10_Neural_Networks.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# SimFin Tutorial 10 - Neural Networks\n",
    "\n",
    "[Original repository on GitHub](https://github.com/simfin/simfin-tutorials)\n",
    "\n",
    "This tutorial was originally written by [Hvass Labs](https://github.com/Hvass-Labs)\n",
    "\n",
    "----\n",
    "\n",
    "\"The attackers can be stopped by removing the head or destroying the brain.\" &ndash; [Shaun of The Dead](https://www.youtube.com/watch?v=ONEZfj6h1xc)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Introduction\n",
    "\n",
    "In the previous tutorial we used a Machine Learning (ML) model to try and predict future stock-returns for 1-3 year investment periods given various signals. This worked quite well on the training-set, but unfortunately it failed to generalize to the test-set.\n",
    "\n",
    "In this tutorial we will use a so-called Neural Network, which has become very fashionable in recent years, because they have been able to solve extremely hard problems, that were previously impossible to solve.\n",
    "\n",
    "Unfortunately, Neural Networks have become so hyped, that people sometimes misunderstand what they really are: A Neural Network is just a function approximator that maps some input x (e.g. signals) to some output y (e.g. stock-returns). It consists of a number of layers of both linear and non-linear mappings, whose parameters will be learned during the training process, so it learns to map from x to y in the training-set, and the learned mapping hopefully generalizes to new data.\n",
    "\n",
    "In theory, a Neural Network is capable of approximating any function to any desired degree of precision - but what is truly important, is not how well it performs on the training-set, but how well it performs on new data that it has not seen during the training.\n",
    "\n",
    "This tutorial shows how to use financial data from SimFin with the TensorFlow library for creating Neural Networks. It is assumed you are already familiar with both."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Imports"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [],
   "source": [
    "%matplotlib inline\n",
    "import pandas as pd\n",
    "import numpy as np\n",
    "import seaborn as sns\n",
    "\n",
    "# TensorFlow and Keras imports.\n",
    "import tensorflow as tf\n",
    "from tensorflow import keras\n",
    "from tensorflow.keras import Sequential\n",
    "from tensorflow.keras.optimizers import RMSprop\n",
    "from tensorflow.keras.layers import Dense, InputLayer\n",
    "from tensorflow.keras.utils import to_categorical\n",
    "\n",
    "# Scikit-learn imports.\n",
    "from sklearn.model_selection import train_test_split\n",
    "from sklearn.metrics import confusion_matrix\n",
    "from sklearn.metrics import r2_score, accuracy_score\n",
    "from sklearn.preprocessing import StandardScaler\n",
    "from sklearn.utils.class_weight import compute_class_weight\n",
    "\n",
    "# Import the main functionality from the SimFin Python API.\n",
    "import simfin as sf\n",
    "\n",
    "# Import names used for easy access to SimFin's data-columns.\n",
    "from simfin.names import *"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'0.3.0'"
      ]
     },
     "execution_count": 2,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Version of the SimFin Python API.\n",
    "sf.__version__"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "'2.0.0'"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# TensorFlow version.\n",
    "tf.__version__"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## SimFin Config"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [],
   "source": [
    "# SimFin data-directory.\n",
    "sf.set_data_dir('~/simfin_data/')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [],
   "source": [
    "# SimFin load API key or use free data.\n",
    "sf.load_api_key(path='~/simfin_api_key.txt', default_key='free')"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Seaborn set plotting style.\n",
    "sns.set_style(\"whitegrid\")"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Data Hub\n",
    "\n",
    "We use a `StockHub` object to easily load and process financial data with these settings:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "# We are interested in the US stock-market.\n",
    "market = 'us'\n",
    "\n",
    "# Add this date-offset to the fundamental data such as\n",
    "# Income Statements etc., because the REPORT_DATE is not\n",
    "# when it was actually made available to the public,\n",
    "# which can be 1, 2 or even 3 months after the Report Date.\n",
    "offset = pd.DateOffset(days=60)\n",
    "\n",
    "# Refresh the fundamental datasets (Income Statements etc.)\n",
    "# every 30 days.\n",
    "refresh_days = 30\n",
    "\n",
    "# Refresh the dataset with shareprices every 10 days.\n",
    "refresh_days_shareprices = 10"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 23 µs, sys: 4 µs, total: 27 µs\n",
      "Wall time: 33.4 µs\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "hub = sf.StockHub(market=market, offset=offset,\n",
    "                  refresh_days=refresh_days,\n",
    "                  refresh_days_shareprices=refresh_days_shareprices)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Signals\n",
    "\n",
    "First we calculate financial signals for the stocks, such as the Current Ratio, Debt Ratio, Net Profit Margin, Return on Assets, etc. These are calculated using data from the financial reports: Income Statements, Balance Sheets and Cash-Flow Statements, which are automatically downloaded and loaded by the data-hub."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset \"us-income-ttm\" on disk (1 days old).\n",
      "- Loading from disk ... Done!\n",
      "Dataset \"us-balance-ttm\" on disk (1 days old).\n",
      "- Loading from disk ... Done!\n",
      "Dataset \"us-shareprices-daily\" on disk (1 days old).\n",
