADS2nsvaRvqumzlOqr6jNgEzDQvNqOtlr891uf2D6yyk66bW8r8EWrOX6NtxNgOmAuG9xO5iZjp7lqN1uTNIufZ7YdVlNJnV2quQ2hdnozNQ7pTo/U9RORSjUYy5cYRvjafCMqYYUTkBKxT9Qcamlxia9SunXV+O9LyWazTeRQ2yuM6zGvh24lcpZN4FPZxaNRJXPT6o7AO2nOoni16EuaY6Q8F46nqZGyPqatkulfCXsINsD6nOZg5qma8d4k4s5TzRKyPqEo5i41FP4kCz0OsQzWtOAwKaVYtOAEtJ813YZ5Xl8fy43fY3IJRmAO0kYn87mT4hVfVuzLi7ol9nOM8vjstayKyR+X8kC/nY7X/aZhp7tVIdqqqfib8vx8bsvq4iGyArXExPJL9DjZq6Fxs7kE/TKHujHksPThJx0RMuf8DaynXDMeMZMuYYeZgnc6XYQ7u3s3TAWJDK2/GKnmvYOba61U1nVFc6tmVff9DhWgmNvHzcMx0/DVV/U9q4q2XCUvsRjIcjZJmGOwFvRFbHGUF6jhyI3upuxqXwUH2JcxJ3DHkNPcj2btD+laPwlbHFE861G11bBGSS7BayxmY6esGMlw+Y2alqzGlPyH83zxDboucbUvg5Q7mcSlzUMb5uflX9JlgJsLnk+0jsh125T2PEzOexwmYieQzUdjznVSGJkX/6w4RxEaa/BnrQ9oxbAeGsIs68H7Fw3N/i01gXAtrkf0hkZ1B8IdF7bKLNevYYi346zGT3lRsHs5RJN47O2Mj38x0IdZiuYPgpystDzl58VnsPXwFM08d2YFnV7i85Zz/I+yD2J+iw0E7O4Ob8MDKKKR+2HCrp7ChkG+GwngeiZ0+OW8k9uV8Jed4qZeKxC95g/t7uugxTMF8B/sgTAkFY3AIu63AtfLszQsxk839Gdu/O5LHZQo12a5zZ5Pdz1L4mVBCOVPfx3rNs6JgxaEdZeiJ6P/ZybHU2VmeG2chWV+W6CODfZCvCGXp9ySeY6lWdJOTY+n+d7AKwc5YZeQizBX4T+igy29gRaxP7BqsVhsfuzTZP5c2z5PDMbcRs7AW2edz8mgE5mRvDuYU7QBq55BkrVdbcQZ5ZQeeXeH3P4TNIHGsiJlWnwReKJSfHXkYXbFRQiF18DrLkuN/vMxLFcKzauRTyaiRU6LmTLkaRBkF2sxOxjIftTI18rLPpKhyblenPY0rDmXTe2aqeEL4eiSdjJjS3ipDdmuSWjblau9zsAlsJ2DKU+JrZlyvcO2dcq2emzFlPirI3kybu/N08mC7O4mx0UP7YOaY15Nj15Uo82WeXdmW888I6xsk4XtklaPM9BW9kcW1UVIhlSxMRc0DhV+qEF64Rk6JmjPlahBlFGjh0R/tyOMyH7UyNfJSzyQ63kg5t6u1GM6tV3FoV3oLviNbUHBVLsrV3k9PtsqM29XJWROhYHovpFyrJ01XrkmDkmames+yI8+kxHXaXd7au3X7jlyxGaJTNUwMSY6N0uqJVRdiX/SrqV7z8hDsK3hsO2W3wBTbCpHsYGyZwKO11gHWJG3zT1PlD0PqLOjdCBE5E/iZ1vpHWQ/r9Il9CzWlo7pMvgX5ftjHbyRt6xlXRj6dp7U+XCqO2V7CFM0TmtEhWvaZJOcui80/mFbsrvMp2oHanvSW6ZwN8qvHsprMVg0yc7DWjmBDA9fVoAQkWsS+PZTIi7KLxM/A1m34OAo7FBvN1ker57KU6iSucy81s6RLdpaXenZl6Gjc3V7pl6FMYSpb8MKxhi9VkHtCw+QnETlbI4+GWS9WswpIUQVaJg3tybd2pn0kNqphiKquXkeu6DPp9Be27AewZHpLxR1GgeyRpDlrRM7pyaUuVdV5IV0/U9VDEvlOzwuxhdmP0GRIqJiTxCs0jAKKwn+G9f/cm4TvgTnPSz8SI7ChqfHopD9gc2I+iuTqzZK+Q6sdI5a5v8KyQb5M2SwVd+bNLQlKv0TBK1yY2lHwCr1UQbZMjfxCmlRAonPqKtBmvrDtTXM4r26NvISiK3N/ZWTL1ljLlKEyFZhDsI/6X0KclTTvCvxEM9acKEKz8qIjrbTORMrN6m5KhbId73/HK12dYSNq5ka5URplbJtlZA/BZtNehvnEORXzA/IsNvGkI/dXZuRF4bzIiK9TOqrL5Ft70kzxfpbCz6Tk/ZWRLdOBWqoMlYz7abJnQ/fLup8SedyUvIiOrU6BReLLpLnk/ZWZ1V3meZSRLdvB3+G+oSVhRu5emv1lux5bzHjRl1BVJwLbFGlCl5HFOo621NoaZD9MAdbUpErUbj8Qka20dvbjVsAHSVjhvMhLg4h0KA0l861UmuvUer4rIntqda2nzDMpk8dlZA8FLhNbxS2tsR6ayJYtQ2XiFrJnoX5MUostmcfNyotKq+fzVJfNvFZP4TSXvL8zKD5Lusz9lZEtk8dl486k25t3mmyGKWoeeAb7umbNph2vtc34C2lCJ3FJ81VT0hDdd9E8LpPmUk1oCj6Tknncng7XIh2opcpQybhHY9P9/0Lbym5rYuads1R1bJyOEnncrLwoZY5qlmmlPZSo7BTNi3aZusqko+bcJUDpl3lhCxemkrKFX6og35RO4pJ50aw0lH1hm/VRK/VMit5fWdkSFYf2pLfMx7UfZtZIZd9M5NrTJ9PZefE0sE1eqycttyXLRdlKYplO1E6vUEbyZcpmqbhrzu/uSr9CQYVUuDC1o+AVeqmCbNNaJ0G+SF40JQ1l861kmsu2OMo8k2a0AMt+AMukt0Ods2LLMM7X5AVvVquuZCWqbMu5KS21kq3hplQoo/suWjY73Gm/RCj9Zphh2tvcjuQyX6pwrCmtk5J50awWUul8a2atJzkvT9E1qwXYrg9go/SWjVtEtsXGpb+BLQhyDbaMYw+skzir1tqprbqS6S3d6ima5pL3V8Zs1JQKZTve/w6VOVgCVs7KyZQRwDkikmbKT4GJYitQ1RSmJOrCsvVeKhGpeam0SZ3EZfKiWWmgXB6XfX6VD0TDDr6Sz6TM/ZWRLdOBWqoMlYkbW8DlFMyXzn3Anqr6iIhsiHmDTIc2F+1EbUpeqOpVYitPxa2eB4CTs1o9JdNcRrZMJ2qZ51FGtmwHf5m4M+n2Sp8SmVKmMJUseKVeqpC+ogWvaQWkGWloxwvblI8a5Z5Js17YMh/AsmWoTNy9tG1lrTPV1jZGVZ+SZK3eknncrLwglJXronStAryVca1SaS55f4dSfCRMUyqUlFfipfI5i25v3mmmGaaorESuE0RkhqpuFB2bpMnyeiWbxaVGXlDcfNW0juqUBmaKMmku0ywu/ExK5nHZTvuiHailylDJuMsso9k0M0yJ9JYyRzXLtBIdK9pRXaZPpmhetKeDv3A6Mu9jCVD6ZV7YwoWppGzhlyqENaWTuGReNCsNZV/YZn3Uyj6TTn9hs6hTcSiV3pJxLwT+hdUMlwXerxzCljtcKpIt24najLwYT1ur53KSVk9GJappfXXSgZEwnVGhDMc6psRLpAPo/jNyw330w2ZxnhC2AwlreyZy47EVgfbHvNVtG8I3JFn7tKTsQuAdzNf2gvC/sv9RRjqeAVbKCF+JAu5PMSUqOceK5kVT0lAm39qR5tG0zVo9JWyVWauHduSZlMnjEnmxLWbaugVbhWsa5kzuNZIZoO0oQ4XjLvkuFc7jJubF5Oj/jORYTRkqWS7KyJaZ1V3m/jr07OqVzc4oF+0qOIt7q1PwChemsgWvZPrKFLymFJBmpaEz8q1BoS70gSj5PJrywtKOD2CJNDcz7qIf4abkBdXunVPXyJmrP5UpFyXur7D7ipL3V0a21PvfGeWi3QWnq7aSBa9wYWpPwSuZ7ma0TsoWkGakoVS+lU1zxvmFa+R14mjWCzs5+t/ZFYemxV00j5uVF3SwlVa2XNS5v8Kt4ZL3V0a2lBLvjHLRaQWnWVvJgle4MHVGwSt5H53ROunQV76T0lDWTNG0j1qJvC9zf2Vkm1ZxaFbcZfK4G+VFs1pqoyneGm5KhbJMHndWPrf7QXTVVjZTusNWsuA1pYA0Kw3NfH40yaTRxBe2aRWHZsVdJo+7UV40paUWwou2hptSoSz77nVGPrf7QXTVVjZTusNWspA2pYA0Kw3NfH406QPfrBd2SdzK5HF3yYuSae5QGaITzIkl763L83hJmJy1uYi8QxiOFv4T9nsvvmTVpfBkGVXtWSLeMnnRrDSUpUyaP47+/zs5pu1NQJn7a3JedAcK53E3yosy5aKwbL3hx5I9S7rTWRx53O2VfjcqeGVY7MqrWWkoSxM/ak77WBLzuEyay8iWnmn/SaDbT85aEikzWeaTnAbH6c60Z5b0J4FuX9NfEukOrZPukAbH6eZ0i9ZwV+M1fcdxWpJWbQ270nccx2kheizuBDiO4zhdhyt9x3GcFsKVvvOJRkQWishkEZkmIjeKyHIdiOsBERnejvP6isi3o/0hIjKtM6/hOEVxpe980vm3qg5V1U2BD4FvxgdFpCtGsPUFvt1IyHG6Alf6Tivxd2A9EdlJRP4utuzjdBHpLSJXishUEZkkIiMARGRZEblORGaIyK3YCA/Csfei/18RkbHh/2oicquIPBG27bFZn+uGFsfP4wTVu4bjNAMfp++0BKFGvydtsyy3ADZV1edF5ARAVfUzYTbmX0RkA+BbwPuqupGIbAZMLHCpi4G/quqXRKQn0Ac4KVxraEjLkEi+PddwnHbjNX3nk86yIjIZc0D3InBFCH9MVZ8P/3cE/g/MNxHwArAB8LkofAowpcD1dsZc9aKqCzVZti+D9lzDcdqN1/SdTzr/rtSwKwSHc//qYLzxBJfu6rPGcWrwmr7jmK3/vwCCWWdNbCm9vwFfC+GbAptF57wqIhuJSA/gS1H4OMxkg4j0DAtvvwuskHPtetdwnE7Hlb7jwKWYO92pwPXYqkn/wcw0fURkBnAmMCE65yTgDuAfwMtR+LHAiBDXBGBjVZ0PPBSGjVZ15Da4huN0Ou6GwXEcp4Xwmr7jOE4L4UrfcRynhXCl7ziO00K40nccx2khXOk7juO0EK70HcdxWghX+o7jOC3E/wdUsTW0UgwBxAAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {
      "needs_background": "light"
     },
     "output_type": "display_data"
    }
   ],
   "source": [
    "most_popular.head(30).plot(kind=\"bar\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "loving-enzyme",
   "metadata": {},
   "source": [
    "#### Analysis\n",
    "- Popular products can be defined by products that are rated the most (not the highest rating)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "rural-timeline",
   "metadata": {},
   "source": [
    "# Part II\n",
    "\n",
    "## Model-based collaborative filtering system"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "id": "handed-harvey",
   "metadata": {},
   "outputs": [],
   "source": [
    "# Subset of Amazon Ratings\n",
    "\n",
    "amazon_ratings_small = amazon_ratings.head(10000)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "latter-scratch",
   "metadata": {},
   "source": [
    "#### Utility matrix based on products sold and user reviews\n",
    "\n",
    "Utility matrix consists of all possible user-item preferences (ratings) represented as a matrix. The utility matrix is sparse - no users would buy all items in the list."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "id": "accompanied-tribe",
   "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>ProductId</th>\n",
       "      <th>0205616461</th>\n",
       "      <th>0558925278</th>\n",
       "      <th>0733001998</th>\n",
       "      <th>0737104473</th>\n",
       "      <th>0762451459</th>\n",
       "      <th>1304139212</th>\n",
       "      <th>1304139220</th>\n",
       "      <th>130414089X</th>\n",
       "      <th>130414643X</th>\n",
       "      <th>1304146537</th>\n",
       "      <th>...</th>\n",
       "      <th>B000052YPE</th>\n",
       "      <th>B000052YPF</th>\n",
       "      <th>B000052YPG</th>\n",
       "      <th>B000052YPH</th>\n",
       "      <th>B000052YPM</th>\n",
       "      <th>B000052YPU</th>\n",
       "      <th>B000052YPV</th>\n",
