Notebooks/Demo_general_purpose_regression_module.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Testing the general-purpose `NN_trainer` module for regression task\n",
    "### Dr. Tirthajyoti Sarkar, Fremont, CA 94536\n",
    "\n",
    "We implemented a [general-purpose trainer module](https://github.com/tirthajyoti/Deep-learning-with-Python/blob/master/utils/NN_trainer.py) for regression task with tabular datasets. The idea is that you can simply read a dataset (e.g. a CSV file), choose the input and target variables, build a densely-connected neural net, train, predict, and save the model for deployment.\n",
    "\n",
    "- create a trainer object\n",
    "- read a dataset\n",
    "- designate input and output variable(s)\n",
    "- prepare the data (scale, test/train split)\n",
    "- build a model (choosing number of layers, number of neurons, activation function, etc.)\n",
    "- train (choosing number of epochs, learning rate, batch size, etc.)\n",
    "- plot training error over epochs, benchmark against test set predictions\n",
    "- save the model for deploying downstream\n",
    "\n",
    "The module exposes simple, intuitive methods which let you do all these tasks one after another. \n",
    "\n",
    "Often deep learning models for regression tasks are used further downstream in some other computing or data science flow. Consequently, the module is able to give you back a prediction function (trained) which can be used for any further prediction, analytics, or optimization task.\n",
    "\n",
    "This notebook tests the general-purpose trainer module and demonstrates various methods available within the module, with a synthetically generated dataset. "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "import numpy as np\n",
    "import pandas as pd\n",
    "from sklearn.datasets import make_regression"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Make some synthetic data using the `make_regression()` function from Scikit-learn"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "data1 = make_regression(n_samples=200, n_features=4, n_informative=2, n_targets=1, \n",
    "                        bias=0.0, effective_rank=None,tail_strength=0.5, \n",
    "                        noise=2.0, shuffle=True, coef=False, random_state=None)\n",
    "df1 = pd.DataFrame(data1[0],columns=['x'+str(i) for i in range(1,5)])\n",
    "df1['y'] = data1[1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>x1</th>\n",
       "      <th>x2</th>\n",
       "      <th>x3</th>\n",
       "      <th>x4</th>\n",
       "      <th>y</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.846714</td>\n",
       "      <td>1.192538</td>\n",
       "      <td>0.222135</td>\n",
       "      <td>0.083209</td>\n",
       "      <td>50.469653</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.401983</td>\n",
       "      <td>0.415436</td>\n",
       "      <td>-0.014497</td>\n",
       "      <td>-0.458806</td>\n",
       "      <td>17.665510</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2.018747</td>\n",
       "      <td>1.743269</td>\n",
       "      <td>0.589840</td>\n",
       "      <td>-0.825264</td>\n",
       "      <td>137.178923</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>-0.059588</td>\n",
       "      <td>0.791043</td>\n",
       "      <td>1.156163</td>\n",
       "      <td>0.730748</td>\n",
       "      <td>102.487704</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.540433</td>\n",
       "      <td>0.895067</td>\n",
       "      <td>1.274116</td>\n",
       "      <td>1.124877</td>\n",
       "      <td>143.243128</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         x1        x2        x3        x4           y\n",
       "0  0.846714  1.192538  0.222135  0.083209   50.469653\n",
       "1  0.401983  0.415436 -0.014497 -0.458806   17.665510\n",
       "2  2.018747  1.743269  0.589840 -0.825264  137.178923\n",
       "3 -0.059588  0.791043  1.156163  0.730748  102.487704\n",
       "4  0.540433  0.895067  1.274116  1.124877  143.243128"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df1.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Create a useless variable\n",
    "\n",
    "We insert a column of a useless variable into the dataset. This is to show a particular feature of the module which detects such _un-interesting_ variable automatically and eliminates them."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "df1['x5'] = -1"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>x1</th>\n",
       "      <th>x2</th>\n",
       "      <th>x3</th>\n",
       "      <th>x4</th>\n",
       "      <th>y</th>\n",
       "      <th>x5</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>0.846714</td>\n",
       "      <td>1.192538</td>\n",
       "      <td>0.222135</td>\n",
       "      <td>0.083209</td>\n",
       "      <td>50.469653</td>\n",
       "      <td>-1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>0.401983</td>\n",
       "      <td>0.415436</td>\n",
       "      <td>-0.014497</td>\n",
       "      <td>-0.458806</td>\n",
       "      <td>17.665510</td>\n",
       "      <td>-1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>2.018747</td>\n",
       "      <td>1.743269</td>\n",
       "      <td>0.589840</td>\n",
       "      <td>-0.825264</td>\n",
       "      <td>137.178923</td>\n",
       "      <td>-1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>-0.059588</td>\n",
       "      <td>0.791043</td>\n",
       "      <td>1.156163</td>\n",
       "      <td>0.730748</td>\n",
       "      <td>102.487704</td>\n",
       "      <td>-1</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>0.540433</td>\n",
       "      <td>0.895067</td>\n",
       "      <td>1.274116</td>\n",
       "      <td>1.124877</td>\n",
       "      <td>143.243128</td>\n",
       "      <td>-1</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "         x1        x2        x3        x4           y  x5\n",
       "0  0.846714  1.192538  0.222135  0.083209   50.469653  -1\n",
       "1  0.401983  0.415436 -0.014497 -0.458806   17.665510  -1\n",
       "2  2.018747  1.743269  0.589840 -0.825264  137.178923  -1\n",
       "3 -0.059588  0.791043  1.156163  0.730748  102.487704  -1\n",
       "4  0.540433  0.895067  1.274116  1.124877  143.243128  -1"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df1.head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Import the `Trainer` class from the `NN_trainer` module"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "from NN_trainer import Trainer"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "t = Trainer()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### What is this object?"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Hello, I am a neural net trainer!\n"
     ]
    }
   ],
   "source": [
    "print(t)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### If we want to train something now, an error message appears!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model has not been built yet. Cannot train!\n"
     ]
    }
   ],
   "source": [
    "t.train_model()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### If we want to build a model right now, an error message appears!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Data preparation not done yet. Should not build a model!\n"
     ]
    }
   ],
   "source": [
    "t.build_model()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### If we want to prepare data (for modeling), an error message appears!"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "No data has been read yet. Cannot do processing!\n"
     ]
    }
   ],
   "source": [
    "t.prepare_data()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Various types of error-checking code has been built into the module to catch and prevent a task to be performed at a wrong time (e.g. trying to build a model without reading in a dataset)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Now we copy the synthetic dataset to the trainer object dataset\n",
    "\n",
    "The `Trainer` object has an internal Pandas DataFrame called `df`. Here, for demo purpose, we are just using an internally generated DataFrame and setting it to this `df`. But this is not usually how you will read in a dataset.\n",
    "\n",
    "Note that the module provides a `read_csv()` method for reading in a CSV file as a Pandas DataFrame. You will, in most situations, use that method to read in an external dataset (a CSV file) for modeling purpose.\n",
    "\n",
    "At the initialization, the `df` is set to `None`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Internal dataset is None\n"
     ]
    }
   ],
   "source": [
    "if t.df is None:\n",
    "    print(\"Internal dataset is None\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [],
   "source": [
    "t.df = df1"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now, the internal dataset is a regular Pandas DataFrame"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>x1</th>\n",
       "      <th>x2</th>\n",
       "      <th>x3</th>\n",
       "      <th>x4</th>\n",
       "      <th>y</th>\n",
       "      <th>x5</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>200.000000</td>\n",
       "      <td>200.000000</td>\n",
       "      <td>200.000000</td>\n",
       "      <td>200.000000</td>\n",
       "      <td>200.000000</td>\n",
       "      <td>200.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>0.070773</td>\n",
       "      <td>0.160090</td>\n",
       "      <td>0.062019</td>\n",
       "      <td>0.059811</td>\n",
       "      <td>8.456266</td>\n",
       "      <td>-1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>0.986712</td>\n",
       "      <td>1.087917</td>\n",
       "      <td>0.959313</td>\n",
       "      <td>1.051857</td>\n",
       "      <td>91.944281</td>\n",
       "      <td>0.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>-2.962772</td>\n",