      "- Loading from disk ... Done!\n",
      "Cache-file 'fin_signals-2a38bb7d.pickle' on disk (1 days old).\n",
      "- Loading from disk ... Done!\n",
      "CPU times: user 14.4 s, sys: 1.05 s, total: 15.5 s\n",
      "Wall time: 15.5 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "df_fin_signals = hub.fin_signals(variant='daily')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we calculate growth signals for the stocks, such as Earnings Growth, FCF Growth, Sales Growth, etc. These are also calculated using data from the financial reports: Income Statements, Balance Sheets and Cash-Flow Statements, which are automatically downloaded and loaded by the data-hub."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Dataset \"us-income-quarterly\" on disk (1 days old).\n",
      "- Loading from disk ... Done!\n",
      "Dataset \"us-cashflow-ttm\" on disk (1 days old).\n",
      "- Loading from disk ... Done!\n",
      "Dataset \"us-cashflow-quarterly\" on disk (1 days old).\n",
      "- Loading from disk ... Done!\n",
      "Cache-file 'growth_signals-2a38bb7d.pickle' on disk (1 days old).\n",
      "- Loading from disk ... Done!\n",
      "CPU times: user 817 ms, sys: 296 ms, total: 1.11 s\n",
      "Wall time: 1.11 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "df_growth_signals = hub.growth_signals(variant='daily')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now we calculate valuation signals for the stocks, such as P/E, P/Sales, etc. These are calculated from the share-prices and data from the financial reports. Because the data-hub has already loaded the required datasets in the function-calls above, the data is merely reused here, and the data-hub can proceed directly to computing the signals."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cache-file 'val_signals-739b68a6.pickle' on disk (1 days old).\n",
      "- Loading from disk ... Done!\n",
      "CPU times: user 137 ms, sys: 276 ms, total: 413 ms\n",
      "Wall time: 413 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "df_val_signals = hub.val_signals(variant='daily')"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We now combine all the signals into a single DataFrame:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 1.39 s, sys: 219 ms, total: 1.61 s\n",
      "Wall time: 1.61 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# Combine the DataFrames.\n",
    "dfs = [df_fin_signals, df_growth_signals, df_val_signals]\n",
    "df_signals = pd.concat(dfs, axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can then show the results:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th>Asset Turnover</th>\n",
       "      <th>Current Ratio</th>\n",
       "      <th>Debt Ratio</th>\n",
       "      <th>Gross Profit Margin</th>\n",
       "      <th>Net Profit Margin</th>\n",
       "      <th>Return on Assets</th>\n",
       "      <th>Return on Equity</th>\n",
       "      <th>Earnings Growth</th>\n",
       "      <th>Earnings Growth QOQ</th>\n",
       "      <th>Earnings Growth YOY</th>\n",
       "      <th>...</th>\n",
       "      <th>Dividend Yield</th>\n",
       "      <th>Earnings Yield</th>\n",
       "      <th>FCF Yield</th>\n",
       "      <th>Market-Cap</th>\n",
       "      <th>P/Book</th>\n",
       "      <th>P/E</th>\n",
       "      <th>P/FCF</th>\n",
       "      <th>P/NCAV</th>\n",
       "      <th>P/NetNet</th>\n",
       "      <th>P/Sales</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Ticker</th>\n",
       "      <th>Date</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th rowspan=\"5\" valign=\"top\">A</th>\n",
       "      <th>2011-06-29</th>\n",
       "      <td>NaN</td>\n",
       "      <td>3.201005</td>\n",
       "      <td>0.24789</td>\n",
       "      <td>0.532001</td>\n",
       "      <td>0.144574</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.049626</td>\n",
       "      <td>0.045221</td>\n",
       "      <td>1.793416e+10</td>\n",
       "      <td>4.527684</td>\n",
       "      <td>20.150736</td>\n",
       "      <td>22.113631</td>\n",
       "      <td>43.956262</td>\n",
       "      <td>-29.990226</td>\n",
       "      <td>2.913281</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2011-06-30</th>\n",
       "      <td>NaN</td>\n",
       "      <td>3.201005</td>\n",
       "      <td>0.24789</td>\n",
       "      <td>0.532001</td>\n",
       "      <td>0.144574</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.049121</td>\n",
       "      <td>0.044761</td>\n",
       "      <td>1.811850e+10</td>\n",
       "      <td>4.574222</td>\n",
       "      <td>20.357860</td>\n",
       "      <td>22.340931</td>\n",
       "      <td>44.408076</td>\n",
       "      <td>-30.298487</td>\n",
       "      <td>2.943225</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2011-07-01</th>\n",
       "      <td>NaN</td>\n",
       "      <td>3.201005</td>\n",
       "      <td>0.24789</td>\n",
       "      <td>0.532001</td>\n",
       "      <td>0.144574</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.048280</td>\n",
       "      <td>0.043995</td>\n",
       "      <td>1.843400e+10</td>\n",
       "      <td>4.653875</td>\n",
       "      <td>20.712360</td>\n",
       "      <td>22.729963</td>\n",
       "      <td>45.181373</td>\n",
       "      <td>-30.826087</td>\n",
       "      <td>2.994477</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2011-07-05</th>\n",
       "      <td>NaN</td>\n",
       "      <td>3.201005</td>\n",
       "      <td>0.24789</td>\n",
       "      <td>0.532001</td>\n",
       "      <td>0.144574</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.048598</td>\n",
       "      <td>0.044284</td>\n",
       "      <td>1.831347e+10</td>\n",
       "      <td>4.623446</td>\n",
       "      <td>20.576933</td>\n",
       "      <td>22.581344</td>\n",
       "      <td>44.885956</td>\n",
       "      <td>-30.624532</td>\n",
