az-item-churn.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "id": "010c871c",
   "metadata": {},
   "source": [
    "# Predicting Item Churn with getML on Amazon Product Dataset\n",
    "\n",
    "## Introduction\n",
    "This notebook showcases how to use [getML](https://getml.com) to predict item churn with the [Amazon e-commerce dataset](https://relbench.stanford.edu/datasets/rel-amazon/#item-churn).\n",
    "The goal is to demonstrate how even a simple, getML model can outperform competing approaches on real-world, multi-table datasets with minimal effort.\n",
    "\n",
    "### Why use getML FastProp?\n",
    "- *Relational Learning.* FastProp (short for Fast Propositionalization) extracts features from related tables without manual SQL or deep business domain expertise.\n",
    "- *Blazing Fast.* FastProp runs [60 to 1000 times faster](https://github.com/getml/getml-community?tab=readme-ov-file#benchmarks) than alternatives like [featuretools](https://www.featuretools.com) or [tsfresh](https://tsfresh.com), scaling easily to millions of rows.\n",
    "- *Superior Performance.* FastProp frequently outperforms graph neural networks and models with handcrafted features on relational learning tasks.\n",
    "\n",
    "The Amazon Product dataset consists of:\n",
    "- 3 Tables (Customers, Product, Review)\n",
    "- 16 Columns in total\n",
    "- ~23 million rows\n",
    "\n",
    "This dataset is part of the [Relational Deep Learning Benchmark (RelBench)](http://relbench.stanford.edu/),\n",
    "maintained at Stanford, which standardizes relational learning tasks.\n",
    "\n",
    "### Outline\n",
    "\n",
    "1. [Setup](#1.-Setup)\n",
    "2. [Loading Data into getML, Inspecting, and Setting Roles (Data Annotation)](#2.-Loading-Data-into-getML,-Inspecting,-and-Setting-Roles-(Data-Annotation))\n",
    "3. [Define the getML DataModel](#3.-Defining-the-getML-DataModel)\n",
    "4. [Create a getML Container referncing the Data](#4.-Creating-a-getML-Container-referencing-the-Data)\n",
    "5. [Define the Pipeline and Check the Data](#5.-Defining-the-getML-Pipeline-and-Checking-the-Data)\n",
    "6. [Fitting and Scoring the Pipeline](#6.-Fitting-and-Scoring-the-getML-Pipeline)\n",
    "7. [Explore Feature Importances](#7.-Feature-Importances)\n",
    "8. [Explore Column Importances](#8.-Column-Importances)\n",
    "9. [Export Features](#9.-Export-Features)\n",
    "10. [Hyperoptimization](#10.-Hyperoptimization)\n",
    "11. [Conclusion](#11.-Conclusion)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "87e72872",
   "metadata": {},
   "source": [
    "---\n",
    "## 1. Setup\n",
    "\n",
    "> <span style=\"font-weight: 500; color: #3b3b3b;\">ⓘ️&nbsp; Note</span>\n",
    ">\n",
    "> We have prepared an [environment with all necessary dependencies](pyproject.toml) for you. To to use it, just start jupyter lab through `uv run jupyter lab` from the repository root.\n",
    "\n",
    "In this section, we:\n",
    "- Import required libraries.\n",
    "- Create a getML project.\n",
    "- Download the \"Amazon\" dataset from RelBench.\n",
    "- \n",
    "The \"rel-amazon\" dataset represents relational data on sales history,\n",
    "and the \"item-churn\" task focuses on predicting item churn."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "id": "1e93edc9-37f6-4101-8c2c-3dcbd2c51669",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[2K  Loading pipelines... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00\n",
      "\u001b[?25h"
     ]
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">Connected to project <span style=\"color: #008000; text-decoration-color: #008000\">'az-item-churn'</span>.\n",
       "</pre>\n"
      ],
      "text/plain": [
       "Connected to project \u001b[32m'az-item-churn'\u001b[0m.\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "import getml\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import matplotlib.pyplot as plt\n",
    "\n",
    "\n",
    "from relbench.datasets import get_dataset\n",
    "from relbench.tasks import get_task\n",
    "\n",
    "# Enable textual output to avoid rendering issues in certain JupyterLab environments\n",
    "getml.utilities.progress.FORCE_TEXTUAL_OUTPUT = True\n",
    "getml.utilities.progress.FORCE_MONOCHROME_OUTPUT = True\n",
    "\n",
    "# Launch getML engine and set project.\n",
    "getml.set_project(\"az-item-churn\")\n",
    "\n",
    "# Download dataset and task from RelBench.\n",
    "dataset = get_dataset(\"rel-amazon\", download=True)\n",
    "task = get_task(\"rel-amazon\", \"item-churn\", download=True)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "33c83711",
   "metadata": {},
   "source": [
    "---\n",
    "## 2. Loading Data into getML, Inspecting, and Setting Roles (Data Annotation)\n",
    "\n",
    "> <span style=\"font-weight: 500; color: #3b3b3b;\">ⓘ️&nbsp; Note</span>\n",
    ">\n",
    "> While the `relbench` library provides a high-level API to load datasets and tasks into pandas dataframes,\n",
    "> we will use getML's native `DataFrame` API to load the data directly into getML to speed up the process.\n",
    "\n",
    "The Amazon item-churn dataset is provided by relbench as a [dataset](https://relbench.stanford.edu/start/) of multiple tables:\n",
    "- `product`\n",
    "- `review`\n",
    "- `customer`\n",
    "\n",
    "The target variable `churn` is provided as a [task](https://relbench.stanford.edu/start/). In getML, we\n",
    "refer to table containing the target variable as the [population table](https://getml.com/latest/user_guide/concepts/data_model/#the-population-table):\n",
    "- population (train/test/val splits) with the target variable `churn`\n",
    "\n",
    "### Objectives:\n",
    "- Load each table from disk\n",
    "- Inspect columns to understand the schema\n",
    "- Assign appropriate roles (join_key, time_stamp, numerical, categorical, etc.)\n",
    "\n",
    "### Why Roles Matter\n",
    "[`Roles`](https://getml.com/latest/user_guide/concepts/annotating_data/#annotating-data-roles) allow getML to:\n",
    "- Efficiently encode links between tables (`join_key`)\n",
    "- Interpret timestamps for time-aware modeling (`time_stamp`)\n",
    "- Encode numerical, categorical, and text features (`numerical`, `categorical`, `text`)\n",
    "- Ignore irrelevant columns (`unused_float` or `unused_string`)\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ce9812e4",
   "metadata": {},
   "source": [
    "### a. Product Table"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "id": "5a576f03-c130-47c0-9f50-57c31537ee7e",
   "metadata": {
    "tags": []
   },
   "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_id</th>\n",
       "      <th>category</th>\n",
       "      <th>brand</th>\n",
       "      <th>title</th>\n",
       "      <th>description</th>\n",
       "      <th>price</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0</td>\n",
       "      <td>[Books, Literature &amp; Fiction, Dramas &amp; Plays]</td>\n",
       "      <td>Visit Amazon's William Shakespeare Page</td>\n",
       "      <td>Love's Labour's Lost: Performed by Derek Jacob...</td>\n",
       "      <td>William Shakespeare is widely regarded as the ...</td>\n",
       "      <td>20.93</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>1</td>\n",
       "      <td>[Books, Children's Books, Literature &amp; Fiction]</td>\n",
       "      <td>Louis (ed) UNTERMEYER</td>\n",
       "      <td>The golden treasury of poetry</td>\n",
       "      <td>Collection of Poetry</td>\n",
       "      <td>23.98</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2</td>\n",
       "      <td>[Books, Mystery, Thriller &amp; Suspense]</td>\n",
       "      <td>Visit Amazon's Agatha Christie Page</td>\n",
       "      <td>The Blue Geranium (Miss Marple)</td>\n",
       "      <td>None</td>\n",
       "      <td>8.82</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>3</td>\n",
       "      <td>[Books, Humor &amp; Entertainment]</td>\n",
       "      <td>Visit Amazon's Tom Baker Page</td>\n",
       "      <td>Who on Earth Is Tom Baker?</td>\n",
       "      <td>Tom Baker's autobiography covers his childhood...</td>\n",
       "      <td>159.95</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>4</td>\n",
       "      <td>[Books, Christian Books &amp; Bibles, Bible Study ...</td>\n",
       "      <td>Visit Amazon's Clarence Larkin Page</td>\n",
       "      <td>The Greatest Book on &amp;quot;Dispensational Trut...</td>\n",
       "      <td>The Greatest Book on Dispensational Truth in t...</td>\n",
       "      <td>44.76</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>...</th>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "      <td>...</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>506007</th>\n",
       "      <td>506007</td>\n",
       "      <td>None</td>\n",
       "      <td>Tricia &amp; Putman, Cara Goyer</td>\n",
       "      <td>Mason Jar Mayhem</td>\n",
       "      <td>Spring has come to Sugarcreek, and the communi...</td>\n",
       "      <td>15.05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>506008</th>\n",
       "      <td>506008</td>\n",
       "      <td>None</td>\n",
       "      <td>Author</td>\n",
       "      <td>The Little Book of Pussy (2013-06-16) [Paperback]</td>\n",
       "      <td>From the time The Big Penis Book was published...</td>\n",
       "      <td>17.00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>506009</th>\n",
       "      <td>506009</td>\n",
       "      <td>None</td>\n",
       "      <td></td>\n",
       "      <td>The Age of Not Believing: A Christmas Tale by ...</td>\n",
       "      <td>None</td>\n",
       "      <td>42.93</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>506010</th>\n",
       "      <td>506010</td>\n",
       "      <td>None</td>\n",
       "      <td></td>\n",
       "      <td>The Charlies Angels' Casebook</td>\n",
       "      <td>None</td>\n",
       "      <td>17.87</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>506011</th>\n",
       "      <td>506011</td>\n",
       "      <td>None</td>\n",
       "      <td></td>\n",
       "      <td>Underwater Photography Masterclass by Alex Mus...</td>\n",
       "      <td>None</td>\n",
       "      <td>46.62</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>506012 rows × 6 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "        product_id                                           category  \\\n",
       "0                0      [Books, Literature & Fiction, Dramas & Plays]   \n",
       "1                1    [Books, Children's Books, Literature & Fiction]   \n",
       "2                2              [Books, Mystery, Thriller & Suspense]   \n",
       "3                3                     [Books, Humor & Entertainment]   \n",
       "4                4  [Books, Christian Books & Bibles, Bible Study ...   \n",
       "...            ...                                                ...   \n",
       "506007      506007                                               None   \n",
       "506008      506008                                               None   \n",
       "506009      506009                                               None   \n",
       "506010      506010                                               None   \n",
       "506011      506011                                               None   \n",
       "\n",
       "                                          brand  \\\n",
       "0       Visit Amazon's William Shakespeare Page   \n",
       "1                         Louis (ed) UNTERMEYER   \n",
       "2           Visit Amazon's Agatha Christie Page   \n",
       "3                 Visit Amazon's Tom Baker Page   \n",
       "4           Visit Amazon's Clarence Larkin Page   \n",
       "...                                         ...   \n",
       "506007              Tricia & Putman, Cara Goyer   \n",
       "506008                                   Author   \n",
       "506009                                            \n",
       "506010                                            \n",
       "506011                                            \n",
       "\n",
       "                                                    title  \\\n",
       "0       Love's Labour's Lost: Performed by Derek Jacob...   \n",
       "1                           The golden treasury of poetry   \n",
       "2                         The Blue Geranium (Miss Marple)   \n",
       "3                              Who on Earth Is Tom Baker?   \n",
       "4       The Greatest Book on &quot;Dispensational Trut...   \n",
       "...                                                   ...   \n",
       "506007                                   Mason Jar Mayhem   \n",
       "506008  The Little Book of Pussy (2013-06-16) [Paperback]   \n",
       "506009  The Age of Not Believing: A Christmas Tale by ...   \n",
       "506010                      The Charlies Angels' Casebook   \n",
       "506011  Underwater Photography Masterclass by Alex Mus...   \n",
       "\n",
       "                                              description   price  \n",
       "0       William Shakespeare is widely regarded as the ...   20.93  \n",
       "1                                    Collection of Poetry   23.98  \n",
       "2                                                    None    8.82  \n",
       "3       Tom Baker's autobiography covers his childhood...  159.95  \n",
       "4       The Greatest Book on Dispensational Truth in t...   44.76  \n",
       "...                                                   ...     ...  \n",
       "506007  Spring has come to Sugarcreek, and the communi...   15.05  \n",
       "506008  From the time The Big Penis Book was published...   17.00  \n",
       "506009                                               None   42.93  \n",
       "506010                                               None   17.87  \n",
       "506011                                               None   46.62  \n",
       "\n",
       "[506012 rows x 6 columns]"
      ]
     },
     "execution_count": 6,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "product_df = pd.read_parquet(f\"{dataset.cache_dir}/db/product.parquet\")\n",
    "\n",
    "# Inspect the structure of the product table\n",
    "product_df\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "id": "a0f24c8b",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "\n",
    "# The category column contains lists with product categories. In that format the information is not accessible to feature learning. \n",
    "\n",
    "# For expediency, we simply pick the first informative element of the list and store it in a new column `category_level_2`.\n",
    "# To keep preprocessing simple, we will drop the columns `brand`, `title`, and `description`.\n",
    "\n",
    "product_df[\"category\"] = product_df[\"category\"].apply(\n",
    "    lambda x: x.tolist() if isinstance(x, np.ndarray) else (x or [])\n",
    ")\n",
    "\n",
    "product_df[\"category_level_2\"] = product_df[\"category\"].apply(\n",
    "    lambda arr: arr[1] if len(arr) > 1 else None\n",
    ")\n",
    "\n",
    "\n",
    "# Role assignment breakdown on `customer`:\n",
    "# - `product_id`: Acts as the primary key, linking to other tables → `join_key`\n",
    "# - `price`: Numerical data representing product price → `numerical`\n",
    "# - `category_level_2`: Categorical data → `categorical`\n",
    "\n",
    "\n",
    "## roles for the product table\n",
    "product_roles = getml.data.Roles(\n",
    "    join_key=[\"product_id\"],\n",
    "    numerical=[\"price\"],\n",
    "    categorical=[\"category_level_2\"]\n",
    ")\n",
    "\n",
    "product_df = product_df.drop(columns=[\"category\",\"brand\",\"title\",\"description\"])\n",
    "\n",
    "\n",
    "product = getml.DataFrame.from_pandas(product_df,\n",
    "    name = 'product',\n",
    "    roles = product_roles\n",
    ")\n",
    "\n"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c4b3d147",
   "metadata": {},
   "source": [
    "### b. Review Table"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "id": "e520b86a-09f5-4464-ab8e-21fb2848b9e5",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "review_roles_common = getml.data.Roles(\n",
    "    time_stamp=[\"review_time\"],\n",
    "    join_key=[\"product_id\",\"customer_id\"],\n",
    "    numerical= [\"rating\"],\n",
    ")\n",
    "\n",
    "review_base = getml.data.DataFrame.from_parquet(f\"{dataset.cache_dir}/db/review.parquet\", name = 'review', roles = review_roles_common)\n",
    "\n",
    "review_all = review_base.with_role([\"verified\"], getml.data.roles.categorical)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "id": "ab3356be-6769-44fe-9e6d-554817236684",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<style>\n",
       "  table {\n",
       "    font-family: Helvetica, sans-serif;\n",
       "  }\n",
       "\n",
       "  th {\n",
       "    text-align: left !important;\n",
       "  }\n",
       "  td {\n",
       "    text-align: left !important;\n",
       "  }\n",
       "  th:nth-child(1) {\n",
       "    text-align: right !important;\n",
       "    border-right: 1px solid LightGray;\n",
       "  }\n",
       "  th.sub-header {\n",
       "    font-weight: normal;\n",
       "    font-style: italic;\n",
       "  }\n",
       "  .join_key,\n",
       "  .numerical,\n",
       "  .target,\n",
       "  .unused_float {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "\n",
       "  .char-align {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  span.left {\n",
       "    text-align: right;\n",
       "    width: 3em;\n",
       "  }\n",
       "  span.right {\n",
       "    float: right;\n",
       "    text-align: left;\n",
       "  }\n",
       "</style>\n",
       "\n",
       "<table class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      \n",
       "        \n",
       "          <th>    name</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"time_stamp\">review_time</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"join_key\">customer_id</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"join_key\">product_id</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"numerical\">   rating</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"unused_string\">verified     </th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"unused_string\">review_text                     </th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"unused_string\">summary                         </th>\n",
       "        \n",
       "      \n",
       "    </tr>\n",
       "    \n",
       "      <tr>\n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header\">    role</th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header time_stamp\"> time_stamp</th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header join_key\">   join_key</th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header join_key\">  join_key</th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header numerical\">numerical</th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header unused_string\">unused_string</th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header unused_string\">unused_string                   </th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header unused_string\">unused_string                   </th>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header\">    unit</th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header time_stamp\"> time stamp</th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header join_key\">           </th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header join_key\">          </th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header numerical\">         </th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header unused_string\">             </th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header unused_string\">                                </th>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <th class=\"sub-header unused_string\">                                </th>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "  </thead>\n",
       "  <tbody>\n",
       "    \n",
       "      <tr>\n",
       "        <th>0</th>\n",
       "        \n",