       "      <th>B000052YPY</th>\n",
       "      <th>B000052YQ0</th>\n",
       "      <th>B000052YQ2</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>UserId</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>A00205921JHJK5X9LNP42</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A024581134CV80ZBLIZTZ</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A03056581JJIOL5FSKJY7</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A03099101ZRK4K607JVHH</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>A0505229A7NSH3FRXRR4</th>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>...</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "      <td>0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 886 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "ProductId              0205616461  0558925278  0733001998  0737104473  \\\n",
       "UserId                                                                  \n",
       "A00205921JHJK5X9LNP42           0           0           0           0   \n",
       "A024581134CV80ZBLIZTZ           0           0           0           0   \n",
       "A03056581JJIOL5FSKJY7           0           0           0           0   \n",
       "A03099101ZRK4K607JVHH           0           0           0           0   \n",
       "A0505229A7NSH3FRXRR4            0           0           0           0   \n",
       "\n",
       "ProductId              0762451459  1304139212  1304139220  130414089X  \\\n",
       "UserId                                                                  \n",
       "A00205921JHJK5X9LNP42           0           0           0           0   \n",
       "A024581134CV80ZBLIZTZ           0           0           0           0   \n",
       "A03056581JJIOL5FSKJY7           0           0           0           0   \n",
       "A03099101ZRK4K607JVHH           0           0           0           0   \n",
       "A0505229A7NSH3FRXRR4            0           0           0           0   \n",
       "\n",
       "ProductId              130414643X  1304146537  ...  B000052YPE  B000052YPF  \\\n",
       "UserId                                         ...                           \n",
       "A00205921JHJK5X9LNP42           0           0  ...           0           0   \n",
       "A024581134CV80ZBLIZTZ           0           0  ...           0           0   \n",
       "A03056581JJIOL5FSKJY7           0           0  ...           0           0   \n",
       "A03099101ZRK4K607JVHH           0           0  ...           0           0   \n",
       "A0505229A7NSH3FRXRR4            0           0  ...           0           0   \n",
       "\n",
       "ProductId              B000052YPG  B000052YPH  B000052YPM  B000052YPU  \\\n",
       "UserId                                                                  \n",
       "A00205921JHJK5X9LNP42           0           0           0           0   \n",
       "A024581134CV80ZBLIZTZ           0           0           0           0   \n",
       "A03056581JJIOL5FSKJY7           0           0           0           0   \n",
       "A03099101ZRK4K607JVHH           0           0           0           0   \n",
       "A0505229A7NSH3FRXRR4            0           0           0           0   \n",
       "\n",
       "ProductId              B000052YPV  B000052YPY  B000052YQ0  B000052YQ2  \n",
       "UserId                                                                 \n",
       "A00205921JHJK5X9LNP42           0           0           0           0  \n",
       "A024581134CV80ZBLIZTZ           0           0           0           0  \n",