       "      <td>2.974898</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2011-07-06</th>\n",
       "      <td>NaN</td>\n",
       "      <td>3.201005</td>\n",
       "      <td>0.24789</td>\n",
       "      <td>0.532001</td>\n",
       "      <td>0.144574</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>NaN</td>\n",
       "      <td>...</td>\n",
       "      <td>NaN</td>\n",
       "      <td>0.048448</td>\n",
       "      <td>0.044148</td>\n",
       "      <td>1.837019e+10</td>\n",
       "      <td>4.637766</td>\n",
       "      <td>20.640663</td>\n",
       "      <td>22.651282</td>\n",
       "      <td>45.024975</td>\n",
       "      <td>-30.719381</td>\n",
       "      <td>2.984111</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>5 rows × 26 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                   Asset Turnover  Current Ratio  Debt Ratio  \\\n",
       "Ticker Date                                                    \n",
       "A      2011-06-29             NaN       3.201005     0.24789   \n",
       "       2011-06-30             NaN       3.201005     0.24789   \n",
       "       2011-07-01             NaN       3.201005     0.24789   \n",
       "       2011-07-05             NaN       3.201005     0.24789   \n",
       "       2011-07-06             NaN       3.201005     0.24789   \n",
       "\n",
       "                   Gross Profit Margin  Net Profit Margin  Return on Assets  \\\n",
       "Ticker Date                                                                   \n",
       "A      2011-06-29             0.532001           0.144574               NaN   \n",
       "       2011-06-30             0.532001           0.144574               NaN   \n",
       "       2011-07-01             0.532001           0.144574               NaN   \n",
       "       2011-07-05             0.532001           0.144574               NaN   \n",
       "       2011-07-06             0.532001           0.144574               NaN   \n",
       "\n",
       "                   Return on Equity  Earnings Growth  Earnings Growth QOQ  \\\n",
       "Ticker Date                                                                 \n",
       "A      2011-06-29               NaN              NaN                  NaN   \n",
       "       2011-06-30               NaN              NaN                  NaN   \n",
       "       2011-07-01               NaN              NaN                  NaN   \n",
       "       2011-07-05               NaN              NaN                  NaN   \n",
       "       2011-07-06               NaN              NaN                  NaN   \n",
       "\n",
       "                   Earnings Growth YOY  ...  Dividend Yield  Earnings Yield  \\\n",
       "Ticker Date                             ...                                   \n",
       "A      2011-06-29                  NaN  ...             NaN        0.049626   \n",
       "       2011-06-30                  NaN  ...             NaN        0.049121   \n",
       "       2011-07-01                  NaN  ...             NaN        0.048280   \n",
       "       2011-07-05                  NaN  ...             NaN        0.048598   \n",
       "       2011-07-06                  NaN  ...             NaN        0.048448   \n",
       "\n",
       "                   FCF Yield    Market-Cap    P/Book        P/E      P/FCF  \\\n",
       "Ticker Date                                                                  \n",
       "A      2011-06-29   0.045221  1.793416e+10  4.527684  20.150736  22.113631   \n",
       "       2011-06-30   0.044761  1.811850e+10  4.574222  20.357860  22.340931   \n",
       "       2011-07-01   0.043995  1.843400e+10  4.653875  20.712360  22.729963   \n",
       "       2011-07-05   0.044284  1.831347e+10  4.623446  20.576933  22.581344   \n",
       "       2011-07-06   0.044148  1.837019e+10  4.637766  20.640663  22.651282   \n",
       "\n",
       "                      P/NCAV   P/NetNet   P/Sales  \n",
       "Ticker Date                                        \n",
       "A      2011-06-29  43.956262 -29.990226  2.913281  \n",
       "       2011-06-30  44.408076 -30.298487  2.943225  \n",
       "       2011-07-01  45.181373 -30.826087  2.994477  \n",
       "       2011-07-05  44.885956 -30.624532  2.974898  \n",
       "       2011-07-06  45.024975 -30.719381  2.984111  \n",
       "\n",
       "[5 rows x 26 columns]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_signals.dropna(how='all').head()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Missing Data\n",
    "\n",
    "Some of the signals have a lot of missing data which cannot be handled by TensorFlow. Let us first see the fraction of each signal-column that is missing:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Dividend Yield         0.533772\n",
       "Debt Ratio             0.324557\n",
       "Sales Growth           0.227931\n",
       "Asset Turnover         0.224232\n",
       "Earnings Growth        0.212780\n",
       "Return on Assets       0.212388\n",
       "Return on Equity       0.212388\n",
       "FCF Growth             0.212388\n",
       "Sales Growth YOY       0.198149\n",
       "Earnings Growth YOY    0.184320\n",
       "FCF Growth YOY         0.183868\n",
       "P/Sales                0.105850\n",
       "Sales Growth QOQ       0.095425\n",
       "Earnings Yield         0.091799\n",
       "P/NetNet               0.091799\n",
       "FCF Yield              0.091799\n",
       "Market-Cap             0.091799\n",
       "P/Book                 0.091799\n",
       "P/E                    0.091799\n",
       "P/FCF                  0.091799\n",
       "P/NCAV                 0.091799\n",
       "Gross Profit Margin    0.083246\n",
       "Net Profit Margin      0.080159\n",
       "Earnings Growth QOQ    0.077892\n",
       "FCF Growth QOQ         0.077380\n",
       "Current Ratio          0.065622\n",
       "dtype: float64"