       "      <td>-2.804898</td>\n",
       "      <td>-2.983775</td>\n",
       "      <td>-2.498102</td>\n",
       "      <td>-285.072441</td>\n",
       "      <td>-1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>-0.516149</td>\n",
       "      <td>-0.613550</td>\n",
       "      <td>-0.579462</td>\n",
       "      <td>-0.642317</td>\n",
       "      <td>-49.594825</td>\n",
       "      <td>-1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>0.105723</td>\n",
       "      <td>0.167176</td>\n",
       "      <td>0.137037</td>\n",
       "      <td>-0.046116</td>\n",
       "      <td>9.884785</td>\n",
       "      <td>-1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>0.776176</td>\n",
       "      <td>0.907556</td>\n",
       "      <td>0.666958</td>\n",
       "      <td>0.882254</td>\n",
       "      <td>66.733337</td>\n",
       "      <td>-1.0</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>2.608495</td>\n",
       "      <td>2.714743</td>\n",
       "      <td>3.333725</td>\n",
       "      <td>3.651494</td>\n",
       "      <td>302.668195</td>\n",
       "      <td>-1.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "               x1          x2          x3          x4           y     x5\n",
       "count  200.000000  200.000000  200.000000  200.000000  200.000000  200.0\n",
       "mean     0.070773    0.160090    0.062019    0.059811    8.456266   -1.0\n",
       "std      0.986712    1.087917    0.959313    1.051857   91.944281    0.0\n",
       "min     -2.962772   -2.804898   -2.983775   -2.498102 -285.072441   -1.0\n",
       "25%     -0.516149   -0.613550   -0.579462   -0.642317  -49.594825   -1.0\n",
       "50%      0.105723    0.167176    0.137037   -0.046116    9.884785   -1.0\n",
       "75%      0.776176    0.907556    0.666958    0.882254   66.733337   -1.0\n",
       "max      2.608495    2.714743    3.333725    3.651494  302.668195   -1.0"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t.df.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### We can even create simple scatter plots"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<Figure size 700x500 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "t.plot_xy('x1','y')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAcEAAAFWCAYAAAAR2CYTAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAIABJREFUeJzsnXl8FOX9+N9PCJBAwikk4YwiKqBQAVFbUEBQ5EatFVu09eAoHighcgmRmyQgWoWKyu9rqkhbNaCAiogpYOuBWJFDAUtAIAmHIhsIhGSf3x+zSzabPWZ2Z7Ozm+f9eu0LsjvzzDMzu/N5PreQUqJQKBQKRU0kJtwTUCgUCoUiXCghqFAoFIoaixKCCoVCoaixKCGoUCgUihqLEoIKhUKhqLEoIahQKBSKGosSgooagRAiVQghhRB/DPdcXBFCxAgh5ggh8oUQ5UKI/4Z7TgpFTUIJQUVEI4T4o0O4OV/lQohCIcQqIcQVJh2jkRAiQwhxkxnjuXEfMA34APgTMDXYAYUQKUKI+UKIjUKIU1YU/sEghLhOCPG8EOJbIUSxEOKIEGKtEKJ7uOemiDxiwz0BhcIkZgN7gbpAN+BBoJ8Q4hopZUGQYzcCZgJlwOYgx3KnH3AKGCfNq1xxJTAZ+B/wX+Bmk8a1Ck8BvYC3gOeBxsAY4HMhxBAp5fpwTk4RWSghqIgWNkgptzr+/6oQ4ntgCfBHYH7YZuWf5sAvJgpAgK+AZlLKE0KInsAWE8e2AouBe6WUpc43hBCvAruBOYASggrdKHOoIlrZ6Pj3Ul8bCSE6CSHWOMyGZ4UQnwkhBrt83hs44PhztovZNcPPuE2EEEuFEEeFEOeFEN8JIdKEEDGOz1OFEBLoD7R1GfePXsa7RAhRIIT4UggR6/J+QyHEQSHETiFEXQAppU1KecLX/HzMu7tjHuM9fNZaCGEXQsxy/F1LCDHFcW5nHdfwayHEuACOW9dh3jwohGjg8n6sEGKb49ybOs7v364C0PHeSSAP6Gj02IqajRKCimjlcse/XoWBw2f4b+A3aFrjdKAe8K4Q4reOzfYATzj+/xYwyvF6x8e4dYFNwGggF3gS2A9koZnvAI47xtnlmKNzXI/mVodQexDoDjzt8tFfgBTgD1LK897mpBcp5TY0s/JIDx/fAwjgTcffM4B5wKfA446/Pwd6BnDc82jnn4x2Tk5moJm3H3IIOl+0APxto1BURkqpXuoVsS80c6cEBgGXoD0IhwD5QDnQ1bFdqmO7P7rs+xaan6+Ty3uJaL60I0Cs277Tdc7pEcf2Y9ze/4fjfdfjbQTyDZzvX4ELwPXAnY7xpvjYvqf7ees4RgZgB9q4vb8d+Nrl76+BdSbfz8mO+d7pOMcy4CUd+/VyzHlRuL+T6hVZL6UJKqKFtWja1RHgXSAOGCWl3O5pYyFELWAAsF5Kucv5vpTSBixDE6ZdA5zLYOBnYIXb+1mOfwcFOC7ARDTz7OvAS2iabGYQ43niTTSN7x7nGw6t+VoqtEDQAno6CSE6mHjsTGAr2rm9jraYmehrByFEimNeh4BZJs5FUQNQQlARLTyB5l/rC1wDtJRSrvSxfTOgPvCdh892O/716U/0QSqwX0p5weRxkVKeQTOLXo42//uklOWBjuflGN+jaX33uLw9Ek1DW+Xy3nQgAdgthPheCPGiw4cazLHtaGkjDdHO8X4pZbG37YUQDdECYRKAIVLKX4I5vqLmoYSgIlrYJqXcKKX8REq5M0jBIBz/hqrZZrDj3u74Nw5oH+RY3ngTuFYIcaXj73uArVLKQ84NpJSfAu2A36P5BYcDnwgh/hrksftTEbl+jbeNhBD10CwAVwKDpZTfBnlcRQ1ECUFFTeU4cAa4ysNnzvfyHf8aFVr5QDvXKE4HHVw+DwghxK+BdOBl4Bu0dJAmgY7ng1VoPraRQohr0a7Jm+4bSSl/kVKulFI+ALQFVgJjhBBtAzmoEKIdsAhNu1sPZDvec9+uDlpw0g3Ab2VFeoxCYQglBBU1Eoem+D5wu6tPSwiRAIwDjqKZBEETlqAlzevhPaAJWgUYV9Ic/64NZM6OueWgBe5MAP4ANAWWBjKeL6SUh9HyC+9xvMqAf7rNp6nbPmWAUxtr5LLdVUKINv6O6fDT5gDn0Uy+DwLngBxnaonLdivRNMb7pZTrjJ6fQuFEJcsrajLTgVuBzUKIF4DTaNGmlwK/czzUkVrS+SHgXiHED2hBLzullDu9jPsK8DCwTAjRGS3N4na0gJkXpZS7veznj8Vo/saeUsqzwE4hxNNAphBitZTyor9OCDHd8V+n8BkihGjl+P9fdPrO3kSLRh0HfCSr5h7uEUJsBb4EitDMko8AO6kQhqCd/7+A3n6O9xTwazTNrtBxHmPRhG86sMCxXTZa9OhHQIwQ4g9u4+Q6fKcKhX/CHZ6qXuoVzIuKFImefrZLxUOqANAJLZr0F6AE+AzNv+S+f280zfC8Y5wMP8drghZlWgCUouXeTQJi3LbTlSKBFlEqgTlu78eg5RaeBFq4vC99vFJ1XtsmjrlLtDxE98+noEWnnkDT2H4AngOau20ngTw/x/qV41ive/jsdcd17+z4O8+M81Mv9ZJSIqQMle9foVAoFApro3yCCoVCoaixKCGoUCgUihqLpYSgECJOCPGFEOIbIcQuIcQz4Z6TQqFQKKIXS/kEhRACqC+lLBZC1EYrn/S4lPKzME9NoVAoFFGIpVIkpCaRnSWSajte1pHSCoVCoYgqLCUE4WIi7FdodQNflFJ+7mGb0WhtaoiPj+/WunXrgI9nt9uJibGUVbhaqcnnr869Zp471Ozzrynnvnfv3hNSymb+trOUOdQVIUQjtF5sj0rvScl0795dbtu2LeDj5OXl0bt374D3j3Rq8vmrc+8d7mmEjZp8/jXl3IUQX0kpu/vbzrLLASnlKbSk2AFhnopCoVAoohRLCUEhRDOHBogQIh7oh+dWNwqFQqFQBI3VfIIpwGsOv2AM8A8pZUDFhhUKhUKh8IelhKCUcgda92qFQqFQKEKOpcyhCoVCoVBUJ0oIKhQKhaLGooSgosZhs9nIycmhqKiInJwcbDZbuKekUCjChBKCihqDlJLszPm0bZXE20vHc8F2hLeXjqdtqySyM+dj1ZxZhUIROiwVGKNQhJJFWQt4bdkcts8qIbUZ5MVJ1jxWTP5xGLJkDgBp6VPCPEuFQlGdKCGoqBHYbDbmzZ19UQC6ktoM3ptwlm4z5zD2z4+SkJAQ0Pi5ubkUFhaSnJzMiBEjSExMNGn2ikBR90XhD2UOVdQIcnNz6dWhVhUB6CS1GfS8Kobc3FxD47qbWI9tnqZMrBbA331RKJwoTVBRIygsLKT9Jed8btO+2TkKCgoMjetuYtVQJtZw4+++tG77ajinp7AQShNU1AiSk5PZdyLO5zb7jseRkpKie0ynifW9CWe9mljnz5tDcXGx5wEUIUHPfSksKFD3RQEoIaioIYwYMYIte8rJP+758/zjsPU7OyNGjNA9ZqhMrIrg0HNfEuJQ90UBKCGoqCEkJiYyddrTDFlSr4og1Exk9ZgydbqhoJhQmVgVwaHnvtSNleq+KADlE1TUICZOmgxA1xmz6dWhFr99SDDslQS27Cln6rTpFz/XS3JyMp+eiAO8m9X2HY+jiwETqyJ49NyX82XCkOlbEb0oTVBRYxBCkJY+hYOHi7hr/FJqJ7bkrvFLOXTkGGnpUxBCGBovFCZWRfDouS/F51D3RQEoIaiogSQmJjJq1CiSkpIYNWpUQHmBznHMNrEqgkfPfUlOSVH3RQEoc6hCERTuJtb2zc6x73hcwCZWhTn4uy9JSclhnqHCKighqFAEgdPEOmbcI6xevZqCggK6pKTwxogRStMII/7uS15eXrinqLAISggqFCbgNLEqrIW6Lwp/KCGoUEQIqg6mQmE+KjBGobA4qj6pQhE6lCaoUFgcVZ9UoQgdSggqQooy4QVHqFtAKRQ1HWUOVYQEZcIzB1WfVKEILUoTVIQEZcIzB1WfVKEILUoTVJiOajFkHqFoARXN2Gw2cnJyyMzMJCcnB5vNFu4pKSyOEoIK01EmPPNQ9Un1oczvikBR5lCF6SgTnnlU1MGcU0WzVvVJK1Dmd0WgKCGoMB3VYshcVH1S36gIWkUwKHOownSUCc9c3FtANe81N6gWUNGGMr8rgkFpggrTUSa80FAT6mAGkleqzO+KYFBCUBESlAlPYQQpJYuyFjBvruP7csk5Pj0Rx4RHxzJ12tNMnDTZq8arzO+KYFBCUBESVIshhRGCCWwZMWIEEx4dS/5xPJpEneb3N5T5XeEBJQQVIaUmmPAUwRFsYEtNM7+rUoTmogJjFApFWDEjsGXipMncP246XWfEM+z5BNLejGXY8wl0nRHP/eOiw/yuciFDg9IEFQpFWDEjsKUmmN9VLmRoUEJQoVCEFTMDW6LV/K5yIUOHpcyhQojWQohPhBB7hBC7hBCPh3tOCoUitDjzSnf+CDlbIPM97V9bifZ5IHml0VZDVOVChg5LCUGgDJgopewA3ACMF0J0DPOcFIqowFUwnDx50jKCISEhgZ49e9FjBvzzczh2Gt7+Ato+DtP+DoOfjdcd2BKtfjOVCxk6LGUOlVIWAAWO/9uEEHuAlsDusE5MEXJUxFvo8JSD17RBJm1bJfnNwasOFmUt4MDurezOpEpk5y3zoFP3XroDW6LVb6ZyIUOH1TTBiwghUoFrgc/DOxNFKInWlbuVcBUMax4rJvveMto1t7N9VgmvLZvDoqwFYZubv7ZbH0+FTz/dypkzZ4IeK5wtvII1z6pShKFDWPEhI4RIAP4FzJVSvuPh89HAaICkpKRuq1atCvhYxcXFNdqRHO7zLyoq5OSxAi5PslPHxS5RWgb7i2Jo2jyFpKTkkBw73OdeHdjtdnbs+IaOLSpf3+KYViTYD1NaBnuOxnBN5y7ExFT/mvjkyZOcOn6Ids3tXrf54VgMjZu3pUmTJqaNVadOnWq790VFhRQUFJAYB3VjJefLBLZzkJJi7Ltt1m+lJnzvAfr06fOVlLK7v+0sZQ4FEELUBt4G3vAkAAGklMuB5QDdu3eXvXv3Dvh4eXl5BLN/pBPO87fZbLRtlaSZrsrQPMIuXGGHbmPqcfBwUUh+tDXh3ufk5PDJqhlMeKy40vXNi8um97k0AF5clUCTpkvDElWZmZnJsc3TePDeMq/brF0XS/Nec0lPTzdtrB49elTLvc/OnM9ryxxJ/A0q3s8/D0Mm1+P+cdN1m2edZu17HvZUivBpxt39O11m7ZrwvTeCpcyhQruDrwJ7pJSLwz0fRWhREW+hx+oBFcnJyew7Eedzm33H40jR4esycywzMNs8q7qJhAZLCUHgN8AooK8Q4r+O18BwT0oRGqz+gI4GrCYY3DHT12U1v1moFnnOXMj09HRGjRpVI0ybocRSQlBKuVVKKaSUnaWUv3K81od7XorQYPUHdDRgNcHgTkXdz3pV5mi07qeZY5mBWuRFBpbzCSpqDqr6f+iJhOLSZrbdslILL5XWEBkoIagIG6F+QEdz7qGRc/MkGDoMiuGOGfGW6O1oZt1PK9UQteoiz263k5OTE5W/i4CQUkb0q1u3bjIYPvnkk6D2j3TCff52u11mLZwnGzeIl0OvT5ATB8fKodcnyMYN4mXWwnnSbrcHP+Ygz2OG+9wDQe+5eeL06dMyJydHLly4UL799tvSZrNV48ytRXXd+6yF8+TVqfXkgSVI+UbF68AS5NWp9WTWwnnVMg8pK747zz672PB3JxIBtkkdMkRpgoqwEoqVeyiqhoRKqzQ6bjDn5lpcOi8vTwVUVANWMs86vzuLFs7SUmaAaKimEyyWTJY3Qvfu3eW2bdsC3r+m58xE2/lXyj30YoLqNlPLPdy2bZvfc5ceSo7tO1GRmxVoybFAxjVybv4EXLTdd6NU9/nbbLaLi7yUlBRGVLN51vW7k9+6IkfUiZHvTqQghIjMZHmFIhiMhKW3bt3a73ihqkUZyLhGzi0a2wlFMuFu8eT63cn38HlN/u5YKkVCoQgWM8PSQ1WLMtBxVci9IlDUd8c7Sggqogozcw9Dlewc6Lgqr1IRKNXx3YnUHo5KCCqiCjOTw0O1eg50XE/nZiupaES7aB1s2VOuOglEKcEImVAWTZAR3glGCUFFVGFm1ZBQrZ5dx3UVYq7d1D2N63puB45B9jqt8ezbX0DRafhgB5SVlfHXpX+x/IOnOolUDcWJGULG9btT6lZfPNicXE+tutY8VmyJVl260JNHYeWXyhMMjmg8f725h/7O/fTp07Jxg/gqOV6uuV5NGtYznG93+vRp2SgxTk4dimxcHzm0K3LiQO3fxvXR3m8Q73Fc57klxsfK9skEnH8WjffdHV85latWvRkxeXFm5RpWyRM0ISc3VL8RM0BnnqBKkVCh4lF7/v7C0vWce6VWOB4q2hhphePK0EG3sfurDWycUrWber/50LHbrby77kOv5xVsqkQ033cnvu7d+pjFnC05Z/m8ODPTYpxs2rSJI0eOmJKukZOTw9tLx7PmMe/BYcOeT+Cu8dXfqkulSChqPGaEpYci2dlms7F16xa2z6paTiu1GWycAt1mbvXa/NQKqRJWL0nnjMD1JDxSm8HlsXZGjp7D2D8/aum8uFDc65iYGNO+F9EQdap8ggqFD0LRwy3YqNNwPnikQf9UuPxx/q5xndjI6FVpdSETDRHLShNUKHRgZrJzsA+2cHYn0JvkLz1UxPn0RBwTHh0bVKUdvVhdeOjF6p0orFok3AhKE1QoqplgV8/h6hFoJMk/3BGD0aChQM3qBxkulBBUKKqZYB9s4Xrw6DXjrly5MiSVdozg7xqXloVXeOglEoTMxEmTuX/cdLrOiGfY8wmkvRnLsOcT6DojnvvHhb9Vlz+UOTTEWD2AQFH9mNFHMRzdCfSaGD/++OOwB+74u8b7Y2LCLjz0YqVOFJ6wUg/HQFBCMESE2yeisDbBPtjC8eDR65+KTxWm++MCWUz6usbLlqcw7u7f6T5+OIkUIRPuIuGBooRgiAhV9wFFdGDWg606Hzx6gyDmP9CXdSvWYUYwRyCLycoCsyU79+zn448/rnSNt23bFvAiNFzWnUgVMlZHCcEQ4C9H6b0JZ+k20/o5SorQE0kPNr1m3JEjRzJ50gRTIgaNLCa9C0ytR+OkIK0vyroTnSghGAKskMysUISC0WPH89X27XSeuoaeV9WiQ0oZ+09UNuMKIYL2eYLxxWSorS/KuhOdKCEYAqIlR0lhfarLNOeuBf2xl52t+2LI2wXDht/OwY9erXRcM4I5jCwmhw8fHlLri7LuRC8qRSIEREuOksK6GKncYkbVFve8v+fvK2f77PPsXljGzs/X8dKyFyptb0alHSOLyVD1fnQS6vEV4UNpgiEgGqooKDxjlZQXPaa5iZMme/VhLVu+Qqugr0MYBaMFBePzNFItpaCgIKTWF2XdiV6UJhgCIiHBVWEMozUzQ4neyi3z5jzjtWrLyWMFuqu2hEsLMlJUIDk5mb3H6/ocz5v1RY+mrKw70YvSBEOE1RNcFcawUlCEHqF0/WXlPDNrNvPvttM0oernRroohEsLSkxM5Mm0dHrOmc8fbiylY0sY0R0S4ysvJuvXr8/+vXv46OszhqwvRqI9lXUnelGaYIgIRfcBRXgwUjOzOtAjlK5ofp4el9nZ/J3WfT57Hbgqq0a6KIRDC3Jq3ouzM7mmNZSUwuufQotHoPPUOlz7dBz3j5vOk2lPMWzwAN54eQFP3g5DstFtfTFS31RZd6IXpQmGmEjKA1N4xmopL3p8ZQeOw5i+MKqX4