       "          \n",
       "            <td class=\"time_stamp\">2016-06-20</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">1</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">14</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            \n",
       "              \n",
       "              \n",
       "          \n",
       "            <td class=\"char-align numerical\">\n",
       "              <span class=\"left\">5</span\n",
       "              ><span class=\"right\" style=\"width: 0ch\"\n",
       "                >&nbsp;</span\n",
       "              >\n",
       "            </td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">true</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">The kids loved it!</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Five Stars</td>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>1</th>\n",
       "        \n",
       "          \n",
       "            <td class=\"time_stamp\">2016-01-24</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">2</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">14</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            \n",
       "              \n",
       "              \n",
       "          \n",
       "            <td class=\"char-align numerical\">\n",
       "              <span class=\"left\">5</span\n",
       "              ><span class=\"right\" style=\"width: 0ch\"\n",
       "                >&nbsp;</span\n",
       "              >\n",
       "            </td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">true</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">My students (3 &amp; 4 year olds) lo...</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Five Stars</td>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>2</th>\n",
       "        \n",
       "          \n",
       "            <td class=\"time_stamp\">2015-07-09</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">3</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">14</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            \n",
       "              \n",
       "              \n",
       "          \n",
       "            <td class=\"char-align numerical\">\n",
       "              <span class=\"left\">5</span\n",
       "              ><span class=\"right\" style=\"width: 0ch\"\n",
       "                >&nbsp;</span\n",
       "              >\n",
       "            </td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">false</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">LOVE IT</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Five Stars</td>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>3</th>\n",
       "        \n",
       "          \n",
       "            <td class=\"time_stamp\">2015-01-18</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">4</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">14</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            \n",
       "              \n",
       "              \n",
       "          \n",
       "            <td class=\"char-align numerical\">\n",
       "              <span class=\"left\">5</span\n",
       "              ><span class=\"right\" style=\"width: 0ch\"\n",
       "                >&nbsp;</span\n",
       "              >\n",
       "            </td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">true</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Great!</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Five Stars</td>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>4</th>\n",
       "        \n",
       "          \n",
       "            <td class=\"time_stamp\">2013-11-03</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">5</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">14</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            \n",
       "              \n",
       "              \n",
       "          \n",
       "            <td class=\"char-align numerical\">\n",
       "              <span class=\"left\">5</span\n",
       "              ><span class=\"right\" style=\"width: 0ch\"\n",
       "                >&nbsp;</span\n",
       "              >\n",
       "            </td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">false</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Over and over the king has probl...</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Not Nice Mice</td>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th></th>\n",
       "        \n",
       "          \n",
       "            <td class=\"time_stamp\">...</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">...</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">...</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            \n",
       "              \n",
       "              \n",
       "          \n",
       "            <td class=\"char-align numerical\">\n",
       "              <span class=\"left\">...</span\n",
       "              ><span class=\"right\" style=\"width: 0ch\"\n",
       "                >&nbsp;</span\n",
       "              >\n",
       "            </td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">...</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">...</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">...</td>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>20862035</th>\n",
       "        \n",
       "          \n",
       "            <td class=\"time_stamp\">2016-10-31</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">1551873</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">506010</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            \n",
       "              \n",
       "              \n",
       "          \n",
       "            <td class=\"char-align numerical\">\n",
       "              <span class=\"left\">5</span\n",
       "              ><span class=\"right\" style=\"width: 0ch\"\n",
       "                >&nbsp;</span\n",
       "              >\n",
       "            </td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">true</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">great book</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Five Stars</td>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>20862036</th>\n",
       "        \n",
       "          \n",
       "            <td class=\"time_stamp\">2012-12-02</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">319196</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">506010</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            \n",
       "              \n",
       "              \n",
       "          \n",
       "            <td class=\"char-align numerical\">\n",
       "              <span class=\"left\">5</span\n",
       "              ><span class=\"right\" style=\"width: 0ch\"\n",
       "                >&nbsp;</span\n",
       "              >\n",
       "            </td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">false</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">This past year fans of &quot;Charlie&#x27;...</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">&quot;NOW THIS IS THE COMPANION TO TH...</td>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>20862037</th>\n",
       "        \n",
       "          \n",
       "            <td class=\"time_stamp\">2008-12-15</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">1407251</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">506010</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            \n",
       "              \n",
       "              \n",
       "          \n",
       "            <td class=\"char-align numerical\">\n",
       "              <span class=\"left\">5</span\n",
       "              ><span class=\"right\" style=\"width: 0ch\"\n",
       "                >&nbsp;</span\n",
       "              >\n",
       "            </td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">true</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">For any fan of Charlie&#x27;s Angels,...</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">An Excellent Case!</td>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>20862038</th>\n",
       "        \n",
       "          \n",
       "            <td class=\"time_stamp\">2018-05-27</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">319324</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">506011</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            \n",
       "              \n",
       "              \n",
       "          \n",
       "            <td class=\"char-align numerical\">\n",
       "              <span class=\"left\">3</span\n",
       "              ><span class=\"right\" style=\"width: 0ch\"\n",
       "                >&nbsp;</span\n",
       "              >\n",
       "            </td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">true</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Great pictures, good recommendat...</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Don&#x27;t buy if your looking for de...</td>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>20862039</th>\n",
       "        \n",
       "          \n",
       "            <td class=\"time_stamp\">2017-09-11</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">1703463</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"join_key\">506011</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            \n",
       "              \n",
       "              \n",
       "          \n",
       "            <td class=\"char-align numerical\">\n",
       "              <span class=\"left\">5</span\n",
       "              ><span class=\"right\" style=\"width: 0ch\"\n",
       "                >&nbsp;</span\n",
       "              >\n",
       "            </td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">true</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Superb book. Very useful.</td>\n",
       "          \n",
       "        \n",
       "          \n",
       "            <td class=\"unused_string\">Superb book.</td>\n",
       "          \n",
       "        \n",
       "      </tr>\n",
       "    \n",
       "  </tbody>\n",
       "</table>\n",
       "\n",
       "  <p>\n",
       "    20862040 rows x 7 columns<br />\n",
       "    memory usage: 12364.26 MB<br />\n",
       "    name: review<br />\n",
       "    type: getml.DataFrame<br />\n",
       "    \n",
       "  </p>\n"
      ],
      "text/plain": [
       "    name   review_time   customer_id   product_id      rating   verified        review_text                     \n",
       "    role    time_stamp      join_key     join_key   numerical   unused_string   unused_string                   \n",
       "    unit    time stamp                                                                                          \n",
       "       0    2016-06-20             1           14           5   true            The kids loved it!              \n",
       "       1    2016-01-24             2           14           5   true            My students (3 & 4 year olds) lo...\n",
       "       2    2015-07-09             3           14           5   false           LOVE IT                         \n",
       "       3    2015-01-18             4           14           5   true            Great!                          \n",
       "       4    2013-11-03             5           14           5   false           Over and over the king has probl...\n",
       "                   ...           ...          ...         ...   ...             ...                             \n",
       "20862035    2016-10-31       1551873       506010           5   true            great book                      \n",
       "20862036    2012-12-02        319196       506010           5   false           This past year fans of \"Charlie'...\n",
       "20862037    2008-12-15       1407251       506010           5   true            For any fan of Charlie's Angels,...\n",
       "20862038    2018-05-27        319324       506011           3   true            Great pictures, good recommendat...\n",
       "20862039    2017-09-11       1703463       506011           5   true            Superb book. Very useful.       \n",
       "\n",
       "    name   summary                         \n",
       "    role   unused_string                   \n",
       "    unit                                   \n",
       "       0   Five Stars                      \n",
       "       1   Five Stars                      \n",
       "       2   Five Stars                      \n",
       "       3   Five Stars                      \n",
       "       4   Not Nice Mice                   \n",
       "           ...                             \n",
       "20862035   Five Stars                      \n",
       "20862036   \"NOW THIS IS THE COMPANION TO TH...\n",
       "20862037   An Excellent Case!              \n",
       "20862038   Don't buy if your looking for de...\n",
       "20862039   Superb book.                    \n",
       "\n",
       "\n",
       "20862040 rows x 7 columns\n",
       "memory usage: 12364.26 MB\n",
       "type: getml.DataFrame"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "review_base"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "81934512",
   "metadata": {},
   "source": [
    "### c. Customer Table"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "id": "a203dd64",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "# The customer table only contains the `customer_id` column and the customer's `name`. It is intuitive to assume that customers` names are non-informative for predicting item churn. We will therefore ignore the entire customer table.\n",
    "\n",
    "# customer = getml.data.DataFrame.from_parquet(f\"{dataset.cache_dir}/db/customer.parquet\", name = 'customer')\n",
    "\n",
    "# customer"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ce3f4fb8",
   "metadata": {},
   "source": [
    "### d. Population tables  (train, test, val)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4f8dced3",
   "metadata": {},
   "source": [
    "The population table holds the statistical population for the inference task at hand. Thus,\n",
    "it is the table that holds the target variable, the possible reference dates (to prevent data leakage),\n",
    "and is split into train, test, and validation sets. The population table is the canonical table data\n",
    "scientist are used to work with in \"classical\", tabular machine learning tasks. It is the table that will be\n",
    "handed over to the machine learning algorithm for training and prediction, and therefore is the target\n",
    "entity for the feature learning process.\n",
    "\n",
    "As we were loading the relbench task, the population table is already split into train, test, and val and contains\n",
    "nothing else than the target variable (`churn`), the primary key (`product_id`), and the reference date (`timestamp`). Accordingly,\n",
    "we assign the following roles:\n",
    "- `product_id` is the join key.\n",
    "- `churn` is the target variable.\n",
    "- `timestamp` represents the prediction or cut-off date to prevent data leakage."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "id": "68fd21d2",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "population_roles = getml.data.Roles(\n",
    "    join_key=[\"product_id\"],\n",
    "    target=[\"churn\"],\n",
    "    time_stamp=[\"timestamp\"],\n",
    ")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "id": "a0bccdd6",
   "metadata": {
    "tags": []
   },
   "outputs": [],
   "source": [
    "population_roles = getml.data.Roles(\n",
    "    join_key=[\"product_id\"], target=[\"churn\"], time_stamp=[\"timestamp\"]\n",
    ")\n",
    "\n",
    "train = getml.data.DataFrame.from_parquet(\n",
    "     f\"{task.cache_dir}/train.parquet\", \"train\", population_roles\n",
    ")\n",
    "\n",
    "val = getml.data.DataFrame.from_parquet(\n",
    "     f\"{task.cache_dir}/val.parquet\", \"val\", population_roles\n",
    ")\n",
    "\n",
    "test = getml.data.DataFrame.from_parquet(\n",
    "     f\"{task.cache_dir}/test.parquet\", \"test\", population_roles\n",
    ")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "33a31d38",
   "metadata": {},
   "source": [
    "---\n",
    "## 3. Defining the getML DataModel\n",
    "\n",
    "In getML, the [`DataModel`](https://getml.com/latest/user_guide/concepts/data_model/) serves as the foundation for relational feature learning. The DataModel acts as the search space for features, similar to the schema of a relational database. It defines the universe of possible feature paths that getML’s algorithms can explore and learn from.\n",
    "\n",
    "The data model is structured as a Directed Acyclic Graph, where each [join path](https://getml.com/latest/user_guide/concepts/data_model/#joins) represents a potential feature learning route. This structure enables automatic extraction of predictive signals (features) from complex, multi-table datasets.\n",
    "\n",
    "Trivial joins ([one-to-one](https://getml.com/latest/reference/data/relationship/#getml.data.relationship.one_to_one) or [many-to-one](https://getml.com/latest/reference/data/relationship/#getml.data.relationship.many_to_one)) onto the population don’t produce new feature learning routes but columns in such tables are added to the main table _as is._ So, these paths are useful for enriching data but aren’t part of the feature learning process.\n",
    "\n",
    "### Amazon DataModel Overview\n",
    "<img src=\"https://relbench.stanford.edu/img/rel-amazon.png\" width=\"500\"/>\n",
    "\n",
    "* The primary feature learning path is the [*one-to-many* relationship](https://getml.com/latest/reference/data/relationship/#getml.data.relationship) between the `population` table and the `review` table. This is where most valuable features are derived.\n",
    "* The `product` table enriches the `population` and `review` tables, through trivial joins and introduces new feature learning paths, that the feature learner can aggregate over. The `customer` table is dropped since it is not conceivable that information is encoded in the customer's name.\n",
    "\n",
    "#### Key Concepts:\n",
    "* The [`DataModel`](https://getml.com/latest/user_guide/concepts/data_model/) is abstract. It defines the relational structure but does not contain any data itself.\n",
    "* [`Placeholders`](https://getml.com/latest/user_guide/concepts/data_model/#placeholders) represent tables in the model. These placeholders mirror the schema of actual DataFrames but are decoupled from raw data until the point of training."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "id": "353efd95",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "            <div style='margin-top: 15px; margin-bottom: 5px;'>\n",
       "            <div style='margin-bottom: 10px; font-size: 1rem;'>diagram</div>\n",