      ]
     },
     "execution_count": 14,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Remove all rows with only NaN values.\n",
    "df = df_signals.dropna(how='all').reset_index(drop=True)\n",
    "\n",
    "# For each column, show the fraction of the rows that are NaN.\n",
    "(df.isnull().sum() / len(df)).sort_values(ascending=False)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let us remove all signals that have more than 25% missing data:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "Index(['Debt Ratio', 'Dividend Yield'], dtype='object')"
      ]
     },
     "execution_count": 15,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# List of the columns before removing any.\n",
    "columns_before = df_signals.columns\n",
    "\n",
    "# Threshold for the number of rows that must be NaN for each column.\n",
    "thresh = 0.75 * len(df_signals.dropna(how='all'))\n",
    "\n",
    "# Remove all columns which don't have sufficient data.\n",
    "df_signals = df_signals.dropna(axis='columns', thresh=thresh)\n",
    "\n",
    "# List of the columns after the removal.\n",
    "columns_after = df_signals.columns\n",
    "\n",
    "# Show the columns that were removed.\n",
    "columns_before.difference(columns_after)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## 1-3 Year Returns\n",
    "\n",
    "We want to try and predict the average 1-3 year returns. We will actually consider the mean log-returns, because that is easier to calculate, but the result is nearly the same as the non-log mean returns."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Name of the new column for the returns.\n",
    "TOTAL_RETURN_1_3Y = 'Total Return 1-3 Years'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Cache-file 'mean_log_change-5cec82bd.pickle' on disk (1 days old).\n",
      "- Loading from disk ... Done!\n",
      "CPU times: user 70.7 ms, sys: 32.4 ms, total: 103 ms\n",
      "Wall time: 105 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# Calculate the mean log-returns for all 1-3 year periods.\n",
    "df_returns_1_3y = \\\n",
    "    hub.mean_log_returns(name=TOTAL_RETURN_1_3Y,\n",
    "                         future=True, annualized=True,\n",
    "                         min_years=1, max_years=3)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Combine Signals & Returns\n",
    "\n",
    "We then combine the signals and returns into a single DataFrame to align the data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 766 ms, sys: 112 ms, total: 878 ms\n",
      "Wall time: 884 ms\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "dfs = [df_signals, df_returns_1_3y]\n",
    "df_sig_rets = pd.concat(dfs, axis=1)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Data Cleaning\n",
    "\n",
    "The data also contains outliers and still has missing values, so let us fix that now. A common method for removing outliers is so-called *Winsorization* of the data. It basically just limits or clips the data between e.g. the 5% and 95% quantiles of the data. We will Winsorize both the stock-returns and all the signals. Then we remove all rows with missing values, and finally we remove tickers that have less than 200 rows of data.\n",
    "\n",
    "There are more than 5 million data-rows before this cleaning:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(5450093, 25)"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_sig_rets.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 4.27 s, sys: 1.06 s, total: 5.33 s\n",
      "Wall time: 5.33 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "\n",
    "# Clip both the signals and returns at their 5% and 95% quantiles.\n",
    "# We do not set them to NaN because it would remove too much data.\n",
    "df_sig_rets = sf.winsorize(df_sig_rets)\n",
    "\n",
    "# Remove all rows with missing values (NaN)\n",
    "# because TensorFlow cannot handle that.\n",
    "df_sig_rets = df_sig_rets.dropna(how='any')\n",
    "\n",
    "# Remove all tickers which have less than 200 data-rows.\n",
    "df_sig_rets = df_sig_rets.groupby(TICKER) \\\n",
    "                .filter(lambda df: len(df)>200)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This data-cleaning has removed about 80% of the rows, which is a lot:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(1109909, 25)"
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_sig_rets.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The reason that so many rows have been removed, is that TensorFlow cannot handle rows with missing values (NaN). Some missing values could be set to a default value, e.g. the Dividend Yield could be set to 0. Other columns could be omitted because they don't have any predictive power anyway."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Split Training / Test-Sets\n",
    "\n",
    "When training a Machine Learning model, we typically split the dataset into training- and test-sets. Only the training-set is used to train the model, after which the model is used on the test-set, to assess how the model performs on data it has not seen during training.\n",
    "\n",
    "If we randomly divide all the data-rows into training- and test-sets, we would most likely get data for every stock-ticker in both the training- and test-sets, which would probably make them highly correlated. To avoid this, we split the dataset according to stock-tickers, so a ticker belongs to either the training- or test-set, but not both. We use 80% of all the tickers in the training-set, and 20% in the test-set."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [],
   "source": [
    "# List of all unique stock-tickers in the dataset.\n",
    "tickers = df_sig_rets.reset_index()[TICKER].unique()\n",
    "\n",
    "# Split the tickers into training- and test-sets.\n",
    "tickers_train, tickers_test = \\\n",