       "A03056581JJIOL5FSKJY7           0           0           0           0  \n",
       "A03099101ZRK4K607JVHH           0           0           0           0  \n",
       "A0505229A7NSH3FRXRR4            0           0           0           0  \n",
       "\n",
       "[5 rows x 886 columns]"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ratings_utility_matrix = amazon_ratings_small.pivot_table(\n",
    "    values=\"Rating\", index=\"UserId\", columns=\"ProductId\", fill_value=0\n",
    ")\n",
    "ratings_utility_matrix.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "id": "martial-malpractice",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(9697, 886)"
      ]
     },
     "execution_count": 11,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "ratings_utility_matrix.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "id": "talented-candidate",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(886, 9697)"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X = ratings_utility_matrix.T\n",
    "X.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "id": "incorrect-journalist",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(886, 10)"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Decomposing the matrix.\n",
    "svd = TruncatedSVD(n_components=10, random_state=10)\n",
    "decomposed_matrix = svd.fit_transform(X)\n",
    "decomposed_matrix.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "durable-monitoring",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(886, 886)"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Correlation matrix\n",
    "correlation_matrix = np.corrcoef(decomposed_matrix)\n",
    "correlation_matrix.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "id": "affected-debate",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'6117036094'"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Assuming the customers buy product id `6117036094` (randomly choosen)\n",
    "product_id = X.index[99]\n",
    "product_id"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "recognized-victory",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['0205616461', '0558925278', '0733001998', '0737104473', '0762451459']"
      ]
     },
     "execution_count": 16,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "product_names = list(X.index)\n",
    "product_names[:5]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "pediatric-swedish",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "99"
      ]
     },
     "execution_count": 17,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "product_idx = product_names.index(product_id)\n",
    "product_idx"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "sapphire-values",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(886,)"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "correlation_product_id = correlation_matrix[product_idx]\n",
    "correlation_product_id.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "marked-angel",
   "metadata": {},
   "source": [
    "Recommending top 10 highly correlated products in sequence."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "earlier-flour",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "['0733001998',\n",
       " '0762451459',\n",
       " '1304139212',\n",
       " '1304139220',\n",
       " '130414643X',\n",
       " '130414674X',\n",
       " '1304174778',\n",
       " '1304174867',\n",
       " '1304174905',\n",
       " '1304196046']"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "recommendations = list(X.index[correlation_product_id > 0.90])\n",