       "            <div style=\"height:320px;width:1660px;position:relative;\"><svg height=\"310\" width=\"1650\"><rect y=\"0\" x=\"1000\" rx=\"10\" ry=\"10\" width=\"150\" height=\"90\" style=\"fill:#6829c2;stroke-width:0;\" /><text y=\"73.8\" x=\"1075.0\" dominant-baseline=\"middle\" text-anchor=\"middle\" fill=\"white\">product</text><rect x=\"1051\" y=\"10\" rx=\"4\" ry=\"4\" width=\"48\" height=\"48\" style=\" fill:#6829c2;stroke:#ffffff;stroke-width:3;\" /><line x1=\"1067.0\" y1=\"10\" x2=\"1067.0\" y2=\"58\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1083.0\" y1=\"10\" x2=\"1083.0\" y2=\"58\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1051\" y1=\"26.0\" x2=\"1099\" y2=\"26.0\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1051\" y1=\"42.0\" x2=\"1099\" y2=\"42.0\" style=\"stroke:white;stroke-width:3\" /><rect y=\"110\" x=\"1000\" rx=\"10\" ry=\"10\" width=\"150\" height=\"90\" style=\"fill:#6829c2;stroke-width:0;\" /><text y=\"183.8\" x=\"1075.0\" dominant-baseline=\"middle\" text-anchor=\"middle\" fill=\"white\">review_recent</text><rect x=\"1051\" y=\"120\" rx=\"4\" ry=\"4\" width=\"48\" height=\"48\" style=\" fill:#6829c2;stroke:#ffffff;stroke-width:3;\" /><line x1=\"1067.0\" y1=\"120\" x2=\"1067.0\" y2=\"168\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1083.0\" y1=\"120\" x2=\"1083.0\" y2=\"168\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1051\" y1=\"136.0\" x2=\"1099\" y2=\"136.0\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1051\" y1=\"152.0\" x2=\"1099\" y2=\"152.0\" style=\"stroke:white;stroke-width:3\" /><rect y=\"220\" x=\"0\" rx=\"10\" ry=\"10\" width=\"150\" height=\"90\" style=\"fill:#6829c2;stroke-width:0;\" /><text y=\"293.8\" x=\"75.0\" dominant-baseline=\"middle\" text-anchor=\"middle\" fill=\"white\">product</text><rect x=\"51\" y=\"230\" rx=\"4\" ry=\"4\" width=\"48\" height=\"48\" style=\" fill:#6829c2;stroke:#ffffff;stroke-width:3;\" /><line x1=\"67.0\" y1=\"230\" x2=\"67.0\" y2=\"278\" style=\"stroke:white;stroke-width:3\" /><line x1=\"83.0\" y1=\"230\" x2=\"83.0\" y2=\"278\" style=\"stroke:white;stroke-width:3\" /><line x1=\"51\" y1=\"246.0\" x2=\"99\" y2=\"246.0\" style=\"stroke:white;stroke-width:3\" /><line x1=\"51\" y1=\"262.0\" x2=\"99\" y2=\"262.0\" style=\"stroke:white;stroke-width:3\" /><rect y=\"220\" x=\"500\" rx=\"10\" ry=\"10\" width=\"150\" height=\"90\" style=\"fill:#6829c2;stroke-width:0;\" /><text y=\"293.8\" x=\"575.0\" dominant-baseline=\"middle\" text-anchor=\"middle\" fill=\"white\">review_customer</text><rect x=\"551\" y=\"230\" rx=\"4\" ry=\"4\" width=\"48\" height=\"48\" style=\" fill:#6829c2;stroke:#ffffff;stroke-width:3;\" /><line x1=\"567.0\" y1=\"230\" x2=\"567.0\" y2=\"278\" style=\"stroke:white;stroke-width:3\" /><line x1=\"583.0\" y1=\"230\" x2=\"583.0\" y2=\"278\" style=\"stroke:white;stroke-width:3\" /><line x1=\"551\" y1=\"246.0\" x2=\"599\" y2=\"246.0\" style=\"stroke:white;stroke-width:3\" /><line x1=\"551\" y1=\"262.0\" x2=\"599\" y2=\"262.0\" style=\"stroke:white;stroke-width:3\" /><rect y=\"220\" x=\"1000\" rx=\"10\" ry=\"10\" width=\"150\" height=\"90\" style=\"fill:#6829c2;stroke-width:0;\" /><text y=\"293.8\" x=\"1075.0\" dominant-baseline=\"middle\" text-anchor=\"middle\" fill=\"white\">review_all</text><rect x=\"1051\" y=\"230\" rx=\"4\" ry=\"4\" width=\"48\" height=\"48\" style=\" fill:#6829c2;stroke:#ffffff;stroke-width:3;\" /><line x1=\"1067.0\" y1=\"230\" x2=\"1067.0\" y2=\"278\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1083.0\" y1=\"230\" x2=\"1083.0\" y2=\"278\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1051\" y1=\"246.0\" x2=\"1099\" y2=\"246.0\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1051\" y1=\"262.0\" x2=\"1099\" y2=\"262.0\" style=\"stroke:white;stroke-width:3\" /><rect y=\"220\" x=\"1500\" rx=\"10\" ry=\"10\" width=\"150\" height=\"90\" style=\"fill:#6829c2;stroke-width:0;\" /><text y=\"293.8\" x=\"1575.0\" dominant-baseline=\"middle\" text-anchor=\"middle\" fill=\"white\">train</text><rect x=\"1551\" y=\"230\" rx=\"4\" ry=\"4\" width=\"48\" height=\"48\" style=\" fill:#6829c2;stroke:#ffffff;stroke-width:3;\" /><line x1=\"1567.0\" y1=\"230\" x2=\"1567.0\" y2=\"278\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1583.0\" y1=\"230\" x2=\"1583.0\" y2=\"278\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1551\" y1=\"246.0\" x2=\"1599\" y2=\"246.0\" style=\"stroke:white;stroke-width:3\" /><line x1=\"1551\" y1=\"262.0\" x2=\"1599\" y2=\"262.0\" style=\"stroke:white;stroke-width:3\" /><line x1=\"150\" y1=\"263.0\" x2=\"490\" y2=\"263.0\" style=\"stroke:#808080;;stroke-width:4\" /><polygon points=\"500, 263.0 490, 257.0 490, 269.0 \" style=\"fill:#808080;;stroke-width:0;\" /><rect y=\"230.0\" x=\"249.0\" rx=\"10\" ry=\"10\" width=\"150\" height=\"70\" style=\"fill:#6829c2;stroke-width:0;\" /><text dominant-baseline=\"middle\" text-anchor=\"middle\" fill=\"white\"><tspan y=\"260.0\" x=\"324.0\" font-size=\"7pt\" >product_id = product_id</tspan><tspan y=\"270.0\" x=\"324.0\" font-size=\"7pt\" >Relationship: many-to-one</tspan></text><line x1=\"650\" y1=\"263.0\" x2=\"990\" y2=\"263.0\" style=\"stroke:#808080;;stroke-width:4\" /><polygon points=\"1000, 263.0 990, 257.0 990, 269.0 \" style=\"fill:#808080;;stroke-width:0;\" /><rect y=\"230.0\" x=\"749.0\" rx=\"10\" ry=\"10\" width=\"150\" height=\"70\" style=\"fill:#6829c2;stroke-width:0;\" /><text dominant-baseline=\"middle\" text-anchor=\"middle\" fill=\"white\"><tspan y=\"260.0\" x=\"824.0\" font-size=\"7pt\" >customer_id = customer_id</tspan><tspan y=\"270.0\" x=\"824.0\" font-size=\"7pt\" >review_time &lt;= review_time</tspan></text><line x1=\"1150\" y1=\"43.0\" x2=\"1573.0\" y2=\"43.0\" style=\"stroke:#808080;;stroke-width:4\" /><line x1=\"1573.0\" y1=\"41.0\" x2=\"1573.0\" y2=\"210\" style=\"stroke:#808080;;stroke-width:4\" /><polygon points=\"1573.0, 220 1567.0, 210 1579.0, 210 \" style=\"fill:#808080;;stroke-width:0;\" /><rect y=\"10.0\" x=\"1249.0\" rx=\"10\" ry=\"10\" width=\"150\" height=\"70\" style=\"fill:#6829c2;stroke-width:0;\" /><text dominant-baseline=\"middle\" text-anchor=\"middle\" fill=\"white\"><tspan y=\"40.0\" x=\"1324.0\" font-size=\"7pt\" >product_id = product_id</tspan><tspan y=\"50.0\" x=\"1324.0\" font-size=\"7pt\" >Relationship: many-to-one</tspan></text><line x1=\"1150\" y1=\"153.0\" x2=\"1573.0\" y2=\"153.0\" style=\"stroke:#808080;;stroke-width:4\" /><line x1=\"1573.0\" y1=\"151.0\" x2=\"1573.0\" y2=\"210\" style=\"stroke:#808080;;stroke-width:4\" /><polygon points=\"1573.0, 220 1567.0, 210 1579.0, 210 \" style=\"fill:#808080;;stroke-width:0;\" /><rect y=\"120.0\" x=\"1249.0\" rx=\"10\" ry=\"10\" width=\"150\" height=\"70\" style=\"fill:#6829c2;stroke-width:0;\" /><text dominant-baseline=\"middle\" text-anchor=\"middle\" fill=\"white\"><tspan y=\"145.0\" x=\"1324.0\" font-size=\"7pt\" >product_id = product_id</tspan><tspan y=\"155.0\" x=\"1324.0\" font-size=\"7pt\" >review_time &lt;= timestamp</tspan><tspan y=\"165.0\" x=\"1324.0\" font-size=\"7pt\" >Memory: 365.0 days</tspan></text><line x1=\"1150\" y1=\"263.0\" x2=\"1490\" y2=\"263.0\" style=\"stroke:#808080;;stroke-width:4\" /><polygon points=\"1500, 263.0 1490, 257.0 1490, 269.0 \" style=\"fill:#808080;;stroke-width:0;\" /><rect y=\"230.0\" x=\"1249.0\" rx=\"10\" ry=\"10\" width=\"150\" height=\"70\" style=\"fill:#6829c2;stroke-width:0;\" /><text dominant-baseline=\"middle\" text-anchor=\"middle\" fill=\"white\"><tspan y=\"260.0\" x=\"1324.0\" font-size=\"7pt\" >product_id = product_id</tspan><tspan y=\"270.0\" x=\"1324.0\" font-size=\"7pt\" >review_time &lt;= timestamp</tspan></text></svg></div>\n",
       "            </div>\n",
       "\n",
       "            <div style='margin-top: 15px;'>\n",
       "            <div style='margin-bottom: 10px; font-size: 1rem;'>staging</div>\n",
       "            <style>\n",
       "  th {\n",
       "    text-align: left !important;\n",
       "  }\n",
       "  td {\n",
       "    text-align: left !important;\n",
       "  }\n",
       "  th:nth-child(1) {\n",
       "    text-align: right;\n",
       "    border-right: 1px solid LightGray;\n",
       "  }\n",
       "  th.float {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  td.float {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  th.int {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  td.int {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "</style>\n",
       "\n",
       "<table class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      \n",
       "        \n",
       "          <th class=\"int\"> </th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"str\">data frames             </th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"str\">staging table                   </th>\n",
       "        \n",
       "      \n",
       "    </tr>\n",
       "    \n",
       "  </thead>\n",
       "  <tbody>\n",
       "    \n",
       "      <tr>\n",
       "        <th>0</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"str\">train, product</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">TRAIN__STAGING_TABLE_1</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>1</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"str\">review_all</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">REVIEW_ALL__STAGING_TABLE_2</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>2</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"str\">review_customer, product</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">REVIEW_CUSTOMER__STAGING_TABLE_3</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>3</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"str\">review_recent</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">REVIEW_RECENT__STAGING_TABLE_4</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "  </tbody>\n",
       "</table>\n",
       "            </div>\n",
       "            "
      ],
      "text/plain": [
       "train:\n",
       "  columns:\n",
       "  - product_id: join_key\n",
       "  - churn: target\n",
       "  - timestamp: time_stamp\n",
       "\n",
       "  joins:\n",
       "  - right: 'product'\n",
       "    on: \n",
       "    - (train.product_id, product.product_id)\n",
       "    relationship: 'many-to-one'\n",
       "    lagged_targets: False\n",
       "  - right: 'review_recent'\n",
       "    on: \n",
       "    - (train.product_id, review_recent.product_id)\n",
       "    time_stamps: (train.timestamp, review_recent.review_time)\n",
       "    relationship: 'many-to-many'\n",
       "    memory: 31536000.0\n",
       "    lagged_targets: False\n",
       "  - right: 'review_all'\n",
       "    on: \n",
       "    - (train.product_id, review_all.product_id)\n",
       "    time_stamps: (train.timestamp, review_all.review_time)\n",
       "    relationship: 'many-to-many'\n",
       "    lagged_targets: False\n",
       "\n",
       "product:\n",
       "  columns:\n",
       "  - category_level_2: categorical\n",
       "  - product_id: join_key\n",
       "  - price: numerical\n",
       "\n",
       "review_recent:\n",
       "  columns:\n",
       "  - customer_id: join_key\n",
       "  - product_id: join_key\n",
       "  - rating: numerical\n",
       "  - review_time: time_stamp\n",
       "  - verified: unused_string\n",
       "  - ...\n",
       "\n",
       "review_all:\n",
       "  columns:\n",
       "  - verified: categorical\n",
       "  - customer_id: join_key\n",
       "  - product_id: join_key\n",
       "  - rating: numerical\n",
       "  - review_time: time_stamp\n",
       "  - ...\n",
       "\n",
       "  joins:\n",
       "  - right: 'review_customer'\n",
       "    on: \n",
       "    - (review_all.customer_id, review_customer.customer_id)\n",
       "    time_stamps: (review_all.review_time, review_customer.review_time)\n",
       "    relationship: 'many-to-many'\n",
       "    lagged_targets: False\n",
       "\n",
       "review_customer:\n",
       "  columns:\n",
       "  - customer_id: join_key\n",
       "  - product_id: join_key\n",
       "  - rating: numerical\n",
       "  - review_time: time_stamp\n",
       "  - verified: unused_string\n",
       "  - ...\n",
       "\n",
       "  joins:\n",
       "  - right: 'product'\n",
       "    on: \n",
       "    - (review_customer.product_id, product.product_id)\n",
       "    relationship: 'many-to-one'\n",
       "    lagged_targets: False\n",
       "\n",
       "product:\n",
       "  columns:\n",
       "  - category_level_2: categorical\n",
       "  - product_id: join_key\n",
       "  - price: numerical"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# 1. Derive the population placeholder from the “train” DataFrame in our populations dict.\n",
    "#    These rows form our target population for learning and prediction (we could\n",
    "#    also use val or test here, as those share the same schema).\n",
    "dm = getml.data.DataModel(population=train.to_placeholder())\n",
    "\n",
    "# 2. Add placeholders for product and review tables.\n",
    "#    These tables hold valuable information relevant to predictions.\n",
    "#    Without this step, the predictor cannot access the knowledge within\n",
    "#    these tables, leading to poorer model performance.\n",
    "dm.add(    product.to_placeholder(), \n",
    "    review_all.to_placeholder(\"review_all\"), \n",
    "    review_base.to_placeholder(\"review_customer\"),\n",
    "    review_base.to_placeholder(\"review_recent\"),\n",
    "    )\n",
    "\n",
    "# 3. Specify relational links:\n",
    "#    - population -> product on \"product id\" (many-to-one)\n",
    "#    - population -> review_recent on \"product_id\" with time_stamps to prevent data leakage\n",
    "#    - population -> review on \"product_id\" with time_stamps to prevent data leakage\n",
    "#    - population -> review_all on \"product_id\" with time_stamps to prevent data leakage\n",
    "#    - review_all -> review_customer on \"customer_id\" with time_stamps to prevent data leakage\n",
    "#    - review_customer -> product on \"product_id\" (many-to-one)\n",
    "dm.population.join(\n",
    "    dm.product, on=\"product_id\", relationship=getml.data.relationship.many_to_one\n",
    ")\n",
    "\n",
    "dm.population.join(\n",
    "    dm.review_recent, on=\"product_id\", time_stamps=(\"timestamp\", \"review_time\"),\n",
    "    memory = getml.data.time.days(365)\n",
    ")\n",
    "\n",
    "dm.population.join(\n",
    "    dm.review_all, on=\"product_id\", time_stamps=(\"timestamp\", \"review_time\")\n",
    ")\n",
    "\n",
    "dm.review_all.join(\n",
    "    dm.review_customer, on=\"customer_id\", time_stamps=(\"review_time\", \"review_time\")\n",
    ")\n",
    "\n",
    "dm.review_customer.join(\n",
    "    dm.product, on=\"product_id\", relationship=getml.data.relationship.many_to_one\n",
    ")\n",
    "\n",
    "dm"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "455c6b77",
   "metadata": {},
   "source": [
    "---\n",
    "## 4. Creating a getML Container referencing the Data\n",
    "\n",
    "The [`Container`](https://getml.com/latest/reference/data/container/#getml.data.Container) holds _all_ data (i.e. population, split into train, test, val, _and_ all peripheral tables: customer, transaction, article).\n",
    "The actual data in a container will be linked to the abstract [**DataModel**](https://getml.com/latest/reference/data/data_model/) during training.\n",
    "\n",
    "Why a Container is an important construct:\n",
    "- `Container` is a convenience API that eases handling of relational data for Machine Learning tasks\n",
    "- The DataModel defines the relational structure; the Container holds references to the actual data.\n",
    "- During `pipeline.fit(container.train)`, the Container maps [*data*](https://getml.com/latest/reference/data/data_frame/) to [*placeholders*](https://getml.com/latest/reference/data/placeholder/),\n",
    "  allowing getML to apply relational joins and time constraints for feature learning.\n",
    "- No data is duplicated – the Container references the original tables, ensuring efficiency.\n",
    "- A container allows you to easily pass around relational datasets for training, validation, and testing: E.g. `container.train` holds a [Subset](https://getml.com/latest/reference/data/subset/) that contains the training data for the population and all related peripheral tables.\n",
    "\n",
    "This separation enhances reproducibility and keeps data handling modular."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "id": "c488aa9d-6f3d-4105-b7c3-fe11733d9622",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div style='margin-top: 15px;'>\n",
       "<div style='float: left; margin-right: 50px;'>\n",
       "<div style='margin-bottom: 10px; font-size: 1rem;'>population</div>\n",
       "    <style>\n",
       "  th {\n",
       "    text-align: left !important;\n",
       "  }\n",
       "  td {\n",
       "    text-align: left !important;\n",
       "  }\n",
       "  th:nth-child(1) {\n",
       "    text-align: right;\n",
       "    border-right: 1px solid LightGray;\n",
       "  }\n",
       "  th.float {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  td.float {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  th.int {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  td.int {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "</style>\n",
       "\n",
       "<table class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      \n",
       "        \n",
       "          <th class=\"int\"> </th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"str\">subset</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"str\">name </th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"int\">   rows</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"str\">type     </th>\n",
       "        \n",
       "      \n",
       "    </tr>\n",
       "    \n",
       "  </thead>\n",
       "  <tbody>\n",
       "    \n",
       "      <tr>\n",
       "        <th>0</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"str\">train</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">train</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"int\">2536014</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">DataFrame</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>1</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"str\">test</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">test</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"int\">166842</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">DataFrame</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>2</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"str\">val</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">val</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"int\">177689</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">DataFrame</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "  </tbody>\n",
       "</table>\n",
       "</div>\n",
       "<div style='float: left;'>\n",
       "<div style='margin-bottom: 10px; font-size: 1rem;'>peripheral</div>\n",
       "    <style>\n",
       "  th {\n",
       "    text-align: left !important;\n",
       "  }\n",
       "  td {\n",
       "    text-align: left !important;\n",
       "  }\n",
       "  th:nth-child(1) {\n",
       "    text-align: right;\n",
       "    border-right: 1px solid LightGray;\n",
       "  }\n",
       "  th.float {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  td.float {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  th.int {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  td.int {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "</style>\n",
       "\n",
       "<table class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      \n",
       "        \n",
       "          <th class=\"int\"> </th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"str\">alias          </th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"str\">name   </th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"int\">    rows</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"str\">type     </th>\n",
       "        \n",
       "      \n",
       "    </tr>\n",
       "    \n",
       "  </thead>\n",
       "  <tbody>\n",
       "    \n",
       "      <tr>\n",
       "        <th>0</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"str\">product</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">product</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"int\">506012</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">DataFrame</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>1</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"str\">review_all</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">review</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"int\">20862040</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">View</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>2</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"str\">review_customer</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">review</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"int\">20862040</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">DataFrame</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>3</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"str\">review_recent</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">review</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"int\">20862040</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">DataFrame</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "  </tbody>\n",