    "    train_test_split(tickers, train_size=0.8, random_state=1234)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now that we have lists of tickers for the training- and test-sets, we can select those rows from the DataFrame with signals and stock-returns:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 3.07 s, sys: 19.9 ms, total: 3.09 s\n",
      "Wall time: 3.09 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "df_train = df_sig_rets.loc[tickers_train]\n",
    "df_test = df_sig_rets.loc[tickers_test]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can then split these DataFrames into signals X and stock-returns y, that the Machine Learning model must try and find a mapping between. We use a common notation where the capital X indicates a 2-dimensional array or matrix, and the lower-case y indicates a 1-dimensional array."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [],
   "source": [
    "# DataFrames with signals for training- and test-sets.\n",
    "X_train = df_train.drop(columns=[TOTAL_RETURN_1_3Y])\n",
    "X_test = df_test.drop(columns=[TOTAL_RETURN_1_3Y])\n",
    "\n",
    "# DataFrames with returns for training- and test-sets.\n",
    "y_train = df_train[TOTAL_RETURN_1_3Y]\n",
    "y_test = df_test[TOTAL_RETURN_1_3Y]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let us check that the shapes of these arrays look reasonable:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((890400, 24), (890400,))"
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.shape, y_train.shape"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "((219509, 24), (219509,))"
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_test.shape, y_test.shape"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let us also get the number of signals, which will be convenient later:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "24"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "num_signals = X_train.shape[1]\n",
    "num_signals"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Scale Signals\n",
    "\n",
    "Neural Networks require that the input has \"moderate\" values e.g. between 0 and 1, or mean 0 and 1 standard deviation, which is the method we use here. It is not necessary to scale the stock-returns because they have already been Winsorized to remove outliers, and they are centered around 0 with a small standard deviation.\n",
    "\n",
    "If you want to input new data to the Neural Network, then you have to run the signals through the scaler first, because the Neural Network has been tuned to the scaled signals."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [],
   "source": [
    "# Create a scaler and fit it to the signals in the training-data.\n",
    "signal_scaler = StandardScaler()\n",
    "signal_scaler.fit(X_train)\n",
    "\n",
    "def scale_signal(df):\n",
    "    \"\"\"Apply the signal-scaler to the given DataFrame.\"\"\"\n",
    "\n",
    "    # Apply the scaler. This returns a numpy array.\n",
    "    array = signal_scaler.transform(df)\n",
    "    \n",
    "    # Create a DataFrame with the correct column-names and index.\n",
    "    df_scaled = pd.DataFrame(data=array,\n",
    "                             columns=df.columns,\n",
    "                             index=df.index)\n",
    "    return df_scaled\n",
    "\n",
    "# Scale the signals in the training- and test-sets.\n",
    "X_train = scale_signal(X_train)\n",
    "X_test = scale_signal(X_test)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Except for small rounding errors, the signals in the training-set now have 0 mean and 1 standard deviation:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "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>Asset Turnover</th>\n",
       "      <th>Current Ratio</th>\n",
       "      <th>Gross Profit Margin</th>\n",
       "      <th>Net Profit Margin</th>\n",
       "      <th>Return on Assets</th>\n",
       "      <th>Return on Equity</th>\n",
       "      <th>Earnings Growth</th>\n",
       "      <th>Earnings Growth QOQ</th>\n",
       "      <th>Earnings Growth YOY</th>\n",
       "      <th>FCF Growth</th>\n",
       "      <th>...</th>\n",
       "      <th>Sales Growth YOY</th>\n",
       "      <th>Earnings Yield</th>\n",
       "      <th>FCF Yield</th>\n",
       "      <th>Market-Cap</th>\n",
       "      <th>P/Book</th>\n",
       "      <th>P/E</th>\n",
       "      <th>P/FCF</th>\n",
       "      <th>P/NCAV</th>\n",
       "      <th>P/NetNet</th>\n",
       "      <th>P/Sales</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>...</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "      <td>8.904000e+05</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>5.720092e-17</td>\n",
       "      <td>-5.377908e-16</td>\n",
       "      <td>2.553613e-18</td>\n",
       "      <td>9.908017e-17</td>\n",
       "      <td>-2.962191e-16</td>\n",
       "      <td>-2.266331e-18</td>\n",
       "      <td>4.290069e-17</td>\n",
       "      <td>2.719598e-17</td>\n",
       "      <td>2.042890e-17</td>\n",
       "      <td>-3.843187e-17</td>\n",
       "      <td>...</td>\n",
       "      <td>9.448367e-17</td>\n",
       "      <td>-2.088855e-16</td>\n",
       "      <td>-2.911118e-17</td>\n",
       "      <td>-1.082732e-16</td>\n",
       "      <td>-2.493603e-16</td>\n",
       "      <td>-2.694061e-16</td>\n",
       "      <td>-5.822237e-17</td>\n",
       "      <td>9.512207e-18</td>\n",
       "      <td>-1.398103e-17</td>\n",
       "      <td>2.788545e-16</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "      <td>1.000001e+00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>-1.289389e+00</td>\n",
       "      <td>-1.112665e+00</td>\n",
       "      <td>-1.418441e+00</td>\n",
       "      <td>-4.276321e+00</td>\n",
       "      <td>-3.908327e+00</td>\n",
       "      <td>-3.357029e+00</td>\n",
       "      <td>-2.558049e+00</td>\n",
       "      <td>-2.631511e+00</td>\n",
       "      <td>-2.639940e+00</td>\n",