    "\n",
    "# Remove the item already bought by the customer.\n",
    "recommendations.remove(product_id)\n",
    "\n",
    "recommendations[:10]"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "administrative-republic",
   "metadata": {},
   "source": [
    "## Recommendation System - Part III\n",
    "\n",
    "- For a business without any user-item purchase history, a search engine based recommendation engine can be designed for users. THe product recommendations can be based on textual clustering analysis given in product description."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "mounted-tract",
   "metadata": {},
   "outputs": [],
   "source": [
    "from sklearn.cluster import KMeans\n",
    "from sklearn.feature_extraction.text import CountVectorizer, TfidfVectorizer\n",
    "from sklearn.metrics import adjusted_rand_score\n",
    "from sklearn.neighbors import NearestNeighbors"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "corresponding-timber",
   "metadata": {},
   "source": [
    "#### Item to item based recommendation system based on product description\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "id": "frank-screening",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(124428, 2)"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "product_descriptions = pd.read_csv(\"./data/product_descriptions.csv\")\n",
    "product_descriptions.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "adult-remedy",
   "metadata": {},
   "source": [
    "#### Check for missing values"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "drawn-vinyl",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "product_uid            0\n",
       "product_description    0\n",
       "dtype: int64"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "product_descriptions.dropna(inplace=True)\n",
    "product_descriptions.isnull().sum()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "stupid-retro",
   "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>product_uid</th>\n",
       "      <th>product_description</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>100001</td>\n",
       "      <td>Not only do angles make joints stronger, they ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>100002</td>\n",
       "      <td>BEHR Premium Textured DECKOVER is an innovativ...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>100003</td>\n",
       "      <td>Classic architecture meets contemporary design...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>100004</td>\n",
       "      <td>The Grape Solar 265-Watt Polycrystalline PV So...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>100005</td>\n",
       "      <td>Update your bathroom with the Delta Vero Singl...</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   product_uid                                product_description\n",
       "0       100001  Not only do angles make joints stronger, they ...\n",
       "1       100002  BEHR Premium Textured DECKOVER is an innovativ...\n",
       "2       100003  Classic architecture meets contemporary design...\n",
       "3       100004  The Grape Solar 265-Watt Polycrystalline PV So...\n",
       "4       100005  Update your bathroom with the Delta Vero Singl..."
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "product_descriptions.head()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "configured-computer",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0    Not only do angles make joints stronger, they ...\n",