       "</table>\n",
       "</div>\n",
       "</div>"
      ],
      "text/plain": [
       "population\n",
       "    subset   name       rows   type     \n",
       "0   train    train   2536014   DataFrame\n",
       "1   test     test     166842   DataFrame\n",
       "2   val      val      177689   DataFrame\n",
       "\n",
       "peripheral\n",
       "    alias             name          rows   type     \n",
       "0   product           product     506012   DataFrame\n",
       "1   review_all        review    20862040   View     \n",
       "2   review_customer   review    20862040   DataFrame\n",
       "3   review_recent     review    20862040   DataFrame"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "\n",
    "# The constructor of the container receives the population tables (in this case already split into train, test, val)\n",
    "container = getml.data.Container(train=train, val=val, test=test)\n",
    "\n",
    "# `add` adds peripheral tables to the container\n",
    "container.add(product, review_all=review_all, review_customer=review_base, review_recent=review_base)\n",
    "\n",
    "# Display the container to verify all splits and tables are correctly loaded.\n",
    "container"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "id": "65142f47-2cd7-46d7-8cb6-a6f27b4010e6",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">Checking data model<span style=\"color: #808000; text-decoration-color: #808000\">...</span>\n",
       "</pre>\n"
      ],
      "text/plain": [
       "Checking data model\u001b[33m...\u001b[0m\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[2K  Staging... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:01\n",
      "\u001b[2K  Checking... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:46\n",
      "\u001b[?25h"
     ]
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">OK.\n",
       "</pre>\n"
      ],
      "text/plain": [
       "OK.\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "pipe = getml.Pipeline(\n",
    "    tags=[\"base_pipeline\"],\n",
    "    data_model=dm,\n",
    "    feature_learners=[getml.feature_learning.FastProp(\n",
    "        num_threads = 32,\n",
    "        n_most_frequent=0,\n",
    "        num_features=1000,\n",
    "        aggregation=(\n",
    "            getml.feature_learning.FastProp.agg_sets.default \n",
    "        ),\n",
    "    )],\n",
    "    predictors=[getml.predictors.XGBoostClassifier(max_depth = 5, n_jobs = 32)],\n",
    "    loss_function=getml.feature_learning.loss_functions.CrossEntropyLoss,\n",
    ")\n",
    "\n",
    "# .check(...) does an analysis on the pipeline, data model and container\n",
    "# to see if everything is consistent (time relationships, roles, table joins, etc.).\n",
    "# We pass the 'train' subset for verification\n",
    "pipe.check(container.train)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "b6dc8039",
   "metadata": {},
   "source": [
    "---\n",
    "## 6. Fitting and Scoring the getML Pipeline"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "id": "d72225bf",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[2K  Staging... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:27\n",
      "\u001b[2K  FastProp: Trying 702 features... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00\n",
      "\u001b[2K  FastProp: Building subfeatures... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 01:48\n",
      "\u001b[2K  FastProp: Building features... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 32:54\n",
      "\u001b[2K  XGBoost: Training as predictor... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 14:53\n",
      "\u001b[?25h"
     ]
    },
    {
     "data": {
      "text/html": [
       "<pre style=\"white-space:pre;overflow-x:auto;line-height:normal;font-family:Menlo,'DejaVu Sans Mono',consolas,'Courier New',monospace\">Trained pipeline.\n",
       "</pre>\n"
      ],
      "text/plain": [
       "Trained pipeline.\n"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Time taken: 0:50:08.082392.\n",
      "\n"
     ]
    },
    {
     "data": {
      "text/html": [
       "<pre>Pipeline(data_model='train',\n",
       "         feature_learners=['FastProp'],\n",
       "         feature_selectors=[],\n",
       "         include_categorical=False,\n",
       "         loss_function='CrossEntropyLoss',\n",
       "         peripheral=['product', 'review_all', 'review_customer', 'review_recent'],\n",
       "         predictors=['XGBoostClassifier'],\n",
       "         preprocessors=[],\n",
       "         share_selected_features=0.5,\n",
       "         tags=['base_pipeline', 'container-OxFq8x'])</pre>"
      ],
      "text/plain": [
       "Pipeline(data_model='train',\n",
       "         feature_learners=['FastProp'],\n",
       "         feature_selectors=[],\n",
       "         include_categorical=False,\n",
       "         loss_function='CrossEntropyLoss',\n",
       "         peripheral=['product', 'review_all', 'review_customer', 'review_recent'],\n",
       "         predictors=['XGBoostClassifier'],\n",
       "         preprocessors=[],\n",
       "         share_selected_features=0.5,\n",
       "         tags=['base_pipeline', 'container-OxFq8x'])"
      ]
     },
     "execution_count": 22,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# .fit(...) orchestrates learning the features, and training the prediction model.\n",
    "\n",
    "pipe.fit(container.train, check=False)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "id": "1db8c155",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "\u001b[2K  Staging... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:01\n",
      "\u001b[2K  Preprocessing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00\n",
      "\u001b[2K  FastProp: Building subfeatures... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 01:49\n",
      "\u001b[2K  FastProp: Building features... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 02:51\n",
      "\u001b[2K  Staging... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:01\n",
      "\u001b[2K  Preprocessing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00\n",
      "\u001b[2K  FastProp: Building subfeatures... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 01:49\n",
      "\u001b[2K  FastProp: Building features... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 02:45\n",
      "\u001b[?25h"
     ]
    },
    {
     "data": {
      "text/html": [
       "<style>\n",
       "  th {\n",
       "    text-align: left !important;\n",
       "  }\n",
       "  td {\n",
       "    text-align: left !important;\n",
       "  }\n",
       "  th:nth-child(1) {\n",
       "    text-align: right;\n",
       "    border-right: 1px solid LightGray;\n",
       "  }\n",
       "  th.float {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  td.float {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  th.int {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "  td.int {\n",
       "    text-align: right !important;\n",
       "  }\n",
       "</style>\n",
       "\n",
       "<table class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr>\n",
       "      \n",
       "        \n",
       "          <th class=\"int\"> </th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"datetime\">date time          </th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"str\">set used</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"str\">target</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"float\">accuracy</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"float\">    auc</th>\n",
       "        \n",
       "      \n",
       "        \n",
       "          <th class=\"float\">cross entropy</th>\n",
       "        \n",
       "      \n",
       "    </tr>\n",
       "    \n",
       "  </thead>\n",
       "  <tbody>\n",
       "    \n",
       "      <tr>\n",
       "        <th>0</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"datetime\">2025-02-07 00:22:34</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">train</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">churn</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"float\">0.7382</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"float\">0.8181</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"float\">0.508</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>1</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"datetime\">2025-02-07 00:27:20</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">val</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">churn</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"float\">0.7454</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"float\">0.824</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"float\">0.504</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "      <tr>\n",
       "        <th>2</th>\n",
       "          \n",
       "            \n",
       "              <td class=\"datetime\">2025-02-07 00:32:01</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">test</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"str\">churn</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"float\">0.7544</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"float\">0.828</td>\n",
       "            \n",
       "          \n",
       "            \n",
       "              <td class=\"float\">0.4845</td>\n",
       "            \n",
       "          \n",
       "      </tr>\n",
       "    \n",
       "  </tbody>\n",
       "</table>"
      ],
      "text/plain": [
       "    date time             set used   target   accuracy       auc   cross entropy\n",
       "0   2025-02-07 00:22:34   train      churn      0.7382    0.8181          0.508 \n",
       "1   2025-02-07 00:27:20   val        churn      0.7454    0.824           0.504 \n",
       "2   2025-02-07 00:32:01   test       churn      0.7544    0.828           0.4845"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# We now evaluate the pipeline on the validation and test sets:\n",
    "\n",
    "pipe.score(container.val)\n",
    "pipe.score(container.test)"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "9192cf6c",
   "metadata": {},
   "source": [
    "Observing the results in pipe.scores, we find the current pipeline\n",
    "meets our performance needs. getML performs marginally higher than RelBench's Deep Learning algorithm (AUC val = 0.8239, AUC test = 0.8281). Therefore, we will skip further optimizations\n",
    "of the data model, feature learner, or pipeline hyperparameters at this stage.\n",
    "\n",
    "Instead, we will explore feature importances, column importances,\n",
    "and examine the logical structure of our learned features. This helps us\n",
    "understand how the pipeline is deriving its predictions and whether\n",
    "certain aspects of the data are particularly influential."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "4ee566c2",
   "metadata": {},
   "source": [
    "---\n",
    "## 7. Feature Importances\n",
    "\n",
    "Since we used XGBoost, we can query each engineered feature's importance.\n",
    "This helps us to understand which features matter for churn."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "id": "2530a1c4",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAp0AAAGdCAYAAABQPcijAAAAOnRFWHRTb2Z0d2FyZQBNYXRwbG90bGliIHZlcnNpb24zLjEwLjAsIGh0dHBzOi8vbWF0cGxvdGxpYi5vcmcvlHJYcgAAAAlwSFlzAAAPYQAAD2EBqD+naQAA62tJREFUeJzs3XtYVNX+P/D3MDIwchkUwRkuMqMB3hBJghj1oIWBUWnnJGUa4NfEVLxxqJQQ0VRSI8ETihwVxVQEM/UoQopCioSIeLwVCqJTys0MUJAZmlm/P3zYP0cuDhcTO5/X8+yHZu3PXnutDU/zce291+IxxhgIIYQQQgh5ivSedQMIIYQQQshfHyWdhBBCCCHkqaOkkxBCCCGEPHWUdBJCCCGEkKeOkk5CCCGEEPLUUdJJCCGEEEKeOko6CSGEEELIU0dJJyGEEEIIeep6POsGkOeDRqPB7du3YWJiAh6P96ybQwghhBAdMMZw7949WFlZQU/v2Y41UtJJdHL79m3Y2to+62YQQgghpAN++eUX2NjYPNM2UNJJdGJiYgLg4R+tqanpM24NIYQQQnRRW1sLW1tb7nv8WaKkk+ik6Za6qakpJZ2EEELIc6Y7PBpHLxKRTomLi4NUKoWhoSHc3d1x5syZNuNTU1MxcOBAGBoawsnJCWlpaVr7GWOIiIiARCKBUCiEl5cXrl27xu2/ceMGpk+fDplMBqFQiAEDBmDp0qVQqVQtnq+4uBgmJiYwMzPrdF8JIYQQ0nHdPulkjCEoKAi9e/cGj8fD+fPnn3WT/qcNXZrB/feePXsQEhKCpUuX4ty5c3B2doa3tzcqKytbPPb06dOYPHkypk+fjsLCQkycOBETJ07EpUuXuJg1a9Zg/fr1iI+PR15eHoyMjODt7Y2GhgYAwM8//wyNRoNNmzbh8uXLWLduHeLj4xEWFtbsfI2NjZg8eTJGjx7dxVeBEEIIIe3Gurm0tDSmr6/PcnJyWFlZGWtsbOx0nQEBAWzChAmdb1wXePDgAQsICGBDhw5lfD6/3e06ceIEA9BsKysrazE+KiqKAWDz589v13lqamoYAGa7IIUrc3NzY3PmzOE+q9VqZmVlxaKiolqsw8/Pj/n6+mqVubu7s5kzZzLGGNNoNEwsFrO1a9dy+6urq5mBgQHbvXt3q21bs2YNk8lkzco/+eQTNnXqVJaYmMhEIpFO/SSEEEL+Spq+v2tqap51U1i3f6azpKQEEokEcrn8WTelGbVaDR6P16kpCNRqNYRCIebNm4dvv/22w/UUFRVpPWtpaWnZLCY/Px+bNm3CsGHDOnyeJiqVCgUFBVi8eDFXpqenBy8vL+Tm5rZ4TG5uLkJCQrTKvL29sX//fgBAaWkpysvL4eXlxe0XiURwd3dHbm4u3nvvvRbrrampQe/evbXKjh8/jtTUVJw/fx779u3rSBcJIaRbYYzhjz/+gFqtftZNId0In89Hjx49usUzm0/SrZPOwMBAbN++HcDDB2Dt7Oxw/fp1rF69GgkJCSgvL4eDgwOWLFmCd955B8DDJC4oKAjHjx9HeXk5+vXrh9mzZ2P+/PkAgMjISK06AeDEiRMAgLFjx+L333/nnv87f/48XFxcUFpaCqlUim3btmHBggVISkrCokWLcPXqVRQXF0MikeCzzz7D7t27UV1djaFDh2L16tUYM2bME/toZGSEjRs3AgBycnJQXV3doWtlaWnZ5nOL9+/fx5QpU/Dvf/8bK1as6NA5HnXnzh2o1Wr07dtXq7xv3774+eefWzymvLy8xfjy8nJuf1NZazGPKy4uxr/+9S98+eWXXNlvv/2GwMBAfPPNN/TSEyHkL0GlUqGsrAz19fXPuimkG+rZsyckEgkEAsGzbkqbunXSGRsbiwEDBiAhIQH5+fng8/mIiorCN998g/j4eNjb2+OHH37A1KlTYWFhAU9PT2g0GtjY2CA1NRXm5uY4ffo0goKCIJFI4Ofnh9DQUPz000+ora1FYmIiAKB37944ffq0Tm2qr6/H6tWrsXnzZpibm8PS0hLBwcG4cuUKkpOTYWVlhe+++w4+Pj64ePEi7O3tn+Yl4gwfPhxKpRJDhw5FZGQkRo4cqbV/zpw58PX1hZeXl05Jp1KphFKp5D7X1tZ2eZs769atW/Dx8cGkSZMwY8YMrnzGjBl4//338be//e0Zto4QQrqGRqNBaWkp+Hw+rKysIBAInotRLfL0McagUqlQVVWF0tJS2NvbP/MJ4NvSrZNOkUgEExMT8Pl8iMViKJVKrFq1CseOHYOHhwcAoH///jh16hQ2bdoET09P6OvrY9myZVwdMpkMubm5SElJgZ+fH4yNjSEUCqFUKiEWi9vdpsbGRmzYsAHOzs4AAIVCgcTERCgUClhZWQEAQkNDkZ6ejsTERKxataoLrkTrJBIJ4uPj4erqCqVSic2bN2PMmDHIy8vDiy++CABITk7GuXPnkJ+fr3O9UVFRWtfxcX369AGfz0dFRYVWeUVFRavXVSwWtxnf9LOiogISiUQrZvjw4VrH3b59G2PHjoVcLkdCQoLWvuPHj+PgwYPc6CdjDBqNBj169EBCQgL+7//+r42eE0JI96JSqaDRaGBra4uePXs+6+aQbkYoFEJfXx83b96ESqWCoaHhs25Sq7p10vm44uJi1NfXY9y4cVrlKpUKLi4u3Oe4uDhs3boVCoUCDx48gEqlapa0dJRAINB6JvLixYtQq9VwcHDQilMqlTA3N++Sc7bF0dERjo6O3Ge5XI6SkhKsW7cOO3bswC+//IL58+fj6NGj7fpDXLx4sdbzl02TyzYRCAQYMWIEMjMzMXHiRAAP/zWemZmJ4ODgFuv08PBAZmYmFixYwJUdPXqU+weETCaDWCxGZmYm9/uqra1FXl4eZs2axR1z69YtjB07FiNGjEBiYmKzf9Xl5uZqPfN04MABrF69GqdPn4a1tbXO14AQQrqT7jyCRZ6t5+Vv47lKOu/fvw8AOHz4cLPkwcDAAMDDUb3Q0FBER0fDw8MDJiYmWLt2LfLy8tqsu+kXxhjjyhobG5vFCYVCrdsa9+/fB5/PR0FBAfh8vlassbFxO3rXddzc3HDq1CkAQEFBASorK7lRT+Dhc68//PADvv76ayiVymbtBh5ez6Zr2pqQkBAEBATA1dUVbm5uiImJQV1dHaZNmwYA8Pf3h7W1NaKiogAA8+fPh6enJ6Kjo+Hr64vk5GScPXuWG6nk8XhYsGABVqxYAXt7e8hkMixZsgRWVlZcYnvr1i2MGTMGdnZ2+PLLL1FVVcW1p2mkdNCgQVrtPHv2LPT09DB06ND2XEZCCCGEdKHnKukcPHgwDAwMoFAo4Onp2WJMTk4O5HI5Zs+ezZWVlJRoxQgEgmZv/1lYWAAAysrK0KtXLwDQaU5QFxcXqNVqVFZWdpv5IM+fP8/dnn711Vdx8eJFrf3Tpk3DwIED8emnn7aYcOrq3XffRVVVFSIiIlBeXo7hw4cjPT2dexFIoVBo/etLLpdj165dCA8PR1hYGOzt7bF//36tZPCTTz5BXV0dgoKCUF1djVGjRiE9PZ0bpT169CiKi4tRXFzcbA3ZR//BQAgh5NkbM2YMhg8fjpiYmGfdFNINPFdJp4mJCUJDQ7Fw4UJoNBqMGjUKNTU1yMnJgampKQICAmBvb4+kpCRkZGRAJpNhx44dyM/Ph0wm4+qRSqXIyMhAUVERzM3NIRKJ8MILL8DW1haRkZFYuXIlrl69iujo6Ce2ycHBAVOmTIG/vz+io6Ph4uKCqqoqZGZmYtiwYfD19X1iHVeuXIFKpcLdu3dx7949LtnV5ZGAmJgYyGQyDBkyBA0NDdi8eTOOHz+O77//nrtmj4/wGRkZwdzcvEtG/oKDg1u9nZ6VldWsbNKkSZg0aVKr9fF4PCxfvhzLly9vcX9gYCACAwPb1caOHEMIIc8D6aLDf+r5bnzx5O+0R+3btw/6+vpPqTWdk5WV1WzWGvJ0PVdJJwB8/vnnsLCwQFRUFK5fvw4zMzO8+OKL3Io0M2fORGFhId59913weDxMnjwZs2fPxpEjR7g6ZsyYgaysLLi6uuL+/fs4ceIExowZg927d2PWrFkYNmwYXnrpJaxYsaLNBKlJYmIiVqxYgX/+85+4desW+vTpg5dffhlvvPGGTn16/fXXcfPmTe5z0/OpuozcqVQq7rw9e/bEsGHDcOzYMYwdO1anc7fXpWXeT6VeQgghfz2Pz6HcXbT0+Bx5+niM7kkSHdTW1kIkEqGmpobmviSEkD9RQ0MDSktLIZPJmr0Q2t1HOh+9vS6VSvHhhx/i6tWr2LdvH8zNzfGvf/0LHh4e+PDDD5GZmYn+/ftj69atcHV1BQBufuxt27bh448/xi+//AJPT09s3rxZ6+XWjRs34ssvv8Qvv/wCmUyG8PBwfPDBB9x+Ho+HDRs24MiRI8jMzMSkSZO4ObubBAQEYNu2bUhPT8eKFStw6dIl8Pl8eHh4cFM4AsCNGzcgk8nw7bff4l//+hfy8vJgb2+P+Ph47sVY4OHjfp999hnOnDkDAwMDuLm5ITk5Gb169YJGo2lzzvHff/8dwcHB+P7773H//n3Y2NggLCyMe2ficW39jXSn7+/n43UnQgghhDz31q1bh5EjR6KwsBC+vr744IMP4O/vj6lTp+LcuXMYMGAA/P39te701dfXY+XKlUhKSuIWUXl0hbrvvvsO8+fPxz//+U9cunQJM2fOxLRp07iFX5pERkbi7bffxsWLF7Fs2TJuFcCioiKUlZUhNjYWAFBXV4eQkBCcPXsWmZmZ0NPTw9tvvw2NRqNV32effYbQ0FCcP38eDg4OmDx5Mv744w8AD9+tePXVVzF48GDk5ubi1KlTePPNN7n3SaKiopCUlIT4+HhcvnwZCxcuxNSpU5GdnQ0AWLJkCa5cuYIjR47gp59+wsaNG9GnT58u/m38+Z672+vPm/Hjx+PkyZMt7gsLC+MeC2jNkCFDtG69P2rTpk2YMmVKp9tICCGE/Blef/11zJw5EwAQERGBjRs34qWXXuIeZfv000/h4eGhNYdzY2Mjvv76a7i7uwMAtm/fjkGDBuHMmTNwc3PDl19+icDAQO4F4pCQEPz444/48ssvtR41e//997VGCktLSwE0X9HvH//4h1abt27dCgsLC1y5ckXrXYjQ0FDuvY1ly5ZhyJAhKC4uxsCBA7FmzRq4urpiw4YNXPyQIUMAQKc5xxUKBVxcXLgRX6lU2pHL3e1Q0vmUbd68GQ8ePGhxny7PuqSlpbX67Mnjy0USQggh3dmj81w3fYc5OTk1K6usrOSSzh49euCll17iYgYOHAgzMzP89NNPcHNzw08//YSgoCCt84wcOZIbuWzSlMA9ybVr1xAREYG8vDzcuXOHG+FUKBRaSeejfWmaMaayshIDBw7E+fPnW30nRJc5x2fNmoV//OMfOHfuHF577TVMnDgRcrlcp/Z3Z5R0PmWdnYzczs6ui1pCCCGEPFuPvsneNOd1S2WP38ruCkZGRjrFvfnmm7Czs8O///1vWFlZQaPRYOjQoVCpVFpxbbVbKBS2Wr8uc46PHz8eN2/eRFpaGo4ePYpXX30Vc+bM4Vbae17RM52EEEII6bb++OMPnD17lvtcVFSE6upqbiGQQYMGIScnR+uYnJwcDB48uM16BQIBAGjN2/3bb7+hqKgI4eHhePXVVzFo0CD8/vvv7W7zsGHDkJmZ2eK+R+ccf+GFF7S2R1+OsrCwQEBAAL755hvExMQ0W/L5eUQjnYQQQgjptvT19TF37lysX78ePXr0QHBwMF5++WW4ubkBAD7++GP4+fnBxcUFXl5e+M9//oN9+/bh2LFjbdZrZ2cHHo+HQ4cO4fXXX4dQKESvXr1gbm6OhIQESCQSKBQKLFq0qN1tXrx4MZycnDB79mx89NFHEAgEOHHiBCZNmoQ+ffo8cc7xiIgIjBgxAkOGDIFSqcShQ4earbb3POr2I52MMQQFBaF3797g8Xg6rRJE/jxxcXGQSqUwNDSEu7s7zpw502Z8amoqBg4cCENDQzg5OSEtLU1rP2MMERERkEgkEAqF8PLywrVr17j9N27cwPTp0yGTySAUCjFgwAAsXbq02W0Pxhi+/PJLODg4wMDAANbW1li5cmXXdZwQQsifomfPnvj000/x/vvvY+TIkTA2NsaePXu4/RMnTkRsbCy+/PJLDBkyBJs2bUJiYiLGjBnTZr3W1tZYtmwZFi1ahL59+yI4OBh6enpITk5GQUEBhg4dioULF2Lt2rXtbrODgwO+//57/Pe//4Wbmxs8PDxw4MAB9OjxcKzv888/x5IlSxAVFYVBgwbBx8cHhw8f5hayEQgEWLx4MYYNG4a//e1v4PP5SE5Obnc7uh3WzaWlpTF9fX2Wk5PDysrKWGNjY6frDAgIYBMmTOh847rAgwcPWEBAABs6dCjj8/ntbtft27fZ5MmTmb29PePxeGz+/Pktxv3+++9s9uzZTCwWM4FAwOzt7dnhw4d1Pk9NTQ0DwGwXpHBlycnJTCAQsK1bt7LLly+zGTNmMDMzM1ZRUdFiHTk5OYzP57M1a9awK1eusPDwcKavr88uXrzIxXzxxRdMJBKx/fv3s//+97/srbfeYjKZjD148IAxxtiRI0dYYGAgy8jIYCUlJezAgQPM0tKS/fOf/9Q619y5c5mjoyM7cOAAu379Ojt79iz7/vvvde4vIYR0Fw8ePGBXrlzh/j/4vyQxMZGJRKJn3Yxur62/kabv75qammfQMm3d/vZ6SUkJJBJJt3xrS61Wg8fjaa0v3pE6hEIh5s2bx80Z1h5KpRIWFhYIDw/HunXrWoxRqVQYN24cLC0tsXfvXlhbW+PmzZudXvbrq6++wowZM7gpKOLj43H48GFs3bq1xdsRsbGx8PHxwccffwzg4b/0jh49iq+//hrx8fFgjCEmJgbh4eGYMGECACApKQl9+/bF/v378d5778HHxwc+Pj5cnf3790dRURE3MTAAbk6zS5cuwdHREQC0lkElhBBCyJ+vW99eDwwMxNy5c6FQKMDj8SCVSqHRaBAVFcXdXnV2dsbevXu5Y9RqtdbtV0dHR61pEyIjI7F9+3YcOHAAPB4PPB4PWVlZyMrKAo/HQ3V1NRd7/vx58Hg83LhxA8DDlRHMzMxw8OBBrQeBlUolQkNDYW1tDSMjI7i7u7e47nhLjIyMsHHjRsyYMYObHqI9pFIpYmNj4e/vD5FI1GLM1q1bcffuXezfvx8jR46EVCqFp6cnnJ2d232+JiqVCgUFBfDy8uLK9PT04OXlhdzc3BaPyc3N1YoHAG9vby6+tLQU5eXlWjEikQju7u6t1gkANTU1WtNP/ec//0H//v1x6NAhyGQybhWMu3fvdqivhBBCCOm8bp10xsbGYvny5bCxsUFZWRny8/OfOIu/RqOBjY0NUlNTceXKFURERCAsLAwpKSkAHk7m6ufnBx8fH5SVlaGsrKxdo6j19fVYvXo1Nm/ejMuXL8PS0hLBwcHIzc1FcnIyLly4gEmTJsHHx0frWcRn6eDBg/Dw8MCcOXPQt29fDB06FKtWrdJ6Y+9xSqUStbW1Wtuj7ty5A7Va3Wyu0L59+6K8vLzFOsvLy9uMb/rZnjqLi4vxr3/9i5tsGACuX7+OmzdvIjU1FUlJSdi2bRsKCgq45cUIIYQ8HwIDA7UGg8jzrVvfXheJRDAxMQGfz4dYLNZpFn99fX0sW7aMq0MmkyE3NxcpKSnw8/ODsbExhEIhlEplh0YWGxsbsWHDBm6UUKFQIDExEQqFAlZWVgAeJrbp6elITEzEqlWruuBKdM7169dx/PhxTJkyBWlpaSguLsbs2bPR2NiIpUuXtnhMVFSU1nXsjm7dugUfHx9MmjQJM2bM4Mo1Gg2USiWSkpLg4OAAANiyZQtGjBiBoqIi7pY7IYQQQv483TrpfJwus/gDD9+o3rp1KxQKBR48eACVSoXhw4d3SRsEAoHWKgQXL16EWq3mkpsmSqUS5ubmXXLOztJoNLC0tERCQgL4fD5GjBiBW7duYe3ata0mnYsXL0ZISAj3uba2Vmv+sD59+oDP56OiokLruEeXLnucWCxuM77pZ0VFBbe6Q9Pnx39/t2/fxtixYyGXy5vNXSaRSNCjRw+t30nTVBMKhYKSTkIIIeQZeK6STl1m8U9OTkZoaCiio6Ph4eEBExMTrF27Fnl5eW3W3fQyEGOMK2tp+UmhUMitPNDUJj6fj4KCAvD5fK1YY2PjdvTu6ZFIJNDX19dq36BBg1BeXg6VSsVNkPsoAwMD7pq2RCAQYMSIEcjMzMTEiRMBPExuMzMzERwc3OIxHh4eyMzMxIIFC7iyo0ePcqPWMpkMYrEYmZmZXJJZW1uLvLw8zJo1izvm1q1bGDt2LEaMGIHExMRmL3KNHDkSf/zxB0pKSjBgwAAAwNWrVwHQCk+EkOfXo99PhDzqefnbeK6Szkdf3vH09GwxJicnB3K5HLNnz+bKSkpKtGIEAkGz5xktLCwAAGVlZejVqxcA6DQnqIuLC9RqNSorKzF69Oj2dOdPM3LkSOzatQsajYZL0K5evQqJRNJiwqmrkJAQBAQEwNXVFW5uboiJiUFdXR33Nru/vz+sra0RFRUFAJg/fz48PT0RHR0NX19fJCcn4+zZs9xIJY/Hw4IFC7BixQrY29tDJpNhyZIlsLKy4hLbW7duYcyYMbCzs8OXX36Jqqoqrj1NI6VeXl548cUX8X//93+IiYmBRqPBnDlzMG7cuGYj0oQQ0t01LbdYX1/f5vKK5H9XfX09AO2lObuj5yrpNDExeeIs/vb29khKSkJGRgZkMhl27NiB/Px8rSlzpFIpMjIyUFRUBHNzc4hEIm75qcjISKxcuRJXr15FdHT0E9vk4OCAKVOmwN/fH9HR0XBxcUFVVRUyMzMxbNgw+Pr6PrGOK1euQKVS4e7du7h37x6X7Or6SEBT/P3791FVVYXz589DIBBwS4DNmjULX3/9NebPn4+5c+fi2rVrWLVqFebNm6dT/a159913UVVVhYiICJSXl2P48OFIT0/nXgRSKBRao5ByuRy7du1CeHg4wsLCYG9vj/3792Po0KFczCeffIK6ujoEBQWhuroao0aNQnp6OgwNDQE8HBktLi5GcXExbGxstNrT9C89PT09/Oc//8HcuXPxt7/9DUZGRhg/frxOv09CCOlu+Hw+zMzMUFlZCeDhZOmP3nEj/7sYY6ivr0dlZSXMzMya3XHtbnism4/JxsTEICYmhpu2iDGG9evXY+PGjbh+/TrMzMzw4osvIiwsDH/729+gVCrx0Ucf4bvvvgOPx8PkyZMhEolw5MgRLjmrqqrClClTkJubi/v37+PEiRMYM2YMcnJyMGvWLFy7dg0vvfQS5s2bh0mTJqG0tBRSqRTbtm3DggULmr1J19jYiBUrViApKQm3bt1Cnz598PLLL2PZsmVwcnJ6Yh+lUilu3rzZrFzXX01L//Oxs7PjrhnwcLqihQsX4vz587C2tsb06dPx6aef6vwHWltbC5FIhJqaGpiamup0DCGEkK7BGEN5eTm9yU1aZGZmBrFY3GI+0J2+v7t90km6h+70R0sIIf+r1Gp1i+8bkP9dj7+z8bju9P39XN1eJ4QQQv6X8fn8bn8LlZDWdOvJ4f8Kxo8fD2Nj4xY3XebwHDJkSKvH79y580/oASGEEEJI59FI51O2efNmPHjwoMV9jy7d2Jq0tLRWb6U8vnIPIYQQQkh3RUnnU/b4fKLtRfNKEkIIIeSvgG6vE0IIIYSQp46STkIIIYQQ8tRR0kk6JS4uDlKpFIaGhnB3d8eZM2fajE9NTcXAgQNhaGgIJycnpKWlae1njCEiIgISiQRCoRBeXl64du2aVsxbb72Ffv36wdDQEBKJBB988AFu377N7b9x4wZ4PF6z7ccff+y6jhNCCCGkXbp90skYQ1BQEHr37g0ej6fT0pTk6Rm6NIP77z179iAkJARLly7FuXPn4OzsDG9vb27VjMedPn0akydPxvTp01FYWIiJEydi4sSJuHTpEhezZs0arF+/HvHx8cjLy4ORkRG8vb3R0NDAxYwdOxYpKSkoKirCt99+i5KSErzzzjvNznfs2DGUlZVx24gRI7rwShBCCCGkXVg3l5aWxvT19VlOTg4rKytjjY2Nna4zICCATZgwofON6wIPHjxgAQEBbOjQoYzP53eoXQ0NDSwsLIz169ePCQQCZmdnx7Zs2cLtV6lUbNmyZax///7MwMCADRs2jB05cqRd56ipqWEAmO2CFK7Mzc2NzZkzh/usVquZlZUVi4qKarEOPz8/5uvrq1Xm7u7OZs6cyRhjTKPRMLFYzNauXcvtr66uZgYGBmz37t2ttu3AgQOMx+MxlUrFGGOstLSUAWCFhYXt6iMhhBDyV9P0/V1TU/Osm8K6/UhnSUkJJBIJ5HI5xGIxevToPi/cq9VqaDSaTtchFAoxb948eHl5dagOPz8/ZGZmYsuWLSgqKsLu3bvh6OjI7Q8PD8emTZvwr3/9C1euXMFHH32Et99+G4WFhR1ut0qlQkFBgVab9fT04OXlhdzc3BaPyc3NbdZHb29vLr60tBTl5eVaMSKRCO7u7q3WeffuXezcuRNyuRz6+vpa+9566y1YWlpi1KhROHjwYIf6SQghhJCu0a2TzsDAQMydOxcKhQI8Hg9SqRQajQZRUVGQyWQQCoVwdnbG3r17uWPUajWmT5/O7Xd0dERsbCy3PzIyEtu3b8eBAwe4Z/2ysrKQlZUFHo+nta7t+fPnwePxuDXMt23bBjMzMxw8eBCDBw+GgYEBFAoFlEolQkNDYW1tDSMjI7i7uyMrK0unPhoZGWHjxo2YMWMGxGJxu69Reno6srOzkZaWBi8vL0ilUnh4eGDkyJFczI4dOxAWFobXX38d/fv3x6xZs/D6668jOjq63edrcufOHajV6mZzhfbt2xfl5eUtHlNeXt5mfNNPXer89NNPYWRkBHNzcygUChw4cIDbZ2xsjOjoaKSmpuLw4cMYNWoUJk6cSIknIYQQ8gx1n2HDFsTGxmLAgAFISEhAfn4++Hw+oqKi8M033yA+Ph729vb44YcfMHXqVFhYWMDT0xMajQY2NjZITU2Fubk5Tp8+jaCgIEgkEvj5+SE0NBQ//fQTamtrkZiYCODhJO2nT5/WqU319fVYvXo1Nm/eDHNzc1haWiI4OBhXrlxBcnIyrKys8N1338HHxwcXL16Evb3907xEOHjwIFxdXbFmzRrs2LEDRkZGeOutt/D5559DKBQCAJRKJQwNDbWOEwqFOHXqVKv1KpVKKJVK7nNtbe3T6UAHffzxx5g+fTpu3ryJZcuWwd/fH4cOHQKPx0OfPn0QEhLCxb700ku4ffs21q5di7feeusZtpoQQgj539Wtk06RSAQTExPw+XyIxWIolUqsWrUKx44dg4eHBwCgf//+OHXqFDZt2gRPT0/o6+tj2bJlXB0ymQy5ublISUmBn58fjI2NIRQKoVQqOzSy2NjYiA0bNsDZ2RkAoFAokJiYCIVCASsrKwBAaGgo0tPTkZiYqNNSl51x/fp1nDp1CoaGhvjuu+9w584dzJ49G7/99huXVHt7e+Orr77C3/72NwwYMACZmZnYt28f1Gp1q/VGRUVpXcfH9enTB3w+HxUVFVrlFRUVrV5XsVjcZnzTz4qKCkgkEq2Y4cOHNzt/nz594ODggEGDBsHW1hY//vgj93fxOHd3dxw9erTV/hBCCCHk6erWt9cfV1xcjPr6eowbN05rDfKkpCSUlJRwcXFxcRgxYgQsLCxgbGyMhIQEKBSKLmmDQCDAsGHDuM8XL16EWq2Gg4ODVpuys7O12vS0aDQa8Hg87Ny5E25ubnj99dfx1VdfYfv27dzym7GxsbC3t8fAgQMhEAgQHByMadOmQU+v9V//4sWLUVNTw22//PKL1n6BQIARI0YgMzNTqy2ZmZmtJn4eHh5a8QBw9OhRLl4mk0EsFmvF1NbWIi8vr9U6m84LQGtk9nHnz5/XSmQJIYQQ8ufq1iOdj7t//z4A4PDhw82WlzQwMAAAJCcnIzQ0FNHR0fDw8ICJiQnWrl2LvLy8NutuSsAYY1xZS2ueC4VC8Hg8rTbx+XwUFBSAz+drxRobG7ejdx0jkUhgbW0NkUjElQ0aNAiMMfz666+wt7eHhYUF9u/fj4aGBvz222+wsrLCokWL0L9//1brNTAw4K5pa0JCQhAQEABXV1e4ubkhJiYGdXV1mDZtGgDA398f1tbWiIqKAgDMnz8fnp6eiI6Ohq+vL5KTk3H27FkkJCQAAHg8HhYsWIAVK1bA3t4eMpkMS5YsgZWVFSZOnAgAyMvLQ35+PkaNGoVevXqhpKQES5YswYABA7jEdPv27RAIBHBxcQEA7Nu3D1u3bsXmzZs7dpEJIYQQ0mnPVdL56Ms7np6eLcbk5ORALpdj9uzZXNnjI44CgaDZrWULCwsAQFlZGXr16gUAOs0J6uLiArVajcrKSowePbo93ekSI0eORGpqKu7fv88luVevXoWenh5sbGy0Yg0NDWFtbY3GxkZ8++238PPz69S53333XVRVVSEiIgLl5eUYPnw40tPTuReBFAqF1miqXC7Hrl27EB4ejrCwMNjb22P//v0YOnQoF/PJJ5+grq4OQUFBqK6uxqhRo5Cens49k9qzZ0/s27cPS5cuRV1dHSQSCXx8fBAeHq6VJH/++ee4efMmevTogYEDB2LPnj0tzuVJCCGEkD/Js56z6UnWrVvH7OzsuM+fffYZMzc3Z9u2bWPFxcWsoKCArV+/nm3bto0xxlhsbCwzNTVl6enprKioiIWHhzNTU1Pm7OzM1bFy5UrWr18/9vPPP7OqqiqmUqmYSqVitra2bNKkSezq1avs0KFDzNHRkQFgpaWljDHGEhMTmUgkatbGKVOmMKlUyr799lt2/fp1lpeXx1atWsUOHTqkUx8vX77MCgsL2ZtvvsnGjBnDCgsLdZ5j8t69e8zGxoa988477PLlyyw7O5vZ29uzDz/8kIv58ccf2bfffstKSkrYDz/8wF555RUmk8nY77//rtM5GOte83wRQgghRDfd6fv7uUs6NRoNi4mJYY6OjkxfX59ZWFgwb29vlp2dzRh7OFF6YGAgE4lEzMzMjM2aNYstWrRIK+msrKxk48aNY8bGxgwAO3HiBGOMsVOnTjEnJydmaGjIRo8ezVJTU3VKOlUqFYuIiGBSqZTp6+sziUTC3n77bXbhwgWd+mhnZ8cANNt09dNPPzEvLy8mFAqZjY0NCwkJYfX19dz+rKwsNmjQIGZgYMDMzc3ZBx98wG7duqVz/Yx1rz9aQgghhOimO31/8xh75CFGQlpRW1sLkUiEmpoamJqaPuvmEEIIIUQH3en7+7l6e50QQgghhDyfKOl8ysaPH681ldKjmy5zeA4ZMqTV43fu3Pkn9IAQQgghpPOeq7fXn0ebN2/m5st8XO/evZ94fFpaWotTNwHNl4skhBBCCOmuKOl8yh6fT7S97OzsuqglhBBCCCHPDt1eJ4QQQgghTx0lnYQQQggh5Kl7pkknYwxBQUHo3bs3eDyeTisAke4nLi4OUqkUhoaGcHd3x5kzZ9qMT01NxcCBA2FoaAgnJyekpaVp7WeMISIiAhKJBEKhEF5eXrh27ZpWzFtvvYV+/frB0NAQEokEH3zwAW7fvq0Vk5KSguHDh6Nnz56ws7PD2rVru6bDhBBCCGm3Z5p0pqenY9u2bTh06BDKysq0lkPsqMDAQG6d7metoaEBgYGBcHJyQo8ePdrdrlOnTmHkyJEwNzeHUCjEwIEDsW7dumZxt27dwtSpU7k4JycnnD17FsDD9eM//fRTODk5wcjICFZWVvD392+WoOlq6NIMSBcd5j7v2bMHISEhWLp0Kc6dOwdnZ2d4e3ujsrKyxeNPnz6NyZMnY/r06SgsLMTEiRMxceJEXLp0iYtZs2YN1q9fj/j4eOTl5cHIyAje3t5oaGjgYsaOHYuUlBQUFRXh22+/RUlJidYyl0eOHMGUKVPw0Ucf4dKlS9iwYQPWrVuHr7/+ukP9JoQQQkgnPcuZ6f/1r3+xfv36dWmdAQEBbMKECZ2u548//mBqtbpTddy/f5999NFHLCEhgXl7e7e7XefOnWO7du1ily5dYqWlpWzHjh2sZ8+ebNOmTVzM3bt3mZ2dHQsMDGR5eXns+vXrLCMjgxUXFzPGGKuurmZeXl5sz5497Oeff2a5ubnMzc2NjRgxol1taVrRwHZBCrP79P8v7+nm5sbmzJnDfVar1czKyopFRUW1WI+fnx/z9fXVKnN3d2czZ85kjD1ccUosFrO1a9dy+6urq5mBgQHbvXt3q+07cOAA4/F4TKVSMcYYmzx5MnvnnXe0YtavX89sbGyYRqPRsdeEEELI8607rUj0zEY6AwMDMXfuXCgUCvB4PEilUmg0GkRFRUEmk0EoFMLZ2Rl79+7ljlGr1Zg+fTq339HREbGxsdz+yMhIbN++HQcOHACPxwOPx0NWVhaysrLA4/FQXV3NxZ4/fx48Hg83btwAAGzbtg1mZmY4ePAgBg8eDAMDAygUCiiVSoSGhsLa2hpGRkZwd3dHVlaWTn00MjLCxo0bMWPGDIjF4nZfIxcXF0yePBlDhgyBVCrF1KlT4e3tjZMnT3Ixq1evhq2tLRITE+Hm5gaZTIbXXnsNAwYMAACIRCIcPXoUfn5+cHR0xMsvv4yvv/4aBQUFUCgU7W7To1QqFQoKCuDl5cWV6enpwcvLC7m5uS0ek5ubqxUPAN7e3lx8aWkpysvLtWJEIhHc3d1brfPu3bvYuXMn5HI59PX1AQBKpRKGhoZacUKhEL/++itu3rzZ/s4SQgghpFOeWdIZGxuL5cuXw8bGBmVlZcjPz0dUVBSSkpIQHx+Py5cvY+HChZg6dSqys7MBABqNBjY2NkhNTcWVK1cQERGBsLAwpKSkAABCQ0Ph5+cHHx8flJWVoaysDHK5XOc21dfXY/Xq1di8eTMuX74MS0tLBAcHIzc3F8nJybhw4QImTZoEHx+fZs8Y/hkKCwtx+vRpeHp6cmUHDx6Eq6srJk2aBEtLS7i4uODf//53m/XU1NSAx+PBzMys1RilUona2lqt7XF37tyBWq1uNl9o3759UV5e3mK95eXlbcY3/dSlzk8//RRGRkYwNzeHQqHAgQMHuH3e3t7Yt28fMjMzodFocPXqVURHRwMAysrKWu03IYQQQp6OZ5Z0ikQimJiYgM/nQywWw9TUFKtWrcLWrVvh7e2N/v37IzAwEFOnTsWmTZsAAPr6+li2bBlcXV0hk8kwZcoUTJs2jUs6jY2NIRQKYWBgALFYDLFYDIFAoHObGhsbsWHDBsjlcjg6OuLOnTtITExEamoqRo8ejQEDBiA0NBSjRo1CYmLiU7kuLbGxsYGBgQFcXV0xZ84cfPjhh9y+69evY+PGjbC3t0dGRgZmzZqFefPmYfv27S3W1dDQgE8//RSTJ09ucw3WqKgoiEQibrO1te3yfnXWxx9/jMLCQnz//ffg8/nw9/cHYwwAMGPGDAQHB+ONN96AQCDAyy+/jPfeew/Aw9FYQgghhPy5us3k8MXFxaivr8e4ceO0ylUqFVxcXLjPcXFx2Lp1KxQKBR48eACVSoXhw4d3SRsEAgGGDRvGfb548SLUajUcHBy04pRKJczNzbvknLo4efIk7t+/jx9//BGLFi3CCy+8gMmTJwN4OPrr6urKLanp4uKCS5cuIT4+HgEBAVr1NDY2ws/PD4wxbNy4sc1zLl68GCEhIdzn2traZolnnz59wOfzUVFRoVVeUVHR6uMEYrG4zfimnxUVFZBIJFoxj/+e+/Tpgz59+sDBwQGDBg2Cra0tfvzxR3h4eIDH42H16tVYtWoVysvLYWFhgczMTABA//792+w7IYQQQrpet0k679+/DwA4fPhws1V8DAwMAADJyckIDQ1FdHQ0PDw8YGJigrVr1yIvL6/NuptGtppGwQC0uLSkUCgEj8fTahOfz0dBQQH4fL5WrLGxcTt61zkymQwA4OTkhIqKCkRGRnJJp0QiweDBg7XiBw0ahG+//VarrCnhvHnzJo4fP97mKCfw8Jo3XffWCAQCjBgxApmZmdyb+RqNBpmZmQgODm7xGA8PD2RmZmLBggVc2dGjR+Hh4cH1VSwWIzMzk0sya2trkZeXh1mzZrXaFo1GA+DhPwgexefzub+n3bt3w8PDAxYWFm32ixBCCCFdr9sknY++vPPoM4uPysnJgVwux+zZs7mykpISrRiBQAC1Wq1V1pRklJWVoVevXgCg05ygLi4uUKvVqKysxOjRo9vTnadGo9FoJVYjR45EUVGRVszVq1e1ls9sSjivXbuGEydOdOkobUhICAICAuDq6go3NzfExMSgrq4O06ZNAwD4+/vD2toaUVFRAID58+fD09MT0dHR8PX1RXJyMs6ePYuEhAQAAI/Hw4IFC7BixQrY29tDJpNhyZIlsLKy4hLbvLw85OfnY9SoUejVqxdKSkqwZMkSDBgwgEte79y5g71792LMmDFoaGjgHpNoej6YEEIIIX+ubpN0mpiYIDQ0FAsXLoRGo8GoUaNQU1ODnJwcmJqaIiAgAPb29khKSkJGRgZkMhl27NiB/Px8biQQAKRSKTIyMlBUVARzc3OIRCK88MILsLW1RWRkJFauXKn1UklbHBwcMGXKFPj7+yM6OhouLi6oqqpCZmYmhg0bBl9f3yfWceXKFahUKty9exf37t3jkl1dHgmIi4tDv379MHDgQADADz/8gC+//BLz5s3jYhYuXAi5XI5Vq1bBz88PZ86cQUJCApfENTY24p133sG5c+dw6NAhqNVq7oWc3r17t+uZ15a8++67qKqqQkREBMrLyzF8+HCkp6dzLwIpFAqtZyjlcjl27dqF8PBwhIWFwd7eHvv379eao/WTTz5BXV0dgoKCUF1djVGjRiE9PZ17G71nz57Yt28fli5dirq6OkgkEvj4+CA8PFxrdHb79u0IDQ0FYwweHh7IysqCm5tbp/pLCCGEkA56lvM1rVu3jtnZ2XGfNRoNi4mJYY6OjkxfX59ZWFgwb29vlp2dzRhjrKGhgQUGBjKRSMTMzMzYrFmz2KJFi5izszNXR2VlJRs3bhwzNjZmANiJEycYY4ydOnWKOTk5MUNDQzZ69GiWmprKALDS0lLGGGOJiYlMJBI1a6NKpWIRERFMKpUyfX19JpFI2Ntvv80uXLigUx/t7OwYgGabLtavX8+GDBnCevbsyUxNTZmLiwvbsGFDs/lD//Of/7ChQ4cyAwMDNnDgQJaQkMDtKy0tbfH8j14bXXSneb4IIYQQopvu9P3NY+yRBx0JaUVtbS1EIhFqamqe+DwoIYQQQrqH7vT9TXPHEEIIIYSQp46Szk4YP348jI2NW9yapjBqy5AhQ1o9fufOnX9CDwghhBBC/hzd5kWi59HmzZvx4MGDFvf17t37icenpaW1OHUT0HxFHkIIIYSQ5xklnZ3w+Hyi7fXotEaEEEIIIX9ldHudEEIIIYQ8dZR0EkIIIYSQp46STtJpcXFxkEqlMDQ0hLu7O86cOdNmfGpqKgYOHAhDQ0M4OTkhLS1Naz9jDBEREZBIJBAKhfDy8sK1a9e0Yt566y3069cPhoaGkEgk+OCDD3D79m2tmJSUFAwfPhw9e/aEnZ0d1q5d2zUdJoQQQki7dfukkzGGoKAg9O7dGzweT6flK8nTM3RpBqSLDnOf9+zZg5CQECxduhTnzp2Ds7MzvL29UVlZ2eLxp0+fxuTJkzF9+nQUFhZi4sSJmDhxIi5dusTFrFmzBuvXr0d8fDzy8vJgZGQEb29vNDQ0cDFjx45FSkoKioqK8O2336KkpATvvPMOt//IkSOYMmUKPvroI1y6dAkbNmzAunXr8PXXXz+Fq0IIIYSQJ3q2c9M/WVpaGtPX12c5OTmsrKyMNTY2drrOgIAANmHChM43rgs8ePCABQQEsKFDhzI+n9/udp04caLF1YbKysq4mA0bNjAnJydmYmLCTExM2Msvv8zS0tLadZ6mFQ1sF6Qwu08PceVubm5szpw53Ge1Ws2srKxYVFRUi/X4+fkxX19frTJ3d3c2c+ZMxtjDVanEYjFbu3Ytt7+6upoZGBiw3bt3t9q+AwcOMB6Px1QqFWOMscmTJ7N33nlHK2b9+vXMxsaGaTQaHXtNCCGEPN+604pE3X6ks6SkBBKJBHK5HGKxGD16dJ8X7tVqNTQaTafrEAqFmDdvHry8vDpcT1FREcrKyrjN0tKS22djY4MvvvgCBQUFOHv2LF555RVMmDABly9f7lTbVSoVCgoKtNqtp6cHLy8v5ObmtnhMbm5us356e3tz8aWlpSgvL9eKEYlEcHd3b7XOu3fvYufOnZDL5dDX1wcAKJVKbq32JkKhEL/++itu3rzZ/s4SQgghpFO6ddIZGBiIuXPnQqFQgMfjQSqVQqPRICoqCjKZDEKhEM7Ozti7dy93jFqtxvTp07n9jo6OiI2N5fZHRkZi+/btOHDgAHg8Hng8HrKyspCVlQUej4fq6mou9vz58+DxeLhx4wYAYNu2bTAzM8PBgwcxePBgGBgYQKFQQKlUIjQ0FNbW1jAyMoK7uzuysrJ06qORkRE2btyIGTNmQCwWd/haWVpaQiwWc5ue3v//1b755pt4/fXXYW9vDwcHB6xcuRLGxsb48ccfO3w+ALhz5w7UanWzOUX79u2L8vLyFo8pLy9vM77ppy51fvrppzAyMoK5uTkUCgUOHDjA7fP29sa+ffuQmZkJjUaDq1evIjo6GgBQVlbWgd4SQgghpDO6ddIZGxuL5cuXw8bGBmVlZcjPz0dUVBSSkpIQHx+Py5cvY+HChZg6dSqys7MBABqNBjY2NkhNTcWVK1cQERGBsLAwpKSkAABCQ0Ph5+cHHx8fblRQLpfr3Kb6+nqsXr0amzdvxuXLl2FpaYng4GDk5uYiOTkZFy5cwKRJk+Dj49Ps5Zenafjw4ZBIJBg3bhxycnJajVOr1UhOTkZdXR08PDxajVMqlaitrdXaupuPP/4YhYWF+P7778Hn8+Hv7w/GGABgxowZCA4OxhtvvAGBQICXX34Z7733HgBoJeSEEEII+XN0n3vVLRCJRDAxMQGfz4dYLIZSqcSqVatw7NgxLmHq378/Tp06hU2bNsHT0xP6+vpYtmwZV4dMJkNubi5SUlLg5+cHY2NjCIVCKJXKDo0sNjY2YsOGDXB2dgYAKBQKJCYmQqFQwMrKCsDDxDY9PR2JiYk6LYfZGRKJBPHx8XB1dYVSqcTmzZsxZswY5OXl4cUXX+TiLl68CA8PDzQ0NMDY2BjfffcdBg8e3Gq9UVFRWtexJX369AGfz0dFRYVWeUVFRavXViwWtxnf9LOiogISiUQrZvjw4c3O36dPHzg4OGDQoEGwtbXFjz/+CA8PD/B4PKxevRqrVq1CeXk5LCwskJmZCeDh3wwhhBBC/lzP1ZBPcXEx6uvrMW7cOK11ypOSklBSUsLFxcXFYcSIEbCwsICxsTESEhKgUCi6pA0CgQDDhg3jPl+8eBFqtRoODg5abcrOztZq09Pi6OiImTNnYsSIEZDL5di6dSvkcjnWrVvXLO78+fPIy8vDrFmzEBAQgCtXrrRa7+LFi1FTU8Ntv/zyS7MYgUCAESNGcMkc8HCkOTMzs9VRVA8PD614ADh69CgXL5PJIBaLtWJqa2uRl5fX5shs07O1SqVSq5zP58Pa2hoCgQC7d++Gh4cHLCwsWq2HEEIIIU9Htx7pfNz9+/cBAIcPH262BKWBgQEAIDk5GaGhoYiOjoaHhwdMTEywdu1a5OXltVl30y3XptuzAFpcF10oFILH42m1ic/no6CgAHw+XyvW2Ni4Hb3rOm5ubjh16pRWmUAgwAsvvAAAGDFiBPLz8xEbG4tNmza1WIeBgQF3TdsSEhKCgIAAuLq6ws3NDTExMairq8O0adMAAP7+/rC2tkZUVBQAYP78+fD09ER0dDR8fX2RnJyMs2fPIiEhAQDA4/GwYMECrFixAvb29pDJZFiyZAmsrKwwceJEAEBeXh7y8/MxatQo9OrVCyUlJViyZAkGDBjAJaZ37tzB3r17MWbMGDQ0NCAxMRGpqancYxiEEEII+XM9V0nnoy/veHp6thiTk5MDuVyO2bNnc2WPjzgKBAKo1WqtsqbRr7KyMvTq1QsAdJoT1MXFBWq1GpWVlRg9enR7uvPUnD9/XuvWdEs0Gk2zUcGOePfdd1FVVYWIiAiUl5dj+PDhSE9P514EUigUWs9QyuVy7Nq1C+Hh4QgLC4O9vT3279+PoUOHcjGffPIJ6urqEBQUhOrqaowaNQrp6enc2+g9e/bEvn37sHTpUtTV1UEikcDHxwfh4eFaifL27dsRGhoKxhg8PDyQlZUFNze3TveZEEIIIe33XCWdJiYmCA0NxcKFC6HRaDBq1CjU1NQgJycHpqamCAgIgL29PZKSkpCRkQGZTIYdO3YgPz8fMpmMq0cqlSIjIwNFRUUwNzeHSCTCCy+8AFtbW0RGRmLlypVabzu3xcHBAVOmTIG/vz+io6Ph4uKCqqoqZGZmYtiwYfD19X1iHVeuXIFKpcLdu3dx7949Ltl9/BnGlsTExEAmk2HIkCFoaGjA5s2bcfz4cXz//fdczOLFizF+/Hj069cP9+7dw65du5CVlYWMjIwn1v+4S8u8YWpqqlUWHByM4ODgFuNbeot/0qRJmDRpUqvn4PF4WL58OZYvX97ificnJxw/frzNdvbp06fVKZYIIYQQ8ud7rpJOAPj8889hYWGBqKgoXL9+HWZmZnjxxRcRFhYGAJg5cyYKCwvx7rvvgsfjYfLkyZg9ezaOHDnC1TFjxgxkZWXB1dUV9+/fx4kTJzBmzBjs3r0bs2bNwrBhw/DSSy9hxYoVbSZHTRITE7FixQr885//xK1bt9CnTx+8/PLLeOONN3Tq0+uvv641d6SLiwsA7Vv9rVGpVNx5e/bsiWHDhuHYsWMYO3YsF1NZWQl/f3+UlZVBJBJh2LBhyMjIwLhx43RqHyGEEEJIZ/GYLpkN+Z9XW1sLkUiEmpqaZiOdhBBCCOmeutP393P19johhBBCCHk+UdL5lI0fP15rKqVHN13m8BwyZEirx+/cufNP6AEhhBBCSOc9d890Pm82b96MBw8etLivd+/eTzw+LS2txambgOZLRRJCCCGEdFeUdD5lj88n2l52dnZd1BJCCCGEkGeHbq8TQgghhJCnjpJOQgghhBDy1HX7pJMxhqCgIPTu3Rs8Hk+nVYLIsxEXFwepVApDQ0O4u7vjzJkzbcanpqZi4MCBMDQ0hJOTE9LS0rT2M8YQEREBiUQCoVAILy8vXLt2jduflZUFHo/X4pafnw8AKCoqwtixY9G3b18YGhqif//+CA8Pb/U5WUIIIYQ8Hd0+6UxPT8e2bdtw6NAhlJWVaS2X2FGBgYHcOt7PWkNDAwIDA+Hk5IQePXp0ql05OTno0aNHiysZ3bp1C1OnToW5uTmEQiGcnJxw9uzZdp9j6NIMSBcd5rYme/bsQUhICJYuXYpz587B2dkZ3t7eqKysbLGe06dPY/LkyZg+fToKCwsxceJETJw4EZcuXeJi1qxZg/Xr1yM+Ph55eXkwMjKCt7c3GhoaADxcUrOsrExr+/DDDyGTyeDq6goA0NfXh7+/P77//nsUFRUhJiYG//73v7F06dJ2950QQgghHdftXyQqKSmBRCKBXC5/1k1pRq1Wg8fjaa0t3pE6hEIh5s2bh2+//bbD9VRXV8Pf3x+vvvoqKioqtPb9/vvvGDlyJMaOHYsjR47AwsIC165d49aY7wpfffUVZsyYgWnTpgEA4uPjcfjwYWzduhWLFi1qFh8bGwsfHx98/PHHAB6uNHX06FF8/fXXiI+PB2MMMTExCA8Px4QJEwAASUlJ6Nu3L/bv34/33nsPAoEAYrGYq7OxsREHDhzA3LlzwePxAAD9+/dH//79uRg7OztkZWXh5MmTXdZ3QgghhDxZtx7pDAwMxNy5c6FQKMDj8SCVSqHRaBAVFQWZTAahUAhnZ2fs3buXO0atVmP69OncfkdHR8TGxnL7IyMjsX37dhw4cIC7FZuVlcXdqq2uruZiz58/Dx6Phxs3bgAAtm3bBjMzMxw8eBCDBw+GgYEBFAoFlEolQkNDYW1tDSMjI7i7u7e45nhLjIyMsHHjRsyYMUMrgWqvjz76CO+//z48PDya7Vu9ejVsbW2RmJgINzc3yGQyvPbaaxgwYECHz/colUqFgoICeHl5cWV6enrw8vJqdf3z3NxcrXgA8Pb25uJLS0tRXl6uFSMSieDu7t5qnQcPHsRvv/3GJb4tKS4uRnp6Ojw9PXXuHyGEEEI6r1snnbGxsVi+fDlsbGxQVlaG/Px8REVFISkpCfHx8bh8+TIWLlyIqVOnIjs7GwCg0WhgY2OD1NRUXLlyBREREQgLC0NKSgoAIDQ0FH5+fvDx8eFuybZnFLW+vh6rV6/G5s2bcfnyZVhaWiI4OBi5ublITk7GhQsXMGnSJPj4+Gg9f/g0JSYm4vr1663eMj548CBcXV0xadIkWFpawsXFBf/+97/brFOpVKK2tlZra82dO3egVqubzRvat29flJeXt3hMeXl5m/FNP9tT55YtW+Dt7Q0bG5tm++RyOQwNDWFvb4/Ro0dj+fLlrfaHEEIIIV2vW99eF4lEMDExAZ/Ph1gshlKpxKpVq3Ds2DFuRK9///44deoUNm3aBE9PT+jr62PZsmVcHTKZDLm5uUhJSYGfnx+MjY0hFAqhVCo7NLLY2NiIDRs2wNnZGQCgUCiQmJgIhUIBKysrAA8T2/T0dCQmJuq06lBnXLt2DYsWLcLJkyfRo0fLv87r169j48aNCAkJQVhYGPLz8zFv3jwIBAIEBAS0eExUVJTWdezufv31V2RkZHD/uHjcnj17cO/ePfz3v//Fxx9/jC+//BKffPLJn9xKQggh5H9Xt046H1dcXIz6+nqMGzdOq1ylUsHFxYX7HBcXh61bt0KhUODBgwdQqVQtvlzTEQKBAMOGDeM+X7x4EWq1Gg4ODlpxSqUS5ubmXXLO1qjVarz//vtYtmxZs/M/SqPRwNXVlUuAXVxccOnSJcTHx7eadC5evBghISHc59raWtja2rYY26dPH/D5/GbPklZUVLSa2IvF4jbjm35WVFRAIpFoxbT0u0xMTIS5uTneeuutFs/X1PbBgwdDrVYjKCgI//znP8Hn81uMJ4QQQkjXeq6Szvv37wMADh8+3GylHwMDAwBAcnIyQkNDER0dDQ8PD5iYmGDt2rXIy8trs+6ml4EYY1xZS9PqCIVC7iWVpjbx+XwUFBQ0S2CMjY3b0bv2u3fvHs6ePYvCwkIEBwcDeJhgMsbQo0cPfP/993jllVcgkUgwePBgrWMHDRrU5otLBgYG3DV9EoFAgBEjRiAzM5N7+16j0SAzM5Nr1+M8PDyQmZmJBQsWcGVHjx7lRrBlMhnEYjEyMzO5JLO2thZ5eXmYNWuWVl2MMSQmJsLf3x/6+vpPbK9Go0FjYyM0Gg0lnYQQQsif5LlKOh99eae1F0FycnIgl8sxe/ZsrqykpEQrRiAQQK1Wa5VZWFgAAMrKyri3unWZE9TFxQVqtRqVlZUYPXp0e7rTaaamprh48aJW2YYNG3D8+HHs3bsXMpkMADBy5EgUFRVpxV29erVLl9gMCQlBQEAAXF1d4ebmhpiYGNTV1XEv9fj7+8Pa2hpRUVEAgPnz58PT0xPR0dHw9fVFcnIyzp49i4SEBAAAj8fDggULsGLFCtjb20Mmk2HJkiWwsrJqNq3U8ePHUVpaig8//LBZu3bu3Al9fX04OTnBwMAAZ8+exeLFi/Huu+/qlKASQgghpGs8V0mniYkJQkNDsXDhQmg0GowaNQo1NTXIycmBqakpAgICYG9vj6SkJGRkZEAmk2HHjh3Iz8/nEjAAkEqlyMjIQFFREczNzSESifDCCy/A1tYWkZGRWLlyJa5evYro6OgntsnBwQFTpkyBv78/oqOj4eLigqqqKmRmZmLYsGHw9fV9Yh1XrlyBSqXC3bt3ce/ePS7ZfdIjAXp6es3mLbW0tIShoaFW+cKFCyGXy7Fq1Sr4+fnhzJkzSEhI4BK8rvDuu++iqqoKERERKC8vx/Dhw5Gens69CKRQKLSmlpLL5di1axfCw8MRFhYGe3t77N+/X6vdn3zyCerq6hAUFITq6mqMGjUK6enpMDQ01Dr3li1bIJfLMXDgwGbt6tGjB1avXo2rV6+CMQY7OzsEBwdj4cKFXdZ3QgghhOiAdXPr1q1jdnZ23GeNRsNiYmKYo6Mj09fXZxYWFszb25tlZ2czxhhraGhggYGBTCQSMTMzMzZr1iy2aNEi5uzszNVRWVnJxo0bx4yNjRkAduLECcYYY6dOnWJOTk7M0NCQjR49mqWmpjIArLS0lDHGWGJiIhOJRM3aqFKpWEREBJNKpUxfX59JJBL29ttvswsXLujURzs7Owag2dYRS5cu1eprk//85z9s6NChzMDAgA0cOJAlJCS0q96amhoGgNXU1HSoXYQQQgj583Wn728eY488xEhIK2prayESiVBTUwNTU9Nn3RxCCCGE6KA7fX9363k6CSGEEELIXwMlnU/Z+PHjYWxs3OKmyxyeQ4YMafX4nTt3/gk9IIQQQgjpvOfqRaLn0ebNm/HgwYMW9/Xu3fuJx6elpbU4dRPQfLUeQgghhJDuipLOp+zx+UTbqyunNSKEEEIIeVbo9johhBBCCHnqKOkkhBBCCCFPHSWdpMvExcVBKpXC0NAQ7u7uOHPmTJvxqampGDhwIAwNDeHk5IS0tDSt/YwxREREQCKRQCgUwsvLC9euXeP2Z2Vlgcfjtbjl5+dzcRkZGXj55ZdhYmICCwsL/OMf/8CNGze6tO+EEEIIadszTToZYwgKCkLv3r3B4/F0WnaSPFtDl2ZAuugwtzXZs2cPQkJCsHTpUpw7dw7Ozs7w9vZGZWVli/WcPn0akydPxvTp01FYWIiJEydi4sSJuHTpEhezZs0arF+/HvHx8cjLy4ORkRG8vb3R0NAA4OGqRmVlZVrbhx9+CJlMBldXVwBAaWkpJkyYgFdeeQXnz59HRkYG7ty5g7///e9P8SoRQgghpJlnOTN9Wloa09fXZzk5OaysrIw1NjZ2us6AgAA2YcKEzjeuCzx48IAFBASwoUOHMj6f3+52nTx5ksnlcta7d29maGjIHB0d2VdffaUVs2rVKubq6sqMjY2ZhYUFmzBhAvv555+b1XX69Gk2duxY1rNnT2ZiYsJGjx7N6uvrdW5L04oGtgtSmN2nh7itiZubG5szZw73Wa1WMysrKxYVFdVifX5+fszX11erzN3dnc2cOZMx9nDlKbFYzNauXcvtr66uZgYGBmz37t0t1qlSqZiFhQVbvnw5V5aamsp69OjB1Go1V3bw4EHG4/GYSqXSuf+EEELI86g7rUj0TEc6S0pKIJFIIJfLIRaL0aNH93mZXq1WQ6PRdLoOoVCIefPmwcvLq93HGxkZITg4GD/88AN++uknhIeHIzw8XGvN9OzsbMyZMwc//vgjjh49isbGRrz22muoq6vjYnJzc+Hj44PXXnsNZ86cQX5+PoKDg7XWQu8MlUqFgoICrT7q6enBy8sLubm5LR6Tm5vb7Jp4e3tz8aWlpSgvL9eKEYlEcHd3b7XOgwcP4rfffsO0adO4shEjRkBPTw+JiYlQq9WoqanBjh074OXlBX19/Q73mRBCCCHt88ySzsDAQMydOxcKhQI8Hg9SqRQajQZRUVGQyWQQCoVwdnbG3r17uWPUajWmT5/O7Xd0dERsbCy3PzIyEtu3b8eBAwe4Z/uysrK4Z/+qq6u52PPnz4PH43HP9m3btg1mZmY4ePAgBg8eDAMDAygUCiiVSoSGhsLa2hpGRkZwd3dHVlaWTn00MjLCxo0bMWPGDIjF4nZfIxcXF0yePBlDhgyBVCrF1KlT4e3tjZMnT3Ix6enpCAwMxJAhQ+Ds7Ixt27ZBoVCgoKCAi1m4cCHmzZuHRYsWYciQIXB0dISfnx8MDAza3aaW3LlzB2q1utm8oX379kV5eXmLx5SXl7cZ3/SzPXVu2bIF3t7esLGx4cpkMhm+//57hIWFwcDAAGZmZvj111+RkpLSvk4SQgghpFOeWdIZGxuL5cuXw8bGBmVlZcjPz0dUVBSSkpIQHx+Py5cvY+HChZg6dSqys7MBABqNBjY2NkhNTcWVK1cQERGBsLAwLoEIDQ2Fn58ffHx8uGf85HK5zm2qr6/H6tWrsXnzZly+fBmWlpYIDg5Gbm4ukpOTceHCBUyaNAk+Pj5aL7T8WQoLC3H69Gl4enq2GlNTUwPg/088X1lZiby8PFhaWkIul6Nv377w9PTEqVOn2jyXUqlEbW2t1tad/frrr8jIyMD06dO1ysvLyzFjxgwEBAQgPz8f2dnZEAgEeOedd8AYe0atJYQQQv73PLP72SKRCCYmJuDz+RCLxVAqlVi1ahWOHTsGDw8PAED//v1x6tQpbNq0CZ6entDX18eyZcu4OmQyGXJzc5GSkgI/Pz8YGxtDKBRCqVR2aGSxsbERGzZsgLOzMwBAoVAgMTERCoUCVlZWAB4mtunp6UhMTNRpGcuuYGNjg6qqKvzxxx+IjIzEhx9+2GKcRqPBggULMHLkSAwdOhQAcP36dQAPR4G//PJLDB8+HElJSXj11Vdx6dIl2Nvbt1hXVFSU1rVuS58+fcDn81FRUaFVXlFR0ervQSwWtxnf9LOiogISiUQrZvjw4c3qS0xMhLm5Od566y2t8ri4OIhEIqxZs4Yr++abb2Bra4u8vDy8/PLLOvWREEIIIZ3TbaZMKi4uRn19PcaNG6e1vnhSUhJKSkq4uLi4OIwYMQIWFhYwNjZGQkICFApFl7RBIBBg2LBh3OeLFy9CrVbDwcFBq03Z2dlabXraTp48ibNnzyI+Ph4xMTHYvXt3i3Fz5szBpUuXkJyczJU1PZc6c+ZMTJs2DS4uLli3bh0cHR2xdevWVs+5ePFi1NTUcNsvv/zSaqxAIMCIESOQmZmpdd7MzEzuHxCP8/Dw0IoHgKNHj3LxMpkMYrFYK6a2thZ5eXnN6mSMITExEf7+/s2e06yvr2/27Cqfz+faSAghhJA/R7d5c+f+/fsAgMOHDzdbOrLp2cPk5GSEhoYiOjoaHh4eMDExwdq1a5GXl9dm3U1Jx6O3U1taz1woFILH42m1ic/no6CggEtUmhgbG7ejd50jk8kAAE5OTqioqEBkZCQmT56sFRMcHIxDhw7hhx9+0HqmsWmUcPDgwVrxgwYNajNZNzAwaNcznyEhIQgICICrqyvc3NwQExODuro67qUef39/WFtbIyoqCgAwf/58eHp6Ijo6Gr6+vkhOTsbZs2e5l6R4PB4WLFiAFStWwN7eHjKZDEuWLIGVlRUmTpyode7jx4+jtLS0xRFgX19frFu3DsuXL8fkyZNx7949hIWFwc7ODi4uLjr3jxBCCCGd022Szkdf3mntmcWcnBzI5XLMnj2bK3t8xFEgEECtVmuVWVhYAADKysrQq1cvANBpTlAXFxeo1WpUVlZi9OjR7enOU6PRaKBUKrnPjDHMnTsX3333HbKysrgEtYlUKoWVlRWKioq0yq9evYrx48d3WbveffddVFVVISIiAuXl5Rg+fDjS09O5F4EUCoXWiKNcLseuXbsQHh6OsLAw2NvbY//+/dxjAQDwySefoK6uDkFBQaiursaoUaOQnp4OQ0NDrXNv2bIFcrkcAwcObNauV155Bbt27cKaNWuwZs0a9OzZEx4eHkhPT4dQKOyy/hNCCCHkCZ7lfE3r1q1jdnZ23OfPPvuMmZubs23btrHi4mJWUFDA1q9fz7Zt28YYYyw2NpaZmpqy9PR0VlRUxMLDw5mpqSlzdnbm6li5ciXr168f+/nnn1lVVRVTqVRMpVIxW1tbNmnSJHb16lV26NAh5ujoyACw0tJSxhhjiYmJTCQSNWvjlClTmFQqZd9++y27fv06y8vLY6tWrWKHDh1qFtuSy5cvs8LCQvbmm2+yMWPGsMLCQlZYWKjTsV9//TU7ePAgu3r1Krt69SrbvHkzMzExYZ999hkXM2vWLCYSiVhWVhYrKyvjtkfn4Fy3bh0zNTVlqamp7Nq1ayw8PJwZGhqy4uJindrBWPea54sQQgghuulO39/dKunUaDQsJiaGOTo6Mn19fWZhYcG8vb1ZdnY2Y4yxhoYGFhgYyEQiETMzM2OzZs1iixYt0ko6Kysr2bhx45ixsTEDwE6cOMEYY+zUqVPMycmJGRoastGjR7PU1FSdkk6VSsUiIiKYVCpl+vr6TCKRsLfffptduHBBpz7a2dkxAM02Xaxfv54NGTKE9ezZk5mamjIXFxe2YcMGrYnOW6obAEtMTNSqKyoqitnY2LCePXsyDw8PdvLkSZ3a0KQ7/dESQgghRDfd6fubxxjNG0OerLa2FiKRCDU1NTA1NX3WzSGEEEKIDrrT93e3eXudEEIIIYT8dVHS2Qnjx4/Xmkrp0U2XOTyHDBnS6vE7d+78E3pACCGEEPLn6DZvrz+PNm/ejAcPHrS4r2lFoLakpaW1OHUT0Hz5R0IIIYSQ5xklnZ3w+Hyi7WVnZ9dFLSGEEEII6d7o9johhBBCCHnqKOkkhBBCCCFP3TNNOhljCAoKQu/evcHj8XRaJYh0X3FxcZBKpTA0NIS7uzvOnDnTZnxqaioGDhwIQ0NDODk5IS0tTWs/YwwRERGQSCQQCoXw8vLCtWvXuP1ZWVng8Xgtbvn5+VxcRkYGXn75ZZiYmMDCwgL/+Mc/cOPGjS7tOyGEEELa9kyTzvT0dGzbtg2HDh1CWVmZ1hKIHRUYGNhsbe5npaGhAYGBgXByckKPHj3a3a7AwMAWE6ohQ4ZoxT0p2UtISMCYMWNgamoKHo+H6urqDvdp6NIMSBcd5rYme/bsQUhICJYuXYpz587B2dkZ3t7eqKysbLGe06dPY/LkyZg+fToKCwsxceJETJw4EZcuXeJi1qxZg/Xr1yM+Ph55eXkwMjKCt7c3GhoaADxcSrOsrExr+/DDDyGTyeDq6goAKC0txYQJE/DKK6/g/PnzyMjIwJ07d/D3v/+9w9eAEEIIIe33TJPOkpISSCQSyOVyiMVi9OjRfd5rUqvV0Gg0na5DKBRi3rx58PLyavfxsbGxWgnVL7/8gt69e2PSpElcjC7JXn19PXx8fBAWFtap/rTlq6++wowZMzBt2jQMHjwY8fHx6NmzJ7Zu3dpq33x8fPDxxx9j0KBB+Pzzz/Hiiy/i66+/BvBwlDMmJgbh4eGYMGEChg0bhqSkJNy+fRv79+8HAAgEAojFYm4zNzfHgQMHMG3aNPB4PABAQUEB1Go1VqxYgQEDBuDFF19EaGgozp8/3+rMAYQQQgjpes8s6QwMDMTcuXOhUCjA4/EglUqh0WgQFRUFmUwGoVAIZ2dn7N27lztGrVZj+vTp3H5HR0fExsZy+yMjI7F9+3YcOHCAGxXMysribsM+OsJ3/vx58Hg87jbrtm3bYGZmhoMHD2Lw4MEwMDCAQqGAUqlEaGgorK2tYWRkBHd3d2RlZenURyMjI2zcuBEzZsyAWCxu9zUSiURaSdXZs2fx+++/Y9q0aVyMLsneggULsGjRIrz88svtboMuVCoVCgoKtBJrPT09eHl5ITc3t8VjcnNzmyXi3t7eXHxpaSnKy8u1YkQiEdzd3Vut8+DBg/jtt9+0rs+IESOgp6eHxMREqNVq1NTUYMeOHfDy8oK+vn6H+0wIIYSQ9nlmQ4uxsbEYMGAAEhISkJ+fDz6fj6ioKHzzzTeIj4+Hvb09fvjhB0ydOhUWFhbw9PSERqOBjY0NUlNTYW5ujtOnTyMoKAgSiQR+fn4IDQ3FTz/9hNraWiQmJgJ4OF/m6dOndWpTfX09Vq9ejc2bN8Pc3ByWlpYIDg7GlStXkJycDCsrK3z33Xfw8fHBxYsXYW9v/zQvUTNbtmyBl5cXN9VSU7K3ePFiLuZJyZ6ulEollEol97m2trbV2Dt37kCtVjebW7Rv3774+eefWzymvLy8xfjy8nJuf1NZazGP27JlC7y9vWFjY8OVyWQyfP/99/Dz88PMmTOhVqvh4eHR7PlRQgghhDxdzyzpFIlEMDExAZ/Ph1gshlKpxKpVq3Ds2DF4eHgAAPr3749Tp05h06ZN8PT0hL6+PpYtW8bVIZPJkJubi5SUFPj5+cHY2BhCoRBKpbJDI4uNjY3YsGEDnJ2dAQAKhQKJiYlQKBSwsrICAISGhiI9PR2JiYk6rTrUVW7fvo0jR45g165dXFlHkj1dRUVFaV3r7u7XX39FRkYGUlJStMrLy8sxY8YMBAQEYPLkybh37x4iIiLwzjvv4OjRo9xteEIIIYQ8Xd3mIcri4mLU19dj3LhxWuUqlQouLi7c57i4OGzduhUKhQIPHjyASqXC8OHDu6QNAoEAw4YN4z5fvHgRarUaDg4OWnFKpRLm5uZdck5dbd++HWZmZn/aS1KLFy9GSEgI97m2tha2trYtxvbp0wd8Ph8VFRVa5RUVFa0m/2KxuM34pp8VFRWQSCRaMS39vhMTE2Fubo633npLqzwuLg4ikQhr1qzhyr755hvY2toiLy/vqT1yQAghhBBt3SbpvH//PgDg8OHDzVb6MTAwAAAkJycjNDQU0dHR8PDwgImJCdauXYu8vLw269bTe/joKmOMK2vpJRKhUKg18nX//n3w+XwUFBSAz+drxRobG7ejd53DGMPWrVvxwQcfQCAQcOUdSfZ0ZWBgwF33JxEIBBgxYgQyMzO5pFij0SAzMxPBwcEtHuPh4YHMzEwsWLCAKzt69Cg3yi2TySAWi5GZmcklmbW1tcjLy8OsWbO06mKMITExEf7+/s2e06yvr+d+/02afpedfVGMEEIIIbrrNknnoy/veHp6thiTk5MDuVyO2bNnc2UlJSVaMQKBAGq1WqvMwsICAFBWVoZevXoBgE5zgrq4uECtVqOyshKjR49uT3e6VHZ2NoqLizF9+nSt8o4ke09LSEgIAgIC4OrqCjc3N8TExKCuro57qcff3x/W1taIiooCAMyfPx+enp6Ijo6Gr68vkpOTcfbsWSQkJAAAeDweFixYgBUrVsDe3h4ymQxLliyBlZVVs9He48ePo7S0FB9++GGzdvn6+mLdunVYvnw5d3s9LCwMdnZ2WiPohBBCCHm6uk3SaWJigtDQUCxcuBAajQajRo1CTU0NcnJyYGpqioCAANjb2yMpKQkZGRmQyWTYsWMH8vPzIZPJuHqkUikyMjJQVFQEc3NziEQivPDCC7C1tUVkZCRWrlyJq1evIjo6+oltcnBwwJQpU+Dv74/o6Gi4uLigqqoKmZmZGDZsGHx9fZ9Yx5UrV6BSqXD37l3cu3ePS3bb80jAli1b4O7u3uI8pk9K9oCHzzWWl5ejuLgYwMPHBkxMTNCvXz/07t1b53a05d1330VVVRUiIiJQXl6O4cOHIz09nXveVKFQaI04yuVy7Nq1C+Hh4QgLC4O9vT3279+v1cdPPvkEdXV1CAoKQnV1NUaNGoX09HQYGho2uz5yuRwDBw5s1q5XXnkFu3btwpo1a7BmzRr07NkTHh4eSE9Ph1Ao7JK+E0IIIUQH7Blat24ds7Oz4z5rNBoWExPDHB0dmb6+PrOwsGDe3t4sOzubMcZYQ0MDCwwMZCKRiJmZmbFZs2axRYsWMWdnZ66OyspKNm7cOGZsbMwAsBMnTjDGGDt16hRzcnJihoaGbPTo0Sw1NZUBYKWlpYwxxhITE5lIJGrWRpVKxSIiIphUKmX6+vpMIpGwt99+m124cEGnPtrZ2TEAzTZdVVdXM6FQyBISElqN+de//sX69evHBAIBc3NzYz/++KPW/qVLl7bYhsTERJ3bUVNTwwCwmpoanY8hhBBCyLPVnb6/eYw98qAjIa2ora2FSCRCTU0NTE1Nn3VzCCGEEKKD7vT9/UxXJCKEEEIIIf8bKOnshPHjx8PY2LjFTZc5PIcMGdLq8Tt37vwTekAIIYQQ8ufoNi8SPY82b96MBw8etLhPlxd00tLSWl3/+/EJ3wkhhBBCnmeUdHbC4/OJtlfTcpaEEEIIIX91dHudEEIIIYQ8dZR0EkIIIYSQp46STtKl4uLiIJVKYWhoCHd3d5w5c6bN+NTUVAwcOBCGhoZwcnJCWlqa1n7GGCIiIiCRSCAUCuHl5YVr165pxVy9ehUTJkxAnz59YGpqilGjRuHEiRNaMfn5+Xj11VdhZmaGXr16wdvbG//973+7ptOEEEIIeaJun3QyxhAUFITevXuDx+PptHwleXqGLs2AdNFhra3Jnj17EBISgqVLl+LcuXNwdnaGt7c3KisrW6zr9OnTmDx5MqZPn47CwkJMnDgREydOxKVLl7iYNWvWYP369YiPj0deXh6MjIzg7e2NhoYGLuaNN97AH3/8gePHj6OgoADOzs544403UF5eDgC4f/8+fHx80K9fP+Tl5eHUqVMwMTGBt7d3qy9yEUIIIaSLPdu56Z8sLS2N6evrs5ycHFZWVsYaGxs7XWdAQACbMGFC5xvXBR48eMACAgLY0KFDGZ/P71C7GhoaWFhYGLcqkZ2dHduyZYtWzLp165iDgwMzNDRkNjY2bMGCBezBgwc6n6NpRQPbBSnM7tNDWlsTNzc3NmfOHO6zWq1mVlZWLCoqqsU6/fz8mK+vr1aZu7s7mzlzJmPs4QpVYrGYrV27lttfXV3NDAwM2O7duxljjFVVVTEA7IcffuBiamtrGQB29OhRxhhj+fn5DABTKBRczIULFxgAdu3aNZ2vASGEEPK86U4rEnX7kc6SkhJIJBLI5XKIxWL06NF9XrhXq9XQaDSdrkMoFGLevHnw8vLqUB1+fn7IzMzEli1bUFRUhN27d8PR0ZHbv2vXLixatAhLly7FTz/9hC1btmDPnj0ICwvrVNsfpVKpUFBQoNUHPT09eHl5ITc3t8VjcnNzm/XZ29ubiy8tLUV5eblWjEgkgru7Oxdjbm4OR0dHJCUloa6uDn/88Qc2bdoES0tLjBgxAgDg6OgIc3NzbNmyBSqVCg8ePMCWLVswaNAgSKXSLrsGhBBCCGldt046AwMDMXfuXCgUCvB4PEilUmg0GkRFRUEmk0EoFMLZ2Rl79+7ljlGr1Zg+fTq339HREbGxsdz+yMhIbN++HQcOHACPxwOPx0NWVhaysrLA4/FQXV3NxZ4/fx48Hg83btwAAGzbtg1mZmY4ePAgBg8eDAMDAygUCiiVSoSGhsLa2hpGRkZwd3dHVlaWTn00MjLCxo0bMWPGDIjF4nZfo/T0dGRnZyMtLQ1eXl6QSqXw8PDAyJEjuZjTp09j5MiReP/99yGVSvHaa69h8uTJT3zesj3u3LkDtVrdbH7Rvn37cre5H1deXt5mfNPPtmJ4PB6OHTuGwsJCmJiYwNDQEF999RXS09PRq1cvAICJiQmysrLwzTffQCgUwtjYGOnp6Thy5Ei3+kcMIYQQ8lfWrZPO2NhYLF++HDY2NigrK0N+fj6ioqKQlJSE+Ph4XL58GQsXLsTUqVORnZ0NANBoNLCxsUFqaiquXLmCiIgIhIWFISUlBQAQGhoKPz8/+Pj4oKysDGVlZZDL5Tq3qb6+HqtXr8bmzZtx+fJlWFpaIjg4GLm5uUhOTsaFCxcwadIk+Pj4NHvh5Wk4ePAgXF1dsWbNGlhbW8PBwQGhoaFak9bL5XIUFBRwSeb169eRlpaG119/vdV6lUolamtrtbbuiDGGOXPmwNLSEidPnsSZM2cwceJEvPnmmygrKwMAPHjwANOnT8fIkSPx448/IicnB0OHDoWvr2+rk/sTQgghpGt162EekUgEExMT8Pl8iMViKJVKrFq1CseOHYOHhwcAoH///jh16hQ2bdoET09P6OvrY9myZVwdMpkMubm5SElJgZ+fH4yNjSEUCqFUKjs0stjY2IgNGzbA2dkZAKBQKJCYmAiFQgErKysADxPb9PR0JCYm6rQcZmdcv34dp06dgqGhIb777jvcuXMHs2fPxm+//YbExEQAwPvvv487d+5g1KhRYIzhjz/+wEcffdTm7fWoqCit6/gkffr0AZ/PR0VFhVZ5RUVFq9dZLBa3Gd/0s6KiAhKJRCtm+PDhAIDjx4/j0KFD+P3332FqagoA2LBhA44ePYrt27dj0aJF2LVrF27cuIHc3Fzo6T38d9auXbvQq1cvHDhwAO+9957O/SSEEEJIx3Trkc7HFRcXo76+HuPGjdNapzwpKQklJSVcXFxcHEaMGAELCwsYGxsjISEBCoWiS9ogEAgwbNgw7vPFixehVqvh4OCg1abs7GytNj0tGo0GPB4PO3fuhJubG15//XV89dVX2L59OzeKl5WVhVWrVmHDhg04d+4c9u3bh8OHD+Pzzz9vtd7FixejpqaG23755Zc22yEQCDBixAhkZmZqtS0zM5P7B8LjPDw8tOIB4OjRo1y8TCaDWCzWiqmtrUVeXh4XU19fDwBcMtlET0+Pe962vr4eenp64PF4Wvt5PF6nn8klhBBCiG669Ujn4+7fvw8AOHz4cLMlKA0MDAAAycnJCA0NRXR0NDw8PGBiYoK1a9ciLy+vzbqbkhbGGFfW0nQ6QqFQK3m5f/8++Hw+CgoKwOfztWKNjY3b0buOkUgksLa2hkgk4soGDRoExhh+/fVX2NvbY8mSJfjggw/w4YcfAgCcnJxQV1eHoKAgfPbZZ80SNuDh9Wy6proKCQlBQEAAXF1d4ebmhpiYGNTV1WHatGkAAH9/f1hbWyMqKgoAMH/+fHh6eiI6Ohq+vr5ITk7G2bNnkZCQAODh85oLFizAihUrYG9vD5lMhiVLlsDKygoTJ04E8DBx7dWrFwICAhAREQGhUIh///vfKC0tha+vLwBg3Lhx+PjjjzFnzhzMnTsXGo0GX3zxBXr06IGxY8e274ITQgghpEOeq6Tz0Zd3PD09W4zJycmBXC7H7NmzubLHRxwFAgHUarVWmYWFBQCgrKyMewFFlzlBXVxcoFarUVlZidGjR7enO11i5MiRSE1Nxf3797kk9+rVq9DT04ONjQ2A/z/S96imBPnRJLuz3n33XVRVVSEiIgLl5eUYPnw40tPTuReBFAqFVjvkcjl27dqF8PBwhIWFwd7eHvv378fQoUO5mE8++YRLkKurqzFq1Cikp6fD0NAQwMPb+unp6fjss8/wyiuvoLGxEUOGDMGBAwe4RyAGDhyI//znP1i2bBk8PDygp6cHFxcXpKena922J4QQQshT9EwnbNLBunXrmJ2dHff5s88+Y+bm5mzbtm2suLiYFRQUsPXr17Nt27YxxhiLjY1lpqamLD09nRUVFbHw8HBmamrKnJ2duTpWrlzJ+vXrx37++WdWVVXFVCoVU6lUzNbWlk2aNIldvXqVHTp0iDk6OjIArLS0lDHGWGJiIhOJRM3aOGXKFCaVStm3337Lrl+/zvLy8tiqVavYoUOHmsW25PLly6ywsJC9+eabbMyYMaywsJAVFhbqdOy9e/eYjY0Ne+edd9jly5dZdnY2s7e3Zx9++CEXs3TpUmZiYsJ2797Nrl+/zr7//ns2YMAA5ufnp9M5GOte83wRQgghRDfd6fv7uUs6NRoNi4mJYY6OjkxfX59ZWFgwb29vlp2dzRh7OFF6YGAgE4lEzMzMjM2aNYstWrRIK+msrKxk48aNY8bGxgwAO3HiBGOMsVOnTjEnJydmaGjIRo8ezVJTU3VKOlUqFYuIiGBSqZTp6+sziUTC3n77bXbhwgWd+mhnZ8cANNt09dNPPzEvLy8mFAqZjY0NCwkJYfX19dz+xsZGFhkZyQYMGMAMDQ2Zra0tmz17Nvv99991Pkd3+qMlhBBCiG660/c3j7EuvL9K/rJqa2shEolQU1PDvSVOCCGEkO6tO31/P1dvrxNCCCGEkOdTh5LO9PR0nDp1ivscFxeH4cOH4/3338fvv//eZY37Kxg/frzWVEqPbrrM4TlkyJBWj9+5c+ef0ANCCCGEkM7r0O11JycnrF69Gq+//jouXryIl156CSEhIThx4gQGDhzITUpOgFu3brW66k3v3r3Ru3fvNo+/efNmi1M3AQ+XgzQxMel0G3XRnYbnCSGEEKKb7vT93aEpk0pLSzF48GAAwLfffos33ngDq1atwrlz59pcWvF/0ePzibaXnZ1dF7WEEEIIIeTZ6dDtdYFAwK0Ec+zYMbz22msAHo7cddc1ugkhhBBCyLPToZHOUaNGISQkBCNHjsSZM2ewZ88eAA8nJW+akJwQQgghhJAmHRrp/Prrr9GjRw/s3bsXGzdu5G4hHzlyBD4+Pl3aQMYYgoKC0Lt3b/B4PJ1WCSLPVlxcHKRSKQwNDeHu7o4zZ860GZ+amoqBAwfC0NAQTk5OSEtL09rPGENERAQkEgmEQiG8vLxw7do1rZhz585h3LhxMDMzg7m5OYKCgrhlUwFg27Zt4PF4LW6VlZVd13lCCCGEtOxZThKqi7S0NKavr89ycnJYWVkZa2xs7HSdAQEBbMKECZ1vXBd48OABCwgIYEOHDmV8Pr9D7WpoaGBhYWGsX79+TCAQMDs7O7ZlyxZu/6VLl9jf//53bhL6devWtfscTZPL2i5IYXafHmq2NUlOTmYCgYBt3bqVXb58mc2YMYOZmZmxioqKFuvNyclhfD6frVmzhl25coWFh4czfX19dvHiRS7miy++YCKRiO3fv5/997//ZW+99RaTyWTswYMHjDHGbt26xXr16sU++ugj9vPPP7MzZ84wuVzO/vGPf3B11NfXs7KyMq3N29ubeXp6tvtaEEIIIc+L7jQ5fIfn6SwpKUF4eDgmT57MjRQdOXIEly9f7pps+JHzSCQSyOVyiMVi9OjRfZaLV6vV0Gg0na5DKBRi3rx58PLy6lAdfn5+yMzMxJYtW1BUVITdu3fD0dGR219fX4/+/fvjiy++gFgs7lR7n+Srr77CjBkzMG3aNAwePBjx8fHo2bMntm7d2mJ8bGwsfHx88PHHH2PQoEH4/PPP8eKLL+Lrr78G8HCUMyYmBuHh4ZgwYQKGDRuGpKQk3L59G/v37wcAHDp0CPr6+oiLi4OjoyNeeuklxMfH49tvv0VxcTEAQCgUQiwWcxufz8fx48cxffr0p3o9CCGEEPJQh5LO7OxsODk5IS8vD/v27eNuY/73v//F0qVLu6xxgYGBmDt3LhQKBXg8HqRSKTQaDaKioiCTySAUCuHs7Iy9e/dyx6jVakyfPp3b7+joiNjYWG5/ZGQktm/fjgMHDnC3V7OyspCVlQUej4fq6mou9vz58+DxeLhx4waAh7dozczMcPDgQQwePBgGBgZQKBRQKpUIDQ2FtbU1jIyM4O7ujqysLJ36aGRkhI0bN2LGjBkdSgjT09ORnZ2NtLQ0eHl5QSqVwsPDAyNHjuRiXnrpJaxduxbvvfceDAwM2n0OXalUKhQUFGglz3p6evDy8kJubm6Lx+Tm5jZLtr29vbn40tJSlJeXa8WIRCK4u7tzMUqlEgKBAHp6///PWSgUAoDWfLKPSkpKQs+ePfHOO+90oKeEEEIIaa8OJZ2LFi3CihUrcPToUQgEAq78lVdewY8//thljYuNjcXy5cthY2ODsrIy5OfnIyoqCklJSYiPj8fly5excOFCTJ06FdnZ2QAAjUYDGxsbpKam4sqVK4iIiEBYWBhSUlIAAKGhofDz84OPjw/KyspQVlYGuVyuc5vq6+uxevVqbN68GZcvX4alpSWCg4ORm5uL5ORkXLhwAZMmTYKPj0+z5w6fhoMHD8LV1RVr1qyBtbU1HBwcEBoa2urcoLpSKpWora3V2p7kzp07UKvV6Nu3r1Z53759UV5e3uIx5eXlbcY3/Wwr5pVXXkF5eTnWrl0LlUqF33//HYsWLQIAlJWVtXjeLVu24P333+eSU0IIIYQ8XR26V33x4kXs2rWrWbmlpSXu3LnT6UY1EYlEMDExAZ/Ph1gshlKpxKpVq3Ds2DF4eHgAAPr3749Tp05h06ZN8PT0hL6+PpYtW8bVIZPJkJubi5SUFPj5+cHY2BhCoRBKpbJDI4uNjY3YsGEDnJ2dAQAKhQKJiYlQKBSwsrIC8DCxTU9PR2Jiok6rDnXG9evXcerUKRgaGuK7777DnTt3MHv2bPz222+dmqQ/KipK6zp2Z0OGDMH27dsREhKCxYsXg8/nY968eejbt6/W6GeT3Nxc/PTTT9ixY8czaC0hhBDyv6lDSaeZmRnKysogk8m0ygsLCzs9GXpbiouLUV9fj3HjxmmVq1QquLi4cJ/j4uKwdetWKBQKPHjwACqVCsOHD++SNggEAgwbNoz7fPHiRajVajg4OGjFKZVKmJubd8k526LRaMDj8bBz506IRCIAD5+rfOedd7Bhw4YOj+QtXrwYISEh3Ofa2lrY2tq2eUyfPn3A5/NRUVGhVV5RUdFqgi8Wi9uMb/pZUVEBiUSiFfPo7/T999/H+++/j4qKChgZGYHH4+Grr75C//79m51z8+bNGD58OEaMGNFmfwghhBDSdTqUdL733nv49NNPkZqaCh6PB41Gg5ycHISGhsLf37+r28hpenb08OHDzZLbpmcVk5OTERoaiujoaHh4eMDExARr165FXl5em3U3jYixR1YFbWn5SaFQCB6Pp9UmPp+PgoIC8Pl8rVhjY+N29K5jJBIJrK2tuYQTAAYNGgTGGH799VfY29t3qF4DA4N2P/8pEAgwYsQIZGZmYuLEiQAeJsWZmZkIDg5u8RgPDw9kZmZiwYIFXNnRo0e5kWyZTAaxWIzMzEwuyaytrUVeXh5mzZrVrL6m2/Bbt26FoaFhs3+g3L9/HykpKYiKimpX3wghhBDSOR1KOletWoU5c+bA1tYWarUagwcPhlqtxvvvv4/w8PCubiPn0Zd3PD09W4zJycmBXC7H7NmzubKSkhKtGIFAALVarVVmYWEB4OEzgL169QIAneYEdXFxgVqtRmVlJUaPHt2e7nSJkSNHIjU1Fffv3+eS3KtXr0JPT++ZTNQfEhKCgIAAuLq6ws3NDTExMairq8O0adMAAP7+/rC2tuaSvvnz58PT0xPR0dHw9fVFcnIyzp49i4SEBAAAj8fDggULsGLFCtjb20Mmk2HJkiWwsrLiElvg4dyxcrkcxsbGOHr0KD7++GN88cUXMDMz02rfnj178Mcff2Dq1Kl/yvUghBBCyEMdSjoFAgH+/e9/Y8mSJbh06RLu378PFxeXDo+q6crExAShoaFYuHAhNBoNRo0ahZqaGuTk5MDU1BQBAQGwt7dHUlISMjIyIJPJsGPHDuTn52s9CiCVSpGRkYGioiKYm5tDJBLhhRdegK2tLSIjI7Fy5UpcvXoV0dHRT2yTg4MDpkyZAn9/f0RHR8PFxQVVVVXIzMzEsGHD4Ovr+8Q6rly5ApVKhbt37+LevXtcsqvLIwHvv/8+Pv/8c0ybNg3Lli3DnTt38PHHH+P//u//uFvrKpUKV65c4f771q1bOH/+PIyNjfHCCy888Rzt8e6776KqqgoREREoLy/H8OHDkZ6ezo1AKhQKrecs5XI5du3ahfDwcISFhcHe3h779+/H0KFDuZhPPvkEdXV1CAoKQnV1NUaNGoX09HQYGhpyMWfOnMHSpUtx//59DBw4EJs2bcIHH3zQrH1btmzB3//+92bJKCGEEEKesmc8T+gTrVu3jtnZ2XGfNRoNi4mJYY6OjkxfX59ZWFgwb29vlp2dzRh7OFF6YGAgE4lEzMzMjM2aNYstWrSIOTs7c3VUVlaycePGMWNjYwaAnThxgjHG2KlTp5iTkxMzNDRko0ePZqmpqQwAKy0tZYwxlpiYyEQiUbM2qlQqFhERwaRSKdPX12cSiYS9/fbb7MKFCzr1sWnS9sc3Xf3000/My8uLCYVCZmNjw0JCQlh9fT23v7S0tMX62zMxeneaXJYQQgghuulO3988xh55iFH3RBV79+7FiRMnUFlZ2WyC9H379nUiDSbdUW1tLUQiEWpqamBqavqsm0MIIYQQHXSn7+8O3V5fsGABNm3ahLFjx6Jv375aL9YQQgghhBDyuA4lnTt27MC+ffvw+uuvd3V7/nLGjx+PkydPtrgvLCwMYWFhbR4/ZMgQ3Lx5s8V9mzZtwpQpUzrdRkIIIYSQp61DSadIJGpx/kPS3ObNm1tdHah3795PPD4tLa3FqZuA5qv0EEIIIYR0Vx16pnP79u1IT0/H1q1baRnB/xHd6ZkQQgghhOimO31/d2ik08/PD7t374alpSWkUin09fW19p87d65LGkcIIYQQQv4aOpR0BgQEoKCgAFOnTqUXiQghhBBCyBPpPTmkucOHD+O7777Dxo0bERkZiaVLl2pthABAXFwcpFIpDA0N4e7ujjNnzrQZn5qaioEDB8LQ0BBOTk5IS0vT2r9v3z689tprMDc3B4/Ha7Zi1N27dzF37lw4OjpCKBSiX79+mDdvHmpqarTiMjMzIZfLYWJiArFYjE8//RR//PFHl/SZEEIIIS3r0Einra3tn/ZcAGMMM2fOxN69e/H777+jsLBQp5V6yNMxdGkG9Ax6trr/xhcPV2Das2cPQkJCEB8fD3d3d8TExMDb2xtFRUWwtLRsdtzp06cxefJkREVF4Y033sCuXbswceJEnDt3jludqK6uDqNGjYKfnx9mzJjRrI7bt2/j9u3b+PLLLzF48GDcvHkTH330EW7fvo29e/cCAP773//i9ddfx2effYakpCTcunULH330EdRqNb788suuuESEEEIIaUlHZpQ/dOgQ8/b25lbqeZrS0tKYvr4+y8nJYWVlZayxsbHTdQYEBLAJEyZ0vnFd4MGDBywgIIANHTqU8fn8drfr5MmTTC6Xs969ezNDQ0Pm6OjIvvrqK62Y7Oxs9sYbbzCJRMIAsO+++67d7Wxa0cB2QQqz+/RQq1sTNzc3NmfOHO6zWq1mVlZWLCoqqsX6/fz8mK+vr1aZu7s7mzlzZrPYphWWCgsLn9julJQUJhAIuL+bxYsXM1dXV62YgwcPMkNDQ1ZbW/vE+gghhJDnSXdakahDt9enTp2KEydOYMCAATAxMUHv3r21tq5UUlICiUQCuVwOsViMHj06NDj7VKjV6marMXWkDqFQiHnz5sHLy6vdxxsZGSE4OBg//PADfvrpJ4SHhyM8PBwJCQlcTF1dHZydnREXF9eptupKpVKhoKBAqz96enrw8vJCbm5ui8fk5uY267+3t3er8bpqeluv6e9GqVRqrdkOAEKhEA0NDSgoKOjUuQghhBDSug5lcDExMV3cjJYFBgZi+/btAAAejwc7Oztcv34dq1evRkJCAsrLy+Hg4IAlS5bgnXfeAfAwiQsKCsLx48dRXl6Ofv36Yfbs2Zg/fz4AIDIyUqtOADhx4gQAYOzYsfj9999hZmYGADh//jxcXFxQWloKqVSKbdu2YcGCBUhKSsKiRYtw9epVFBcXQyKR4LPPPsPu3btRXV2NoUOHYvXq1RgzZswT+2hkZISNGzcCAHJyclBdXd2ua+Ti4gIXFxfus1Qqxb59+3Dy5EkEBQUBeDhB/fjx49tVb2fcuXMHarW62Tyiffv2xc8//9ziMeXl5S3Gl5eXd6odn3/+OXcdgIeJbExMDHbv3g0/Pz+Ul5dj+fLlAICysrIOn4sQQgghbevw2+t/htjYWAwYMAAJCQnIz88Hn89HVFQUvvnmG8THx8Pe3h4//PADpk6dCgsLC3h6ekKj0cDGxgapqakwNzfH6dOnERQUBIlEAj8/P4SGhuKnn35CbW0tEhMTATycpP306dM6tam+vh6rV6/G5s2bYW5uDktLSwQHB+PKlStITk6GlZUVvvvuO/j4+ODixYuwt7d/mpeomcLCQpw+fRorVqzoVD1KpRJKpZL7XFtb29mm/alqa2vh6+uLwYMHIzIykit/7bXXsHbtWnz00Uf44IMPYGBggCVLluDkyZPQ0+vQwD8hhBBCdNDpe9UNDQ1QqVRaZV31kpFIJIKJiQn4fD7EYjGUSiVWrVqFY8eOwcPDAwDQv39/nDp1Cps2bYKnpyf09fWxbNkyrg6ZTIbc3FykpKTAz88PxsbGEAqFUCqVEIvF7W5TY2MjNmzYAGdnZwCAQqFAYmIiFAoFrKysAAChoaFIT09HYmIiVq1a1QVX4slsbGxQVVWFP/74A5GRkfjwww87VV9UVJTWdWyPPn36gM/no6KiQqu8oqKi1WsuFovbFd+We/fuwcfHByYmJvjuu++azSMbEhKChQsXoqysDL169cKNGzewePFiWmWLEEIIeYo6NLRTV1eH4OBgWFpawsjICL169dLanpbi4mLU19dj3LhxMDY25rakpCSUlJRwcXFxcRgxYgQsLCxgbGyMhIQEKBSKLmmDQCDAsGHDuM8XL16EWq2Gg4ODVpuys7O12vS0nTx5EmfPnkV8fDx3+7gzFi9ejJqaGm775ZdfdD5WIBBgxIgRyMzM5Mo0Gg0yMzO5fyw8zsPDQyseAI4ePdpqfGtqa2vx2muvQSAQ4ODBg82e32zC4/FgZWUFoVCI3bt3w9bWFi+++GK7zkUIIYQQ3XVopPOTTz7BiRMnsHHjRnzwwQeIi4vDrVu3sGnTJnzxxRdd3UbO/fv3ATycJ9Ta2lprn4GBAQAgOTkZoaGhiI6OhoeHB0xMTLB27Vrk5eW1WXfTrVX2yKqgLa15LhQKtSbDv3//Pvh8PgoKCsDn87VijY2N29G7zpHJZAAAJycnVFRUIDIyEpMnT+5wfQYGBtw17YiQkBAEBATA1dUVbm5uiImJQV1dHaZNmwYA8Pf3h7W1NaKiogAA8+fPh6enJ6Kjo+Hr64vk5GScPXtW64Wou3fvQqFQ4Pbt2wCAoqIiAA9HScViMZdw1tfX45tvvkFtbS33WICFhQX3+1m7di18fHygp6eHffv24YsvvkBKSkqz3x8hhBBCuk6Hks7//Oc/SEpKwpgxYzBt2jSMHj0aL7zwAuzs7LBz505MmTKlq9sJABg8eDAMDAygUCjg6enZYkxOTg7kcjlmz57NlT0+4igQCKBWq7XKLCwsAIC75Qqg2eTjLXFxcYFarUZlZSVGjx7dnu48NRqNRut5zGfh3XffRVVVFSIiIlBeXo7hw4cjPT2de1lIoVBoPUMpl8uxa9cuhIeHIywsDPb29ti/fz83RycAHDx4kEtaAeC9994DACxduhSRkZE4d+4c94+LF154Qas9TS+DAcCRI0ewcuVKKJVKODs748CBA3/qi1aEEELI/6IOJZ13797lnn8zNTXF3bt3AQCjRo3CrFmzuq51jzExMUFoaCgWLlwIjUaDUaNGoaamBjk5OTA1NUVAQADs7e2RlJSEjIwMyGQy7NixA/n5+dxIIPDwDe+MjAwUFRXB3NwcIpEIL7zwAmxtbREZGYmVK1fi6tWriI6OfmKbHBwcMGXKFPj7+yM6OhouLi6oqqpCZmYmhg0bBl9f3yfWceXKFahUKty9exf37t3jkl1dJsGPi4tDv379MHDgQADADz/8gC+//BLz5s3jYu7fv4/i4mLuc2lpKc6fP4/evXujX79+TzzHoy4t89b5md3g4GAEBwe3uC8rK6tZ2aRJkzBp0qRW6wsMDERgYGCr+8eMGaM1Ut2a48ePPzGGEEIIIV2sI5N7Ojk5saysLMYYY6+++ir75z//yRhjLDY2lllbW3fVHKKMMcbWrVvH7OzsuM8ajYbFxMQwR0dHpq+vzywsLJi3tzfLzs5mjDHW0NDAAgMDmUgkYmZmZmzWrFls0aJFzNnZmaujsrKSjRs3jhkbGzMA7MSJE4wxxk6dOsWcnJyYoaEhGz16NEtNTWUAuEnwExMTmUgkatZGlUrFIiIimFQqZfr6+kwikbC3336bXbhwQac+2tnZMQDNNl2sX7+eDRkyhPXs2ZOZmpoyFxcXtmHDBqZWq7mYEydOtFh/QECATudgrHtNLksIIYQQ3XSn728eYzoMDT1m3bp14PP5mDdvHo4dO4Y333wTjDE0Njbiq6++4ubEJH8dtbW1EIlE3GTrhBBCCOn+utP3d4eSzsfdvHkTBQUFeOGFF7Te7CZ/Hd3pj5YQQgghuulO398dnqczMzMTmZmZqKysbLYU5NatWzvdsL+K8ePH4+TJky3uCwsLQ1hYWJvHDxkyBDdv3mxx36ZNm57aS1uEEEIIIV2pQ0nnsmXLsHz5cri6ukIikWhNIUS0bd68GQ8ePGhxny7r1KelpbU4dROAZstGEkIIIYR0Vx1KOuPj47Ft2zZ88MEHXd2ev5zH5xNtLzs7uy5qCSGEEELIs9OhFYlUKhXkcnlXt4UQQgghhPxFdSjp/PDDD7Fr166ubgshhBBCCPmL6lDS2dDQgK+++gqenp6YO3cuQkJCtLauxBhDUFAQevfuDR6Pp9MqQaR7iIuLg1QqhaGhIdzd3XHmzJk241NTU/9fe/ceF1W1/4//NYwMM3EZlIsDSNzBS9zygpCEFYqmpZ6TkGmAmZqKN850UvOCmU5mHKEjiqTiLUPB60cJUhQTIUWEIwphkDaV3EwBARloZv3+8Ov+OTLoIPd6Px+PeTyavdda814Lct6svfda6N+/P4RCIVxdXZGcnKx2/tChQxg9ejRMTExa/F2YPXs2HBwcIBKJYGZmhgkTJuDHH3/kzv/vf//DlClTYG1tDZFIhAEDBiA6Orpd+ksIIYSQlj1T0nnlyhV4eHhAR0cHV69eRW5uLvdq76QwJSUFO3fuxPHjx1FaWqq2LeKzCg0NxcSJE9seXDtoaGhAaGgoXF1d0atXr1bHdejQIYwaNQpmZmYwMjKCt7c3UlNT1crcu3cPixYtgo2NDUQiEXx8fJCdnf1M8b6wKhW2S060+Hpo//79CA8Px6pVq3D58mW4u7sjICAAFRUVGtvNzMzElClTMGPGDOTm5mLixImYOHEirl69ypWpq6vDiBEjsH79+hbjGzx4MOLj41FYWIjU1FQwxjB69Ghu29OcnByYm5tj7969uHbtGj7++GMsXboUmzZteqbxIIQQQoh22mWdzo60adMmbNiwocVlg55FaGgoqqqqcOTIkTa1o1QqwePx1PYQb626ujpIpVK8+OKLOHjwIIRCYaviWrRoESwtLfHKK6/A2NgY8fHx+OKLL3DhwgV4enoCeLAP+tWrV7FlyxZYWlpi79692LhxIwoKCrR+0OnhOl/Wiw5AR++5Fsvd/OzBtp9eXl4YOnQol8ypVCpYW1tj/vz5WLJkSbN6QUFBqKurw/Hjx7ljw4cPh4eHB2JjY9U/4+ZN2NnZITc396lbhV65cgXu7u4oLi6Gg4ODxjLz5s1DYWEhbY9JCCHkL6c7rdP57NlSJwgNDcX8+fMhl8vB4/Fga2sLlUoFmUwGOzs7iEQiuLu7IykpiaujVCoxY8YM7ryLi4va5dOIiAjs2rULR48eBY/HA4/HQ3p6OtLT08Hj8VBVVcWVzcvLA4/Hw82bNwEAO3fuhLGxMY4dO4aBAwdCT08PcrkcCoUCUqkUVlZW0NfXh5eXl8a9xTXR19fHli1bMHPmTEgkklaPUVRUFP79739j6NChcHJywrp16+Dk5IT/+7//AwDcv38fBw8exOeff46XX34Zjo6OiIiIgKOjI7Zs2dLqz9NGY2MjcnJy4O/vzx3T0dGBv78/srKyNNbJyspSKw8AAQEBLZbXRl1dHeLj42FnZwdra+sWy1VXV2u1fBUhhBBCnt0zLw7fGaKjo+Hg4IC4uDhkZ2eDz+dDJpNh7969iI2NhZOTE77//ntMmzYNZmZm8PPzg0qlQr9+/ZCYmAgTExNkZmZi1qxZsLCwQGBgIKRSKQoLC1FTU4P4+HgAD9bLzMzM1Cqm+vp6rF+/Htu2bYOJiQnMzc0RFhaGgoICJCQkwNLSEocPH8aYMWOQn58PJyenjhyiZlQqFe7du8clUX/++SeUSiWEQqFaOZFIhIyMjBbbUSgUUCgU3PuamhqtY7h9+zaUSmWzdUT79u2rdn/lo8rKyjSWLysr0/pzH9q8eTP+/e9/o66uDi4uLjh58iQEAoHGspmZmdi/fz9OnDih8TwhhBBC2ke3TjrFYjEMDQ3B5/MhkUigUCiwbt06nDp1Ct7e3gAAe3t7ZGRkYOvWrfDz84Ouri5Wr17NtWFnZ4esrCwcOHAAgYGBMDAwgEgkgkKheKaZxaamJmzevBnu7u4AALlcjvj4eMjlclhaWgIApFIpUlJSEB8fj3Xr1rXDSGjviy++QG1tLQIDAwEAhoaG8Pb2xpo1azBgwAD07dsX33zzDbKysuDo6NhiOzKZTG0ce5KpU6di1KhRKC0txRdffIHAwECcP3++WeJ99epVTJgwAatWrcLo0aO7KFpCCCHk76FbJ52PKy4uRn19PUaNGqV2vLGxkbt/EXjw1PSOHTsgl8tx//59NDY2PvXeP20JBAK1/eXz8/OhVCrh7OysVk6hUMDExKRdPlNb+/btw+rVq3H06FGYm5tzx/fs2YP33nsPVlZW4PP5ePHFFzFlyhTk5OS02NbSpUvVViKoqal54iXqR5mamoLP56O8vFzteHl5eYuJvkQiaVX5JxGLxRCLxXBycsLw4cPRu3dvHD58GFOmTOHKFBQU4LXXXsOsWbOwfPnyVn8GIYQQQlqnRyWdtbW1AIATJ040ewBGT08PAJCQkACpVIrIyEh4e3vD0NAQGzZswIULF57Y9sOHgR59rkrT9pMikUht28/a2lrw+Xzk5OSAz+erlTUwMGhF79omISEB77//PhITE5vdG+ng4ICzZ8+irq4ONTU1sLCwQFBQEOzt7VtsT09PjxvT1hIIBBg8eDDS0tK4p/FVKhXS0tIQFhamsY63tzfS0tKwaNEi7tjJkye5Ge1nxRgDY0ztVoFr167h1VdfRUhICNauXdum9gkhhBCinR6VdD768I6fn5/GMufPn4ePjw/mzp3LHSspKVErIxAIuCV0HjIzMwMAlJaWonfv3gCg1fJPnp6eUCqVqKiogK+vb2u6026++eYbvPfee0hISMC4ceNaLKevrw99fX3cvXsXqamp+PzzzzsspvDwcISEhGDIkCEYNmwYoqKiUFdXh+nTpwMAgoODYWVlBZlMBgBYuHAh/Pz8EBkZiXHjxiEhIQGXLl1CXFwc1+adO3cgl8tx69YtAEBRURGAB7OkEokEP//8M/bv34/Ro0fDzMwMv/32Gz777DOIRCK8/vrrAB5cUn/11VcREBCA8PBw7p5RPp/P/Q4QQgghpP31qKTT0NAQUqkUixcvhkqlwogRI1BdXY3z58/DyMgIISEhcHJywu7du5Gamgo7Ozvs2bMH2dnZsLOz49qxtbVFamoqioqKYGJiArFYDEdHR1hbWyMiIgJr167F9evXERkZ+dSYnJ2dMXXqVAQHByMyMhKenp6orKxEWloa3NzcnpgEPlRQUIDGxkbcuXMH9+7d45JdbW4J2LdvH0JCQhAdHQ0vLy8uiRKJRBCLxQDArVfp4uKC4uJifPjhh+jfvz+XAHaEoKAgVFZWYuXKlSgrK4OHhwdSUlK4h4XkcrnaUlM+Pj7Yt28fli9fjmXLlsHJyQlHjhxRW5f12LFjajG//fbbAIBVq1YhIiICQqEQ586dQ1RUFO7evYu+ffvi5ZdfRmZmJne7QVJSEiorK7F3717s3buXa8vGxoZbpYAQQgghHYB1cxs3bmQ2Njbce5VKxaKiopiLiwvT1dVlZmZmLCAggJ09e5YxxlhDQwMLDQ1lYrGYGRsbszlz5rAlS5Ywd3d3ro2Kigo2atQoZmBgwACwM2fOMMYYy8jIYK6urkwoFDJfX1+WmJjIALAbN24wxhiLj49nYrG4WYyNjY1s5cqVzNbWlunq6jILCws2adIkduXKFa36aGNjwwA0e2nDz89PY92QkBCuzP79+5m9vT0TCARMIpGwefPmsaqqKq3af6i6upoBYNXV1a2qRwghhJCu052+v7v94vCke+hOi8sSQgghRDvd6fu7Wy8OTwghhBBC/hoo6exgY8eOhYGBgcaXNmt4Dho0qMX6X3/9dSf0gBBCCCGk7XrUg0Q90bZt23D//n2N57TZejE5OVnj0k0Amu3gQwghhBDSXVHS2cEeX0+0tWxsbNopEkIIIYSQrkOX1wkhhBBCSIejpJMQQgghhHQ4SjpJh4qJiYGtrS2EQiG8vLxw8eLFJ5ZPTExE//79IRQK4erqiuTkZLXzhw4dwujRo2FiYgIej/fEXaMYYxg7dix4PB6OHDnCHf/f//6HKVOmwNraGiKRCAMGDEB0dHRbukkIIYSQp+j293QyxjB79mwkJSXh7t27yM3N1WqnHtIxXliVCh29555Y5uZnD3Zh2r9/P8LDwxEbGwsvLy9ERUUhICAARUVF3A5Bj8rMzMSUKVMgk8kwfvx47Nu3DxMnTsTly5e5nYnq6uowYsQIBAYGYubMmU+MIyoqCjwer9nxnJwcmJubY+/evbC2tkZmZiZmzZoFPp/f4t7whBBCCGmbbj/TmZKSgp07d+L48eMoLS1V2xbxWYWGhmLixIltD64dNDQ0IDQ0FK6urujVq1er4yotLcU777wDZ2dn6OjoYNGiRc3KHDp0CEOGDIGxsTH09fXh4eGBPXv2tE8HnuA///kPZs6cienTp2PgwIGIjY3Fc889hx07dmgsHx0djTFjxuDDDz/EgAEDsGbNGrz44ovYtGkTV+bdd9/FypUr4e/v/8TPzsvLQ2RkpMbPeu+99xAdHQ0/Pz/Y29tj2rRpmD59Og4dOtS2DhNCCCGkRd0+6SwpKYGFhQV8fHwgkUjQq1f3mZxVKpVQqVRtbkMkEmHBggVPTaQ0USgUMDMzw/Lly+Hu7q6xTJ8+ffDxxx8jKysLV65cwfTp0zF9+nSkpqa2KfYnaWxsRE5OjlqfdHR04O/vj6ysLI11srKymo1BQEBAi+VbUl9fj3feeQcxMTGQSCRa1amurtZqCStCCCGEPJtunXSGhoZi/vz5kMvl4PF4sLW1hUqlgkwmg52dHUQiEdzd3ZGUlMTVUSqVmDFjBnfexcVF7X69iIgI7Nq1C0ePHgWPxwOPx0N6ejrS09PB4/FQVVXFlc3LywOPx8PNmzcBADt37oSxsTGOHTuGgQMHQk9PD3K5HAqFAlKpFFZWVtDX14eXlxfS09O16qO+vj62bNmCmTNnap0gPcrW1hbR0dEIDg6GWCzWWGbkyJGYNGkSBgwYAAcHByxcuBBubm7IyMho9edp6/bt21Aqlc3WEu3bty/Kyso01ikrK2tV+ZYsXrwYPj4+mDBhglblMzMzsX//fsyaNatVn0MIIYQQ7XWfaUMNoqOj4eDggLi4OGRnZ4PP50Mmk2Hv3r2IjY2Fk5MTvv/+e0ybNg1mZmbw8/ODSqVCv379kJiYCBMTE+5+PQsLCwQGBkIqlaKwsBA1NTWIj48H8GAmMDMzU6uY6uvrsX79emzbtg0mJiYwNzdHWFgYCgoKkJCQAEtLSxw+fBhjxoxBfn4+nJycOnKIWo0xhtOnT6OoqAjr169vsZxCoYBCoeDe19TUdEZ4bXbs2DGcPn0aubm5WpW/evUqJkyYgFWrVmH06NEdHB0hhBDy99Wtk06xWAxDQ0Pw+XxIJBIoFAqsW7cOp06dgre3NwDA3t4eGRkZ2Lp1K/z8/KCrq4vVq1dzbdjZ2SErKwsHDhxAYGAgDAwMIBKJoFAonmlmsampCZs3b+YuZcvlcsTHx0Mul8PS0hIAIJVKkZKSgvj4eK22uuwM1dXVsLKygkKhAJ/Px+bNmzFq1KgWy8tkMrVxbC1TU1Pw+XyUl5erHS8vL29x3CUSSavKa3L69GmUlJTA2NhY7fg///lP+Pr6qs1AFxQU4LXXXsOsWbOwfPlyrT+DEEIIIa3XrZPOxxUXF6O