       "      <td>-2.512918e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>-1.836140e+00</td>\n",
       "      <td>-4.257038e+00</td>\n",
       "      <td>-3.393204e+00</td>\n",
       "      <td>-6.855607e-01</td>\n",
       "      <td>-1.102352e+00</td>\n",
       "      <td>-2.636409e+00</td>\n",
       "      <td>-2.457002e+00</td>\n",
       "      <td>-2.251178e+00</td>\n",
       "      <td>-2.297130e+00</td>\n",
       "      <td>-8.794930e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>-7.636819e-01</td>\n",
       "      <td>-6.884963e-01</td>\n",
       "      <td>-7.695973e-01</td>\n",
       "      <td>-1.134641e-01</td>\n",
       "      <td>-2.152307e-01</td>\n",
       "      <td>-2.446826e-01</td>\n",
       "      <td>-3.020091e-01</td>\n",
       "      <td>-3.061496e-01</td>\n",
       "      <td>-3.094782e-01</td>\n",
       "      <td>-3.337451e-01</td>\n",
       "      <td>...</td>\n",
       "      <td>-5.360124e-01</td>\n",
       "      <td>-8.971649e-03</td>\n",
       "      <td>-2.344496e-01</td>\n",
       "      <td>-6.328744e-01</td>\n",
       "      <td>-6.496472e-01</td>\n",
       "      <td>-3.117574e-01</td>\n",
       "      <td>-3.297937e-01</td>\n",
       "      <td>-3.420120e-01</td>\n",
       "      <td>-2.983202e-01</td>\n",
       "      <td>-6.456978e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>-2.548889e-01</td>\n",
       "      <td>-3.129784e-01</td>\n",
       "      <td>-2.323587e-01</td>\n",
       "      <td>1.299692e-01</td>\n",
       "      <td>1.061240e-01</td>\n",
       "      <td>7.290451e-02</td>\n",
       "      <td>3.647427e-02</td>\n",
       "      <td>4.296695e-02</td>\n",
       "      <td>4.286700e-02</td>\n",
       "      <td>1.840948e-02</td>\n",
       "      <td>...</td>\n",
       "      <td>-1.232627e-01</td>\n",
       "      <td>2.363103e-01</td>\n",
       "      <td>1.365973e-01</td>\n",
       "      <td>-4.356544e-01</td>\n",
       "      <td>-3.040989e-01</td>\n",
       "      <td>2.220294e-02</td>\n",
       "      <td>1.934255e-02</td>\n",
       "      <td>-8.798291e-02</td>\n",
       "      <td>-5.791425e-02</td>\n",
       "      <td>-3.515289e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>5.391371e-01</td>\n",
       "      <td>3.496738e-01</td>\n",
       "      <td>6.911947e-01</td>\n",
       "      <td>4.196976e-01</td>\n",
       "      <td>5.122451e-01</td>\n",
       "      <td>4.369811e-01</td>\n",
       "      <td>2.758701e-01</td>\n",
       "      <td>3.026659e-01</td>\n",
       "      <td>2.745033e-01</td>\n",
       "      <td>3.263090e-01</td>\n",
       "      <td>...</td>\n",
       "      <td>3.922545e-01</td>\n",
       "      <td>4.308233e-01</td>\n",
       "      <td>4.652162e-01</td>\n",
       "      <td>1.275690e-01</td>\n",
       "      <td>2.687979e-01</td>\n",
       "      <td>3.688783e-01</td>\n",
       "      <td>3.747462e-01</td>\n",
       "      <td>3.765118e-01</td>\n",
       "      <td>1.092467e-01</td>\n",
       "      <td>2.030451e-01</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>2.359459e+00</td>\n",
       "      <td>3.038246e+00</td>\n",
       "      <td>2.062386e+00</td>\n",
       "      <td>1.264991e+00</td>\n",
       "      <td>1.421640e+00</td>\n",
       "      <td>2.050068e+00</td>\n",
       "      <td>2.644496e+00</td>\n",
       "      <td>2.565916e+00</td>\n",
       "      <td>2.617264e+00</td>\n",
       "      <td>2.465788e+00</td>\n",
       "      <td>...</td>\n",
       "      <td>2.736179e+00</td>\n",
       "      <td>1.081616e+00</td>\n",
       "      <td>1.703814e+00</td>\n",
       "      <td>2.910829e+00</td>\n",
       "      <td>3.171427e+00</td>\n",
       "      <td>2.434734e+00</td>\n",
       "      <td>2.342218e+00</td>\n",
       "      <td>2.425320e+00</td>\n",
       "      <td>2.479116e+00</td>\n",
       "      <td>3.322991e+00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>8 rows × 24 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       Asset Turnover  Current Ratio  Gross Profit Margin  Net Profit Margin  \\\n",
       "count    8.904000e+05   8.904000e+05         8.904000e+05       8.904000e+05   \n",
       "mean     5.720092e-17  -5.377908e-16         2.553613e-18       9.908017e-17   \n",
       "std      1.000001e+00   1.000001e+00         1.000001e+00       1.000001e+00   \n",
       "min     -1.289389e+00  -1.112665e+00        -1.418441e+00      -4.276321e+00   \n",
       "25%     -7.636819e-01  -6.884963e-01        -7.695973e-01      -1.134641e-01   \n",
       "50%     -2.548889e-01  -3.129784e-01        -2.323587e-01       1.299692e-01   \n",
       "75%      5.391371e-01   3.496738e-01         6.911947e-01       4.196976e-01   \n",
       "max      2.359459e+00   3.038246e+00         2.062386e+00       1.264991e+00   \n",
       "\n",
       "       Return on Assets  Return on Equity  Earnings Growth  \\\n",
       "count      8.904000e+05      8.904000e+05     8.904000e+05   \n",
       "mean      -2.962191e-16     -2.266331e-18     4.290069e-17   \n",
       "std        1.000001e+00      1.000001e+00     1.000001e+00   \n",
       "min       -3.908327e+00     -3.357029e+00    -2.558049e+00   \n",
       "25%       -2.152307e-01     -2.446826e-01    -3.020091e-01   \n",
       "50%        1.061240e-01      7.290451e-02     3.647427e-02   \n",
       "75%        5.122451e-01      4.369811e-01     2.758701e-01   \n",
       "max        1.421640e+00      2.050068e+00     2.644496e+00   \n",
       "\n",
       "       Earnings Growth QOQ  Earnings Growth YOY    FCF Growth  ...  \\\n",
       "count         8.904000e+05         8.904000e+05  8.904000e+05  ...   \n",
       "mean          2.719598e-17         2.042890e-17 -3.843187e-17  ...   \n",
       "std           1.000001e+00         1.000001e+00  1.000001e+00  ...   \n",
       "min          -2.631511e+00        -2.639940e+00 -2.512918e+00  ...   \n",
       "25%          -3.061496e-01        -3.094782e-01 -3.337451e-01  ...   \n",