       "1    BEHR Premium Textured DECKOVER is an innovativ...\n",
       "2    Classic architecture meets contemporary design...\n",
       "3    The Grape Solar 265-Watt Polycrystalline PV So...\n",
       "4    Update your bathroom with the Delta Vero Singl...\n",
       "Name: product_description, dtype: object"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "product_descriptions_small = product_descriptions.head(500)\n",
    "product_descriptions_small[\"product_description\"].head()"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "adjusted-thomson",
   "metadata": {},
   "source": [
    "#### Feature extraction from product descriptions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "piano-determination",
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<500x8932 sparse matrix of type '<class 'numpy.float64'>'\n",
       "\twith 34817 stored elements in Compressed Sparse Row format>"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "vectorizer = TfidfVectorizer(stop_words=\"english\")\n",
    "X = vectorizer.fit_transform(product_descriptions_small[\"product_description\"])\n",
    "X"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "distributed-gibson",
   "metadata": {},
   "source": [
    "#### Visualizing product clusters in subset of data"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "id": "equipped-german",
   "metadata": {},
   "outputs": [
    {
     "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": [
    "# Fitting k-means to dataset.\n",
    "\n",
    "kmeans = KMeans(n_clusters=10, init=\"k-means++\", random_state=0)\n",
    "y_kmeans = kmeans.fit_predict(X)\n",
    "plt.plot(y_kmeans, \".\")\n",
    "plt.show()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "id": "identical-extra",
   "metadata": {},
   "outputs": [],
   "source": [
    "def print_cluster(i):\n",
    "    print(\"Cluster {}:\".format(i))\n",
    "    for ind in order_centroids[i, :10]:\n",
    "        print(\" \", terms[ind])\n",
    "    print"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "id": "talented-tucson",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Top term per cluster:\n",
      "Cluster 0:\n",
      "  concrete\n",
      "  stake\n",
      "  used\n",
      "  steel\n",
      "  grip\n",
      "  pliers\n",
      "  head\n",
      "  sprinkler\n",
      "  cement\n",
      "  garden\n",
      "Cluster 1:\n",
      "  door\n",
      "  ft\n",
      "  wood\n",
      "  painted\n",
      "  proposition\n",
      "  nbsp\n",
      "  residents\n",
      "  california\n",
      "  65\n",
      "  post\n",
      "Cluster 2:\n",
      "  light\n",
      "  water\n",
      "  watt\n",
      "  led\n",
      "  installation\n",
      "  heater\n",
      "  energy\n",
      "  year\n",
      "  bulb\n",
      "  easy\n",
      "Cluster 3:\n",
      "  paint\n",
      "  water\n",
      "  easy\n",
      "  roller\n",
      "  handle\n",
      "  ft\n",
      "  brush\n",
      "  use\n",
      "  smooth\n",
      "  provides\n",
      "Cluster 4:\n",
      "  finish\n",
      "  furniture\n",
      "  paint\n",
      "  available\n",
      "  wood\n",
      "  using\n",
      "  includes\n",
      "  look\n",
      "  waxes\n",
      "  dog\n",
      "Cluster 5:\n",
      "  metal\n",
      "  storage\n",
      "  shelves\n",
      "  shelf\n",
      "  unit\n",
      "  steel\n",
      "  durable\n",
      "  adjustable\n",
      "  lbs\n",
      "  kitchen\n",
      "Cluster 6:\n",
      "  power\n",
      "  air\n",
      "  cutting\n",