yGdrb2fNqhiG73aWzi0IM+at3as27MhLX0yaelTmDM7g40bN7A7E9peAo0ToOs06HUltE+G7wtg406YPiO9kvXFbrcb9nMq6050ojRBhcIPVguK0OUr2wsjrtP+Tm0G76XB/Heh2OU09Gpv1a0F+dO8308rZXF2JgUFBcyfN49bOmnvC6EJ+YPPwV09oHkDuPt66NO5Pm3aXlbJ+vLzzz8b9nMq6050ojRBRViwSpSlHszs6Xb06FEyMjI4fPgwrVq1IiMjgxYtWhiaj9/KLdkwZSgkuFgwU5tBzysg90tNO3R2UdCrvVWnFqRX887IyKBtM8GVbpc9MV47RydfH6q6QCkrKwt4YaOsO9GF0gQV1YqVoiz1YkZPN7vdztBBt9EutSX5n73MlfJ98j97mXapLRk66DbsdruhObm3r3ni9RgGLNRMgfffBBMHVt2nfTIUnHJ0USgy1kXBVQsa9MAifqx3B/Gpg1iQtYQx4x4xVQvSq3n/+OOPtG1Sxr5C3+PtKahVReONjY1V0Z4KQGmCimrGSlGWejGj9dHwIbez+6sN7Mmiyv795m9g+JDbeXfdh7rn5F6Ae+3atRwrXsOh589X0gBd2XkYjhXXZcH6mCpdFPRo5lJKXlr2QqVoynUr1jF50gRTa2fq1bxbd2nN4ePxbNnjOyr0P/vgn24LlMaNG19c2Khoz5qNEoKKaiOSS08FYw48evQoH31UVQCC9vfGKdAxfcNFAWQEp2lu+PDhtG31HidseBSC+cfhPz/UJvvZvzBy5MiLXRT8pQmMHjue1atXU1hYyPavvmTHZ+uCWsDoEbZ6A3F25mbQ6aq/Me4WzQT8XlrVBcYt82DSU1OrfJ9iYmJMqW+qiAKklBH96tatmwyGTz75JKj9I53qPP/XXntNDr0+Qco38Poaen2CzMnJqZb5BHLup0+fljk5OXLhwoUyJydH2mw2v/s8/PDDsv813s9ZvoHsfw1yzJgxAZxFBVkL58lOqfXkgSWVxz6wBHl1aj2ZtXDexW2d5561cJ682sM+/3sW2bJprEysV1sOvT5BPjkoVvbthGxUD5l1L9L+etVjNGlYz+v1sNvtMmvhPNm4Qbwcen2CnDgoVg69PkE2bhAvsxbOk3a7vcq5eJqX+7lkLZwnO7WNl1OHIhvXRw7tipw4EDn4WmR8HeSQgbdWGdt5/lXmNNj3nKKFmvLMA7ZJHTJEaYKKasNqUZaBEEhQxOHDh+nU0vc2nVrC94cO6R7TXaMaPnw4UkoOFZXS6SktReCqFPj2MHx5oDZPz6yqrfrSzN/+EhrFlbH1aUhtduHi+97SLfyliRg1g+vVvJ3/zps7mxs7xHChrIT138Zy8Lhk2vSpTJ0+06uJNlKa1SpCi+WEoBBiBTAYOCalvDrc81GYh5lRlmZRHVGqrVq1YtdnvrfZdQQu+3Ubv/MsKCjg6+3b+PD9tdzUMfai+fKRsQ/Soolgx/wymibA6m1aEEybS6DAFouAKsLAWxSmrQTmrYHtcz2bb99Lg27TYewtlU2v3hYwgZjB9QooT9ulpKQwwoAgU9GeNRvLCUHg/4AXgJwwz0NhMlYqPSWlpKiokLatkkLeIDUjI4N2qS/7Pu/vIWdDhsd5uvrs2jU5y8kf7VAOvdppUaDF54pp8xh8MKlifNcUgTuuK/Hoa/Wmmedu0zRJnykKLukWTvYdj+OKRo3IycmptKgIptiAXgGlBJkiUCwnBKWUm4UQqeGeh8J8zIiyNItFWQuoFx9XLVGqLVq0oH//W+k3fwMbp3iKDoV+/W71GBTjq3KK0yzZvAHcdJUfoeVByHjTzAtPaekUvnCmWzg5cAw++fYcWyZNqLKouPE3N3NVhJvBFdGLkBbMy3IIwbXezKFCiNHAaICkpKRuq1atCvhYxcXFNdr+H47zLyoqpKCggMQ4qBsrOV8msJ2DlJQUkpKMRUcGgt1uZ8eOb7isbQsayMNVPi8tgz1HY7imcxdiYsxLpd2/fx+nT58mMQ7ia0PJBbCdgwYNGnD55e29zrNjCzt1PCxXS8tgzxFo3hDK7dCqifdjH/5JUDuxJUlJSYB23+vVq+dx/JPFcOoMtEvyPt4PRdC4PjRJ0ObxXYGgloD2ybLSWM7P4utA+yTvz5r9RYJGzdpwySWXeD+oidTk331NOfc+ffp8JaXs7m87y2mCepBSLgeWA3Tv3l327t074LE2bdrEoUOHIqJySSjIy8sjmOsXKDabjdWrV3PU4cN56A/VF4yQk5PDJ6tm8Kv0DHqfS/O4zYurEmjSdKmpJrbevXtz9OhRZs2axaFDh2jTpg0ZGRle0yKc85zwWDGUeR7zxdchuSEU/gJrJno/9rBXErhr/NKL99p537dv+5yJT1XWzG0XoE0afD3Pu/l28FPwQJ9aHPw5ns27yygvL2fHvAukllFlri3OwXWT8Jgi4jpe7TpxTJs+w1RTtDfC9b23AjX53D0RkULQDJy+ltg6cXyyakZIfUKKqlSnD8c9+OXgwYNhi1Jt0aIFf/3rX3VtqyuaNhkaxsM/vyAgX6u3KMxyeY4BWYIPJl2oYoYdvKQeQ0cMotW13bkuJYVBJSWsWzGxUhSpK1e3hvYt6nB7tlb305NZN+NOuKvHOcsWTFBELzVWCDp9LYsWztJW2oDVK5cojOEeVOJc6OTtLKXtJTEM9rFvdUepesLps7OVFJO7TfPVJTeq6KkHsK9QKxY9dRgMzoK1k6r6HH35Wr1FYb4+fDgvLXvBZ4qCc5GYmZnpV1j3v7qcvbI/XWf8i2tbl/CrtrC/ELZ8r8194kCtALaVCyYoohPLCUEhxJtAb+ASIcRhYKaU8lUzj+Easp0fSyXzjdUrlyj04ys37ZZ5cPRnIL7qflYpmTV8+HAeGfsgbR7TAl/aJ8One2HC3zTBced1WreIN8bD8dOw5MNYOk8V9LmmruEC1540c705dHpSX/afiOfu8fcyeOhwXsp8nKQG5/lVG23uVQp9V2NbKjPxlm4TScXiayKWE4JSypGhPoZryHa+h88j+Yeo0PCXm/bxVMj9BXadhE6tKj6zUsms559bTP04yf09qdJVfXAWZK2Dzm0Ev19e3yHstDJna9asqSS0pJT87W9/q/QQ1oses7WR1JelS5fSt0M56UO8jxdpkaLeLA6PPzKGXr1uYuvWLSFPw1EEjuWEYHUQDZVLwo3VV7d6ctMSfhH0eLI2/boY15xCiZSSeXOeYe7sZ+h3NZSVw9tfVGiAEwdqZs9rptSiz+9mcOmll1aWX852AAAgAElEQVTS0JxCy9vDecKjY1m2fAVSSlMewkZSX6xYMCFYvFkcpv0d/v75BrbPQlcajtV/U9FKjRSC0fhDrC58PVittLotKCigXZOzPreJrw2Tpz5NamqqpUpmLcpawOvL53vsOOFasqxv53guvfRSr5qaL3Pw+mMFLMpaYJrfW2+ZMysVTDADbxYHWwks+9hH1R0Xl0v9+vUj4jcVrdRIIej6Q6R11c8j7YdYnURKK6Svt2/TKqv44HyZIDU11VIm74qHaqnnh6dLyTJf1gp/5uDLY+2MHG2e31tvmTMrFUwwA28WB11Vdxwul6KCwxHxm4pWaqQQdP0hLlpY+bNAf4g1wZQRKa2QbDYbH76/Fsp9pw0US3jwD9Za6OgqMeYoWebLWuFvnDqxofF76/EhetIa9xTWJW9nKb373ESzpBbYbLaI+P14c63oqrrT7Bz5+fk8mz3f8r+paKbGdpZ3dubefTSGYc8nkPZmLMOeT6DrjHjuH6ffJyQjsFN6oBipARlOcnNzualjLNOGa+ZD947w+ceh/wJo0LCh5R4senMDdx3x3c3eyn5v1y71d/75Rb6392Prngvc0rkOHWI28s6yRyLm95OcnOyxQ31yI/x2vN93PI7Dhw9HxG8qmqmRmiBU/BA3bdpE06ZLA/YJRYp50Ays/GB1xTnPiQO1v7tO00xT7ZO1B9OW76FLW0F8fL2wztMTevzV3xfA1z/W8WmtqA6/txnWj/fXv8e+bz7mm7kXSG1W6ng3cn4/3nycI7prgUz+fJ8TBrUi8Yj1f1PRTI3VBJ3ExMQwatQo0tPTGTVqlGET6Ly5s6v4NqDClDF/3hyKi70/iCIJb6teV/YdjyPFAknm+07EIYQWQHLwOS2hvHkD7d9Dz0ODxPrUqVMnJMe32Wzk5OSQmZlJTk4ONptN974jRoxgy57yKtqrk/zj8PFOuKZrhdkwkHFKy3xrkr5wtX689eKfKfzXFN5c8jCtUi5h7uwMv9qbc/82LZvz7uq3qlSlgcj5/VS4VupVutaJ8TDuFq04uidLhNPl0rZt24j4TUUzNV4IBkOkmAfNQs8DOtAHq5m4zzMxXmv5kz5E+/eETZtno0aNquwbjAAzwzTu7aEK2vXtOw8uS6lNp9p5Ps2G/sbZXxQTcACKq/Xj3cfPsPj3dt5PK+WbuaWseOEZhg0e4PNcnftPH3KOfp30BY9YGadrpeuM+EqulaWfxNGx261V3nd1uUTKbyqaqbHmUDOIFPOgWURKZJ/eebp2iPCU+vGvorqMG/MQvfvcwt2/u4c77rjDp7nPLNO4p8CRXUdi+NeuUp4YAHPuvoAWMe97bF9pC8uWpzDu7t/5nYs7eooQdErfwLw5zzDt6Qyf+//jM33BI1b//fiLjHUWi/fkcomU31RUI6WM6Fe3bt1kMHzyyScB7/vaa6/JodcnSPkGXl9Dr0+QOTk5Qc0xlBg9f7vdLrMWzpONG8TLodcnyImDY+XQ6xNk4wbxMmvhPGm320MzUYPomafruWctnCevTq0nDyxB2l9HZt2LbFwfOfha5GO3IQdcW8fnOZ4+fVo2bhAvDyzx/D04sATZpGE9abPZdJ/D6dOnZU5Ojpw1a5asF1db7lwY2NjOcRYuXChzcnKkzWYL+Hv/2muvySE96vv8zg++Flk/vo7H+bj+Zl4bixza1fs4ofz9BPO7N5vq/k1Z6dxDCbBN6pAhShMMgmhL/NWD3nywcGNknu7aTfY6eG2ze6JzqU+tK5ju6d5wphvk5OTQr0tdOrXy3KXBfWxPAStmpUEUFhZy+SUlPre5MgV+OIHHc3W1nugNHjH6+4m0dKVI+U1FK0oIBkFNNmVUZyukYNAzT1cBZiuBeWv0Vfpwva+hNI3rHfvo0aNkZ86vZNLderwuj/75YW66uS933e3fpOuP5ORkPimIBUq9brOvEFKblnk8V9eo1cR4rQzckGytCID772dAVm1uHTBQd5qEjJBqRt6IlN9UtKGEYJDoLRelCC3BrP5dhYyRSh+uD6xQpiToHfvof7/i28/Wufnrysg/Drct/IBFuzbxxGPjmDrtabr3uNHwPECzfjz654d9a2974Yar4j1GNLpbT9zTWC5rDt8cgs9/gB7tyjhz4H3atkryK8RsNhsPP/QAeR+tYdrgC4zu62w3FXi6RaRplIrAUNGhQeKa+HvX+KU07zWXu8Yv5dCRY6SlT7H0yjMakCZEZLqmfuit9OGu5ZgZ5eceodqvXz+/Y2/ZU84H69d6Tdf58Ck4crKULdNKeG3ZHIqK/GRyeyExMZH0p6Zwyzwvof/ZMLYvfLZPejxX96hV1zQWKSFnKwy+FoqWwifTJO8+fobts7Q5L8paUGU81/t/fNdb/P7GC2z+Dto+rpm1pTSebmHGd0oROShN0CSUKSM8mBGROWLECB4Z9xD5x7VKH5/u9X1MTxqdGaZx7+a8cnr27MXgZ7ew9okSj2PfOmAgJQc/8FtubfsBTSCsKSiguLg4IFP91Okz+fyz/9AxfQO3dNJ8gM4iBONugTXfxPs8V29l0zbtPFOlaLhz7t7M0J7vf9Vi40Z8stFSAENpsvpQmqAiYjG1WIGUDM6CrqnawzwQjc5bvpjeUnyuD981jxWTfW8Zax4rZvusEg7s3splnXp5Hfvart11lVsrOOVoIxVHwPl3QgjWrP2AqdNnsmlPbdZ/W4cL9hhu7FCfZXnx/PHPWl9Db/mWnqwnrbr+nv7X1jeUM+j3/qfB/Heh2HFZ9Phko6EAhj9NVlEZpQkqIhazIjJzc3PpfU1dbrq8jJtmQ6vGcNtCzYRoRKMLJspPX3Hyrez67gc2btxYZeycnBz+7c9vWAhd2mj/rxtrZ+WbbyKlDEhDEEIw/ekMHp8w8eK5pqSksGr4cF5a9gKprZP9Bqe4Wk8yMzNJPHLe5zHdhZiRYuOjeunzyYYiyre68afJtm77ajinZzmUEFRELGZFZBYWFnJFs/OkDYIxfbWH5j+/gC5T4OYOcEUy7DwM/95fmxkz/Wt0gZjG9T58N27c6HFsXek6e+GN8drfJaXA0Q95e+mWgKIn3U1t48aNIzExkezM+QGZEgMJLNJbbLzglP50i0gvgKFnMRWMKTwaUeZQRcRiVi1T13ES4+G+mzRT2uG/wG8dNUePFddl0ZIXQxbsFOzD11+ZtCHZMGWoZgbNPw5nz8M/HrNfNLd6Czxxx5epbe7sDObOmRWQKTGQwCJd978QYmvpT1eKlPq43tCzmArGFB6NKCGoiFjMisj0No6z5ujdN8DBk7UYOXKk7rkZrUFqxsPX1Sd5e3YdHs/RhF/XaXD/TVo6glMgJjfSHoZgzNfly2/5+vL5tG5iD6gWqF8h7kGI6So2vgtmvxunuz1apNfy1LOYqhsrLavJhgMlBBURSyAPzlCOA4GF19tsNkpKSvjov+dYtE5L2HdHz8PXNeBk5ISX2R8zgE17anNjh/ocPVWL2xZWCMSkhpX31VOs2l/QyPtppeQXlV4MRPGEL23WaGCRv/s2IKs2Q0f8lh+PHtetwZv5XQgHehZT58uEZTXZcKB8goqIxlO4/d5jcfxrVyk33ay/S7lZRQ+MhNe7p0SM7Wvngx0wZzVMG65pbkIYf/gmJiZy3333cd99910s3vzmm28SG/Mhh563kxAHeR728+fr0mNqu6F9RSCKJ3wFp+gJLHL3RY4eqzk5Pd+3wKrERHIBDD2+4WIJD/7BWppsONM5lBBURDRCCMaMe4TEhk3YtGkTH323h/17v6P31XW4UmzknWVbL1ZJ8fVANKN+o74Iz4pcN185brcugHX/FTRIrB/Uw9cZpCOl5O2lW0iIC7yijR5T21Up8PfPYHh3Z8WWCvQGp3gKLPJXEi3/x0LWrFljSt3NSK7lqSdfdfqCFMuchxVK3SkhqIhY3H9A7Zqc5Se7nUMx0OeKCw5NyliSs7fITj0rVSPh9cOHD/cpMDdMhi7TYsl+djFvjBwZ9EPLjGLveiI49xbA4Z+gzWPBabPu6NGwx4x7hNzcXAoKCnjnnXeC1iYitQCGP002KclPSaRqxAqFCZQQVEQsRqqFeKs44g8jK1UjEZ56BGbvq+sSFxdn2qr91gGDuOGZNUy6/QLXusg6TwLKk9AfPnz4xco63gTplwe0EmgnbOZps3o07C7TZjJ3zixu6hhL20YlrNoXy9jRDzJ02HCWv/wqDRo0MHzcSMWfJpuXlxfuKQLGLSehQglBRUTi9weUBt2mw9hbtCjIQJOcjaxUjeS6FRQU+BWYl18SfD6auxC/94Zy1n8DMd1g9quCxIR6bP3OflFA+RL6PXv2IjHOzuAsWDvJQ9eHhRVpGAlx5mmzehYM1116gYGdL2BH6wLS68py+lwFO3a/RauUNcyY+YwpprVgfFfV7feyuiZrlcIESggqIhKj1ULAeJKz0ZWqEZPjO++841dgfn2gjJ9+fJ2kpKSAWyB505Y3xMJLv8SS0nkghz5ecVFA+Up2v2XeBu7vCU0SKro+tE/WcvE2fwfldq3YgOs1MkOb1aNhd2kDH++GQyeqtsHKP34haNNaML4rK/i9rIhVChOoFAlFRGKkWogTo0nORlaqYCy8Xk8+2hc/wI3NvmXVc6MD6mDgK6WhTix8MOkCH76/Ttf2qc3g46nw0iZNuz74HNzlKCRwVw/48S/QpyOs3lZ5PzMeYnqT4vN2V+1L6Jy7rzxIPTmdvvIj/RUaCGbfaMYqhQmUEIwAjCZeRzs2m439+/ez66jvr+++QkhppP0/kCTnQFaqenPdnAJzsBeBOTgLMu6EZQ9o+XeBPDCNCnEj2rWzkED6EO3fhLiqiw4w5yGmZ8GweQ/07qCvD6QTvTmdwRTVjoaC3KHCKoUJlBC0MKqvWWVcr8eRr1eydU+p7x/QXhhxXeCRiYGsVI30l5w4aTLXXD+QTk/BsEUw8Q24ZS5cOxX+eHNFw1kI7IFpVIgHol274rroAPMeYv407P4LoNdV0KGl73HcFyx6NTSjiwlXgtk32rFKYQLlE7QwVggfthLu1yN7nRYF6m4Ccz4YO7cR/H554JGJwaQV6AlKEEJwbdfraHYql+suLWft1/DzGc20mOBB9hoNFDBalFrP9t8XVHSicMW9QLfRh5i/oBFfYf9lFy4wpk8ZW773fYxdR2LonKylB9jtdt3+3mB8V1bxe1kVKxQmUELQolglfNgqeLoeTk2p6zT4zRVwaTPIP5XA1u/KGTBwMF1+1Y0WLVoEnORsRqNcfyQnJ/PpqXie71VMwSlo3dSzAHRi5IHpSYjbSiB3G9TtAYs2ah3pnUJcj9DftAsW3FP1/VvmQcumdcjItRt6iHkKGvlXUV3GjXmI3n1u4e7f3XMxKMhb2P9fl/6FFS/OpuBEic+5b95VSs+D/wPg559/1q2hBdLhwkkw+1YH4W68a4XCBMocalGUGaUynq6HEFoe4MHn4O7rYeOeOrS49vccOnKMlW/+g6eeeopRo0YF9WMKtlGuP1z9IsmNNJOiL4z42FzNTQeOaZpz28fh7S/gQjl8sAPKysr469K/IKXUZZ66pd+t9Jpb9Vo88MhM0ma9TMMeM0juci9PpE2heXJLQwW5Vz9aTHLDMv6z5wx9O1zgcvsHVYKCEhMTGT58OMnJyReT4kePHc8D45+mXMbSf37VhsjOvNEJA2DxoiyKi4spKyvTraEF47uyit/LHau5WpyWk/T09KB/s0ZRmqBFUWaUyvi6Hs4gjW8O22l+2WWm/oBCvVJ11TbfHHuWCX8jqKou7jiFdJepM0huWHYxfSAvDj6aUjV9QI95qri4uMq1qF+/PouyFvBs9nx6dahF4pFzvL3OdxqAu3afvQ5e2+ye4lB60fwvpUQI4TXVYOGi51kw41E6Ty6n55Waj3B/IWz5HqYO0ywHO4u0hWO9evUc/l7/GlowFgEj+1anVqZcLRUoIWhRrG5GqW5CeT30PHzMTDz2VgT6prmzaXVJObctLDXc1d4bztqq8+bOZsPkMl2mdX9C39O1CKSZrqt2byvRktzdc/xc59h56kzaNq/j8RiDl8yhTqN23HGdZNadWqpGwSn4VRvNT+k0MTsXjt27d7+ooelZcATju/K375NpT5GdOb/a8giN+ENrgqvFrzlUCDFYCPGxEOJbIcTfhRA3edjmeiFEuRkTEkIMEEJ8L4TYL4Swbrn2EGNVM0q40HM98naep6SkxFAKSXWahLyZoFJba8Ea+T8WkjbrZa64dgBdptVh6HPmmF8DMa3rNU/ZbDZeeuklZj0z03AagKt2n7tNS773NsemCVBeXub1GGsnnOX73d/y7SG7x/QNJ67m5NtuH8SArNq6IhONRP2642/fxdkLqzWP0Ig/tCbgUxMUQvQH1gCfAf8CbgQ+EUIsAdKkyU8JIUQt4EWgP3AY+FII8a6UcreZx4kEqiMoI5Lwdz36L4CubctYt2IikydN0LWCLioqrFaTkF4TlGsLJDPMr6EwrbsGtKReYufXl18wXP7KVbsvPKWlX3jDn5BMbQa3XA0f79S+D00TtH0KT2m+1hHd4WSxtnDskf8DJ0+e4MyB90lpWEanp+CGy6Fzmxj+91M9n9pdMBYBT/uGIwDOiD+0JuDPHDoTyJFS/sn5hhDiAeB54DIhxEgppe+raYwewH4p5f8cx1oFDANqnBAEa4QPWwn363FZk7N8+6Odr/7n7FggdXeNsNlsFBQU+NRezHz4GH3YmWl+