       "50%           4.296695e-02         4.286700e-02  1.840948e-02  ...   \n",
       "75%           3.026659e-01         2.745033e-01  3.263090e-01  ...   \n",
       "max           2.565916e+00         2.617264e+00  2.465788e+00  ...   \n",
       "\n",
       "       Sales Growth YOY  Earnings Yield     FCF Yield    Market-Cap  \\\n",
       "count      8.904000e+05    8.904000e+05  8.904000e+05  8.904000e+05   \n",
       "mean       9.448367e-17   -2.088855e-16 -2.911118e-17 -1.082732e-16   \n",
       "std        1.000001e+00    1.000001e+00  1.000001e+00  1.000001e+00   \n",
       "min       -1.836140e+00   -4.257038e+00 -3.393204e+00 -6.855607e-01   \n",
       "25%       -5.360124e-01   -8.971649e-03 -2.344496e-01 -6.328744e-01   \n",
       "50%       -1.232627e-01    2.363103e-01  1.365973e-01 -4.356544e-01   \n",
       "75%        3.922545e-01    4.308233e-01  4.652162e-01  1.275690e-01   \n",
       "max        2.736179e+00    1.081616e+00  1.703814e+00  2.910829e+00   \n",
       "\n",
       "             P/Book           P/E         P/FCF        P/NCAV      P/NetNet  \\\n",
       "count  8.904000e+05  8.904000e+05  8.904000e+05  8.904000e+05  8.904000e+05   \n",
       "mean  -2.493603e-16 -2.694061e-16 -5.822237e-17  9.512207e-18 -1.398103e-17   \n",
       "std    1.000001e+00  1.000001e+00  1.000001e+00  1.000001e+00  1.000001e+00   \n",
       "min   -1.102352e+00 -2.636409e+00 -2.457002e+00 -2.251178e+00 -2.297130e+00   \n",
       "25%   -6.496472e-01 -3.117574e-01 -3.297937e-01 -3.420120e-01 -2.983202e-01   \n",
       "50%   -3.040989e-01  2.220294e-02  1.934255e-02 -8.798291e-02 -5.791425e-02   \n",
       "75%    2.687979e-01  3.688783e-01  3.747462e-01  3.765118e-01  1.092467e-01   \n",
       "max    3.171427e+00  2.434734e+00  2.342218e+00  2.425320e+00  2.479116e+00   \n",
       "\n",
       "            P/Sales  \n",
       "count  8.904000e+05  \n",
       "mean   2.788545e-16  \n",
       "std    1.000001e+00  \n",
       "min   -8.794930e-01  \n",
       "25%   -6.456978e-01  \n",
       "50%   -3.515289e-01  \n",
       "75%    2.030451e-01  \n",
       "max    3.322991e+00  \n",
       "\n",
       "[8 rows x 24 columns]"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_train.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Using the same scaler on the test-data, also gives means that are close to 0 and standard deviations close to 1:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 30,
   "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>Asset Turnover</th>\n",
       "      <th>Current Ratio</th>\n",
       "      <th>Gross Profit Margin</th>\n",
       "      <th>Net Profit Margin</th>\n",
       "      <th>Return on Assets</th>\n",
       "      <th>Return on Equity</th>\n",
       "      <th>Earnings Growth</th>\n",
       "      <th>Earnings Growth QOQ</th>\n",
       "      <th>Earnings Growth YOY</th>\n",
       "      <th>FCF Growth</th>\n",
       "      <th>...</th>\n",
       "      <th>Sales Growth YOY</th>\n",
       "      <th>Earnings Yield</th>\n",
       "      <th>FCF Yield</th>\n",
       "      <th>Market-Cap</th>\n",
       "      <th>P/Book</th>\n",
       "      <th>P/E</th>\n",
       "      <th>P/FCF</th>\n",
       "      <th>P/NCAV</th>\n",
       "      <th>P/NetNet</th>\n",
       "      <th>P/Sales</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>...</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "      <td>219509.000000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>mean</th>\n",
       "      <td>-0.073532</td>\n",
       "      <td>0.101292</td>\n",
       "      <td>0.082474</td>\n",
       "      <td>-0.040152</td>\n",
       "      <td>-0.015602</td>\n",
       "      <td>-0.087059</td>\n",
       "      <td>0.036525</td>\n",
       "      <td>-0.019496</td>\n",
       "      <td>-0.008481</td>\n",
       "      <td>-0.018176</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.012192</td>\n",
       "      <td>-0.050717</td>\n",
       "      <td>-0.015696</td>\n",
       "      <td>-0.169045</td>\n",
       "      <td>0.039496</td>\n",
       "      <td>-0.052480</td>\n",
       "      <td>-0.020469</td>\n",
       "      <td>-0.007880</td>\n",
       "      <td>-0.072173</td>\n",
       "      <td>0.073102</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>std</th>\n",
       "      <td>0.937071</td>\n",
       "      <td>1.055220</td>\n",
       "      <td>0.978228</td>\n",
       "      <td>1.072309</td>\n",
       "      <td>1.028253</td>\n",
       "      <td>1.115918</td>\n",
       "      <td>1.011748</td>\n",
       "      <td>1.029467</td>\n",
       "      <td>1.023818</td>\n",
       "      <td>1.049672</td>\n",
       "      <td>...</td>\n",
       "      <td>1.014780</td>\n",
       "      <td>1.035817</td>\n",
       "      <td>0.966917</td>\n",
       "      <td>0.799262</td>\n",
       "      <td>1.002008</td>\n",
       "      <td>1.081207</td>\n",
       "      <td>1.015491</td>\n",
       "      <td>1.046830</td>\n",
       "      <td>0.996451</td>\n",
       "      <td>1.041276</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>min</th>\n",
       "      <td>-1.289389</td>\n",
       "      <td>-1.112665</td>\n",
       "      <td>-1.418441</td>\n",
       "      <td>-4.276321</td>\n",
       "      <td>-3.908327</td>\n",
       "      <td>-3.357029</td>\n",
       "      <td>-2.558049</td>\n",
       "      <td>-2.631511</td>\n",
       "      <td>-2.639940</td>\n",
       "      <td>-2.512918</td>\n",
       "      <td>...</td>\n",
       "      <td>-1.836140</td>\n",
       "      <td>-4.257038</td>\n",
       "      <td>-3.393204</td>\n",
       "      <td>-0.685561</td>\n",
       "      <td>-1.102352</td>\n",
       "      <td>-2.636409</td>\n",
       "      <td>-2.457002</td>\n",
       "      <td>-2.251178</td>\n",
       "      <td>-2.297130</td>\n",
       "      <td>-0.879493</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>25%</th>\n",