      "  volt\n",
      "  control\n",
      "  cooling\n",
      "  20\n",
      "  easy\n",
      "  plan\n",
      "  tool\n",
      "Cluster 7:\n",
      "  rug\n",
      "  bamboo\n",
      "  shades\n",
      "  vary\n",
      "  order\n",
      "  slightly\n",
      "  natural\n",
      "  product\n",
      "  collection\n",
      "  size\n",
      "Cluster 8:\n",
      "  piece\n",
      "  toilet\n",
      "  tile\n",
      "  design\n",
      "  flush\n",
      "  water\n",
      "  bowl\n",
      "  insulation\n",
      "  use\n",
      "  installation\n",
      "Cluster 9:\n",
      "  wall\n",
      "  plate\n",
      "  posts\n",
      "  fence\n",
      "  post\n",
      "  panels\n",
      "  center\n",
      "  connect\n",
      "  speaker\n",
      "  screws\n"
     ]
    }
   ],
   "source": [
    "# Optimal clusters can be obtained from elbow method.\n",
    "true_k = 10\n",
    "model = KMeans(n_clusters=true_k, init=\"k-means++\", max_iter=100, n_init=1)\n",
    "model.fit(X)\n",
    "\n",
    "print(\"Top term per cluster:\")\n",
    "order_centroids = model.cluster_centers_.argsort()[:, ::-1]\n",
    "terms = vectorizer.get_feature_names()\n",
    "for i in range(true_k):\n",
    "    print_cluster(i)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "crude-tampa",
   "metadata": {},
   "source": [
    "#### Predicting clusters based on key search words."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "id": "tight-bishop",
   "metadata": {},
   "outputs": [],
   "source": [
    "def show_recommendations(product):\n",
    "    X = vectorizer.transform([product])\n",
    "    preds = model.predict(X)\n",
    "    print_cluster(preds[0])"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "united-costs",
   "metadata": {},
   "source": [
    "**Keyword**: cutting tool"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "id": "indie-meeting",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cluster 6:\n",
      "  power\n",
      "  air\n",
      "  cutting\n",
      "  volt\n",
      "  control\n",
      "  cooling\n",
      "  20\n",
      "  easy\n",
      "  plan\n",
      "  tool\n"
     ]
    }
   ],
   "source": [
    "show_recommendations(\"cutting tool\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "hidden-manchester",
   "metadata": {},
   "source": [
    "**Keyword**: spray paint"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "id": "still-desperate",
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cluster 3:\n",
      "  paint\n",
      "  water\n",
      "  easy\n",
      "  roller\n",
      "  handle\n",
      "  ft\n",
      "  brush\n",
      "  use\n",
      "  smooth\n",
      "  provides\n"
     ]
    }
   ],
   "source": [
    "show_recommendations(\"spray paint\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "quantitative-fellow",
   "metadata": {},
   "source": [
    "In case a word appears in multiple clusters, the algorithm chooses the cluster with the highest frequency of occurance of the word"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "secondary-dragon",
   "metadata": {},
   "source": [
    "Once a cluster is identified based on the user's search words, the recommendations system can display items from the corresponding product cluster based on the product descriptions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "clean-shift",
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "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": "3.9.10"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}