DYUp2VWg/+MzOHba9/aeHqqu0ajJjeDTvd73LzwFV7XwfYwrkqG8HHo+A2fOw01XaYL1073weA7Uj4/l17/+DatWZLFo4SwmPH4G0EyxL38Cmetr0efW2zm0cUW1LS7DUT8zNjZWtz+0JuBPCF6NJggvIqVcIYT4BlgLbBJCDDZxPi2BH13+Pgxcb+L4EYUVwoethOv1ePPNN5nw2Hjm3GVn9ROVTV56hFhubi6JcZDqpbmA2Q8fvQ+7lStXEhcXF3RwhKvfsWHDhob8X3rGdhXo/gQYeH6oGgkKqhUD3/5Y9f1KxyjUgqQS4mDrTA8FvrMEeXmb+HZ+OfmxQJljHvHw5EC447oLdJu5zuPYoSIcAXCNGzc29fsQ6QhfFk0hRCEwWkr5rofPLgM+BMqBDOANKWWtoCYjxG+B26SUDzn+HgX0kFI+6rbdaGA0QFJSUrdVq1YFfMzi4uIaKVCcROr5nzx5klPHD9Guud3rNj8ci6Fx87Y0adKkymdFRUXUrgVNYg573f/wT4LaiS1JSkoKer5FRUVcsB2hVRPvv7fDP8Hx04IG9QR1YyXnywS2c5CSkmKoB1xRUSEFBQUkxnFxnNMlkjq1oH2ypE4sFMe0IsF+mNIy2F8UQ9Pm+o/hfu3tdtjxI3RsqdUkdae0DPYcjeGazl2IiYlx7GPn559/5sKFC5w9e5bTv5yidqwEqWlvruOUlsG+ohhKyySdWkrvxzgCkoqqNbG1oHE9cByS0jLYdRi6tIWztbTzd8fXdyYUBPs9DoTi4mLOnCnm5LECLk+yV7nWRr8PVqVPnz5fSSm7+9vOnya4A7gdqCIEpZT/E0L8BlgP/F8gk/TAYaC1y9+tgKMejr0cWA7QvXt32bt374APmJeXRzD7RzqRev6ZmZkc2zyNB+8t87rN2nWxNO81l/T09Cqf5eTkUHz2JHc0SPO6/7BXErhr/FLD18dTtGlubi5vvzKTNY95N0H1ew4GdoEHXcql5Z+HgenxdL5hEF27XedXO3RGab434WwlLffAOeg1K5bTJYI+19Tltw89w8uvZDhM608z7u7f6Y5A9HTtt2+DiYs9V+8ZsqQe94+bTt++faskx19+yTl2H40hb2cpzRrGUCfGztFTcGN7zUe393g8eTtL6d3nFhITE3j5s3Wsf9Jz/8g6sfBdAfTrVNHdwjU9QghY/Bpc8mto3T+b3ueq3ntf35lQYLPZaNsqyaOZHLRzu3NmPQ4eLjJtsZqXl8egQYNYlLWAex725Gox9n2IdPwJwbeBqUKIJlLKn9w/lFIeE0LcDOQC/UyYz5dAeyHEpcAR4B7gXhPGVUQZwfq5RowYwauvvkL+efNMQr5a5jyZlu7XBLX9AKx+wnU8eOsLKDhRQsqutyg6s9pn6Lwvv+OlzWHr02V0nRHP4AcXUzuxAXeNXxqQad3TtXet3tPrSq16z97jcXy2X1TyX8+b8wyvL5/P9lmlHgSZnftv0toxLd8Ec9bAhfJz3NI5jitiNrLvYByHT5TReUosva+pS/tm5/lvfhnbD2i1Q78/CrsWehDCLg2Wr0rRUidcV9quVLcvLFwBcMrVUoFPISilfAl4CUAIEecpCEZKeQa4VQjhp3ytf6SUZUKIR9DMrLWAFVLKXcGOq4g+gqnrCdrDJyUlhSGT65n28FmUtYAVL85m2uASyu2ar2zmiGJOFsOQJVn07NmLIUu2eo1unTqssm9z0Xotefzrec5zLMNX9Koev2OvDrWIi4sjKSkpYAuAp2vvrN4zpq8WZDL97do89/zz/MPRTFdKyZzZGcyd/Qx7srzkAro0QpZAy0Z2R/PeM46tnDmB9Ui4bCBJ13anecI2kn55jy3fnXO5Tt7H/fYwtLnE83mFyxcWzgA4qzferQ6MJMv/Vwhxv5Tyc/cPhBD3AUuAoI3WUsr1aCZWhcIrZqygk5KSHSXRgn/4nD59mlkZM6glytj8XUVU4oS/acLt3cfP0m3mFiZOeoquMxZWOt4n356nW2oZEwdW+Av1JI+7B/4YCbJo3dqbLlQVT+Zdb9f+ZDH8v0/rMeuZ6Tz88MMX31+UtYC/Pj+fflf7TnP4zRXw6GtaabcdCzyf+9oJZ+k2cx3LX15B/fr16fqrznRtsNNvC6jlm2Bbfm0KbLEsGVp5m3CmHSmtLLwYEYJ7ga1CiCxgppTyghCiOZpvbjCwOBQTVCi8YcYK2qyHz5jRD5LcsIwNk72b43p1qEWbtpdx8HBRpeMNKilh3YqJCFFhXtSTF+cevWp2KoQv8+6UqdO5T+cCwmmmHX1zKWV+SmqkXqIJwF9f4f/cV65ciZSSnbt280h/3+NengyZ62szfcYzCGD30RheXJVgqbQjpZWFB91CUEo5VAjxJzRhN0gIsQKYDvwM9JJS/idEc1QoPGLWCjrYh4/NZmPde2vY4ccc9/ueJVrUptvxbDYbkydNqGRe9Jc8DlVD542YiLdt21Z1Azd8J/fP5f5x06sIdE/X3mmm7dhSE3C+OHBcE/6tmvre7rImZ5nw2Hh6d6pF3452dnkP8gXgv/nQ59bhpDn8qJs2baJp06VK61IYqx0qpfx/Qoivgf+gCcOvgZ4mJ8wrwkS426oESrhX0Lm5ufS8KsavOW7r3lieuLeqFubJtBtI7p173t32/IqKKV1TYeRfjfX305vc7+/aO820I7rjt0D4v/fB/N/Buq99z+/bH+3MucvOkwPLsJVo3TF8jftlfm2Ofr7iYiBRTEyM0roUgMFWSkKIocD7aGkLLwCdgb87zKKKCMVqbVUijcLCQjqk+LbzXZ4M3x31XuvVvWXTtvxabNxVtS2QE2+1Y59Me4pLO/bkuqfh9U+1SMjXP4XrnoZLO/bkybSndJ2Tma28kpOT2XcijsR4zT86JNt7u6MpQ2HkjVpqg69z334ARvfV/vY3bv8FMHjocKXpKTyiWxMUQrwGjELzAU6UUp4RQvwNyAF2CyHGSyn/HqJ5KkJIpLZVsYrmqscX980hGDas8oPYff5jxj1SybQ7rNE2hixZZyjwZ3H2Qg7s3lolAlPbZyuLsxfqupdmVjJxNdO6p1K0T4adh+HT72HmnRX5fE6h5inv0FMkreu43S+Fq1vDD0Ww+TsoJ5aXlr/qd56KmokRTbAvMEBKOdaRFoGUchtwLVqy/OvmT08RapxtVYx2AAgnVtNc9XS4+PJAbZa/rD2Ifc3/pWUv8Ic//IH09HTeWPl3Qw19nSZMM+6lU3vzhd4Gv67Neg+eqGiEfFcPrarLF/+rxQ1X1CZtkCYAQRNq99+kCbVhi2BCDgxcFEfnqbVp1URcFHpOXBssx9aC747CzR0g5ZJ4ZsycFRFmfUV4MOITvFpK+Yv7m1LK80CaEOId86alqC6MtFWxig/FapqrnnSNp2dMv/gg1jt/o4E/RkyY/lIkgs3DdMdXJO+T6ekszs4k/3hFJwpveYfDpawSSetKYjzYJRwrrsuc92IuFhWAypp3u3btsNlsSjgqDEWHVhGAbp//O/jpKKqbSGurEo7WM3rQm64RyPz1Bv6YmScYTB6mNzO1L4EeV7eurrxDT5G0ruQfh//8UJvsZ//CSJdEffemtU0bZNK2VVJImtYqIgvVWb6GE2ltVcLRekYPerW2UM7f7DxBo3mYnuqC5hXU4pFxDzFoyDBeWv4qDRo08Hheeo+lV+t2T9R317zz4uxsn1ViaZ+3onpQQrCGE2ltVcLResYI/rS2UM7f7DxBo+ZYz2beMvKPw60L3qJV8mpmZMzyqHkZOZYR4WxVy4HCOighWMOJiYmJqA72oWgQW52Ecv4JCQn07NmLW+Zt4OOpnrs5BHIv9Zhj/QmbDZOh67QyVrw4G/Cueek5lhGBaVXLgcI6KCGoCLj8WDhSFMwO2KhuzJq/p2v/0rIX+N+uLdxzQ+UUhO8L4ONd0K9fT0NlwYzcX13Fu6+EmzuUMHeeOZqXHoFpdcuBIvwYSpZXRCfOlfXBw0XcNX4pzXvN5a7xSzl05NjFKEVXwpmi4Bpu7zHh2mKaqzvBzt/btW/TsjmznpnJ2idKmPu7ihSE5g3g7uvhy9nw6adbOXPmjMdx9RzD1/3VJWySoaxcf5K9GZiZ6qGITpQmqLiI3ijEcKcohLP1jBkEM39v137ROvhgR4V2mRgPo3pV3lev2S+Q+6vLzFsIXdpUr+ZlFcuBVQo7KJiqoj4AABzvSURBVKqiNEGFIcxMyA4Uo5qr1Qh0/r6ufbkdOrfxfVw9wifQ+6unYMDWvTDiuurVvMJtObBaYQdFVZQmqDCElQINwl042x/uq/9+/fqxcePGStqAkfn7uvaBFNw2egzwfn+dwmbwkjms9RRg5agLesJW/T5bT5p3h0Ex3DEjPuSWg3BbTRT+UUJQYQgVaOCfKvlyTUt4fVctxjxUSu+r69Cxhf1iXz4jydqerr2tROs9ePA4bNwJO3/U6ma6o9fsF8z9dQqTzlNn0uOyC3RpA/sLtWLYU4fBndfB7dl1eHLipGr12XqKJm3cvC2HjhwL6TxUekZkoISgwhDBhvjXBN+I++o/ex3I8nJHUetSx1bGtQHXay8lLFqvdZ93RoHe1AF6zIAnBsCcuyvqcBox+wVzf53CZvTY8YwZ/RDLVudyebLk7uvL2fAtPPMOXJoE2VkLqVu3brVXanG1HOTl5YVc8OjVqleuXElcXFxU/yasjPIJKgyhy/fjocWPEd+IzWYjJyeHzMxMcnJysNlsoTwlU3H3qdlKNEHl3g0BjPtQXa/9ovXw2mbYPhfWTITs38P76bA7E1Z9Bl2m1fFbcNvfMTzh7f660qBBA95c9Q+mTpvGT2dq0ag+jPoNFLwIO+aV8vXsc7y2bA6Lshb4nU8ko0erdjYHVv7C8KE0QYUhAq0pqcc3MnHS5EpmxPaXnAvIbBhOVq5cyQ2X2y+eY+42TVMzw4fqvPYDF8+m4EQJX3vpZP/xVPjVdGjYYwZ3paYa6poeTM1QV2w2G4uzM9k+q7TGmgL1aNUVzYGd2/i3ENQEa0p1ooSgwjBGQ/z1+kbOl5ay6tVMSwcReHsAOf2AM2bMYHSfsovbF57STJW+MOJDnThpMtu+2k7Krrd8CtabO9UhNTU1oMAhM1JQrBRABVXvW6tWrUJ+TD3pGdsPwOonKr/vbZHg7muOxEWiFVFCUOEXTw9+s1v83NBekLlgHt/Mtabm4O8BJIGcZXOYfWcZm7+r2M+sqE0nQgi6druOojOrgTKv23kTrEePHiUjI4MePXqwcuVKMjIyaNGiRZVjGLm/nrBKAJW3+9bnnlls3/Z5SAWHP63aU3NgJ54WCSrSNDQoIajwip6Vp1ktfuxlJfz6iljLaA7u+HoADVw8m8Mnytgx7wJNE2DuGi6u/kd0hwl/w9Rk7UCCV+x2O8OH3M5HH22g11Vw03VXkv/Zy7RLfZn+/W9l9XvvExNTOUQgmBQUq9R49XbfNsTamfhU+Ao7fPLtebqlljFxoHe/n+siQUWahg4VGKPwiusDZM1jxWTfW8aax4rZPqvEUGCDntJVB0/G0iHFu2YD4Uu98JdA/mDPEq67VGsImxivre6HZGsCzv1vVwJN1g4keGX4kNvZ/dUG9mRpxaxbNdX+3ZMFu7/awPAht+s+fqjmaDY2m425c2bxx1+f5R+fQc4WLVAJoE5seAs7ZD/7Ag0S6+NLCXUtKmDEvKwwhtIEFR4xsvKUUvp01OvxjRw6AftPxAPea1uGqzuEvwdQuR2uccnNmzhQ+9dZxPryJG2bDpPQ8gRb2tkfRJk3o8ErR48e5aOPNjhSNCqPldoMNk6BjukbLt4/M/A0R2dO4+4j8Pp/6vBECPMFpZSMfvgBSkvPsfk7zS/76V5NK586DLrfGd7CDnqaA7taCKxiXo5GlBBUeETPyvM3VwpGP/wAH6xf69NRr+ehPXnKJBZnZ4a9xqMn/D2AkhvBhm81TaPwlPb3mL7aa/U2KDgFcXF1yVyURaNGjSgoKOBXfnxs/iIAjQSvZGRk0OsqPxGqV8LUqVPp3bu3aVGHzjlc+/QsWjexc6ColBvbQ4cW2qJhcXYmcSHKF1yUtYCvt65h10IPLaWyoXU/oG74BIfRhYxVzMvRiBKCCo/oWXnais+yd8tqts+64NdRr+ehHVe3riX7Gvp6AEkJP57UtIz4OnBlSmWNY+JAOHgCFr5fiz/96U9+5683AtBI8Mrhw4fp1NLXMaH0Avx95Wuc3P1P06IOnXM8f/48ry+fz7cLXAVSacgCOiqsGBc8WzHSYM0pKG4YXsExeux4tn21nY5PreaqlrXoeUUZB3+O97iQsUoh8GhECUGFR/ytPG0l8MUPkl0LvTxo3Bz1eh7aVu0O4esBtGg9vPlvqpganRrHz8Xw7g79uXUPP/QAeR+tYdrgC4zuq/kUfUUA6gleadWqFbs+8/75ovVw5GfYtdBOajP9+Wp6sNlsLKrmfEE9VoyEX2D5pvAIDveFzrh+sKcA/m9zDIOH3s7Bj169qIG7WgRuHTCIgYvXsf7JEkstEiMdFRij8Ii/wIblm+CGy/UlgbvifGinp6czfPhw3nnnnYuVYYqLiy3ZHcJbJwJbCcxdDWsneakGkwbPvg/3PDDJpwB3raZzfNdb/P7GC2z+Dto+7ii5JoPr0JGRkcGW76oG5riew0dTgq9o44lwBHTosWLUrQ2Z62uHRXC4B5wtureM9RPPs2PeBb79bB0vLXvBY4WlkoMfcPhEGZ2nxDL0ufoBVQRSVEVpggqP+PNZZL1fm3tvKAfsXsfw5m8xK/XCbHz54TxpqZt216Lrped9PuD7XZtAm9R2PgW45zD+Cm0SIG1Q4IEcLVq0oH//W+k3fwMbHcLObtd8mP/4DLpeak5FG0+EI6BDj//sbCn0uXV4tQsOMwpHDF5Sj4TLBtL82u6G8zcVVVFCUOEVX+bJWwcM4oeDHxCIo95qSb96/XDu5twr6m+nZfHb+Epav6K57we834diGnSbDmNv0ZKqAxUYq997n+FDbueqtA1c2hxGT4T3t2rCoPulvvcNRkgFE9ARaHkwPf6zEhnDy6+sqHbrghmFI9ZOOEu3metY/vIKJfxMQJlDFV7x1fz15VdWBJQHZoWmvO4YyYd0NecOHDiQ/X7yH/01kNVlLrwCcr/UN543YmJieHfdh6Q/NRk7sXRqBRumwOi+cMDLPdR7Dr4IJF8w2Ea0ehrpJqekhEWA6C8cYdzVoAgMpQkq/OIt+CKQQsu5ubnceEUMm7/TTHHJjbSqKloASOhzt1y1i3bt2nH06NGAK3EEG7Fns9lYt24dP/9ylpwtla+DK+2TtTQLT+MZ0ZZsNhsv/OU5ts8qIz8WKAtNRRtXAinIbYalwF+QVVJS4PmQwRSw1qMZHzwZy4BrrFk4IhpRmqAiYCZOmsz946bTdUY8w55P8Ouol1Lyz7+/yeadZ3j7Czh2Gt7+onIACITmB+5Juzh1/BAdrriUVo3LaXuJ5/18rbr1aByeFgKuc7HtX0PXVLvH6+BkXyHE1qo8XiDakietMxQVbdwx8j0xy1Lgy4oRqKk9WA0V9GnGFYUjvBOMdq6ojNIEFQFjtNDyoqwF7N+xyS1frGoASChytzxpF3lxdr6ZW8ptC7U0gbRBnvf1JZQDSevQGwjjfO/jXbB5XxzTpleMF4i25M0U51rhpkc76NAyhv/9VM+01BQj3xM95uEel5UxceJEsrOz/WpgwdQ/dccMDTXSC0dEI0oIKoJGz4OmIgDES76YIwDk9i7m/8D9BZ98+FTl4BN3fAllowsBI4EwJ2wwIKs2Q0cMrxQEEWgxZW+mOCE0oTumL9w8ry6NGgznrpGDTI861PM90eMzuyqplA/f/z/arvpbtbUQMrOAdSQXjohGLCMEhRC/BTKADkAPKeW28M5IYSZ6Vvi/bg+3ZdUx/Qeu99i5X8KoXpU/07vq1qtx6JnLdZfBTXPrkn8ixuNDPtBefa4+TFpX3e9kMRw8WYvNr7wStgesHp/Z/47BlEGl9Lqq+qKJ9V7zlStXEhcX59NfGMmFI6IRywhBYCdwB/BSuCeiMB89K/xLm4E9uW9AP3BfwQp6jn1Zc9h1pPJ7oVh165lLh5YxNGow3KswCjT3ztUUt2hh5e2tomHoCjbaC2+M17T26moh5O+aSwmnbWdIe+IR+lxTV1fpOV8LJzN6Oir0YRkhKKXcA6jOyFGKnhV+/qkE7nnkXkPfAT05fnqOvfdYHTbvkew5Vjekq25dms5P9bhr5CCvD7tgcu+c57L7aAwvrkqwnIbh12eWDVOGVpitq6sThL9rvmg9HPlJsmPeBVKbXXC8G3zuq5k+TYVnhJ6IpupECJEHpPkyhwohRgOjAZKSkrqtWrUq4OMVFxfX6JVVdZ2/3W5nx45v6NjCTh0PS6/SMthzNIZrOncB4Oeff6asrIzY2FgaN25cpeGrk6KiQk4eK+DypMrjlpbB/qIYmjZPoVmz5h6PXRzTigT74YvH7tjpamw2GxcuXKB27do0atTI63EDxch18HZsM8ZwnmcozzUYiooKKSgooH5dSXys5HwZ2M5BSiNIalh528M/CWontiQpKUn3+Ea/976uud0OO36Eji0J+H5UJzXlmdenT5+vpJTd/W1XrZqgEGIj4ClBZ5qUco3ecaSUy4HlAN27d5e9e/cOeE55eXkEs3+kU53nv33b50x8yruz/76x09i+7fNKWt2+E04tpapJyWaz0bZVkhasUEaVwi1X2KHbmHocPFzk8dh5cdmk/pjGkCX1uH/cdPr37x/6i4D/63D/uOn07ds3pGNEwvfeZrORlpbGlvf/jymDSnnoOkioC7hZJYe9ksBd45caOp9Azt/bNV+0Dj7YAROm4LV40IurEmjSdKkltLpIuPfVSbUKQSllv+o8nsJa+HP2SzAUgm4kQMTTsTsMiuGOGfHVbgY0I+ihJgROJCYmkp2dTdtVf6PXVZ4jd6szXcDbNf/ov+cY29dOIHV0FeHHMj5BRfTjy9kvpazQ6nSGoBsJEPF07MbN23LoyLFqNw2ZEfRQUwInAqk4Eyq8XfNBJSWsWzER1fA2MrGMEBRCjAD+AjQD1gkh/iulvC3M01KEAE/O/pycHMNh/4EEiLgeOy8vL6wCw4ygh5oQOGGW1uteMs9ms+kud+aK+zW32WxMnjRBJbdHKOH30jqQUuZKKVtJKetKKZOUAKxZBBL2H0hxZkXk4a8Emr9oYm8l84yUO/NFoOXzFNbAMpqgomYTqFZnFVOZP4IpuhzKsSKJQLVebyXzts8qMS3Zvib4aKMVy2iCippNoFqd0SLe1Y0ZRZdDMZZVsNls5OTkkJmZSU5ODjabzfTxq6N1V7DaqiJ8KE1QYQkC1eqsHiBiZgNhqzUjDga9jYyDJdASc4FSE3y00YYSggrLEIxJyYoPHzOLLps5lhWoLoEeaIk5Rc1BmUMVliHaTEpGtJDqHMuVUJsjvR2zOkyUoPma953wkGDogurNV7NRQlBhOZxaXXp6OqNGjYoIzcYTZmohodBowuVfDJVA94SKIFb4Q5lDFYoQEUyh61COBVptznD5F6vTRBlJEcSK8KA0QYUiRJiphZg5ls1mo6CgoFrMkZ6obhOlpwjiH47FWCaCWBFelBBUKEKEmUnUZo6Vm5tLYpzn6iZgrjnSE9VtovTka3aWzItEX7PCXJQ5VKEIIWYmUZs1VmFhIS2a+P7pt2taQn5+vu65GSFcJkorlcxTWAclBBWKEGJmHqNZYyUnJ3P+7Emf2+z8sZxPN80mvm6saTl7rqgKKwqroISgQlENmJ3H6IzelFIajuQcMWIEr776CvnnvRd8/uoAfPHMBe5ZFpogGasXOVDUHJRPUKGIEMwqm5aYmEhKSop3/2I2TBkKnVqFPkgmWtJhFJGL0gQVigjBzCorSUnJ3D9uOl2mZXD9ZaV0bg37CmHL9zB1GEwcqG1ndlkxb/z/9u4/xu/6LuD489XVtusojsjosY60LkwDa5ikDZhsyVqYEd1WuQUVYyZsRsKc0yVlRSxCglBX6g+izkQSFtqEWYyDbJkmA2IbcA60m1MxZVidVba2wBbW68ho6r38467hVu7H9+6+38/78/m+n4/k0vtev9+71zt3vVffv16vWouCqzxnglIHDKLKyk3bbuGmm29l6ZIlnHc2XHMZ/M+fwE3vhalbgIMsKzaMRcHVLc4EpQ4YVCHotWvXcmDFSra9v0xX9GEqCq5uciYodcCgqqyULCvWZA1RaSYmQamlpha3PnToEM++sHzW5y+kykrJruhN1hCVZuJyqNQy0/Xa++ax5fzdv3yP/35h5msNf//MOA8sYMbWzzt78zngYpsjtYFJUGqZ6ffJTrH9QXjP78Njt9DXKiv9uLO3kCa5/S4KLi2ESVBqkdma5975CxN/XvQJ+KlLz+LHzutvlZXFXOhfyAGX0dFRPv6xGwcyu5V65Z6g1CKz7ZNFwF2/CFe84w28+dJfbk3T4YUecCm5Hymd5kxQapFe9skuGnmF8976VrZt29ZQVLNbzPUNa4iqNJOg1CJd3CdbzAEXa4iqNJdDpRYpeW9vofrRJHfVqlVcffXVjIyMcOTIER566CHGxsb6Har0GiZBqUW6uE+22MRt6TSV5HKo1DKl9skWWsR6sU1yLZ2mkkyCUss0vU+2kDt+Z1po4p7tSsjpk6Ubbr+TG3/9Y62a/Wp4mARVhK1z5tbvRrwz6cdMbKGJe1CFwaVemQTVqH7MOtQ//Z6JzTdxWzpNpZkE1Sj3f9ql9Eysi1dCNFw8HarG2DqnfUrNxE53yDh8+DD7nz7ZqSshGi4mQTXG1jnt0487fvNx5nWI7z51B2vPhSt30JkrIRouLoeqMe7/tE/TRaynWw7PhFv/Ci7eBu9++zLevmbc0mlqjDNBNabpWYfm1uTl/JmWw08XBv/HO+CJZ+CHL7utFYXBVYfWzAQjYhfwfuAk8J/AhzLzpbJRqZ9sndNOTV3On2s5fP0FcOUly1i3bp3XIdSYNs0EHwXWZ+YlwLOARwSHTBdLgtXg9B2/w88d45qP/vnAWjS5HK42as1MMDMfmfLwSeCaUrFocGyd016DvpzvdQi1UWuS4Bk+DDxYOgj1n61z6tWl5XArGtUjmqzQHhGPASPT/NX2zPzc5HO2AxuBD+QMwUXEDcANAKtXr96wd+/eBcd04sSJqn/51jx+x9782I8dO8q3nz/ChavHWTblv+AnT8GhY0v4kfPOZ/Xq6X5F9Nds4z927ChHjhxh1QpYvjR55VQw9n04//xmYhu0Wn7uN2/e/JXM3DjnEzOzNW/AdcCXgZW9vmbDhg25GPv27VvU67uu5vE79uaNj4/nrp078pyzX59bLj8rt75vaW65/Kw85+zX566dO3J8fLynz3P8+PHcvXt37ty5M3fv3p3Hjx+fVxwzjX/Xzh25ft3K/MY9ZD7w6ts37iHXr1uZu3bumNfXaaNafu6BA9lDDmnNwZiIuAq4GdiSmS+XjkdS/y32EE4OsPegFY3q1KY9wT8DlgOPTv5DeDIzbywbkqRBWOghnEHWni1dR1VltCYJZuaFpWOQ1F6D7j3oFY46tWY5VJJmM+jas1Y0qpNJUFInDHqmNjo6yhMH/8+OFpUxCUrqhEHP1KxoVKfW7AlK0myauGxvRaP6OBOU1AlNzNSaqqOq9nAmKKkzmpqpDbqOqtrDJCipM6w9q35zOVRSJ52uDpOvll2U5s0kKKkzBlk2TXVyOVRSZwyybJrq5ExQUidY4FqDYBKU1AmDLpumOpkEJXWCBa41CO4JSmJsbIyHH36Yo0ePMjIywujoKKtWrSod1g8YGRnhSy+uAGZe7vyPF1bwDgtcax6cCUoV69Jpy8UUuB4bG2PPnj3cfffd7Nmzh/Hx8QFHq65wJihVrEunLV8tm3bnaw7HzFQ2LTP5w12fZMddkxVmzv0+X3pxBZuvvYOvHniKrZ/4bUuhVc4kKFVq0E1qB2G+ZdNmSvKPLB1n683tSvIqw+VQqVJdPG05nwLXs12pWLbUKxWa4ExQqlSXT1v2UuB6PkneYtn1ciYoVWrQTWpL63KSV3NMglKlFnPasguGPcmrP0yCUqWaaFJb0rAnefWHe4JSxZpqUlvCbFcqTp7qfpJXf5gEpYoNe5PamZL85muXcN1Hup3k1R8mQUk9nbbsopmS/Jo1a7jiiitKh6cWMAlKGnpnJvn9+/eXC0atYhKUNG9dKLgt9cLToZJ61qWC21IvnAlK6lmXCm5LvXAmKKkns9XiPF1w21qc6hqToKSedLHgtjQXk6CknliLU8PIJCipJ9bi1DAyCUrqibU4NYxMgpJ6MuwFt1Unr0hI6tkwF9xWnVqTBCPi94CfA8aB54HrM/NbZaOSNNWwF9xWfVqTBIFdmfm7ABHxm8BtwI1lQ5I0nWEtuK36tGZPMDOPT3n4BsD6S5KkgYo21fqLiLuAXwG+C2zOzGnPoUXEDcANAKtXr96wd+/eBX/NEydOVL2MU/P4HXudY4e6x1/L2Ddv3vyVzNw41/MaTYIR8RgwMs1fbc/Mz0153i3Aisy8fa7PuXHjxjxw4MCCY9q/fz+bNm1a8Ou7rubxO/ZNpcMopubx1zL2iOgpCTa6J5iZ7+nxqZ8B/gaYMwlKkrRQrdkTjIi3TXm4BXimVCySpDq06XToJyPix5m4InEYT4ZKkgasVQdjFiIiXmAiaS7UucCLfQqni2oev2OvV83jr2XsazNzhp4nr+p8ElysiDjQy+bpsKp5/I69zrFD3eOveezTac2eoCRJTTMJSpKqZRKEe0sHUFjN43fs9ap5/DWP/TWq3xOUJNXLmaAkqVomQSbaOEXEv0bE1yLikYh4c+mYmhIRuyLimcnxPxwRbywdU5Mi4ucj4t8jYjwiqjgxFxFXRcTXI+JQRFTVADAiPh0Rz0fE06VjaVpEXBAR+yLi4OTP/G+VjqkNTIITdmXmJZn5E8AXmGjjVItHgfWZeQnwLHBL4Xia9jTwAeDx0oE0ISJeB3wK+BngYuCXIuLislE16n7gqtJBFHIK2JqZFwE/CXy0su/9tEyC1N3GKTMfycxTkw+fBN5SMp6mZebBzPx66TgadBlwKDP/KzNPAnuZaGZdhcx8HPhO6ThKyMwjmfnVyffHgIPAmrJRldemsmlFndnGqXA4pXwYeLB0EBqoNcD/Tnn8HHB5oVhUSESsAy4FniobSXnVJMG52jhl5nZg+2Qbp99giDpY9NLCKiK2M7Fc8kCTsTWh1xZelYhpPlbNyocgIs4CPgt8/IxVsCpVkwRrbuM019gj4jrgfcCVOYR3Zubxva/Bc8AFUx6/BfhWoVjUsIj4ISYS4AOZ+VDpeNrAPUHqbuMUEVcBNwNbMvPl0vFo4P4JeFtE/GhELAOuBT5fOCY1ICICuA84mJl/VDqetvCyPBARnwV+oI1TZn6zbFTNiIhDwHLg25MfejIzq2ljFRGjwJ8CbwJeAr6WmT9dNqrBioifBe4BXgd8OjPvKhxSYyLiL4FNTHRSOAbcnpn3FQ2qIRHxLuAJ4N+Y+F0H8DuZ+bfloirPJChJqpbLoZKkapkEJUnVMglKkqplEpQkVcskKEmqlklQGhIRsTEi7p/sEDEeEfeXjklqu2oqxkgVeCfwLiYKoa8qHIvUCd4TlIZERCzJzPHJ9w8AT2fm9WWjktrN5VCpAyLijRHxXETsOePjn4+IZyNi5ekEKKl3JkGpAzLzJeBXgQ9GxNUAEfEh4L3A9dZ9lRbGPUGpIzLzixFxL/AXEXEY+GPgDzLzHwqHJnWWM0GpW7YC3wO+zERbpNvKhiN1m0lQ6pDMPAF8gYnOH/dl5iuFQ5I6zSQodUhEbAQ+AvwzcGtEjBQOSeo0k6DUERGxAtgDfJGJ+4DfAe4tGpTUcR6MkbrjTmAEuDIzX46I64AnIuL6zLw/It4EvHvyuecAayPiGoDM/OsyIUvt5mV5qQMi4p3A48AHM/MzUz6+C/g1YD1wIbBvutdnZjQRp9Q1JkFJUrXcE5QkVcskKEmqlklQklQtk6AkqVomQUlStUyCkqRqmQQlSdUyCUqSqmUSlCRV6/8B/dxUSlE5P1oAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 504x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "t.plot_xy('x1','x2')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Now drop the _un-interesting_ variable"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dropped the following column: x5\n"
     ]
    }
   ],
   "source": [
    "t.drop_static_cols()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Here are the features: x1, x2, x3, x4, and y."
     ]
    }
   ],
   "source": [
    "t.show_cols()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Handling missing values in the dataset\n",
    "The `show_missing()` and `drop_missing()` methods are used to detect and drop missing values (rows of data) from the dataset before user starts modeling. Internally, they just use Pandas `isna()` and `dropna()` functions."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [],
   "source": [
    "t.df.iloc[2]['x1']=np.nan\n",
    "t.df.iloc[4]['x4']=np.nan"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "The following table shows the number of missing values in the dataset (by the features)\n",
      "\n",
      "x1    1\n",
      "x2    0\n",
      "x3    0\n",
      "x4    1\n",
      "y     0\n",
      "dtype: int64\n"
     ]
    }
   ],
   "source": [
    "t.show_missing()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [],
   "source": [
    "t.drop_missing()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "<class 'pandas.core.frame.DataFrame'>\n",
      "Int64Index: 198 entries, 0 to 199\n",
      "Data columns (total 5 columns):\n",
      "x1    198 non-null float64\n",
      "x2    198 non-null float64\n",
      "x3    198 non-null float64\n",
      "x4    198 non-null float64\n",
      "y     198 non-null float64\n",
      "dtypes: float64(5)\n",
      "memory usage: 9.3 KB\n"
     ]
    }
   ],
   "source": [
    "t.df.info()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Trying to set the output variable - spelled it wrong (`v` instead of `y`)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Following feature is set as output (to be modeled): v\n",
      "Warning! The designated output variable v is not in the list of columns of the dataset!\n"
     ]
    }
   ],
   "source": [
    "t.set_single_output('v')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Setting the correct output variable `y`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Following feature is set as output (to be modeled): y\n"
     ]
    }
   ],
   "source": [
    "t.set_single_output('y')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Setting the input variables"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Following features are set as input: x1, x2, x3, and x4."
     ]
    }
   ],
   "source": [
    "t.set_inputs(['x1','x2','x3','x4'])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Preparing the dataset\n",
    "\n",
    "This is a somewhat automated function. With a single line of code, user can instruct the software to intelligently scale all the input and output variables as per the range of their values, create test/train splits for DNN modeling, and prepare the dataset to be fed into a DNN."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [],
   "source": [
    "t.prepare_data()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Build the model\n",
    "This lets the user build a DNN model of her choice with various configurable parameters (e.g. number of hidden layers, number of neurons in each layer, activation function, etc.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<keras.engine.sequential.Sequential at 0x13b2d895940>"
      ]
     },
     "execution_count": 41,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t.build_model(n_hidden_layers=2,\n",
    "              layer_architecture=[20, 20])"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Train (fit) the model\n",
    "\n",
    "This lets the user train the model with configurable parameters (number of epochs, learning rate, batch size, etc.)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Epoch 1/30\n",
      "138/138 [==============================] - 0s 3ms/step - loss: 0.0608 - mean_squared_error: 0.0608\n",
      "Epoch 2/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 0.0075 - mean_squared_error: 0.0075\n",
      "Epoch 3/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 0.0051 - mean_squared_error: 0.0051\n",
      "Epoch 4/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 0.0036 - mean_squared_error: 0.0036\n",
      "Epoch 5/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 0.0028 - mean_squared_error: 0.0028\n",
      "Epoch 6/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 0.0023 - mean_squared_error: 0.0023\n",
      "Epoch 7/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 0.0019 - mean_squared_error: 0.0019\n",
      "Epoch 8/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 0.0015 - mean_squared_error: 0.0015\n",
      "Epoch 9/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 0.0014 - mean_squared_error: 0.0014\n",
      "Epoch 10/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 0.0012 - mean_squared_error: 0.0012\n",
      "Epoch 11/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 9.0884e-04 - mean_squared_error: 9.0884e-04\n",
      "Epoch 12/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 7.7148e-04 - mean_squared_error: 7.7148e-04\n",
      "Epoch 13/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 6.4336e-04 - mean_squared_error: 6.4336e-04\n",
      "Epoch 14/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 4.8863e-04 - mean_squared_error: 4.8863e-04\n",
      "Epoch 15/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 4.1388e-04 - mean_squared_error: 4.1388e-04\n",
      "Epoch 16/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 4.0460e-04 - mean_squared_error: 4.0460e-04\n",
      "Epoch 17/30\n",