+vr5ZstTY2AhPT0/ufUxMDHbs2AG5XI779++jsbGx3Z54FwgEcHNz497n5+dDqVTC2dlZrZxCoYCJiUm7fGZ7MDQ0RF5eHmpra5GWlobw8HDY29tj5MiRGssvXboU4eHh3PuamhpYW1tr/XkCgQCDBw9GWloa93CUSqVCWlpai0+Ie3t7Iy0tTe1hqJMnT3J/YGhjyZIleP/999WOubq6YuPGjXjjjTe4Y9euXcOrr76KkJAQrF27Vuv2CSGEEPJselTSWVtbCwA4ceJEs+0l9fT0AAAJCQmQSqWIjIyEt7c3DA0NsWHDBly4cOGJbevoPLi9lTHGHdO057lIJFJbhqe2thZ8Ph85OTng8/lqZQ0MDFrRu46lo6MDR0dHAICHhwcKCwshk8laTDr19PS4MX1W4eHhCAkJwZAhQzBs2DBERUWhrq4O06dPBwAEBwfDysoKMpkMALBw4UL4+fkhMjIS48aNQ0JCAi5duoS4uDiuzTt37kAul+PWrVsAgKKiIgAPZkkffT3u+eefh52dHYAHl9RfffVVBAQEIDw8nLtnlM/nw8zMrE19JoQQQohmPSrpfPThHT8/P41lzp8/Dx8fH8ydO5c7VlJSolZGIBBAqVSqHXuYbJSWlqJ3794A8MSFxx/y9PSEUqlERUUFfH19W9OdLqVSqdTu2ewIQUFBqKysxMqVK1FWVgYPDw+kpKRwDwvJ5XIu2QcAHx8f7Nu3D8uXL8eyZcvg5OSEI0eOqC2TdezYMS5pBYC3334bALBq1SpERERoFVdSUhIqKyuxd+9e7N27lztuY2PDPTRGCCGEkPbVo5JOQ0NDSKVSLF68GCqVCiNGjEB1dTXOnz8PIyMjhISEwMnJCbt370Zqairs7OywZ88eZGdnc7NcwIMnvlNTU1FUVAQTExOIxWI4OjrC2toaERERWLt2La5fv47IyMinxuTs7IypU6ciODgYkZGR8PT0RGVlJdLS0uDm5oZx48Y9tY2CggI0Njbizp07uHfvHpfsantLwMPytbW1qKysRF5eHgQCAQYOHAjgwf2ZQ4YMgYODAxQKBZKTk7Fnzx5s2bJFq/YfdXV1AIyMjLQuHxYW1uLldE1P+E+ePBmTJ09usb3Q0FCEhoZq/fmA+uw18GAFA20TVEIIIYS0E9bNbdy4kdnY2HDvVSoVi4qKYi4uLkxXV5eZmZmxgIAAdvbsWcYYYw0NDSw0NJSJxWJmbGzM5syZw5YsWcLc3d25NioqKtioUaOYgYEBA8DOnDnDGGMsIyODubq6MqFQyHx9fVliYiIDwG7cuMEYYyw+Pp6JxeJmMTY2NrKVK1cyW1tbpquryywsLNikSZPYlStXtOqjjY0NA9DspS1NdR8ds48//pg5OjoyoVDIevfuzby9vVlCQoLW7TPGWHV1NQPAqqurW1WPEEIIIV2nO31/8xh7bBqIEA1qamogFotRXV3dqplOQgghhHSd7vT93a0XhyeEEEIIIX8NlHR2sLFjx8LAwEDjS5s1PAcNGtRi/a+//roTekAIIYQQ0nY96kGinmjbtm24f/++xnPa7PWdnJyscekmAM22jCSEEEII6a4o6exgj68n2lo2NjbtFAkhhBBCSNehy+uEEEIIIaTDUdJJOk1MTAxsbW0hFArh5eWFixcvPrF8YmIi+vfvD6FQCFdXVyQnJ6udP3ToEEaPHg0TExPweDyNi/mXlZXh3XffhUQigb6+Pl588UUcPHhQ4+cpFAp4eHi02BYhhBBCnl23TzoZY5g1axb69OlDyUAPtn//foSHh2PVqlW4fPky3N3dERAQgIqKCo3lMzMzMWXKFMyYMQO5ubmYOHEiJk6ciKtXr3Jl6urqMGLECKxfv77Fzw0ODkZRURGOHTuG/Px8/OMf/0BgYCByc3Oblf33v/8NS0vLtneWEEIIIc10+3U6v/32W0yYMAHp6emwt7eHqakpevVq262ooaGhqKqqwpEjR9onyDZoaGjABx98gJycHBQWFmL8+PGtiisjIwMfffQRfvzxR9TX18PGxgazZ8/G4sWLuTIRERFYvXq1Wj0XFxf8+OOPWn/Ow3W+rBcdgI7ec1rXu/nZgx2ZvLy8MHToUGzatAnAg204ra2tMX/+fCxZsqRZvaCgINTV1eH48ePcseHDh8PDwwOxsbHqn3HzJuzs7JCbm9tsFycDAwNs2bIF7777LnfMxMQE69evx/vvv88d+/bbbxEeHo6DBw9i0KBBGtsihBBCeprutE5nt3+QqKSkBBYWFvDx8enqUJpRKpXg8Xhq+4c/SxsikQgLFixo8bLvk+jr6yMsLAxubm7Q19dHRkYGZs+eDX19fcyaNYsrN2jQIJw6dYp739bEvTUaGxuRk5ODpUuXcsd0dHTg7++PrKwsjXWysrIQHh6udiwgIKDVfyj4+Phg//79GDduHIyNjXHgwAE0NDRg5MiRXJny8nLMnDkTR44cwXPPaZ9QE0IIIUR73fryemhoKObPnw+5XA4ejwdbW1uoVCrIZDLY2dlBJBLB3d0dSUlJXB2lUokZM2Zw511cXBAdHc2dj4iIwK5du3D06FHweDzweDykp6cjPT0dPB4PVVVVXNm8vDzweDzcvHkTALBz504YGxvj2LFjGDhwIPT09CCXy6FQKCCVSmFlZQV9fX14eXlp3FdcE319fWzZsgUzZ86ERCJp9Rh5enpiypQpGDRoEGxtbTFt2jQEBATg3LlzauV69eoFiUTCvUxNTVv9Wc/q9u3bUCqVzZZ46tu3L8rKyjTWKSsra1X5lhw4cABNTU0wMTGBnp4eZs+ejcOHD8PR0RHAg9s3QkND8cEHH2DIkCGtapsQQggh2uvWM53R0dFwcHBAXFwcsrOzwefzIZPJsHfvXsTGxsLJyQnff/89pk2bBjMzM/j5+UGlUqFfv35ITEyEiYkJMjMzMWvWLFhYWCAwMBBSqRSFhYWoqalBfHw8gAfrZWZmZmoVU319PdavX49t27bBxMQE5ubmCAsLQ0FBARISEmBpaYnDhw9jzJgxyM/Ph5OTU0cOUTO5ubnIzMzEp59+qnb8p59+gqWlJYRCIby9vSGTyfD888+32I5CoYBCoeDe19TUdFjMHWnFihWoqqrCqVOnYGpqiiNHjiAwMBDnzp2Dq6sr/vvf/+LevXtqs7CEEEIIaX/dOukUi8UwNDQEn8+HRCKBQqHAunXrcOrUKXh7ewMA7O3tkZGRga1bt8LPzw+6urpq9y/a2dkhKysLBw4cQGBgIAwMDCASiaBQKJ5pZrGpqQmbN2+Gu7s7AEAulyM+Ph5yuZx7CEUqlSIlJQXx8fFa7TrUHvr164fKykr8+eefiIiIULtf0cvLCzt37oSLiwtKS0uxevVq+Pr64urVqzA0NNTYnkwma3Yf6LMyNTUFn89HeXm52vHy8vIWfwYSiaRV5TUpKSnBpk2bcPXqVQwaNAgA4O7ujnPnziEmJgaxsbE4ffo0srKyoKenp1Z3yJAhmDp1Knbt2qX15xFCCCGkZd066XxccXEx6uvrMWrUKLXjjY2N8PT05N7HxMRgx44dkMvluH//PhobG9vtoRCBQAA3NzfufX5+PpRKJZydndXKKRQKmJiYtMtnauPcuXOora3FDz/8gCVLlsDR0RFTpkwB8GArzofc3Nzg5eUFGxsbHDhwADNmzNDY3tKlS9XuqaypqYG1tfUzxSYQCDB48GCkpaVh4sSJAB48SJSWloawsDCNdby9vZGWloZFixZxx06ePMn9saGN+vp6AGh2zy2fz4dKpQIAfPnll2qzwrdu3UJAQAD2798PLy8vrT+LEEIIIU/Wo5LO2tpaAMCJEyea7fTzcKYqISEBUqkUkZGR8Pb2hqGhITZs2IALFy48se2HicmjD/Nr2n5SJBKBx+OpxcTn85GTkwM+n69W1sDAoBW9axs7OzsAgKurK8rLyxEREcElnY8zNjaGs7MziouLW2xPT0+v2exfW4SHhyMkJARDhgzBsGHDEBUVhbq6OkyfPh3Ag6WNrKysIJPJAAALFy6En58fIiMjMW7cOCQkJODSpUuIi4vj2rxz5w7kcjlu3boFACgqKgIA7r7V/v37w9HREbNnz8YXX3wBExMTHDlyBCdPnuSein/8FoOHPzMHBwf069ev3fpPCCGE/N31qKTz0Yd3/Pz8NJY5f/48fHx8MHfuXO5YSUmJWhmBQAClUql2zMzMDABQWlqK3r17A4BWa4J6enpCqVSioqICvr6+relOh1GpVGr3Yz6utrYWJSUlassIdbSgoCBUVlZi5cqVKCsrg4eHB1JSUriHheRyudqMpI+PD/bt24fly5dj2bJlcHJywpEjR/DCCy9wZY4dO8YlrQDw9ttvAwBWrVqFiIgI6OrqIjk5GUuWLMEbb7yB2tpaODo6YteuXXj99dc7qeeEEEIIAXpY0mloaAipVIrFixdDpVJhxIgRqK6uxvnz52FkZISQkBA4OTlh9+7dSE1NhZ2dHfbs2YPs7GxuJhAAbG1tkZqaiqKiIpiYmEAsFsPR0RHW1taIiIjA2rVrcf36dURGRj41JmdnZ0ydOhXBwcGIjIyEp6cnKisrkZaWBjc3N4wbN+6pbRQUFKCxsRF37tzBvXv3uGRXm1sCYmJi8Pzzz6N///4AgO+//x5ffPEFFixYwJWRSqV44403YGNjg1u3bmHVqlXg8/ktzoR2lLCwsBYvp2t62n/y5MmYPHlyi+2FhoYiNDT0iZ/p5OTUqqWobG1t0c2XriWEEEJ6JtbNbdy4kdnY2HDvVSoVi4qKYi4uLkxXV5eZmZmxgIAAdvbsWcYYYw0NDSw0NJSJxWJmbGzM5syZw5YsWcLc3d25NioqKtioUaOYgYEBA8DOnDnDGGMsIyODubq6MqFQyHx9fVliYiIDwG7cuMEYYyw+Pp6JxeJmMTY2NrKVK1cyW1tbpquryywsLNikSZPYlStXtOqjjY0NA9DspY0vv/ySDRo0iD333HPMyMiIeXp6ss2bNzOlUsmVCQoKYhYWFkwgEDArKysWFBTEiouLtWr/oerqagaAVVdXt6oeIYQQQrpOd/r+7vY7EpHuoTvtaEAIIYQQ7XSn7+9uvTg8IYQQQgj5a6Cks4ONHTsWBgYGGl/arOE5aNCgFut//fXXndADQgghhJC261EPEvVE27Ztw/379zWe69Onz1PrJycna1y6CUCzbSIJIYQQQrorSjo72OPribaWjY1NO0VCCCGEENJ16PI6IYQQQgjpcJR0EkIIIYSQDkdJJ+lUMTExsLW1hVAohJeXFy5evPjE8omJiejfvz+EQiFcXV2RnJzMnWtqasJHH30EV1dX6Ovrw9LSEsHBwdy2mA/duXMHU6dOhZGREYyNjTFjxgxuS1XgwfaZr7zyCvr27QuhUAh7e3ssX768xXtpCSGEENJ6XXpPJ2MMs2fPRlJSEu7evYvc3FytduEhXeeFVanQ0XuuVXVufvZgV6b9+/cjPDwcsbGx8PLyQlRUFAICAlBUVARzc/Nm9TIzMzFlyhTIZDKMHz8e+/btw8SJE3H58mW88MILqK+vx+XLl7FixQq4u7vj7t27WLhwId58801cunSJa2fq1KkoLS3FyZMn0dTUhOnTp2PWrFnYt28fAEBXVxfBwcF48cUXYWxsjP/973+YOXMmVCqVVisMEEIIIeTpunRx+G+//RYTJkxAeno67O3tYWpqil692pYHh4aGoqqqCkeOHGmfINugoaEBH3zwAXJyclBYWIjx48e3Kq709HS88sorzY6XlpZCIpEAAO7du4cVK1bg8OHDqKiogKenJ6KjozF06FCuPGMMq1atwldffYWqqiq89NJL2LJlC5ycnLSO5eHistaLDjxz0unl5YWhQ4di06ZNAB7sEW9tbY358+djyZIlzeoFBQWhrq4Ox48f544NHz4cHh4eiI2N1fhZ2dnZGDZsGH755Rc8//zzKCwsxMCBA5GdnY0hQ4YAAFJSUvD666/jt99+g6WlpcZ2wsPDkZ2djXPnzrWqr4QQQkh3QovD/z8lJSWwsLCAj48PJBJJmxPO9qRUKqFSqdrchkgkwoIFC+Dv7//M7RQVFaG0tJR7PTor+P777+PkyZPYs2cP8vPzMXr0aPj7++P333/nynz++ef48ssvERsbiwsXLkBfXx8BAQFoaGhoU/9ao7GxETk5OWrjoKOjA39/f2RlZWmsk5WV1WzcAgICWiwPANXV1eDxeDA2NubaMDY25hJOAPD394eOjg4uXLigsY3i4mKkpKTAz89P2+4RQggh5Cm6LOkMDQ3F/PnzIZfLwePxYGtrC5VKBZlMBjs7O4hEIri7uyMpKYmro1QqMWPGDO68i4sLoqOjufMRERHYtWsXjh49Ch6PBx6Ph/T0dKSnp4PH46Gqqoorm5eXBx6Ph5s3bwIAdu7cCWNjYxw7dgwDBw6Enp4e5HI5FAoFpFIprKysoK+vDy8vL6Snp2vVR319fWzZsgUzZ87kZiafhbm5OSQSCffS0XnwY7t//z4OHjyIzz//HC+//DIcHR0REREBR0dHbNmyBcCDWc6oqCgsX74cEyZMgJubG3bv3o1bt2516mzw7du3oVQqm60t2rdvX5SVlWmsU1ZW1qryDQ0N+OijjzBlyhTur7mysrJml+579eqFPn36NGvHx8cHQqEQTk5O8PX1xSeffNKqPhJCCCGkZV02tRgdHQ0HBwfExcUhOzsbfD4fMpkMe/fuRWxsLJycnPD9999j2rRpMDMzg5+fH1QqFfr164fExESYmJggMzMTs2bNgoWFBQIDAyGVSlFYWIiamhrEx8cDeLAAe2ZmplYx1dfXY/369di2bRtMTExgbm6OsLAwFBQUICEhAZaWljh8+DDGjBmD/Pz8Vl2ebgsPDw8oFAq88MILiIiIwEsvvQQA+PPPP6FUKiEUCtXKi0QiZGRkAABu3LiBsrIytRlDsVgMLy8vZGVl4e2339b4mQqFAgqFgntfU1PT3t1qV01NTQgMDARjjEu4W2v//v24d+8e/ve//+HDDz/EF198gX//+9/tHCkhhBDy99RlSadYLIahoSH4fD4kEgkUCgXWrVuHU6dOwdvbGwBgb2+PjIwMbN26FX5+ftDV1cXq1au5Nuzs7JCVlYUDBw4gMDAQBgYGEIlEUCgUzzSz2NTUhM2bN8Pd3R0AIJfLER8fD7lczt37J5VKkZKSgvj4+A5/yMTCwgKxsbEYMmQIFAoFtm3bhpEjR+LChQt48cUXYWhoCG9vb6xZswYDBgxA37598c033yArKwuOjo4AwM3mtWbGEABkMpnaWLeVqakp+Hw+ysvL1Y6Xl5e3+LOSSCRalX+YcP7yyy84ffq02j0rEokEFRUVauX//PNP3Llzp1k71tbWAICBAwdCqVRi1qxZ+Ne//gU+n9+6zhJCCCGkmW6zZFJxcTHq6+sxatQotf3Fd+/ejZKSEq5cTEwMBg8eDDMzMxgYGCAuLg5yubxdYhAIBHBzc+Pe5+fnQ6lUwtnZWS2ms2fPqsXUUVxcXDB79mwMHjwYPj4+2LFjB3x8fLBx40auzJ49e8AYg5WVFfT09PDll19iypQp3CX4Z7V06VJUV1dzr19//bVN7QkEAgwePBhpaWncMZVKhbS0NO6PjMd5e3urlQeAkydPqpV/mHD+9NNPOHXqFExMTJq1UVVVhZycHO7Y6dOnoVKp4OXl1WK8KpUKTU1Nbb6vlxBCCCEPdJsndx6um3jixIlmW0fq6ekBABISEiCVShEZGQlvb28YGhpiw4YNLT4Q8tDDBOzRB/U1rcEoEonA4/HUYuLz+cjJyWk222VgYNCK3rWfYcOGcZfOAcDBwQFnz55FXV0dampqYGFhgaCgINjb2wMAN5tXXl4OCwsLrl55efkTl6fS09Pjxr29hIeHIyQkBEOGDMGwYcMQFRWFuro6TJ8+HQAQHBwMKysryGQyAMDChQvh5+eHyMhIjBs3DgkJCbh06RLi4uIAPPgZvvXWW7h8+TKOHz8OpVLJzd726dMHAoEAAwYMwJgxYzBz5kzExsaiqakJYWFhePvtt7nZ66+//hq6urpwdXWFnp4eLl26hKVLlyIoKAi6urrtOgaEEELI31W3SToffXinpaeGz58/Dx8fH8ydO5c79viMo0AggFKpVDtmZmYG4MFSQ7179wbw4EGip/H09IRSqURFRQV8fX1b050Ok5eXp5Y8PqSvrw99fX3cvXsXqamp+PzzzwE8uAVBIpEgLS2NSzJrampw4cIFzJkzpzNDR1BQECorK7Fy5UqUlZXBw8MDKSkp3KV/uVyuNkPr4+ODffv2Yfny5Vi2bBmcnJxw5MgRvPDCCwCA33//HceOHQOAZgn0mTNnMHLkSAAPksqwsDC89tpr0NHRwT//+U98+eWXXNlevXph/fr1uH79OhhjsLGxQVhYGBYvXtyBo0EIIYT8vXSbpNPQ0BBSqRSLFy+GSqXCiBEjUF1djfPnz8PIyAghISFwcnLC7t27kZqaCjs7O+zZswfZ2dmws7Pj2rG1tUVqaiqKiopgYmICsVgMR0dHWFtbIyIiAmvXrsX169cRGRn51JicnZ0xdepUBAcHIzIyEp6enqisrERaWhrc3Nwwbty4p7ZRUFCAxsZG3LlzB/fu3eOSXW0WwY+KioKdnR0GDRqEhoYGbNu2DadPn8Z3333HlUlNTQVjDC4uLiguLsaHH36I/v37c7OHPB4PixYtwqeffgonJyfY2dlhxYoVsLS0xMSJE58aw+Ourg5o0zpfYWFhCAsL03hO06oAkydPxuTJkzWWt7W1hTbLzPbp04dbCF6ToKAgBAUFPbUdQgghhLQB60IbN25kNjY23HuVSsWioqKYi4sL09XVZWZmZiwgIICdPXuWMcZYQ0MDCw0NZWKxmBkbG7M5c+awJUuWMHd3d66NiooKNmrUKGZgYMAAsDNnzjDGGMvIyGCurq5MKBQyX19flpiYyACwGzduMMYYi4+PZ2KxuFmMjY2NbOXKlczW1pbp6uoyCwsLNmnSJHblyhWt+mhjY8MANHtpY/369czBwYEJhULWp08fNnLkSHb69Gm1Mvv372f29vZMIBAwiUTC5s2bx6qqqtTKqFQqtmLFCta3b1+mp6fHXnvtNVZUVKRVDA9VV1czAKy6urpV9QghhBDSdbrT93eX7khEeo7utKMBIYQQQrTTnb6/u83T64QQQggh5K+Lks42GDt2rNpSSo++tFnDc9CgQS3W//rrrzuhB4QQQgghnaPbPEjUE23btg3379/XeK5Pnz5PrZ+cnKxx6Sag+WLuhBBCCCE9GSWdbfD4eqKtZWNj006REEIIIYR0b3R5nRBCCCGEdDhKOkm3EBMTA1tbWwiFQnh5eeHixYtPLJ+YmIj+/ftDKBTC1dUVycnJ3LmmpiZ89NFHcHV1hb6+PiwtLREcHIxbt26ptXH9+nVMmDABpqamMDIywogRI3DmzBnu/M6dO8Hj8TS+Ht/PnRBCCCFP1qVLJjHGMHv2bCQlJeHu3bvIzc3VatF00vkeLrlgvegAdPSea5c2b372YHH9/fv3Izg4GLGxsfDy8kJUVBQSExNRVFQEc3PzZvUyMzPx8ssvQyaTYfz48di3bx/Wr1+Py5cv44UXXkB1dTXeeustzJw5E+7u7rh79y4WLlwIpVKJS5cuce04OzvDyckJMpkMIpEIUVFR2LlzJ0pKSiCRSHD//n1UV1erfXZoaCgaGho0LmRPCCGEdDe0ZNL/k5KSgp07d+L48eMoLS3ltjdsi9DQ0GfaaacjNDQ0IDQ0FK6urujVq1eb4jp//jx69erVLCmXyWQYOnQoDA0NYW5ujokTJ6KoqIg7f+fOHcyfPx8uLi4QiUR4/vnnsWDBgmbJVFf6z3/+g5kzZ2L69OkYOHAgYmNj8dxzz2HHjh0ay0dHR2PMmDH48MMPMWDAAKxZswYvvvgiNm3aBAAQi8U4efIkAgMD4eLiguHDh2PTpk3IycmBXC4HANy+fRs//fQTlixZAjc3Nzg5OeGzzz5DfX09rl69CgAQiUSQSCTci8/n4/Tp05gxY0bnDAwhhBDyF9KlSWdJSQksLCzg4+MDiUSCXr26z3NNSqUSKpWqzW2IRCIsWLAA/v7+z9xOVVUVgoOD8dprrzU7d/bsWcybNw8//PADTp48iaamJowePRp1dXUAgFu3buHWrVv44osvcPXqVezcuRMpKSndJnFqbGxETk6O2vjo6OjA398fWVlZGutkZWU1G8+AgIAWywNAdXU1eDwejI2NAQAmJiZwcXHB7t27UVdXhz///BNbt26Fubk5Bg8erLGN3bt347nnnsNbb73Vyl4SQgghpMuSztDQUMyfPx9yuRw8Hg+2trZQqVSQyWSws7ODSCSCu7s7kpKSuDpKpRIzZszgzru4uCA6Opo7HxERgV27duHo0aPcvXfp6elIT08Hj8dDVVUVVzYvLw88Hg83b94E8OD+PWNjYxw7dgwDBw6Enp4e5HI5FAoFpFIprKysoK+vDy8vL60vrerr62PLli2YOXMmJBLJM4/VBx98gHfeeQfe3t7NzqWkpCA0NBSDBg2Cu7s7du7cCblcjpycHADACy+8gIMHD+KNN96Ag4MDXn31Vaxduxb/93//hz///POZY2ovt2/fhlKpbLZEVN++fVFWVqaxTllZWavKNzQ04KOPPsKUKVO4Sws8Hg+nTp1Cbm4uDA0NIRQK8Z///AcpKSno3bu3xna2b9+Od955ByKRqLXdJIQQQv72umxqMTo6Gg4ODoiLi0N2djb4fD5kMhn27t2L2NhYODk54fvvv8e0adNgZmYGPz8/qFQq9OvXD4mJiTAxMUFmZiZmzZoFCwsLBAYGQiqVorCwEDU1NYiPjwfwYL3MzMxMrWKqr6/H+vXrsW3bNpiYmMDc3BxhYWEoKChAQkICLC0tcfjwYYwZMwb5+flwcnLqyCECAMTHx+Pnn3/G3r178emnnz61/MPL5k9aJ/ThfR1PmllWKBRQKBTc+5qamlZE3X00NTUhMDAQjDFs2bKFO84Yw7x582Bubo5z585BJBJh27ZteOONN5CdnQ0LCwu1drKyslBYWIg9e/Z0dhcIIYSQv4QuSzrFYjEMDQ3B5/MhkUigUCiwbt06nDp1ipvRs7e3R0ZGBrZu3Qo/Pz/o6upi9erVXBt2dnbIysrCgQMHEBgYCAMDA4hEIigUimeaWWxqasLmzZvh7u4OAJDL5YiPj4dcLoelpSUAQCqVIiUlBfHx8VrtOtQWD+85PHfunFa3HqhUKixatAgvvfRSi/fH3r59G2vWrMGsWbOe2JZMJlMb645iamoKPp+P8vJytePl5eUt/gwlEolW5R8mnL/88gtOnz6tdgP16dOncfz4cdy9e5c7vnnzZpw8eRK7du3CkiVL1Nratm0bPDw8Wrz0TgghhJAn6zZLJhUXF6O+vh6jRo1S2w5y9+7dKCkp4crFxMRg8ODBMDMzg4GBAeLi4riHQ9pKIBDAzc2Ne5+fnw+lUglnZ2e1mM6ePasWU0dQKpV45513sHr1ajg7O2tVZ968ebh69SoSEhI0nq+pqcG4ceMwcOBAREREPLGtpUuXorq6mnv9+uuvre2CVgQCAQYPHoy0tDTumEqlQlpamsbbCQDA29tbrTwAnDx5Uq38w4Tzp59+wqlTp2BiYqJWvr6+HsCD+0cfpaOj0+xe3traWhw4cKDb3AdLCCGE9ETd5smd2tpaAMCJEyea7fSjp6cHAEhISIBUKkVkZCS8vb1haGiIDRs24MKFC09s+2Fi8ejqUJq2nxSJRODxeGox8fl85OTkgM/nq5U1MDBoRe9a7969e7h06RJyc3MRFhYG4EEyxhhDr1698N133+HVV1/lyoeFheH48eP4/vvv0a9fP43tjRkzBoaGhjh8+DB0dXWf+Pl6enrcuHe08PBwhISEYMiQIRg2bBiioqJQV1eH6dOnAwCCg4NhZWUFmUwGAFi4cCH8/PwQGRmJcePGISEhAZcuXUJcXByABz/bt956C5cvX8bx48ehVCq5+z379OkDgUAAb29v9O7dGyEhIVi5ciVEIhG++uor3LhxA+PGjVOLb//+/fjzzz8xbdq0ThkPQggh5K+o2ySdjz684+fnp7HM+fPn4ePjg7lz53LHHp9xFAgEUCqVasfMzMwAAKWlpdxDInl5eU+NydPTE0qlEhUVFfD19W1Nd9rMyMgI+fn5asc2b96M06dPIykpCXZ2dgAeJNLz58/H4cOHkZ6ezh1/VE1NDQICAqCnp4djx45BKBR2Sh+0FRQUhMrKSqxcuRJlZWXw8PBASkoK97CQXC5Xm5H08fHBvn37sHz5cixbtgxOTk44cuQId0vB77//jmPHjgFAsyWmzpw5g5EjR8LU1BQpKSn4+OOP8eqrr6KpqQmDBg3C0aNHudsrHtq+fTv+8Y9/cE++E0IIIaT1uk3SaWhoCKlUisWLF0OlUmHEiBGorq7G+fPnYWRkhJCQEDg5OWH37t1ITU2FnZ0d9uzZg+zsbLVEy9bWFqmpqSgqKoKJiQnEYjEcHR1hbW2NiIgIrF27FtevX0dkZORTY3J2dsbUqVMRHByMyMhIeHp6orKyEmlpaXBzc2s2I6ZJQUEBGhsbcefOHdy7d49Ldp+2CL6Ojk6z+zLNzc0hFArVjs+bNw/79u3D0aNHYWhoyM3oicViiEQi1NTUYPTo0aivr8fevXtRU1PDPRRkZmbWbAa3q4SFhXEzuo/TtFrA5MmTMXnyZI3lbW1toc2eB0OGDEFqaupTy2n7IBohhBBCnoB1oY0bNzIbGxvuvUqlYlFRUczFxYXp6uoyMzMzFhAQwM6ePcsYY6yhoYGFhoYysVjMjI2N2Zw5c9iSJUuYu7s710ZFRQUbNWoUMzAwYADYmTNnGGOMZWRkMFdXVyYUCpmvry9LTExkANiNGzcYY4zFx8czsVjcLMbGxka2cuVKZmtry3R1dZmFhQWbNGkSu3LlilZ9tLGxYQCavZ7FqlWr1PrKGNPYNgAWHx/PGGPszJkzLZZ52HdtVFdXMwCsurr6mWInhBBCSOfrTt/fXboNJuk5utM2WoQQQgjRTnf6/u42T68TQgghhJC/Lko622Ds2LFqSyk9+tJmDc9Bgwa1WP/rr7/uhB4QQgghhHSObvMgUU+0bds23L9/X+O5J+0I9FBycrLGpZsANNvmkRBCCCGkJ6Oksw0eX0+0tWxsbNopEkIIIYSQ7o0urxNCCCGEkA5HSSchhBBCCOlwlHSSbicmJga2trYQCoXw8vLCxYsXn1g+MTER/fv3h1AohKurK5KTk9XOR0REoH///tDX10fv3r3h7+/fbOvUN998E88//zyEQiEsLCzw7rvv4tatW9z5hoYGhIaGwtXVFb169cLEiRPbrb+EEELI30G3v6eTMYbZs2cjKSkJd+/eRW5u7lN38yEd54VVqdDRe67d27352YPdnfbv34/w8HDExsbCy8sLUVFRCAgIQFFREczNzZvVy8zMxJQpUyCTyTB+/Hjs27cPEydOxOXLl7mdm5ydnbFp0ybY29vj/v372LhxI0aPHo3i4mJui9RXXnkFy5Ytg4WFBX7//XdIpVK89dZb3G5ESqUSIpEICxYswMGDB9u9/4QQQshfXbdfHP7bb7/FhAkTkJ6eDnt7e5iamqJXr7blyqGhoaiqqsKRI0faJ8g2aGhowAcffICcnBwUFhZi/PjxrYorNDQUu3btanZ84MCBuHbtGoAHM32rV69WO+/i4oIff/xR6895uLis9aIDHZp0enl5YejQodi0aRMAQKVSwdraGvPnz8eSJUua1QsKCkJdXR2OHz/OHRs+fDg8PDwQGxv7xL6cOnUKr732msYyx44dw8SJE6FQKKCrq6t2rjv9/hBCCCFPQovDt0JJSQksLCzg4+MDiUTS5oSzPSmVSqhUqja38XAGzd/fv9X1o6OjUVpayr1+/fVX9OnTp9m+5IMGDVIrl5GR0aa4O0JjYyNycnLUxkFHRwf+/v7IysrSWCcrK6vZuAUEBLRYvrGxEXFxcRCLxXB3d9dY5s6dO/j666/h4+PTLOEkhBBCyLPp1klnaGgo5s+fD7lcDh6PB1tbW6hUKshkMtjZ2UEkEsHd3R1JSUlcHaVSiRkzZnDnXVxcEB0dzZ2PiIjArl27cPToUfB4PPB4PKSnpyM9PR08Hg9VVVVc2by8PPB4PNy8eRMAsHPnThgbG+PYsWMYOHAg9PT0IJfLoVAoIJVKYWVlBX19fXh5eSE9PV2rPurr62PLli2YOXMmJBJJq8dILBZDIpFwr0uXLuHu3buYPn26WrlevXqplTM1NW31Z3W027dvQ6lUNlujtG/fvigrK9NYp6ysTKvyx48fh4GBAYRCITZu3IiTJ082G4OPPvoI+vr6MDExgVwux9GjR9uhV4QQQggBunnSGR0djU8++QT9+vVDaWkpsrOzIZPJsHv3bsTGxuLatWtYvHgxpk2bhrNnzwJ4cDm2X79+SExMREFBAVauXIlly5bhwIEDAACpVIrAwECMGTOGm/Xz8fHROqb6+nqsX78e27Ztw7Vr12Bubo6wsDBkZWUhISEBV65cweTJkzFmzBj89NNPHTIuT7J9+3b4+/s3WwP0p59+gqWlJezt7TF16lTI5fIntqNQKFBTU6P26sleeeUV5OXlITMzE2PGjEFgYCAqKirUynz44YfIzc3Fd999Bz6fj+DgYHTzu08IIYSQHqP7XKvWQCwWw9DQEHw+HxKJBAqFAuvWrcOpU6fg7e0NALC3t0dGRga2bt0KPz8/6Orqqt2/aGdnh6ysLBw4cACBgYEwMDCASCSCQqF4ppnFpqYmbN68mbs0K5fLER8fD7lcDktLSwAPEtuUlBTEx8drtR1me7l16xa+/fZb7Nu3T+24l5cXdu7cCRcXF5SWlmL16tXw9fXF1atXYWhoqLEtmUzW7D7QjmZqago+n4/y8nK14+Xl5S3+rCQSiVbl9fX14ejoCEdHRwwfPhxOTk7Yvn07li5dqvb5pqamcHZ2xoABA2BtbY0ffviB+10jhBBCyLPr1jOdjysuLkZ9fT1GjRqltk/57t27UVJSwpWLiYnB4MGDYWZmBgMDA8TFxT11Zk9bAoEAbm5u3Pv8/HwolUo4OzurxXT27Fm1mDrDrl27YGxs3Gw5n7Fjx2Ly5Mlwc3NDQEAAkpOTUVVVxc3+arJ06VJUV1dzr19//bWDo38wtoMHD0ZaWhp3TKVSIS0trcXEz9vbW608AJw8efKpiaJKpYJCoXjieQBPLEMIIYQQ7XXrmc7H1dbWAgBOnDjRbAtKPT09AEBCQgKkUikiIyPh7e0NQ0NDbNiwodm6jI/T0XmQfz96OVXTvugikQg8Hk8tJj6fj5ycHPD5fLWyBgYGrehd2zDGsGPHDrz77rsQCARPLGtsbAxnZ2cUFxe3WEZPT48b084UHh6OkJAQDBkyBMOGDUNUVBTq6uq4e1SDg4NhZWUFmUwGAFi4cCH8/PwQGRmJcePGISEhAZcuXUJcXBwAoK6uDmvXrsWbb74JCwsL3L59GzExMfj999+5h60uXLiA7OxsjBgxAr1790ZJSQlWrFgBBwcHteS1oKAAjY2NuHPnDu7du4e8vDwAoCW8CCGEEC30qKTz0Yd3/Pz8NJY5f/48fHx8MHfuXO7Y4zOOAoEASqVS7djD9RpLS0vRu3dvAOCSiifx9PSEUqlERUUFfH19W9OddnX27FkUFxdjxowZTy1bW1uLkpISvPvuu50QWesEBQWhsrISK1euRFlZGTw8PJCSksI9LCSXy7k/EADAx8cH+/btw/Lly7Fs2TI4OTnhyJEj3BqdfD4fP/74I3bt2oXbt2/DxMQEQ4cOxblz5zBo0CAAwHPPPYdDhw5h1apVqKurg4WFBcaMGYPly5erJd6vv/46fvnlF+69p6cnANB9n4QQQogWelTSaWhoCKlUisWLF0OlUmHEiBGorq7G+fPnYWRkhJCQEDg5OWH37t1ITU2FnZ0d9uzZg+zsbNjZ2XHt2NraIjU1FUVFRTAxMYFYLIajoyOsra0RERGBtWvX4vr164iMjHxqTM7Ozpg6dSqCg4MRGRkJT09PVFZWIi0tDW5ubhg3btxT22iPGbTt27fDy8uLS7YeJZVK8cYbb8DGxga3bt3CqlWrwOfzMWXKFK3bf+jq6oAOX+crLCwMYWFhGs9pWhVg8uTJzZaIekgoFOLQoUNP/DxXV1ecPn36qXE9XMWAEEIIIa3Xo5JOAFizZg3MzMwgk8nw888/w9jYGC+++CKWLVsGAJg9ezZyc3MRFBQEHo+HKVOmYO7cufj222+5NmbOnIn09HQMGTIEtbW1OHPmDEaOHIlvvvkGc+bMgZubG4YOHYpPP/20xWTmUfHx8fj000/xr3/9C7///jtMTU0xfPhwjB8/Xqs+tXUGrbq6GgcPHlRbGupRv/32G6ZMmYI//vgDZmZmGDFiBH744QdudpcQQgghpKN1+x2JSPfQnXY0IIQQQoh2utP3d496ep0QQgghhPRMlHR2sLFjx6otpfToS5s1PAcNGtRi/a+//roTekAIIYQQ0nY97p7Onmbbtm24f/++xnN9+vR5av3k5GSNSzcBaLb9IyGEEEJId0VJZwd7fD3R1np8O0tCCCGEkJ6ILq8TQgghhJAOR0kn6fZiYmJga2sLoVAILy8vXLx48YnlExMT0b9/fwiFQri6uiI5OVntfEREBPr37w99fX307t0b/v7+ajtWpaeng8fjaXxlZ2dz5Q4cOAAPDw8899xzsLGxwYYNG9q344QQQshfSLe/vM4Yw+zZs5GUlIS7d+8iNzeXth3sQi+sSoWO3nMd/jk3P3uwqP7+/fsRHh6O2NhYeHl5ISoqCgEBASgqKoK5uXmzepmZmZgyZQpkMhnGjx+Pffv2YeLEibh8+TK3cL6zszM2bdoEe3t73L9/Hxs3bsTo0aNRXFwMMzMz+Pj4oLS0VK3dFStWIC0tDUOGDAEAfPvtt5g6dSr++9//YvTo0SgsLMTMmTMhEolaXNieEEII+Tvr9ut0fvvtt5gwYQLS09Nhb28PU1NT9OrVtlw5NDQUVVVVOHLkSPsE2QYNDQ344IMPkJOTg8LCQowfP75VcZWWluJf//oXLl26hOLiYixYsABRUVFqZXbu3MntXf6Qnp4eGhoatP6ch+t8WS860KlJp5eXF4YOHYpNmzYBAFQqFaytrTF//nwsWbKkWb2goCDU1dXh+PHj3LHhw4fDw8MDsbGxGj/rYd9OnTqF1157rdn5pqYmWFlZYf78+VixYgUA4J133kFTUxMSExO5cv/973/x+eefQy6Xg8fjPXvnCSGEkHZC63S2QklJCSwsLODj4wOJRNLmhLM9KZVKqFSqNrchEomwYMEC+Pv7t7q+QqGAmZkZli9fDnd39xbLGRkZobS0lHs9ugNSd9XY2IicnBy1cdHR0YG/vz+ysrI01snKymo2jgEBAS2Wb2xsRFxcHMRicYvjd+zYMfzxxx9qibtCoYBQKFQrJxKJ8Ntvv/WIsSWEEEI6W7dOOkNDQzF//nxu5sjW1hYqlQoymQx2dnYQiURwd3dHUlISV0epVGLGjBnceRcXF7XtISMiIrBr1y4cPXqUu08vPT2du4+vqqqKK5uXlwcej8ftub1z504YGxvj2LFjGDhwIPT09CCXy6FQKCCVSmFlZQV9fX14eXlp3CNcE319fWzZsgUzZ86ERCJp9RjZ2toiOjoawcHBEIvFLZbj8XiQSCTcqycst3T79m0olcpmsfbt2xdlZWUa65SVlWlV/vjx4zAwMIBQKMTGjRtx8uRJmJqaamxz+/btCAgIQL9+/bhjAQEBOHToENLS0qBSqXD9+nVERkYCQLNL84QQQgjp5vd0RkdHw8HBAXFxccjOzgafz4dMJsPevXsRGxsLJycnfP/995g2bRrMzMzg5+cHlUqFfv36ITExESYmJsjMzMSsWbNgYWGBwMBASKVSFBYWoqamBvHx8QAerJeZmZmpVUz19fVYv349tm3bBhMTE5ibmyMsLAwFBQVISEiApaUlDh8+jDFjxiA/Px9OTk4dOURaq62thY2NDVQqFV588UWsW7cOgwYNarG8QqGAQqHg3tfU1HRGmJ3mlVdeQV5eHm7fvo2vvvoKgYGBuHDhQrP7RH/77TekpqbiwIEDasdnzpyJkpISjB8/Hk1NTTAyMsLChQsREREBHZ1u/bccIYQQ0iW69bejWCyGoaEh+Hw+JBIJjIyMsG7dOuzYsQMBAQGwt7dHaGgopk2bhq1btwIAdHV1sXr1agwZMgR2dnaYOnUqpk+fziUNBgYGEIlE0NPT42b9BAKB1jE1NTVh8+bN8PHxgYuLC27fvo34+HgkJibC19cXDg4OkEqlGDFiBJfUdjUXFxfs2LEDR48exd69e6FSqeDj44PffvutxToymQxisZh7WVtbd2LED5iamoLP56O8vFzteHl5eYuzwhKJRKvy+vr6cHR0xPDhw7F9+3b06tUL27dvb9ZefHw8TExM8Oabb6od5/F4WL9+PWpra/HLL7+grKwMw4YNAwDY29u3uq+EEELIX123TjofV1xcjPr6eowaNUptO8jdu3ejpKSEKxcTE4PBgwfDzMwMBgYGiIuLg1wub5cYBAIB3NzcuPf5+flQKpVwdnZWi+ns2bNqMXUlb29vBAcHw8PDA35+fjh06BDMzMy4RF2TpUuXorq6mnv9+uuvnRjxAwKBAIMHD0ZaWhp3TKVSIS0tDd7e3hrreHt7q5UHgJMnT7ZY/tF2H53ZBR6snBAfH4/g4GDo6upqrMfn82FlZQWBQIBvvvkG3t7eMDMz06Z7hBBCyN9Kt768/rja2loAwIkTJ5rt9KOnpwcASEhIgFQqRWRkJLy9vWFoaIgNGzaorcOoycNLoo8+zK9p+0mRSKT2ZHJtbS34fD5ycnLA5/PVyhoYGLSid51HV1cXnp6eKC4ubrGMnp4eN6ZdKTw8HCEhIRgyZAiGDRuGqKgo1NXVcQ/1BAcHw8rKCjKZDACwcOFC+Pn5ITIyEuPGjUNCQgIuXbqEuLg4AEBdXR3Wrl2LN998ExYWFrh9+zZiYmLw+++/Y/LkyWqfffr0ady4cQPvv/9+s7hu376NpKQkjBw5Eg0NDdxs99mzZzt4RAghhJCeqUclnY8+vOPn56exzPnz5+Hj44O5c+dyxx6fcRQIBFAqlWrHHs5OlZaWonfv3gAePEj0NJ6enlAqlaioqICvr29rutNllEol8vPz8frrr3d1KE8VFBSEyspKrFy5EmVlZfDw8EBKSgr3sJBcLle7h9LHxwf79u3D8uXLsWzZMjg5OeHIkSPcGp18Ph8//vgjdu3ahdu3b8PExARDhw7FuXPnmt3jun37dvj4+KB///4aY9u1axekUikYY/D29kZ6ejp3iZ0QQggh6npU0mloaAipVIrFixdDpVJhxIgRqK6uxvnz52FkZISQkBA4OTlh9+7dSE1NhZ2dHfbs2YPs7GzY2dlx7dja2iI1NRVFRUUwMTGBWCyGo6MjrK2tERERgbVr16o9jfwkzs7OmDp1KoKDgxEZGQlPT09UVlYiLS0Nbm5uGDdu3FPbKCgoQGNjI+7cuYN79+5xya62i+A/LF9bW4vKykrk5eVBIBBg4MCBAIBPPvkEw4cPh6OjI6qqqrBhwwb88ssvGmfwuqOwsLAWF1zXtErA5MmTm81aPiQUCnHo0CGtPnffvn0tnjM1NW1xGSZCCCGEaMC6uY0bNzIbGxvuvUqlYlFRUczFxYXp6uoyMzMzFhAQwM6ePcsYY6yhoYGFhoYysVjMjI2N2Zw5c9iSJUuYu7s710ZFRQUbNWoUMzAwYADYmTNnGGOMZWRkMFdXVyYUCpmvry9LTExkANiNGzcYY4zFx8czsVjcLMbGxka2cuVKZmtry3R1dZmFhQWbNGkSu3LlilZ9tLGxYQCavbSlqe6jY7Zo0SL2/PPPM4FAwPr27ctef/11dvnyZa3bZ4yx6upqBoBVV1e3qh4hhBBCuk53+v7u9jsSke6hO+1oQAghhBDtdKfv7x719DohhBBCCOmZKOnsYGPHjlVbSunR17p1655af9CgQS3W//rrrzuhB4QQQgghbdejHiTqibZt24b79+9rPNenT5+n1k9OTta4dBOAHrGVJSGEEEIIQElnh3t8PdHWsrGxaadICCGEEEK6Dl1eJ4QQQgghHY6STkIIIYQQ0uEo6SR/CTExMbC1tYVQKISXlxcuXrz4xPKJiYno378/hEIhXF1dkZycrHY+NDQUPB5P7TVmzBi1MmvXroWPjw+ee+45GBsbP/Hz/vjjD/Tr1w88Hg9VVVXP0kVCCCGkR+vx93QyxjB79mwkJSXh7t27yM3N1XonH9J6L6xKhY7ec10dBgDg5mcPdnvav38/wsPDERsbCy8vL0RFRSEgIABFRUUwNzdvVi8zMxNTpkyBTCbD+PHjsW/fPkycOBGXL1/mtssEgDFjxiA+Pp57//he9I2NjZg8eTK8vb2xffv2J8Y6Y8YMuLm54ffff29LlwkhhJAeq8fPdKakpGDnzp04fvw4SktL1ZKGZxUaGoqJEye2Pbh20NDQgNDQULi6uqJXr17PFJdCocDHH38MGxsb6OnpwdbWFjt27Gj/YLvIf/7zH8ycORPTp0/HwIEDERsbi+eee67FPkZHR2PMmDH48MMPMWDAAKxZswYvvvgiNm3apFZOT08PEomEe/Xu3Vvt/OrVq7F48WK4uro+Mb4tW7agqqoKUqm0bR0lhBBCerAen3SWlJTAwsICPj4+kEgk6NWr+0zeKpVKqFSqNrchEomwYMEC+Pv7P1MbgYGBSEtLw/bt21FUVIRvvvkGLi4ubYqru2hsbEROTo7a2Ojo6MDf37/FvdGzsrKajWVAQECz8unp6TA3N4eLiwvmzJmDP/74o9XxFRQU4JNPPsHu3buho9Pj/3cjhBBCnlmP/hYMDQ3F/PnzIZfLwePxYGtrC5VKBZlMBjs7O4hEIri7uyMpKYmro1QqMWPGDO68i4sLoqOjufMRERHYtWsXjh49yt3Ll56ejvT09Gb34+Xl5YHH4+HmzZsAgJ07d8LY2BjHjh3DwIEDoaenB7lcDoVCAalUCisrK+jr68PLywvp6ela9VFfXx9btmzBzJkzIZFIWj1GKSkpOHv2LJKTk+Hv7w9bW1t4e3vjpZdeanVb3dHt27ehVCqbrVnat29flJWVaaxTVlb21PJjxozB7t27kZaWhvXr1+Ps2bMYO3YslEql1rEpFApMmTIFGzZswPPPP9+KXhFCCCF/Pd1nWvAZREdHw8HBAXFxccjOzgafz4dMJsPevXsRGxsLJycnfP/995g2bRrMzMzg5+cHlUqFfv36ITExESYmJsjMzMSsWbNgYWGBwMBASKVSFBYWoqamhrufr0+fPsjMzNQqpvr6eqxfvx7btm2DiYkJzM3NERYWhoKCAiQkJMDS0hKHDx/GmDFjkJ+fDycnp44cIhw7dgxDhgzB559/jj179kBfXx9vvvkm1qxZA5FI1GI9hUIBhULBva+pqenQOLubt99+m/tvV1dXuLm5wcHBAenp6Xjttde0amPp0qUYMGAApk2b1lFhEkIIIT1Gj046xWIxDA0NwefzIZFIoFAosG7dOpw6dQre3t4AAHt7e2RkZGDr1q3w8/ODrq4uVq9ezbVhZ2eHrKwsHDhwAIGBgTAwMIBIJIJCoXimmcWmpiZs3rwZ7u7uAAC5XI74+HjI5XJYWloCAKRSKVJSUhAfH6/VVpht8fPPPyMjIwNCoRCHDx/G7du3MXfuXPzxxx9qD8k8TiaTqY1Td2Vqago+n4/y8nK14+Xl5S3+/CQSSavKAw9+j0xNTVFcXKx10nn69Gnk5+dzM+2MMS7mjz/+uEeMLyGEENJeenTS+bji4mLU19dj1KhRascbGxvh6enJvY+JicGOHTsgl8tx//59NDY2ttsT7wKBAG5ubtz7/Px8KJVKODs7q5VTKBQwMTFpl898EpVKBR6Ph6+//hpisRjAgwdv3nrrLWzevLnF2c6lS5ciPDyce19TUwNra+sOj7e1BAIBBg8ejLS0NO4hK5VKhbS0NISFhWms4+3tjbS0NCxatIg7dvLkSe4PFU1+++03/PHHH7CwsNA6toMHD6ptgZqdnY333nsP586dg4ODg9btEEIIIX8Ff6mks7a2FgBw4sSJZttPPlzuJiEhAVKpFJGRkfD29oahoSE2bNiACxcuPLHthw+BPJytAqBxT3SRSAQej6cWE5/PR05ODvh8vlpZAwODVvTu2VhYWMDKyopLOAFgwIABYIzht99+a/Hyvp6eXrMlgrqr8PBwhISEYMiQIRg2bBiioqJQV1eH6dOnAwCCg4NhZWUFmUwGAFi4cCH8/PwQGRmJcePGISEhAZcuXUJcXByABz+z1atX45///CckEglKSkrw73//G46OjggICOA+Vy6X486dO5DL5VAqlcjLywMAODo6wsDAoFliefv2bQAPxv9p63oSQgghfzV/qaTz0Yd3/Pz8NJY5f/48fHx8MHfuXO5YSUmJWhmBQNDsgREzMzMAQGlpKbd0zsMk40k8PT2hVCpRUVEBX1/f1nSnXbz00ktITExEbW0tl+Rev34dOjo66NevX6fH0xGCgoJQWVmJlStXoqysDB4eHkhJSeEeFpLL5WpPjvv4+GDfvn1Yvnw5li1bBicnJxw5coRbbovP5+PKlSvYtWsXqqqqYGlpidGjR2PNmjVqifjKlSuxa9cu7v3D2fQzZ85g5MiRndBzQgghpAdhPdzGjRuZjY0N9/7jjz9mJiYmbOfOnay4uJjl5OSwL7/8ku3cuZMxxlh0dDQzMjJiKSkprKioiC1fvpwZGRkxd3d3ro21a9ey559/nv3444+ssrKSNTY2ssbGRmZtbc0mT57Mrl+/zo4fP85cXFwYAHbjxg3GGGPx8fFMLBY3i3Hq1KnM1taWHTx4kP3888/swoULbN26dez48eNa9fHatWssNzeXvfHGG2zkyJEsNzeX5ebmalX33r17rF+/fuytt95i165dY2fPnmVOTk7s/fff16r+Q9XV1QwAq66ublU9QgghhHSd7vT9/ZdLOlUqFYuKimIuLi5MV1eXmZmZsYCAAHb27FnGGGMNDQ0sNDSUicViZmxszObMmcOWLFmilnRWVFSwUaNGMQMDAwaAnTlzhjHGWEZGBnN1dWVCoZD5+vqyxMRErZLOxsZGtnLlSmZra8t0dXWZhYUFmzRpErty5YpWfbSxsWEAmr20VVhYyPz9/ZlIJGL9+vVj4eHhrL6+Xuv6jHWvX1pCCCGEaKc7fX/zGHvkJkVCWlBTUwOxWIzq6moYGRl1dTiEEEII0UJ3+v7u0YvDE0IIIYSQnoGSzi42duxYGBgYaHxps4bnoEGDWqz/9ddfd0IPCCGEEEKe7i/19HpPtG3bNrW1HB/Vp0+fp9ZPTk7WuHQTgGZbPRJCCCGEdBVKOrvY4+uJtpaNjU07RUIIIYQQ0nHo8johhBBCCOlwlHQSQgghhHQTMTExsLW1hVAohJeXFy5evPjE8lVVVZg3bx4sLCygp6cHZ2dnJCcnNys3c+ZMmJiYQCQSwdXVFZcuXVI7X1hYiDfffBNisRj6+voYOnQo5HI5d76kpASTJk2CmZkZjIyMEBgYiPLy8lb1rdtfXmeMYfbs2UhKSsLdu3eRm5vbbvukk9Z7YVUqdPSe6+owCCGEkL+Mm5+NAwDs378f4eHhiI2NhZeXF6KiohAQEICioiKYm5s3q9fY2IhRo0bB3NwcSUlJsLKywi+//KK21fLdu3cBALq6uvj2229hZmaGn376idtdEXiQUI4YMQIzZszA6tWrYWRkhGvXrkEoFAIA6urqMHr0aLi7u+P06dMAgBUrVuCNN97ADz/8oLbr35N0+3U6v/32W0yYMAHp6emwt7eHqakpevVqW64cGhqKqqoqHDlypH2CbIOGhgZ88MEHyMnJQWFhIcaPH9+quEpLS/Gvf/0Lly5dQnFxMRYsWICoqKhm5aKiorBlyxbI5XKYmprirbfegkwm436hnubhOl/Wiw5Q0kkIIYS0o4dJp5eXF4YOHYpNmzYBAFQqFaytrTF//nwsWbKkWb3Y2Fhs2LABP/74I3R1dTW2vXjxYkRFRT1xnc63334burq62LNnj8bz3333HcaOHYu7d+9ybVRXV6N379747rvv4O/vr1U/u/3l9ZKSElhYWMDHxwcSiaTNCWd7UiqVUKlUbW5DJBJhwYIFWv/QHqVQKGBmZobly5fD3d1dY5l9+/ZhyZIlWLVqFQoLC7F9+3bs378fy5Yta1PshBBCCGkfjY2NyMnJUcsFdHR04O/vj6ysLI11jh07Bm9vb8ybNw99+/bFCy+8gHXr1kGpVHJlvv32WwBAcHAwzM3N4enpia+++oo7r1KpcOLECTg7OyMgIADm5ubw8vJSmwBTKBTg8XjQ09PjjgmFQujo6CAjI0PrPnbrpDM0NBTz58+HXC4Hj8eDra0tVCoVZDIZ7OzsIBKJ4O7ujqSkJK6OUqnEjBkzuPMuLi6Ijo7mzkdERGDXrl04evQoeDweeDwe0tPTkZ6eDh6Ph6qqKq5sXl4eeDwebt68CQDYuXMnjI2NcezYMQwcOBB6enqQy+VQKBSQSqWwsrKCvr4+vLy8kJ6erlUf9fX1sWXLFsycORMSiaTVY2Rra4vo6GgEBwdDLBZrLJOZmYmXXnoJ77zzDmxtbTF69GhMmTLlqfeJEEIIIaRz3L59G0qlstlyh3379kVZWZnGOj///DOSkpKgVCqRnJyMFStWIDIyEp9++ilX5mEO4+DggNTUVMyZMwcLFizArl27AAAVFRWora3FZ599hjFjxuC7777DpEmT8I9//ANnz54FAAwfPhz6+vr46KOPUF9fj7q6OkilUiiVSpSWlmrdx+4zbahBdHQ0HBwcEBcXh+zsbPD5fMhkMuzduxexsbFwcnLC999/j2nTpsHMzAx+fn5QqVTo168fEhMTYWJigszMTMyaNQsWFhYIDAyEVCpFYWEhampqEB8fD+DBepiZmZlaxVRfX4/169dj27ZtMDExgbm5OcLCwlBQUICEhARYWlri8OHDGDNmDPLz8+Hk5NSRQ6QVHx8f7N27FxcvXsSwYcPw888/Izk5Ge+++26LdRQKBRQKBfe+pqamM0IlhBBCiJZUKhXMzc0RFxcHPp+PwYMH4/fff8eGDRuwatUqrgwArFq1CkZGRvD09MTVq1cRGxuLkJAQ7vyECROwePFiAICHhwcyMzMRGxsLPz8/mJmZITExEXPmzMGXX34JHR0dTJkyBS+++KLW93MC3TzpFIvFMDQ0BJ/Ph0QigUKhwLp163Dq1Cl4e3sDAOzt7ZGRkYGtW7fCz88Purq6WL16NdeGnZ0dsrKycODAAQQGBsLAwAAikQgKheKZZhabmpqwefNm7lK2XC5HfHw85HI5LC0tAQBSqRQpKSmIj4/XalehjvbOO+/g9u3bGDFiBBhj+PPPP/HBBx888fK6TCZTG0dCCCGEdBxTU1Pw+fxmT4SXl5e3mK9YWFhAV1cXfD6fOzZgwACUlZWhsbERAoEAEokEv/76q1q9AQMG4ODBg9zn9urVCwMHDmxW5tFL56NHj0ZJSQlu376NXr16wdjYGBKJBPb29lr3sVtfXn9ccXEx6uvrMWrUKLXtHnfv3o2SkhKuXExMDAYPHgwzMzMYGBggLi5O7bH/thAIBHBzc+Pe5+fnQ6lUwtnZWS2ms2fPqsXUldLT07Fu3Tps3rwZly9fxqFDh3DixAmsWbOmxTpLly5FdXU193r8F5YQQggh7UcgEGDw4MFIS0vjjqlUKqSlpXETbY976aWXUFxcrPZ8yfXr12FhYQGBQADgwcNJj7t+/Tq3uYxAIMDQoUNRVFTUYplHmZqawtjYGKdPn0ZFRQXefPNNrfvYrWc6H1dbWwsAOHHiRLOdfB7e3JqQkACpVIrIyEh4e3vD0NAQGzZswIULF57Y9sPp4Ucf5te0vaRIJAKPx1OLic/nIycnR+0vDQAwMDBoRe86zooVK/Duu+/i/fffBwC4urqirq4Os2bNwscff6xxalxPT0/thmFCCCGEdKzw8HCEhIRgyJAhGDZsGKKiolBXV4fp06cDePAwkJWVFWQyGQBgzpw52LRpExYuXIj58+fjp59+wrp167BgwQKuzblz5yIpKQlffPEFgoODcfHiRcTFxSEuLo4r8+GHHyIoKAgvv/wyXnnlFaSkpOD//u//1J5PiY+Px4ABA2BmZoasrCwsXLgQixcvhouLi9b961FJ56MP7/j5+Wksc/78efj4+GDu3LncscdnHAUCgdqTXQBgZmYG4MESRA/XrsrLy3tqTJ6enlAqlaioqICvr29rutNp6uvrmyWWDxPkbr5iFiGEEPK3ERQUhMrKSqxcuRJlZWXw8PBASkoK93CRXC5X+z63trZGamoqFi9eDDc3N1hZWWHhwoX46KOPuDKDBw8GABw8eBCff/457OzsEBUVhalTp3JlJk2ahNjYWMhkMixYsAAuLi44ePAgRowYwZUpKirC0qVLcefOHdja2uLjjz/m7gHVVo9KOg0NDSGVSrF48WKoVCqMGDEC1dXVOH/+PIyMjBASEgInJyfs3r0bqampsLOzw549e5CdnQ07OzuuHVtbW6SmpqKoqAgmJiYQi8VwdHSEtbU1IiIisHbtWly/fh2RkZFPjcnZ2RlTp05FcHAwIiMj4enpicrKSqSlpcHNzQ3jxo17ahsFBQVobGzEnTt3cO/ePS7Z1XYR/Ifla2trUVlZiby8PAgEAu7+jDfeeAP/+c9/4OnpCS8vLxQXF3OLuj4+O0sIIYSQrhMWFoawsDCN5zStjOPt7Y0ffvjhqe1mZWW1uE4nALz33nt47733Wjz/2Wef4bPPPnvq5zwR6+Y2btzIbGxsuPcqlYpFRUUxFxcXpqury8zMzFhAQAA7e/YsY4yxhoYGFhoaysRiMTM2NmZz5sxhS5YsYe7u7lwbFRUVbNSoUczAwIABYGfOnGGMMZaRkcFcXV2ZUChkvr6+LDExkQFgN27cYIwxFh8fz8RicbMYGxsb2cqVK5mtrS3T1dVlFhYWbNKkSezKlSta9dHGxoYBaPbSlqa6j45ZU1MTi4iIYA4ODkwoFDJra2s2d+5cdvfuXa0/o7q6mgFg1dXVWtchhBBCSNfqTt/f3X5HItI9PNyR6Ek7GhBCCCGke+lO39896ul1QgghhBDSM/Woezp7orFjx+LcuXMazy1btuypW1EOGjQIv/zyi8ZzW7duVbsRuCM9nBCnReIJIYSQnuPh93Z3uLBNl9c72O+//4779+9rPNenTx/06dPnifV/+eUXjUs3AQ+2xjI0NGxzjNr4+eef4eDg0CmfRQghhJD2VVJS0qqF3DsCJZ1EK1VVVejduzfkcnmLe7z/XdXU1MDa2hq//vprl98v093Q2GhG49IyGpuW0di0jMamZdXV1Xj++edx9+5dGBsbd2ksdHmdaOXhumBisZj+h26BkZERjU0LaGw0o3FpGY1Ny2hsWkZj07LW7JHeYTF0dQCEEEIIIeSvj5JOQgghhBDS4SjpJFrR09PDqlWraD92DWhsWkZjoxmNS8tobFpGY9MyGpuWdaexoQeJCCGEEEJIh6OZTkIIIYQQ0uEo6SSEEEIIIR2Okk5CCCGEENLhKOkkhBBCCCEdjpJOwomJiYGtrS2EQiG8vLxw8eLFJ5ZPTExE//79IRQK4erqiuTk5E6KtPO1ZmyuXbuGf/7zn7C1tQWPx0NUVFTnBdrJWjMuX331FXx9fdG7d2/07t0b/v7+T/0d68laMzaHDh3CkCFDYGxsDH19fXh4eGDPnj2dGG3nau2/NQ8lJCSAx+Nh4sSJHRtgF2rN2OzcuRM8Hk/tJRQKOzHaztXa35uqqirMmzcPFhYW0NPTg7Oz81/2e6o1YzNy5Mhmvzc8Hg/jxo3r+EAZIYyxhIQEJhAI2I4dO9i1a9fYzJkzmbGxMSsvL9dY/vz584zP57PPP/+cFRQUsOXLlzNdXV2Wn5/fyZF3vNaOzcWLF5lUKmXffPMNk0gkbOPGjZ0bcCdp7bi88847LCYmhuXm5rLCwkIWGhrKxGIx++233zo58o7X2rE5c+YMO3ToECsoKGDFxcUsKiqK8fl8lpKS0smRd7zWjs1DN27cYFZWVszX15dNmDChc4LtZK0dm/j4eGZkZMRKS0u5V1lZWSdH3TlaOzYKhYINGTKEvf766ywjI4PduHGDpaens7y8vE6OvOO1dmz++OMPtd+Zq1evMj6fz+Lj4zs8Vko6CWOMsWHDhrF58+Zx75VKJbO0tGQymUxj+cDAQDZu3Di1Y15eXmz27NkdGmdXaO3YPMrGxuYvm3S2ZVwYY+zPP/9khoaGbNeuXR0VYpdp69gwxpinpydbvnx5R4TXpZ5lbP7880/m4+PDtm3bxkJCQv6ySWdrxyY+Pp6JxeJOiq5rtXZstmzZwuzt7VljY2Nnhdhl2vrvzcaNG5mhoSGrra3tqBA5dHmdoLGxETk5OfD39+eO6ejowN/fH1lZWRrrZGVlqZUHgICAgBbL91TPMjZ/B+0xLvX19WhqakKfPn06Kswu0daxYYwhLS0NRUVFePnllzsy1E73rGPzySefwNzcHDNmzOiMMLvEs45NbW0tbGxsYG1tjQkTJuDatWudEW6nepaxOXbsGLy9vTFv3jz07dsXL7zwAtatWwelUtlZYXeK9vi3ePv27Xj77behr6/fUWFyKOkkuH37NpRKJfr27at2vG/fvigrK9NYp6ysrFXle6pnGZu/g/YYl48++giWlpbN/njp6Z51bKqrq2FgYACBQIBx48bhv//9L0aNGtXR4XaqZxmbjIwMbN++HV999VVnhNhlnmVsXFxcsGPHDhw9ehR79+6FSqWCj48Pfvvtt84IudM8y9j8/PPPSEpKglKpRHJyMlasWIHIyEh8+umnnRFyp2nrv8UXL17E1atX8f7773dUiGp6dcqnEELIIz777DMkJCQgPT39L/3gQ2sYGhoiLy8PtbW1SEtLQ3h4OOzt7TFy5MiuDq3L3Lt3D++++y6++uormJqadnU43Y63tze8vb259z4+PhgwYAC2bt2KNWvWdGFkXU+lUsHc3BxxcXHg8/kYPHgwfv/9d2zYsAGrVq3q6vC6je3bt8PV1RXDhg3rlM+jpJPA1NQUfD4f5eXlasfLy8shkUg01pFIJK0q31M9y9j8HbRlXL744gt89tlnOHXqFNzc3DoyzC7xrGOjo6MDR0dHAICHhwcKCwshk8n+Uklna8empKQEN2/exBtvvMEdU6lUAIBevXqhqKgIDg4OHRt0J2mPf2t0dXXh6emJ4uLijgixyzzL2FhYWEBXVxd8Pp87NmDAAJSVlaGxsRECgaBDY+4sbfm9qaurQ0JCAj755JOODFENXV4nEAgEGDx4MNLS0rhjKpUKaWlpan9FP8rb21utPACcPHmyxfI91bOMzd/Bs47L559/jjVr1iAlJQVDhgzpjFA7XXv9zqhUKigUio4Iscu0dmz69++P/Px85OXlca8333wTr7zyCvLy8mBtbd2Z4Xeo9vi9USqVyM/Ph4WFRUeF2SWeZWxeeuklFBcXc3+kAMD169dhYWHxl0k4gbb93iQmJkKhUGDatGkdHeb/r8MfVSI9QkJCAtPT02M7d+5kBQUFbNasWczY2JhbfuPdd99lS5Ys4cqfP3+e9erVi33xxRessLCQrVq16i+9ZFJrxkahULDc3FyWm5vLLCwsmFQqZbm5ueynn37qqi50iNaOy2effcYEAgFLSkpSW67j3r17XdWFDtPasVm3bh377rvvWElJCSsoKGBffPEF69WrF/vqq6+6qgsdprVj87i/8tPrrR2b1atXs9TUVFZSUsJycnLY22+/zYRCIbt27VpXdaHDtHZs5HI5MzQ0ZGFhYayoqIgdP36cmZubs08//bSrutBhnvX/qREjRrCgoKBOjZWSTsL573//y55//nkmEAjYsGHD2A8//MCd8/PzYyEhIWrlDxw4wJydnZlAIGCDBg1iJ06c6OSIO09rxubGjRsMQLOXn59f5wfewVozLjY2NhrHZdWqVZ0feCdozdh8/PHHzNHRkQmFQta7d2/m7e3NEhISuiDqztHaf2se9VdOOhlr3dgsWrSIK9u3b1/2+uuvs8uXL3dB1J2jtb83mZmZzMvLi+np6TF7e3u2du1a9ueff3Zy1J2jtWPz448/MgDsu+++69Q4eYwx1nnzqoQQQggh5O+I7ukkhBBCCCEdjpJOQgghhBDS4SjpJIQQQgghHY6STkIIIYQQ0uEo6SSEEEIIIR2Okk5CCCGEENLhKOkkhBBCCCEdjpJOQgghhBDS4SjpJIQQQgghHY6STkIIIYQQ0uEo6SSEEEIIIR2Okk5CCCGEENLh/j+ClpSQyyKFjgAAAABJRU5ErkJggg==",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot the top 20 features by importance\n",
    "ax = (\n",
    "    pipe.features.sort(by=\"importances\")[:20]\n",
    "    .to_pandas()\n",
    "    .plot.barh(y=\"importances\", x=\"names\")\n",
    ")\n",
    "_ = ax.bar_label(ax.containers[0], fmt=\"%.4f\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "5d3a4f95",
   "metadata": {},
   "source": [
    "If you want to see the underlying SQL logic for the top features:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "id": "2c4b75e5",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "```sql\n",
       "DROP TABLE IF EXISTS \"FEATURE_1_18\";\n",
       "\n",
       "CREATE TABLE \"FEATURE_1_18\" AS\n",
       "SELECT SUM( t2.\"review_time\" ) AS \"feature_1_18\",\n",
       "       t1.rowid AS rownum\n",
       "FROM \"TRAIN__STAGING_TABLE_1\" t1\n",
       "INNER JOIN \"REVIEW_RECENT__STAGING_TABLE_4\" t2\n",
       "ON t1.\"product_id\" = t2.\"product_id\"\n",
       "WHERE t2.\"review_time\" <= t1.\"timestamp\"\n",
       "AND ( t2.\"review_time, '+365.000000 days'\" > t1.\"timestamp\" OR t2.\"review_time, '+365.000000 days'\" IS NULL )\n",
       "GROUP BY t1.rowid;\n",
       "```"
      ],
      "text/plain": [
       "'DROP TABLE IF EXISTS \"FEATURE_1_18\";\\n\\nCREATE TABLE \"FEATURE_1_18\" AS\\nSELECT SUM( t2.\"review_time\" ) AS \"feature_1_18\",\\n       t1.rowid AS rownum\\nFROM \"TRAIN__STAGING_TABLE_1\" t1\\nINNER JOIN \"REVIEW_RECENT__STAGING_TABLE_4\" t2\\nON t1.\"product_id\" = t2.\"product_id\"\\nWHERE t2.\"review_time\" <= t1.\"timestamp\"\\nAND ( t2.\"review_time, \\'+365.000000 days\\'\" > t1.\"timestamp\" OR t2.\"review_time, \\'+365.000000 days\\'\" IS NULL )\\nGROUP BY t1.rowid;'"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Top 1\n",
    "pipe.features.sort(by=\"importances\")[0].sql"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "c5517945",
   "metadata": {},
   "source": [
    "Review time is encoded in UNIX time. Hence this feature can be interpreted that churn is correlated with the combination of frequency of purchases and their respective recency."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "id": "8e2497d7",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "text/markdown": [
       "```sql\n",
       "DROP TABLE IF EXISTS \"FEATURE_1_6\";\n",
       "\n",
       "CREATE TABLE \"FEATURE_1_6\" AS\n",
       "SELECT SUM( t2.\"rating\" ) AS \"feature_1_6\",\n",
       "       t1.rowid AS rownum\n",
       "FROM \"TRAIN__STAGING_TABLE_1\" t1\n",
       "INNER JOIN \"REVIEW_RECENT__STAGING_TABLE_4\" t2\n",
       "ON t1.\"product_id\" = t2.\"product_id\"\n",
       "WHERE t2.\"review_time\" <= t1.\"timestamp\"\n",
       "AND ( t2.\"review_time, '+365.000000 days'\" > t1.\"timestamp\" OR t2.\"review_time, '+365.000000 days'\" IS NULL )\n",
       "GROUP BY t1.rowid;\n",
       "```"
      ],
      "text/plain": [
       "'DROP TABLE IF EXISTS \"FEATURE_1_6\";\\n\\nCREATE TABLE \"FEATURE_1_6\" AS\\nSELECT SUM( t2.\"rating\" ) AS \"feature_1_6\",\\n       t1.rowid AS rownum\\nFROM \"TRAIN__STAGING_TABLE_1\" t1\\nINNER JOIN \"REVIEW_RECENT__STAGING_TABLE_4\" t2\\nON t1.\"product_id\" = t2.\"product_id\"\\nWHERE t2.\"review_time\" <= t1.\"timestamp\"\\nAND ( t2.\"review_time, \\'+365.000000 days\\'\" > t1.\"timestamp\" OR t2.\"review_time, \\'+365.000000 days\\'\" IS NULL )\\nGROUP BY t1.rowid;'"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Top 2\n",
    "pipe.features.sort(by=\"importances\")[1].sql"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "761fa525",
   "metadata": {},
   "source": [
    "The features simply states: Churn is correlated with the combination of frequency and magnitude of rating a product."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "776c8300",
   "metadata": {},
   "source": [
    "---\n",
    "## 8. Column Importances\n",
    "\n",
    "Beyond feature importances, getML can also compute *column importances,* showing how each original column (like `rating`, `price`, or `category`) contributed overall to the final model.\n",
    "This is particularly useful if you want to prune entire columns that have negligible relevance for churn. If, for example `review_time` is extremely low, we might decide\n",
    "to remove it. This is valuable when dealing with very large data to speed up the pipeline and reduce noise."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "id": "b6699a6c",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "data": {
      "image/png": "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",
      "text/plain": [
       "<Figure size 640x480 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Plot the top 10 columns by importance\n",
    "ax = (\n",
    "    pipe.columns.sort(by=\"importances\")[:10]\n",
    "    .to_pandas()\n",
    "    .assign(name=lambda c: c.marker.str.cat(c.table, sep=\" \").str.cat(c.name, sep=\".\"))\n",
    "    .plot.barh(y=\"importance\", x=\"name\")\n",
    ")\n",
    "_ = ax.bar_label(ax.containers[0], fmt=\"%.4f\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "2691c09e",
   "metadata": {},
   "source": [
    "---\n",
    "## 9. Export Features\n",
    "\n",
    "Now that we have gained a deeper understanding about the drivers of the model's performance, we will squeeze the last bit out of it by optimizing its hyperparameters. \n",
    "In order not to generate features for every hyperoptimization run, we generate them now and save them to disk."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "id": "01b7beb1",
   "metadata": {
    "tags": []
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "703\n",
      "\u001b[2K  Staging... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:01\n",
      "\u001b[2K  Preprocessing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00\n",
      "\u001b[2K  FastProp: Building subfeatures... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 02:11\n",
      "\u001b[2K  FastProp: Building features... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 30:40\n",
      "\u001b[2K  Staging... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00\n",
      "\u001b[2K  Preprocessing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00\n",
      "\u001b[2K  FastProp: Building subfeatures... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 01:47\n",
      "\u001b[2K  FastProp: Building features... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 02:13\n",
      "\u001b[2K  Staging... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00\n",
      "\u001b[2K  Preprocessing... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 00:00\n",
      "\u001b[2K  FastProp: Building subfeatures... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 01:45\n",
      "\u001b[2K  FastProp: Building features... ━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━━ 100% • 02:04\n",
      "\u001b[?25h"
     ]
    }
   ],
   "source": [
    "target = \"churn\"\n",
    "\n",
    "# performance on test already above human data science baseline\n",
    "# but we think that there is more room with an hyperopt on top of extracted features\n",
    "# next steps: select top x features, export them,\n",
    "# run lightgbm hyperopt script on 200 most important features \n",
    "\n",
    "important_features=pipe.features.sort(by='importance')[:200].names\n",
    "\n",
    "col_export = important_features + [target]\n",
    "\n",
    "print(len(pipe.features))\n",
    "\n",
    "pipe.transform(container.train, df_name=\"train_transform\")[col_export].to_parquet(fname=f\"train_transform\")\n",
    "pipe.transform(container.val,   df_name=\"val_transform\")[col_export].to_parquet(fname=f\"val_transform\")\n",
    "pipe.transform(container.test,  df_name=\"test_transform\")[col_export].to_parquet(fname=f\"test_transform\")"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "d0bdf5cf",
   "metadata": {},
   "source": [
    "\n",
    "---\n",
    "## 10. Hyperoptimization\n",
    "\n",
    "In this section, we train a LightGBM classifier using the features exported from the FastProp pipeline. We leverage Optuna for hyperparameter optimization (hyperopt) to improve performance.\n",
    "\n",
    "If you want to run the terminal script in the background, use:\n",
    "\n",
    "`python az-item-churn-lgbm_tuning.py &`"
   ]
  },
  {
   "cell_type": "markdown",
   "id": "afc51b28",
   "metadata": {},
   "source": [
    "The best hyperparameter constellation results in an AUC on the test set of 0.8312."
   ]
  },
  {
   "cell_type": "markdown",
   "id": "ccb9d242",
   "metadata": {},
   "source": [
    "---\n",
    "## 11. Conclusion\n",
    "\n",
    "In this notebook, we demonstrated how to build a *time-aware, relational* model for item churn prediction using getML. We:\n",
    "1. Loaded and annotated the data (population, product, review),\n",
    "2. Defined a DataModel to represent their relationships,\n",
    "3. Applied getML's FastProp for automated feature engineering using propositionalization,\n",
    "4. Trained an XGBoostRegressor to predict item-level churn,\n",
    "5. Evaluated performance on validation and test sets,\n",
    "6. Analyzed feature and column importances for interpretability.\n",
    "7. Applied Optuna for simple hyperparameter tuning.\n",
    "\n",
    "This provides a solid foundation for churn prediction. While further tuning and feature engineering can enhance accuracy, this analysis highlights the strength of getML in terms of model performance, ease of use, and interpretability with minimal code.\n",
    "\n",
    "### Next Steps\n",
    "To explore model tuning or integrate custom predictors with getML's features, check out [hm-item.ipynb](hm-item.ipynb) in this repository, which focuses on item sales prediction using regression techniques."
   ]
  }
 ],
 "metadata": {
  "environment": {
   "kernel": "getml-relbench",
   "name": "workbench-notebooks.m127",
   "type": "gcloud",
   "uri": "us-docker.pkg.dev/deeplearning-platform-release/gcr.io/workbench-notebooks:m127"
  },
  "kernelspec": {
   "display_name": "relbench",
   "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": "3.12.3"
  },
  "widgets": {
   "application/vnd.jupyter.widget-state+json": {
    "state": {},
    "version_major": 2,
    "version_minor": 0
   }
  }
 },
 "nbformat": 4,
 "nbformat_minor": 5
}