      "138/138 [==============================] - 0s 2ms/step - loss: 3.2259e-04 - mean_squared_error: 3.2259e-04\n",
      "Epoch 18/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 2.3985e-04 - mean_squared_error: 2.3985e-04\n",
      "Epoch 19/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 2.1647e-04 - mean_squared_error: 2.1647e-04\n",
      "Epoch 20/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 2.2491e-04 - mean_squared_error: 2.2491e-04\n",
      "Epoch 21/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 1.8680e-04 - mean_squared_error: 1.8680e-04\n",
      "Epoch 22/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 1.5697e-04 - mean_squared_error: 1.5697e-04\n",
      "Epoch 23/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 1.5254e-04 - mean_squared_error: 1.5254e-04\n",
      "Epoch 24/30\n",
      "138/138 [==============================] - 0s 2ms/step - loss: 1.5703e-04 - mean_squared_error: 1.5703e-04\n",
      "Epoch 25/30\n",
      "138/138 [==============================] - 0s 2ms/step - loss: 1.4329e-04 - mean_squared_error: 1.4329e-04\n",
      "Epoch 26/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 1.2429e-04 - mean_squared_error: 1.2429e-04\n",
      "Epoch 27/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 1.0929e-04 - mean_squared_error: 1.0929e-04\n",
      "Epoch 28/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 1.2864e-04 - mean_squared_error: 1.2864e-04\n",
      "Epoch 29/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 1.0930e-04 - mean_squared_error: 1.0930e-04\n",
      "Epoch 30/30\n",
      "138/138 [==============================] - 0s 1ms/step - loss: 1.0978e-04 - mean_squared_error: 1.0978e-04\n"
     ]
    },
    {
     "data": {
      "text/plain": [
       "<keras.callbacks.History at 0x13b2d8c1dd8>"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t.train_model(lr=5e-4,n_epochs=30,verbose=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Plot the training error"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAkEAAAFaCAYAAAAZ/YloAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADl0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uIDIuMi4yLCBodHRwOi8vbWF0cGxvdGxpYi5vcmcvhp/UCwAAIABJREFUeJzs3Xl8VOXZ//HPlYUlhMhmErR1waq11f5ckFYfF1ARqyCCImKpYn20LnXBtvaxdW2t2s2tbsX6iIgibqCgggsGLVq2R607FnckYVUIYQnJ9fvjTGgcZiZnJpmFzPf9es0rmXPuc841t1Ny9V7N3RERERHJNwXZDkBEREQkG5QEiYiISF5SEiQiIiJ5SUmQiIiI5CUlQSIiIpKXlASJiIhIXlISJCJxmdkNZrYhxWs7mZmb2f+0dVzSstb8txPJF0qCRLYRkYQizOvqbMeaCyJJQPN6qTezT83sb2ZWHqP8PyPlPjYzi3G+t5ltjpS5LepcLzP7i5m9Y2Z1ZrbSzN4wszvM7FvNyh3Twn+7y9NTG8kxs6PM7GozK8l2LCLpVJTtAEQktB9HvT8b+C/g9Kjj/2rDZ14OXJXKhe6+wcw6A/VtGE8qzgVqgRKC+vpv4BAz29fdo2PbAOwMHAq8FHXuVILPUtj8oJn1Av4P2A64D3gLKAO+A4yK3OffUfe6FZgfI9bXk/lgaXQU8CvgNqAuy7GIpI2SIJFthLtPbP7ezI4CDo4+Hk+kdaOTu69P4pmbgc1JBfr163OhO+ZRd18R+X2cma0GLgKOAaZFlV1M0EI+mq2ToB9Hyo+IOn4O8E3gcHf/2jVmdi5QGiOml9390WQ/iIi0LXWHibRDzcbj3GZmI83sX8BGIq1GZvbfZvaCmVWb2UYzW2xmvzWz4qj7bDWuJHLNdDM7ONKFtCHSzfSzODH8T7Nj50SOHWxmfzKzGjNbb2bPmNlOMT7HyWb2duQZ75nZ6W0w1uXlyM9vxTn/ADDCzDo2i2Nv4P8BsRLObwGNwD+iT7j7Rndf2YpYv6ZZ/R0a+W+73MzWmdk0M9sl5D1+amZvRup0mZmNN7Pezc4/RNAKBLC8WVddZVt9DpFcoZYgkfZtAHAKcDvwBfB25PiFBF0vTwPrgMMIur52BM4Mcd/dgCnAvcAEgq6iv5rZm+4+O8T1t0aeey1QCfycoCtpQFMBMxsKPAS8A/wG6AL8GVgS4v6J7BL5uSrO+YnA74DBwGORYz8GlgIvxCj/McH/oRwD/G/IGLpGutGifRlpfWvJXwm67q4FehP896wys++5+5p4F0XGi10FVAHjCLr+fgYcamb7u/tXBF1gXQg+f1NXIsCXIeIS2aYoCRJp374N7Ovub0Yd/4G7Nx/rcZeZfQj8j5ld5u7LQtz3CHd/EcDMxgOfEYy3CZMErQGOcvfGyPVrgevNbDd3Xxwp8weCxOPgpj/sZjaFIHlLZpxRj8g45xLgYODXBAnY07EKu/snZvYPgi6xx8ysgCDJmwQ0xLjkLoJE4h4zu4zg8/8DeDpBPcZLlg4lRotSDAUE3W8bAcxsDvAkMBa4JtYFkdaeXxMkcoPcvSFy/FXgUYJE9Ep3/4eZDSZIgpp3JYq0O+oOE2nfXo6RANGUAJlZoZl1i7RKzCYY9LtviPu+25QANbvfAqBPyLjuakqAIpoSpz6RuPoAewL3NW/ZiHyWWSGf0eR9YDnwCUEisww41t2XJ7hmInCsmXUnaJ36BrG7wnD3auAA4G8EA6LPJGghW2pm48ysS4zLfgsMjPHa6r9VHHc2JUCRGKYRjGcanOCaQUAxcGNTAhS59rEQ14q0S2oJEmnfFsc6aGaHAL8HfgB0iDrdLcR9P4lxbDWw1biekNevjvzsEfm5c+TnBzGu/YCgxSSsEwlannoSDGL+Hi3PeHqYoMvuZOAg4G13f83MOsUq7O6fRO59TmRszuHAxcBZBOOFzom65E13fz6JzxDt/RjHFgEHJrhml8jP92Kcexc4pBXxiGyTlASJtG9bzQQzsz2A5wn+GI4FPiUYX7ILcDfhWohjdQsBbLW+ThquD/uMJi81demY2VSCqekPm9lezVtTmnP3L83sKYLuvT2B68I+zN0/Bj42s4cJ6ni0mZ3r7p5k3AkfE+NYsvUSfW1bxieyTVB3mEj+GQZ0BI5x9zvcfXqkVaKlcUCZ1NRStHuMc7GOhRJJen4N7Eow6DeRiUBfginuD6TwrPUE3VtdCNYQakvfjnFsd4JB2vE0nYt17Z5R1yohkrygJEgk/zSNxdnyv38zKwQuyU44W3P3Dwm6d043s7Km42a2D81mkKV47+kEyckvm0+Dj+Ep4GrgAnf/LF6hyHT/rdYCMrPtgX5Atbu39cyqc6Km8A8hmLH3VIJrZhIMKL84Mti76dphBNP8m6+ZtC7yM0zXqMg2S91hIvnnaYLxQDPNbBzBYNlRRK2EnAP+B3gceMXM/pegReZnBCtix2rNSMafCabkjyEY0LwVd99EnJlWUX4CnGRmTwALCRKInQnWZNqeoEst2qFxxhd96O6vhHimA7PN7EGCKfIXEczOuznuBe5Lzew6ginyz0Vm2u0MXAB8CNzYrPjCyM8/RsrVA1NzZPFLkTajJEgkz7j725H/9/87gmnoXwKTgfv5zx+/rHP3KWY2iuCP9vUEf6jHEgxU3qWVt59EsMbOr8zsnpBr88TzV4I1h44AjiVoPVlNMFvup+4+I8Y1F8a51wNAmCToAmAkcAXB1P9ZwIUttTi5+9VmVk2QTP6FYMD4ZOB/ImsENZkB/IlgfaQTCMYM9QaqQ8Qmss2wth2rJyKSXmY2E9jB3ffJdiyZZmbnAHcCB7r7gmzHI7Kt05ggEclJZlYcGavU/Nj3CDb3THatIBGRrag7TERy1c7ADDN7APicYPDuuQRdOH/JZmAi0j4oCRKRXLWSYIzSmQQDjOuAF4HfuPun2QxMRNoHjQkSERGRvKQxQSIiIpKX8r47rFevXr7LLrvEPLdu3Tq6dIm196G0RHWXOtVdalRvqVPdpU51l7p01d3ChQtXuPv2YcrmfRK0yy67sGBB7JmmVVVV9O/fP7MBtROqu9Sp7lKjekud6i51qrvUpavuzCzWBs8xqTtMRERE8pKSIBEREclLWUmCzOw8M/vIzDaY2UIzOzRB2eFm9qyZLTeztWY218yOjyozxsw8xivW3jwiIiIimU+CzGwkcAtwHbAfwT45z5jZTnEuOZxgddjjIuWfBqbESJzqCPa22fLSZn8iIiISTzZagi4Bxrv73e7+rrtfACwlWAl2K+5+kbvf4O7z3P3f7n4NwQJqJ2xd1Kubv9L7MWLGyty5c/nxySPpXlpKYUEB3UtLOW3kKcybNw+tySQiIpI7MpoEmVkH4ADg2ahTzwIHJ3GrrgS7NDfX2cw+MbPPzWy6me3XilCTVl9fzxmjRzPi6KOpeGk2UysreX33PZhaWUn57CpOGjiQM0aPpr6+PpNhiYiISByZbgnqBRQCNVHHa4DKMDcws/OBbwD3Nzv8PvATYCgwCtgAzDGz3VsbcBjuzlljxvDhzJlMqajkzG7dKS8qpsiM8qJizuzWnSkVlSyeOZOzxoxRi5CIiEgOyOi2GWa2A7AEOMzdX252/CpglLt/u4XrTyRIfk5x9ycTlCsEXgdedPcLY5w/GzgboKKi4oCHHnoo5n1qa2spLS1t8XO98847/P7SS3lyhx0pKYifV9Y1NnL8F0u4/E9/Yq+99mrxvtuysHUnW1PdpUb1ljrVXepUd6lLV90NGDBgobv3DVM204slrgAa2LrVp5ytW4e+plkCdFqiBAjA3RvMbAEQsyXI3ccB4wD69u3r8RZrCruQ0//eeRejSksTJkAAJQUFjCot5Z8vvcS558YcAtVuaAGx1KnuUqN6S53qLnWqu9TlQt1ltDvM3TcRDGoeGHVqIMEssZjM7GRgIjDG3R9t6TlmZsD3CAZcp920p6YzJGQ2O7hLKdOmT09zRCIiItKSbGybcSNwv5nNA+YA5wA7AHcBmNkEAHc/LfL+FIIWoF8AL5lZUyvSJndfFSlzFfBP4AOgDLiQIAnKSHPLmro6ehSGq8oeRUWsqatLc0QiIiLSkownQe4+2cx6ApcTrOfzFnCsuzft9RG9XtA5BHHeHHk1mQ30j/zejaB7qxL4CniNYNzRvHR8hmhlJSWsathMeVFxi2VXbd5MWUlJBqISERGRRLKygaq73wHcEedc/0Tv41wzFhjbFrGlYshxg5k2u4ozu3Vvsez0dbUMGTw4A1GJiIhIIto7rA2cf8lYHl6/nrrGxoTl1jU2MrmujvPHZi1fExERkQglQW2gX79+HDF4MBevWhk3EVrX2MjYVSs5csgQDjzwwAxHKCIiItGUBLUBM+Pu8ePZbdAghtVUc8+Xq6mpr6fenZr6eu75cjXDqpey26BB3D1+PMHkNREREckmJUFtpLi4mHsnTuTR555j2eH9OX7J5+y76H2O++xTlg8YwOMvvMD4Bx6guLjlwdMiIiKSfkqC2pCZ0a9fPyZMfoj7J0/GgSMGDeK+SZPUBSYiIpJjlASlSWVlsJzR0qUZWa9RREREkqQkKE2akqDq6uosRyIiIiKxKAlKk4qKCgBqampobGHqvIiIiGSekqA06dSpE926dWPz5s2sWrUq2+GIiIhIFCVBaaQuMRERkdylJCiNlASJiIjkLiVBaaQkSEREJHcpCUojJUEiIiK5S0lQGikJEhERyV1KgtJISZCIiEjuUhKURkqCREREcpeSoDRSEiQiIpK7lASlUe/evQElQSIiIrlISVAa9ezZk8LCQlauXMmmTZuyHY6IiIg0oyQojQoLCykvLwdg2bJlWY5GREREmlMSlGYaFyQiIpKblASlmZIgERGR3KQkKM2UBImIiOQmJUFppiRIREQkNykJSjMlQSIiIrlJSVCaKQkSERHJTUqC0kxJkIiISG5SEpRmSoJERERyk5KgNFMSJCIikpuUBKVZ165d6dy5M+vWrWPt2rXZDkdEREQilASlmZmpNUhERCQHKQnKACVBIiIiuUdJUAYoCRIREck9SoIyQEmQiIhI7lESlAFKgkRERHKPkqAMUBIkIiKSe5QEZYCSIBERkdyjJCgDlASJiIjkHiVBGaAkSEREJPcoCcqAiooKAGpqamhsbMxyNCIiIgJKgjKiY8eOdO/enYaGBlauXJntcERERAQlQRnTu3dvQF1iIiIiuSIrSZCZnWdmH5nZBjNbaGaHJig73MyeNbPlZrbWzOaa2fExyp1oZu+Y2cbIz2Hp/RTJ0bggERGR3JJUEmRmHcxsRzPbzcy6p/JAMxsJ3AJcB+wHvAI8Y2Y7xbnkcGAWcFyk/NPAlOaJk5kdBEwGHgD2jfx8xMy+n0qM6aAkSEREJLe0mASZ2XfN7A9mthCoBT4FFgErzGyZmU01s9Fm1jnkMy8Bxrv73e7+rrtfACwFzo1V2N0vcvcb3H2eu//b3a8BFgInNCt2MfCiu/8+cs/fA1WR4zlBSZCIiEhuiZsEmdl/mdmLwL8IWmNmA2cBxwODgJHAH4ENwE3AF2Z2hZmVJrhnB+AA4NmoU88CBycRd1dgdbP3B8W458wk75lWSoJERERyS1GCc48DtwI/dvfPE93EzAqBo/hPy8vv4hTtBRQCNVHHayLXt8jMzge+Adzf7HBlnHtWxrnH2cDZEExfr6qqivms2trauOeStXp1kLO98cYbbXbPXNaWdZdvVHepUb2lTnWXOtVd6nKh7hIlQTu7+4YwN3H3BoKWl5lm1inMJVHvLcaxrZjZicCfgFPc/ZNU7+nu44BxAH379vX+/fvHfF5VVRXxziWrvr6e66+/Hndvs3vmsrasu3yjukuN6i11qrvUqe5Slwt1F7c7LGwClOR1K4AGtm6hKWfrlpyviSRA9wOnufuTUaerU7lnJqk7TEREJLckagnawswOS3C6EVgDvO/uGxPdx903RQZYDwQeaXZqIPBYguefDNwHnO7uj8Yo8mrkHn+KuucrieLJJCVBIiIiuSVUEkQw06p511KsrqYNZvZ34JJI91g8NwL3m9k8YA5wDrADcBeAmU0AcPfTIu9PIWgB+gXwkpk1tfhscvdVkd9viZy7DJgCDAMGAIeE/Hxp17NnTwoLC1m1ahUbN26kY8eO2Q5JREQkr4VdJ+gogqnxdwHHAn0jP/8GfAacDFxPMHvst4lu5O6TCQZQXw68TpCoHNtsjM9OkVeTcwiStZsJptI3vR5vds9XgFOA0wlms50GjHT3uSE/X9oVFBRs2UNs2bJlWY5GREREwrYE/Qy4z92vjjo+08yuBsa4+xAzKwLGAL9JdDN3vwO4I865/oneJ7jno0CsrrKcUVlZyRdffEF1dTXf/OY3sx2OiIhIXgvbEnQ08I845+YQdD0BvAT0bm1Q7ZXGBYmIiOSOsEnQKoJFEmM5PnIeoAT4qrVBtVdNSdDSpUuzHImIiIiE7Q77I3Crme0CTAOWA9sDQwnGBl0QKTcAmN+2IbYfagkSERHJHaGSIHe/zcyWAJcBt0eu20wwsHm4u0+NFL0e2JSOQNsDJUEiIiK5I2xLEO4+hWD39gKCVqDl7t4YVWZFG8fXrigJEhERyR2hk6AmkcQnZ1Zi3pYoCRIREckdoZMgM+sLDCfYvDR6fzB395FtGVh7pCRIREQkd4TdNuNc4DZgJfABGveTkuZJkLtjZlmOSEREJH+FbQn6BXAvcI67b05jPO1aaWkpJSUl1NXVsXbtWsrKyrIdkoiISN4Ku05QOTBJCVDrmJm6xERERHJE2CToGeD76QwkXygJEhERyQ1hu8NuB8aZWTHwHPBldAF3f6ctA2uvevcOdhVREiQiIpJdYZOgFyM/rwKujDpngAOFbRVUe6aWIBERkdwQNgka0HIRCUNJkIiISG4Iu23G7HQHki+UBImIiOSGsAOjpY0oCRIREckNcVuCzGwZMMjdXzOz5QTjfuJy9/K2Dq49UhIkIiKSGxJ1h93Of/YIu50WkiAJR0mQiIhIboibBLn7Nc1+vzoj0eSB8vKgwWzZsmU0NDRQWKhJdSIiItmgMUEZ1qFDB3r27ElDQwMrV67MdjgiIiJ5K5ld5E8i/i7yuHu/NoyrXausrGTlypVUV1dvaRkSERGRzArVEmRmVwMPA3sBnwFvx3hJSBoXJCIikn1hW4LOBG5w91+nM5h8oSRIREQk+8KOCeoKvJDOQPKJkiAREZHsC5sEPQQck85A8omSIBERkewL2x32AvAHM+tF/F3kn27LwNqzpiRo6dKlWY5EREQkf4VNgiZHfu4CnB7jvHaRT4JagkRERLIvbBK0a1qjyDNKgkRERLIv7C7yn6Q7kHyiJEhERCT7Em2gWuLudU2/t3SjprLSsh49elBUVMSXX37Jhg0b6NRpq7UnRUREJM0SzQ5ba2ZNq0DXAmtbeElIBQUFVFRUAFBTU9NCaREREUmHRN1hPwEWN/tdu8i3ocrKSpYsWUJ1dTU777xztsMRERHJO4l2kb+v2e/jMxJNHtG4IBERkezSLvJZoiRIREQku5LZRX4kcBawB7F3kdd26ElQEiQiIpJdYXeRPxW4D/g38A3gSWB65Po1wG3pCrC96t27N6AkSEREJFvCdof9EvgdcH7k/R3u/hOCRRRXAJoenyS1BImIiGRX2CRod2COuzcADUAZgLuvBf4A/Cw94bVfSoJERESyK2wS9BXQMfL7EmCvZucM6NmWQeUDJUEiIiLZFXZg9ALge8BMgvFAV5rZZmATcCUwNz3htV9NiyVWV1fj7phZliMSERHJL2GToOuBphX9roz8fgfBzvHzgbPbPrT2rbS0lNLSUmpra1mzZg3bbbddtkMSERHJK6G6w9z9n+4+OfL7l+4+FCgFurn79939w2QeambnmdlHZrbBzBaa2aEJyvY2swfN7D0zazCz8THKjDEzj/HK6U251CUmIiKSPS0mQWbWycw2mtkJzY+7+0Z3X5PsAyPrDd0CXAfsB7wCPGNmO8W5pCPBDLQbSNztVgf0bv5y9w3JxpdJSoJERESyp8UkKJJILAM2t9EzLwHGu/vd7v6uu18ALAXOjfP8j939wsjWHasSh+rVzV9tFG/aKAkSERHJnrCzw/4GXGhmxa15mJl1AA4Ano069SxwcGvuDXQ2s0/M7HMzm25m+7XyfmmnJEhERCR7wg6M7gbsDXxsZi8ANXx9V3l391+FuE8vgsHUNVHHa4CjQsYSy/sEO92/AXQFLgLmmNn/c/cPogub2dlEBnNXVFRQVVUV86a1tbVxz7WF9evXAzBv3ry0Picb0l137ZnqLjWqt9Sp7lKnuktdLtRd2CToRGBj5PdYg5gdCJMENS/fnMU4Fv5m7q8Cr265mdkrwOvABcCFMcqPA8YB9O3b1/v37x/zvlVVVcQ71xYWL17MPffcQ4cOHdL6nGxId921Z6q71KjeUqe6S53qLnW5UHehkiB337WNnreCYMXpyqjj5WzdOpQyd28wswUEK13nLHWHiYiIZE/YDVRPM7OYq0KbWQ8zOy3Mfdx9E7AQGBh1aiDBLLE2YcHKg98jGHCds5QEiYiIZE/Y7rB7gYOAlTHO7Ro5PyHkvW4E7jezecAc4BxgB+AuADObAODuWxIrM9s38msZ0Bh5v8nd34mcvwr4J/BBpMyFBElQzBlnuaIpCVq6NKdzNRERkXYpbBKUaE+HnkDo9YLcfXKkVelygvV83gKOdfdPIkVirRf0WtT7IcAnwC6R990IxvhUEuxz9hpwmLvPCxtXNpSXlwOwfPlyGhoaKCwszHJEIiIi+SNuEmRmQ4GhzQ5dYWbLo4p1IhgoPT+Zh7r7HQTbbsQ61z/GsYQba7n7WGBsMjHkguLiYnr16sWKFStYvnz5lpYhERERSb9ELUHlwD7N3u/G1gOaNxGs8XNtG8eVNyorK1mxYgXV1dVKgkRERDIobhLk7ncDdwOY2YvAue7+XqYCyxeVlZW89dZbGhwtIiKSYWGnyA9IdyD5SjPEREREsiPuFHkzu8LMtkvmZmZ2hJkNaX1Y+UNJkIiISHYkWieoH/CZmd1vZkPNbPvoAmZWbGb7m9lvzOxfwAP8Z2VpCUFJkIiISHYkGhM0xMy+T7D1xINAJzNbQbDq80aCaek7AMXA28D/AuPcvS7tUbcjSoJERESyI+GYIHefC8w1s1Lgv4D9CWaIdQJWEWxcOifWJqUSTu/evQElQSIiIpkWdmB0LTAz8pI2pJYgERGR7Ai1d5ikj5IgERGR7FASlGXdu3enuLiYr776ivXr12c7HBERkbyhJCjLzGxLa1BNTU2WoxEREckfSoJygLrEREREMq/FJMjMOprZj8xs90wElI+UBImIiGRei0mQu28E/k6wJpCkgZIgERGRzAvbHfYmsEc6A8lnSoJEREQyL9Q6QcBYYLyZLQVmuPvmNMaUd5QEiYiIZF7YJGgqUAI8AbiZrQa8eQF3L2/j2PKGkiAREZHMC5sE3U5U0iNtR0mQiIhI5oXdNuPqNMeR15QEiYiIZF7YliAAzKwDsA/Qg2AD1TfdfVM6AssnFRUVQJAEuTtmluWIRERE2r/QiyWa2aVADTCPYCPV+UCNmf0yTbHljS5dutC1a1c2btzIV199le1wRERE8kKoJMjMLgauBx4EBgB7Af0j7683swvTFWC+UJeYiIhIZoVtCTofuMHdz3f3l9z9/cjP84E/AEqCWqkpCVq6dGmWIxEREckPYZOgbwIvxjlXBXyjTaLJY2oJEhERyaywSdCnwNFxzg2MnJdWUBIkIiKSWWFnh90K3GpmPYBHCQZIlwMjgDGoO6zVlASJiIhkVth1gm4zs43AVcBPCBZONOAL4Bx3/3v6QswPSoJEREQyK/Q6Qe5+t5n9nWD8T29gKfC5u2sl6TagJEhERCSzWhwTZGadzGyRmR3jgc/cfV7kpxKgNqIkSEREJLNaTILcfQPQDWhMfzj5S0mQiIhIZoWdHfYAcEY6A8l35eXlmBnLly9n8+bN2Q5HRESk3Qs7JuhT4GQzWwA8TTA7rHlXmLv7nW0dXD4pKipi++23Z9myZSxfvpzevXtnOyQREZF2LWwS9JfIz97A/jHOO6AkqJUqKytZtmwZ1dXVSoJERETSLFR3mLsXtPAqTHeg+UDjgkRERDIn7Oywu83sB5kIKJ8pCRIREcmcsLPDTgE6pT+c/KYkSEREJHPCzg6bBQxIZyCiJEhERCSTwg6Mvh34u5l1IfbsMNz9nTaOLe8oCRIREcmcsEnQjMjPSyKv5gmQRd5rcHQrKQkSERHJnLBJkLrCMkBJkIiISOaE3UV+droDESVBIiIimRR2YDQAZvZDM7vCzMaZ2U6RY4eZ2Q5J3uc8M/vIzDaY2UIzOzRB2d5m9qCZvWdmDWY2Pk65E83sHTPbGPk5LJmYckG3bt3o0KEDa9asoa6uLtvhiIiItGuhkiAzqzCzucA04HTgTKBX5PQZwBVhH2hmI4FbgOuA/YBXgGeakqoYOgIrgBuAuXHueRAwmWCPs30jPx8xs++HjSsXmNmW1qCamposRyMiItK+hW0J+itQCnw78rJm554HjkzimZcA4939bnd/190vAJYC58Yq7O4fu/uF7j4eWBXnnhcDL7r77yP3/D1QFTm+TVGXmIiISGaETYKOAS53938TNTUe+BzYMcxNzKwDcADwbNSpZ4GDQ8YSy0Ex7jmzlffMCiVBIiIimZHMmKCGOMd7AetD3qMXwVT66L6eGqAyiViiVabhnlmhJEhERCQzwk6Rfxm4wMyeanasqUXoJwQrSicjujXJYhxLVuh7mtnZwNkAFRUVVFVVxbxhbW1t3HPpsmHDBgBeffVV9tprr4w+uy1lo+7aC9VdalRvqVPdpU51l7pcqLuwSdCvgH8AbwFTCJKLs8xsb2BvIOzmqisIWpSiW2jK2bolJxnVydzT3ccB4wD69u3r/fv3j3nTqqoq4p1Ll3fffZcJEybQqVOnjD+7LWWj7toL1V1qVG+pU92lTnWXulyou1DdYe7+FsFYngXAGIJEZjjwGfB9d18U8j6bgIXAwKhTAwlmiaXq1TTcMyvUHSYiIpKA1EloAAAgAElEQVQZYVuCcPfFwI/b4Jk3Aveb2TxgDnAOsANwF4CZTYg877SmC8xs38ivZUBj5P2mZvuV3QK8ZGaXEbRUDSNY5fqQNog3o5QEiYiIZEboJKituPtkM+sJXA70JuhiO9bdP4kUibVe0GtR74cAnwC7RO75ipmdAlwLXAMsBka6e8x1hXKZkiAREZHMyHgSBODudwB3xDnXP8Yxi1E0usyjwKOtDi7LKioqgCAJcnfMWvzoIiIikoKkts2Q9CspKaGsrIz6+npWr16d7XBERETaLSVBOah3796AusRERETSSUlQDnF35s6dy7rVq+lgxj5770330lJOG3kK8+bNw721SymJiIhIk7hJkJkdbGZdWrqBmfU0s1PbNqz8U19fzxmjRzPi6KM52Z1n+/Th9d33YGplJeWzqzhp4EDOGD2a+vr6bIcqIiLSLiRqCXoZ+G7TGzMrNLMGM9s/qty3gPvTEVy+cHfOGjOGD2fOZEpFJf/doyflRcUUmVFeVMyZ3bozpaKSxTNnctaYMWoREhERaQOJkqBY05I0VSkN5s2bx6zp07mpR09KCmL/JykpKODmHj2ZNX068+fPz3CEIiIi7Y/GBOWA22+8iZM7d46bADUpKSjg5M6duf2mmzIUmYiISPulJCgHTHtqOkNKS0OVHdyllGnTp6c5IhERkfavpSQo1uATDUhpY2vq6uhRGG7dyh5FRaypq0tzRCIiIu1fS395x5vZuqhj95tZ87/CLc4gk8TKSkpY1bCZ8qLiFsuu2ryZspKSDEQlIiLSviVKgu6LceztOGXntUEseWvIcYOZNruKM7t1b7Hs9HW1DBk8OANRiYiItG9xkyB3PyOTgeSz8y8Zy4gZzzCqbLuEg6PXNTbyUF0dj48dm8HoRERE2icNjM4B/fr144jBg7l41UrqGhtjllnX2MhPP/8M79iRPfbYI8MRioiItD+JVoz+jpmdEuP4D81soZmtM7MPzezi9IbY/pkZd48fz26DBjGsppp7vlxNTX099e7U1Ndzz5erGbr0Cz4qKOCLFSsYOHAgq1atynbYIiIi27RELUFXAD9pfsDMDgCeAHoBdwJvAX8xsxFpizBPFBcXc+/EiTz63HMsO7w/w5fVsP+/P2D4shqWDxjA1FmzeO3tt+nTpw8LFizgyCOPZMWKFdkOW0REZJuVaGD094Hro45dBDQAh7r7pwBm9vfI8UfSEmEeMTP69evHhMkPxS3z0ksvccQRR/D6668zYMAAnn/+eSoqKjIYpYiISPuQqCWoEvgg6tixwOymBCjiMUCDVDJkxx13pKqqir322ou33nqL/v37s3Tp0myHJSIiss1JlAR9CWyZs21mewE9gJeiyq0DtHBNBvXu3Zuqqir22Wcf3nvvPQ4//HA+//xz3J25c+fy45NH0r20lMKCArqXlnLayFOYN2+eNl4VERFpJlESNBc438yausx+SrBa9JNR5fYClqQhNkmgvLycF198kf32248PPviAww47jJHDhjHi6KOpeGk2UysreX33PZhaWUn57CpOGjiQM0aPpr6+Ptuhi4iI5ISWBkZ/H1hqZh8AFwIPu/tbUeVOBf6RpvgkgZ49e/LCCy/Qt29flnz8MR899xxTKio5s1t3youKKTKjvKiYM7t1Z0pFJYtnzuSsMWPUIiQiIkKCJCiS7OwLjANmAv8NjG5exsy2B94A/pbGGCWB7t27c8MNN1BWVMRdO34j7mKLJQUF3NyjJ7OmT2f+/PkZjlJERCT3JNw7zN0XA79JcH45QQuRZNF94+7m9O49Eq42DUEidHLnztx+0030mzQpQ9GJiIjkJq0Y3Q5Me2o6Q7qWhio7uEsp06ZPT3NEIiIiuS9uS5CZfZjMjdy9T+vDkVSsqaujR2HCRr0tehQVsaauLs0RiYiI5L5Efzl3AdYC04EvMhKNpKSspIRVDZspLypuseyqzZspK9GKBiIiIomSoKuAk4GRwMvAJOBRd9emVTlmyHGDmTa7ijO7dW+x7PR1tQwZPDgDUYmIiOS2RLPDfufu+xDMEJsDXEowXf5pM/uxmXXNVJCS2PmXjOXh9evj7kDfZF1jIw+sXcv5Y8dmKDIREZHc1eLAaHd/y90vd/dvAYcC7wLXATVmdlW6A5SW9evXjyMGD+biVSvjJkLrGhv56eefsWr9eiZNmqRFE0VEJO8lOzvs/4BZBN1jHQhWi5YsMzPuHj+e3QYNYlhNNfd8uZqa+nrq3ampr+eeL1czrHopHb/zHRoKCrj55psZMGAAS5ZooW8REclfLSZBFjjCzO4GqoGJwCZgMPCjNMcnIRUXF3PvxIk8+txzLDu8P8OX1bD/vz9g+LIalg8YwOMvvMArCxbw8ssvs+OOOzJnzhz2339/Zs2aBaB9x0REJO8kmiJ/EDAKOAnoCjxFsGr00+6+KTPhSTLMjH79+jFh8kNxyxx00EG89tprnHrqqTz//PMMHDiQq6++msXvvsusp57i5M6dmVpZSY/CIlY1bGba7CpOmvEMRwwezN3jx1Nc3PIMNBERkW1BotlhcwimyE8DniDYLR7gKDPbqrC7P93m0UlabL/99syYMYNrrrmG3/3ud1x71VXs26ULU3rv8LVVp5v2HRtVth0XR/Ydu3fiRGL99xcREdnWtLTCXleCDVJHAYn+8jlQ2FZBSfoVFhby29/+lu7du3PtLy/ljqgEqLmmfceGRfYd69evX4ajFRERaXuJkqBdMxaFZM1r/5zLGT17at8xERHJO3GTIHf/JJOBSHZMe2o6UysrQ5Ud3KWU4dp3TERE2olWb6BqZgPM7Jm2CEYyT/uOiYhIvkr418/MugHHAN8EPgSedPf6yLkRwK+A/YFFaY5T0kT7jomISL6K2xJkZvsQrA79IPAH4BHgVTPb2czmAA8BHQnWCvpOBmKVNBhy3GCm1daGKvvEmq/o2qULzz333FbrBkWvM3TkEUdonSEREclpibrDrgPWAAcBJQSrQ68C5gN7A6e7+z7uPsndE29aJTkrmX3HJqxezac1NRx99NEccMABPPzwwzQ0NFBfX88Zo0cz4uijqXhpNlMrK3l99z2YWllJ+ewqTho4kDNGj9ZWHSIiklMSJUF9gSvcfa67b3D394FzgV7Az919YkYilLQKu+/Y2FUrOWbYMK677jrKy8t57bXXGDlyJHvuuSdHHHIIH86cyZSKSs7s1p3yomKKzLasMzSlopLFkXWG1CIkIiK5IlESVAF8HHWs6f0b6QhGMi/svmO7DRrE+Acf5LLLLuPjjz/mzjvvpE+fPixevJh3Fi7kph7xp9k3rTM0K7LOkIiISC5oaXZYvP/bvrmtA5HsCbPv2PgHHtiyZUbnzp0555xzeP/99zn8oIM4vUePpNYZEhERyQUtJUEzzWxZ0wtYGjn+QvPjkXOhmdl5ZvaRmW0ws4VmdmgL5Q+PlNtgZh+a2TlR5682M496VScTU75rvu/YqrVr2dzQwKq1a7lv0iQOPPDAmNcUFRXxxr/+xdCyslDPGNyllGlx1hnSBq4iIpJpiabIX5OOB5rZSOAW4DzgH5Gfz5jZd9z90xjldwWeBv4XGA0cAtxhZsvd/bFmRd8H+jd735CO+OXr2mKdofr6es4aM4ZZ06drA1cREcmYRCtGpyUJAi4Bxrv73ZH3F5jZMQSDri+LUf4c4At3vyDy/l0z+z7wC6B5ErTZ3dX6k2HJrjNU5M6YMWMYNWoURx55JIWFhZw1ZsyWgdXawFVERDKl1StGJ8PMOgAHAM9GnXoWODjOZQfFKD8T6Gtmzf/y9jGzJZFutofMrE+bBC0JJbfO0BoA7rvvPo455hh22GEHRowYwXNPPKGB1SIiknGWybEWZrYDsAQ43N1fanb8SuBH7r5njGsWARPd/bfNjh0GzAZ2cPelZvZDgh3v3wPKgcuBbwPfdfeVMe55NnA2QEVFxQEPPfRQzHhra2spLS1N9ePmhXfeeYffX3opT+6wY8LB0esaGzn+iyWc/Ytf8Mknn/DCCy/w2Wef0dGM83r24qyePVt81t9Xr+KdvffmV1dc8bXj7s67777Lk488wqv//Ce1GzdS2rEjBx90EMePGMG3v/3tbar1SN+71KjeUqe6S53qLnXpqrsBAwYsdPe+YcpmKwk6zN1fbnb8KmCUu387xjWLgPvd/XfNjh0OVAG9Y3WBmVkpwTYfN7j7jYli6tu3ry9YsCDmuaqqKvr37x/ik+Uvd+eM0aNZPHMmN8dpzWlaZ2i3QYO2dGe5O6+//jqHfP/7PL3TTqG602rq6xm+rIZVa9duORY9nmhIael/xhPV1vLw+vXb3Hgife9So3pLneoudaq71KWr7swsdBKU0e4wYAXBgOXobcvLgZo411THKb8Z2KqVB8Dda4G3gd1TjlRCSWadobvHj9/SImNm7LfffmzYvDnlgdXu/rXxRFqoUUREkpHRJMjdNwELgYFRpwYCr8S57FXgqBjlFzRt5hrNzDoRdIctjXVe2lbMdYY+WBR3naHmmgZWh7Fq82YK3TnssMO46qqruPPOO3lh+nSNJxIRkZSE+7/gbetG4H4zmwfMIZj9tQNwF4CZTQBw99Mi5e8CfmZmNwN/A/4LGAOMarqhmf0ZmAZ8StBKdAXQBbgv/R9H4OvrDEH4Zs4hxw1m2uwqzuzWvcWyT6xZgwEvv/wyL7/88pbxRMks1Nhv0qQwH0dERPJAprvDcPfJwMUEg5dfJ1j351h3/yRSZKfIq6n8R8CxwGGR8r8BLoxaI+gbwCSCtYIeBzYCP2h2T8lRyWzg+ujGDUx/9lmmTp3KRRddhBUUMHS71i/UKCIi+SnjSRCAu9/h7ru4e0d3P6D5TDF37+/u/aPKz3b3/SPld3X3u6LOn+LuO7h7B3ff0d1PdPd3MvRxpBWS2cD1yCFDOPLIIxk6dCg333wzmxobW71QI2i1ahGRfJWVJEikSaoDqyH58UQdzLj11ltZsmTJluP19fWcMXo0I44+moqXZjO1spLXd9+DqZWVlM+u4qSBAzlj9Gjq62MOPwOURImIbKuUBEnWJbuBa5PkFmr8isbGRi666CK++c1vcsghh3DzzTcz+uSTWzW7rC2SKBERyY5sDIwW2Ur0wOowzr9kLCNmPMOosu1aXKjxkY0b+c011/Daa6/xzDPPMGfOHObMmUOPwkKe7bNbi7PLhkVml/Xr12/Luegp+tryQ0Rk26KWINlmJTOe6KghQ7j88st5/PHHWbZsGZMmTWK3b3yD07v3SGp2WXPz5s1jVhtM0Y/uTjvyiCPUnSYikgFKgmSblep4oq5du3LKKaewcvXqpGaXPfzwIwwcOJAxY8bw61//mrHnncfJnTqlnESButNERLJJ3WGyTWsaTzR//nxu+8uNDH/6KdbU1VFWUsKQwYN5/JJLOPDAA2Neu6auLqnZZRsbG3j++ee3HOtgxvV9wu3TO7hLKcOjpuirO01EJLuUBMk2L5XxRPCf2WVh9i1btXkzZSUlPPTYYyxZsoQvvviCq668Mqkk6qt165g6dSoHHnggO+6445butOgEqLlEY5JERKR1lARJ3kpmterp62oZevzxHHPMMVuO3fiHPySVRBUBw4YNA6B37950LihgRMfkutO04rWISNvRmCDJW8msVj25ro7zx4792vFkpuhPq13Ld/fai6OOOopu3bqxdOlSPv/iC44v6xrq+ngrXmuNIhGR1CkJkryV7GrV0WOLkkmiHl6/nrvuvZfnnnuOlStXsmjRIjZDq1a81qBqEZHWURIkeas1q1VD6klUQUEBu+++e/IrXhcUcNNNN/Huu+/S2Nj4tUHVqSz0CGpJEpH8piRI8lqqq1VD65OopFe8bmjgkksu4Tvf+Q477LADzzz2WKvWKFJLkojkOyVBkveazy5btXYtmxsaWLV2LfdNmhR3en2T1iRRyXSnPbJhI7+++mpGjx7N9ttvz5fLlvGjrmUpr1EUPT0/1ZYkEZFtmZIgkVZKNYlKasXr44dwxRVXcP/991NdXU3HTp2SWujxiSeeYPXq1VuOpWu1a3Wnici2REmQSJak2p1WUFBA7YYNyQ2qXr+eHj16UFFRQf/+/fnJj37ECK12LSJ5TkmQSBbF7E77YFGL3WnJDqruWFhISUkJy5YtY/bs2fz7ww85vmvq0/PVnSYi7YGSIJEsi+5Oe37WrBa705IZVD19XS0njxjB2rVr+fjjj5kxY0bS0/O/qq3lsssuY8KECcyfP58XX3xR3Wkiss3TitEi26DzLxnLiBnPMKpsu4RdWk0LPT42diwFBQXsvPPO7LzzzklvGVJkxg033LDlWEczzuvZq1WrXdfX13PWmDHMmj6dkzt3ZmplJT0Ki1jVsJlps6s4acYzHDF4MHePHx9zYLmISGupJUhkG9TahR6TW+26loMPOogrr7ySESNGsPfee+OQ1MDsdHSnRbciHXnEEUm1IqkVSkSUBIlsg1q7RlFyq13X8YebbuKaa67h4Ycf5s0330ypO+0Xv/gFjz76KJ999hlz585tVXdaawdla1C3iICSIJFtVmvWKGptS1KyA7OLzPjLX/7CiBEj2GmnnfjhkUcyomPHlGantbYVSYO6RaSJkiCRbViqaxRlcrXr6bW1HH7YYVx55ZUMGjSIbt26Ubd+PceXpdad1to1jrRGkog0URIkkqcytdr15PV1XPvHP3LNNdcwY8YMVq5cmXR32pe1tey1114MGjSIM049lREdU1/j6PYbb+Lkzp21RpKIaHaYSD5r3pKUjC3daTNncnOcFpV43WkFBQVJz04rNuO9997jvffeo4MZx/fpEyrOwV1K+eEjj/D2Bx9QVFREYWEh/zd3LjN22SX09cNbGNTd/LM3daeNKtuOiyPdafdOnLhVS1rTfebNm8dtf7mR6U8/xZq6OspKShhy3GB+9vNLOPDAA2NeJyJtRy1BIpK0jHanravlxOHDefPNN3nqqaeSbkXa2NDAwoULmTt3Lq+88gobGxqSHtR9+umnc+211zJ58mQmTJjAC9OmZXXz2tZ2xbV2Zp1Ie6EkSERSkrHutLo6Lr70Uvbee2+OPfbYpAdll5WUMH/+fF599VVefvllunbqlPSg7gkTJnDFFVdwyimn8NMzzmhVd1xrB2bnwsw4jYeS9kJJkIikLBObx0Z3pyXbijT0+OPp27cvP/jBDzjkkEMYevzQpAZ1HzFgAOPGjeOXv/wlJ5xwApgltUbSlKlTefPNN9m4cSPQuoHZuTAzLttJVFu3gimBy29KgkQk41rTnZZsK9L5Y8d+7Xiyg7qvuf56zjrrLP74xz8yZcoU6t2T6k6r3bCB733ve5SUlLDnnnty6kknpdySlO2ZcdlOonKxFUxdids2JUEikhWpdqe1do2jTK+R1KmoiN133x2ARYsW8fmSJRxfFn7z2scff5zbbruN++67j1///Oec3ClkAtWpM3+89loWLVrE22+/zeuvv87vrriCkzulPjMum0mUWsGy3wrWHseiWb5nrX379vUFCxbEPFdVVUX//v0zG1A7obpLnequZdH7jg3uUkqPoiJWbd7M9HW1TK6r48ghQ+LuO9aa608beQrls6s4s1v3FuO858vVLB8wgPsmTWLDhg0sWrSI/fbdl9d334OiEDO/6t3Zd9H7NP0r3cGMZ/v0CTWrrqa+nkEffcimZv/GJ3v9sZ98zODhwykvL6eiooKZTz7JDz79lDO792jx+uafvcncuXMZcfTRW82qi1bX2Miwmmoefe45+vXr1+pr2+J6d+eM0aP5cObMuElgXWMjF69ayW6DBsWcFRj9vRtSWvqf/fJqa3l4/fq4++W15tr2cH0yzGyhu/cNVVZJkJKgdFDdpU51F467M3/+/P9MMV+3jrIuXRgyeDA/u+SShGOSYl7fNEW9hevD/jFd19jIsOqlPPb881/7Y9q9tDRoOQiZiAz+/DNOO/NM1q5dywMTJ/LGHnsmlUDt9q1vUVxcTHFxMW/+619JX9/8L0SySdSQJZ8z7t576dmzJ7169eK6q65il3nzObN7cgkkpJ58Nmnt9dlMolqbgG3r1ydLSVASlASlh+oudaq71GSq3pr+QV8cYo2kWP+gt+aPcbIJ1PBlNaxau3bLsWSvH7r0C+665x6WLVtGTU0NN1x/fUaTqOM++5ThI0dSWFjIIw8+yNM77ZTUtSeNGkVDQwMNDQ088eijPLPzzuETuM8/48prr6WkpITOnTsz4e9/p+/ixXnZCpbt65OVTBKkxRJFRJLQNKj7rDFjGBaiOy3W5rUjZjzDqLLtWmxJmlxXx2PNBnYPOW4w00ImUNPX1TJk8OCvHUv2+hOGDmXUqFFbjt15661JLXJZUlzMsSecwMqVK1m5ciX/euONpAaVr6+vZ+LEiQAYya0Ptb6+ngkTJmw5luz16zZt4tJLL91yrIMZ1yS5SOfSlSspKyujrKyMBXPmJD0gvl8kiUpllfMDH3yQ9evX89VXX3HDNdeEHgs2olNn/vz733PPxImUlpZiZik9v1+zBLC116eTkiARkSQ1Depu6k4bHtWd9niC7rTWrLbdmgSqLa5PNok68cQTv9Ya0r20NKkkqrRTJ277299oaGjgwnPPTe7ajh259a67KCwspLCwkHN+8pPkErgOHTjngguoq6tj/fr13Dd+fNKLdD733HNbjnUw469JJFGDJk/m6eefp0uXLlR/9hkzd9019LXHTJ7Mg48+yubNm7c8+9mQzx5SWsqgadN4pKwsWN29rIz1a9Yk9fzBjz3GdhdeCAQtp49NeZynd9op9PXRq7Snk5IgEZEUpLrlSGtaklqTQLXF9ZlOooadcAKnnXYaAC/OmJnctcOGMWbMmC3HZjzxZHIJ3PDh/PnPf95ybOojjySVRJV17szkKVNYs2YNa9as4eyzzkoqiap3Z8WKFaxYsSLpVqxN7vjmzXTs2JFu3bqxrKYm6WeXlJRQV1fHl19+mfTz6+rr+etf/7rlWLLXr6mrC1W2LWiKvIhIhqW6PEBrtytp7fWtXV6gNWs8ZXR9qBjXJ71I59ChDBo0iBEjRnDmmWcmvbRCty5dqKmp4cMPP6Rr585JXbtdly5s2LCBDRs2UF1dzXZduiT37NJS1q1bx6ZNm1ixYkXSsXfp2JFbbrlly6tLx45Jr/KeKUqCRESyINXVtluzXUnc6z9YFOr6bCZR2V4fKtNJ1JAhQygvL2fXXXdl6JDjk7r2+CFD6NixY+rPjowlKy4upmfPnhw/eEhS1w8fNowLL7xwy2vY0BNSen4mKAkSEdnGpJpAxbv++VmzQl/fmiSsNUmUWsGy1wqW7evTSUmQiIgkpTVJWGuSqLS0gmUggYNtuxUs29enlbvn9euAAw7weF588cW45yQx1V3qVHepUb2lTnUXXmNjo8+dO9d/fPJI715a6oVm3r201E875RSfN29ewms3bdrkp596qn+zrMx/XlHhL/bZzd/YY09/sc9u/vOKCv9G165++qmn+qZNm9r02vZwfTKABR4yB8h6EpLtl5Kg9FDdpU51lxrVW+pUd6lLtu62SqIKCkInUa25NievTyKBTEYySVBWpsib2XnAL4HewNvAxe7+coLyhwM3At8FvgD+6O53teaeIiIimZbq0gqtvTYXr8+F1fEzPibIzEYCtwDXAfsBrwDPmFnMlZTMbFfg6Ui5/YDrgb+a2Ymp3lNEREQkGwOjLwHGu/vd7v6uu18ALAXOjVP+HOALd78gUv5u4D7gF624p4iIiOS5jCZBZtYBOAB4NurUs8DBcS47KEb5mUBfMytO8Z4iIiKS5zI9JqgXUAjURB2vAY6Kc00l8HyM8kWR+1my9zSzs4GzASoqKqiqqor54Nra2rjnJDHVXepUd6lRvaVOdZc61V3qcqHusrV3mEe9txjHWirfdNwSlIl5T3cfB4wD6Nu3r8cbmJULg7a2Vaq71KnuUqN6S53qLnWqu9TlQt1lOglaATQQtO40V87WLTlNquOU3wysJEh2kr3nFgsXLlxhZp/EOd0rErMkT3WXOtVdalRvqVPdpU51l7p01d3OYQtmNAly901mthAYCDzS7NRA4LE4l70KnBB1bCDBOgD1ACncs3lM28c7Z2YL3L1vS/eQranuUqe6S43qLXWqu9Sp7lKXC3WXje6wG4H7zWweMIdg9tcOwF0AZjYBwN1Pi5S/C/iZmd0M/A34L2AMMCrsPUVERESiZTwJcvfJZtYTuJxgYcO3gGPdvalLaqeo8h+Z2bHATQRT3r8ALnT3x5K4p4iIiMjXZGVgtLvfAdwR51z/GMdmA/unes9WGNfG98snqrvUqe5So3pLneoudaq71GW97izYZkNEREQkv2RjxWgRERGRrFMSJCIiInlJSVAcZnaemX1kZhvMbKGZHZrtmHKdmV1tZh71qs52XLnGzA4zsyfNbEmkjsZEnbdIXX5hZuvNrMrMvpulcHNKiLobH+M7+M8shZszzOwyM5tvZmvMbLmZTTOzvaPK6HsXQ8i60/cuBjM738z+Fam7NWb2qpkd1+x81r9zSoJi0K70rfI+wQy9ptc+2Q0nJ5USzGC8CFgf4/ylwM+BC4ADgWXAc2bWNWMR5q6W6g6CbXaafwePzUxoOa0/wcSRg4EjCBabfd7MejQro+9dbP1pue5A37tYPgd+RTCxqS8wC5hqZt+LnM/6d04Do2Mws7nAv9z9rGbHPgAedffLshdZbjOzq4GT3H3vlspKwMxqgZ+5+/jIeyNYBuI2d/995Fhngn8cfuHuf8tWrLkmuu4ix8YDvdx9cLbi2haYWSnwFXCCu0/T9y686LqLHBuPvnehmNkq4DKCmWFZ/86pJSiKdqVvtT6RroqPzOwhM+uT7YC2MbsSbAGz5fvn7uuBl9D3L6xDzGyZmS0ys7vNrDzbAeWgrgT//q+OvNf3LrzouqRyMdAAAAYaSURBVGui710CZlZoZqcQtOa+Qo5855QEbS3RTvfR+5PJ180lWM37h8BZBPX1SmQhSwmn6Tum719qZgCnAUcSNLP3A2aZWcesRpV7bgFeJ9iWCPS9S0Z03YG+d3GZ2T6RVtuNBLs4DHP3N8mR71y2dpHfFiS7033ec/dnmr+PDAz8EDidYGsTCU/fvxS4+0PN3r5pwb6CnwDHAY9nJ6rcYmY3AocAh7h7Q9Rpfe8SiFd3+t4l9D6wL9ANOBG4z8z6Nzuf1e+cWoK2lspO9xKDu9cCbwO7ZzuWbUjTbDp9/9qAu39BMDhT30HAzG4i2HfxCHf/sNkpfe9akKDutqLv3X+4+yZ3/7e7L4iMqX0dGEuOfOeUBEVx901A0670zQ0k6MeUkMysE/BtYGm2Y9mGfETwj8OW71+kHg9F37+kmVkvYEf0HcTMbgFOJfgj/l7UaX3vEmih7mKV1/cuvgKgIznynVN3WGzalT4FZvZnYBrwKUE2fwXQBbgvm3Hlmsjskm9F3hYAO5nZvsAqd//UzG4GfmNm7wGLCDYGrgUezErAOSRR3UVe/7+9ewuxqorjOP79oVC+VCSERQNTRIiF0YUQiXkYorILZEWEGAQVQUkXwuilnC6YFfRSowz2IF3IIHMeDPJSFAnThKGERVNBpuagZWkZTMOM/x7WGtqeOTNOorP3tH8fWJyz9l57nTWLzT7/2fu/z+4A1pG+fFqBF0h3m6yf7LFWiaRO4G7gVuB3SSP/fR+JiCMREd7vmjve3OV9sgPvd6NIWgF8AOwhJZQvIv3kwE2V2eciwqVJAR4EdpGSub4E2soeU9ULsJZ0y+Mg8DPpoDCn7HFVreSDQDQpa/J6kQ6q/cAA8ClwadnjrkIZb+6AGcBG0pfPICknYw3QUva4yy5jzFkAHYU23u9OYO683407d2vyfPyd52cLcH1hfen7nH8nyMzMzGrJOUFmZmZWSw6CzMzMrJYcBJmZmVktOQgyMzOzWnIQZGZmZrXkIMjMzMxqyUGQmZVOUoekGKMsLmE8IWnJZH+umU0u/2K0mVXFYeCGJst/mOyBmFk9OAgys6oYiojPyx6EmdWHL4eZWeVJas2XqBZJelPSn5IOSFrWpG27pF5JA5L2S1qZn+9UbDNTUpek/tyuT9KjDV1Nk7Rc0i/5szolnVbo4yxJr0val/vYLWn1KZoCMzsFfCbIzCpD0qhjUkQMFaovAxuAO4A2YJmkXyOiM28/B/gQ2AzcDrQAK4ALyZfaJM0APiE95PcZ4FvSQ1kv4liPAx8Di4G5pIdi/gS8lNe/AswHHiM9Dbslj8nMpgg/O8zMSiepAxh1Vie7IL/+CGyOiOsK260GbiQ9rPKopLXAlcDsiBjObe4E3gXmR0SPpAeAVcAVEbFjjPEE8FlEtBWWdQOzImJeru8EuiLi1RP9u82sXD4TZGZVcRi4tsnyfcB5+f36hnXvA/cB5wO7gauB90YCoGwdMARcA/QA7cD2sQKggk0N9W+Aqwr1HcBSScPAloj47jj9mVnFOCfIzKpiKCK2NSmDhTYHGrYZqZ9beN1fbJADooPA2XnRTKB/AuM51FAfBE4v1JcA3cDTQJ+k7yXdNYF+zawiHASZ2VRyzhj1/sLrMW0kTSMFPr/lRQf5N2g6YRFxKCIejohZwGVAL/B2zksysynAQZCZTSULG+q3kQKfvbneCyzMgU+xzXRga65/BFwuae7JGlREfAUsJR1TZ5+sfs3s1HJOkJlVxXRJ85os31N4f4mkLlKeTxtwL/BIRBzN658HtgPdklaRcoVeBDZGRE9u8wbwELApJ2T3kZKvL46IJyc6WElbSTlKO4EA7gf+Ar6YaB9mVi4HQWZWFWeSEpcbPQW8ld8/AdxMCoIGgOeA10YaRsTXkhYAy0lJ038A7+TtRtoMSGon3Tr/LHAGsAtY+R/H2wPcA7QCw6Tga0FE7B1nGzOrEN8ib2aVJ6mVdIv8LRGxodzRmNn/hXOCzMzMrJYcBJmZmVkt+XKYmZmZ1ZLPBJmZmVktOQgyMzOzWnIQZGZmZrXkIMjMzMxqyUGQmZmZ1ZKDIDMzM6ulfwD3yYwxFGNYzQAAAABJRU5ErkJggg==\n",
      "text/plain": [
       "<Figure size 648x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "t.plot_training_error()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Run a benchmark test (plots the test set ground truth against the predictions)\n",
    "\n",
    "This function computes and plots the test set performance against the ground truth (the true values of the test set). This feature is provided for a **one-shot visual check of the effectiveness of the model**. If the data scientist/ML engineer does not like the model performance, (s)he can just look at this plot and re-build and re-train the model."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 576x360 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "t.benchmark_test()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Calculate the root-mean-square-error (RMSE) of the test set predictions"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "20.452"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t.rmse_test()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### We can also calculate RMSE for any given test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "# We take a random sample of 25 data points from the dataset\n",
    "df1 = t.df.sample(25)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [],
   "source": [
    "sample_test_x = df1[['x1','x2','x3','x4']]\n",
    "sample_test_y = df1['y']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "20.04168063844275"
      ]
     },
     "execution_count": 48,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "t.rmse_error(sample_test_x,sample_test_y)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "There is also a method for mean-absolute-error (MAE) calculation - `mae_error()` "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Model summary in simple English"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Here is the model description\n",
      "===========================================================================\n",
      "Layer No. 1\n",
      "Number of neurons: 20\n",
      "Activation function: relu\n",
      "----------------------------------------\n",
      "Layer No. 2\n",
      "Number of neurons: 20\n",
      "Activation function: relu\n",
      "----------------------------------------\n",
      "Layer No. 3 (output layer)\n",
      "Layer No. 3\n",
      "Number of neurons: 1\n",
      "Activation function: linear\n",
      "----------------------------------------\n",
      "In total, there are 541 parameters in this model!\n",
      "\n"
     ]
    }
   ],
   "source": [
    "t.model_in_plain_english()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Keras-style model summary"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_4 (Dense)              (None, 20)                100       \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 20)                420       \n",
      "_________________________________________________________________\n",
      "dense_6 (Dense)              (None, 1)                 21        \n",
      "=================================================================\n",
      "Total params: 541\n",
      "Trainable params: 541\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n",
      "None\n"
     ]
    }
   ],
   "source": [
    "t.model_summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Save the trained model (with a default name)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [],
   "source": [
    "t.save_model()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Save the trained model with an user-supplied name"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [],
   "source": [
    "t.save_model(\"My_model.h5\")"
   ]
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.4"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 4
}