       "      <td>-0.778601</td>\n",
       "      <td>-0.634935</td>\n",
       "      <td>-0.713012</td>\n",
       "      <td>-0.147471</td>\n",
       "      <td>-0.257038</td>\n",
       "      <td>-0.342831</td>\n",
       "      <td>-0.336577</td>\n",
       "      <td>-0.352551</td>\n",
       "      <td>-0.336894</td>\n",
       "      <td>-0.393597</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.574930</td>\n",
       "      <td>-0.080457</td>\n",
       "      <td>-0.278461</td>\n",
       "      <td>-0.648587</td>\n",
       "      <td>-0.582690</td>\n",
       "      <td>-0.383735</td>\n",
       "      <td>-0.363888</td>\n",
       "      <td>-0.399488</td>\n",
       "      <td>-0.399617</td>\n",
       "      <td>-0.634951</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>50%</th>\n",
       "      <td>-0.264721</td>\n",
       "      <td>-0.228923</td>\n",
       "      <td>-0.005686</td>\n",
       "      <td>0.111629</td>\n",
       "      <td>0.117563</td>\n",
       "      <td>0.055753</td>\n",
       "      <td>0.037830</td>\n",
       "      <td>0.037644</td>\n",
       "      <td>0.032183</td>\n",
       "      <td>0.023868</td>\n",
       "      <td>...</td>\n",
       "      <td>-0.166401</td>\n",
       "      <td>0.221903</td>\n",
       "      <td>0.119017</td>\n",
       "      <td>-0.497031</td>\n",
       "      <td>-0.273378</td>\n",
       "      <td>-0.003164</td>\n",
       "      <td>0.009331</td>\n",
       "      <td>-0.098033</td>\n",
       "      <td>-0.084065</td>\n",
       "      <td>-0.302029</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>75%</th>\n",
       "      <td>0.362259</td>\n",
       "      <td>0.458257</td>\n",
       "      <td>0.697248</td>\n",
       "      <td>0.453733</td>\n",
       "      <td>0.555522</td>\n",
       "      <td>0.416461</td>\n",
       "      <td>0.311117</td>\n",
       "      <td>0.300353</td>\n",
       "      <td>0.299220</td>\n",
       "      <td>0.348709</td>\n",
       "      <td>...</td>\n",
       "      <td>0.400847</td>\n",
       "      <td>0.424382</td>\n",
       "      <td>0.459260</td>\n",
       "      <td>-0.087066</td>\n",
       "      <td>0.313428</td>\n",
       "      <td>0.348195</td>\n",
       "      <td>0.392931</td>\n",
       "      <td>0.389160</td>\n",
       "      <td>0.085461</td>\n",
       "      <td>0.338123</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>max</th>\n",
       "      <td>2.359459</td>\n",
       "      <td>3.038246</td>\n",
       "      <td>2.062386</td>\n",
       "      <td>1.264991</td>\n",
       "      <td>1.421640</td>\n",
       "      <td>2.050068</td>\n",
       "      <td>2.644496</td>\n",
       "      <td>2.565916</td>\n",
       "      <td>2.617264</td>\n",
       "      <td>2.465788</td>\n",
       "      <td>...</td>\n",
       "      <td>2.736179</td>\n",
       "      <td>1.081616</td>\n",
       "      <td>1.703814</td>\n",
       "      <td>2.910829</td>\n",
       "      <td>3.171427</td>\n",
       "      <td>2.434734</td>\n",
       "      <td>2.342218</td>\n",
       "      <td>2.425320</td>\n",
       "      <td>2.479116</td>\n",
       "      <td>3.322991</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>8 rows × 24 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "       Asset Turnover  Current Ratio  Gross Profit Margin  Net Profit Margin  \\\n",
       "count   219509.000000  219509.000000        219509.000000      219509.000000   \n",
       "mean        -0.073532       0.101292             0.082474          -0.040152   \n",
       "std          0.937071       1.055220             0.978228           1.072309   \n",
       "min         -1.289389      -1.112665            -1.418441          -4.276321   \n",
       "25%         -0.778601      -0.634935            -0.713012          -0.147471   \n",
       "50%         -0.264721      -0.228923            -0.005686           0.111629   \n",
       "75%          0.362259       0.458257             0.697248           0.453733   \n",
       "max          2.359459       3.038246             2.062386           1.264991   \n",
       "\n",
       "       Return on Assets  Return on Equity  Earnings Growth  \\\n",
       "count     219509.000000     219509.000000    219509.000000   \n",
       "mean          -0.015602         -0.087059         0.036525   \n",
       "std            1.028253          1.115918         1.011748   \n",
       "min           -3.908327         -3.357029        -2.558049   \n",
       "25%           -0.257038         -0.342831        -0.336577   \n",
       "50%            0.117563          0.055753         0.037830   \n",
       "75%            0.555522          0.416461         0.311117   \n",
       "max            1.421640          2.050068         2.644496   \n",
       "\n",
       "       Earnings Growth QOQ  Earnings Growth YOY     FCF Growth  ...  \\\n",
       "count        219509.000000        219509.000000  219509.000000  ...   \n",
       "mean             -0.019496            -0.008481      -0.018176  ...   \n",
       "std               1.029467             1.023818       1.049672  ...   \n",
       "min              -2.631511            -2.639940      -2.512918  ...   \n",
       "25%              -0.352551            -0.336894      -0.393597  ...   \n",
       "50%               0.037644             0.032183       0.023868  ...   \n",
       "75%               0.300353             0.299220       0.348709  ...   \n",
       "max               2.565916             2.617264       2.465788  ...   \n",
       "\n",
       "       Sales Growth YOY  Earnings Yield      FCF Yield     Market-Cap  \\\n",
       "count     219509.000000   219509.000000  219509.000000  219509.000000   \n",
       "mean          -0.012192       -0.050717      -0.015696      -0.169045   \n",
       "std            1.014780        1.035817       0.966917       0.799262   \n",
       "min           -1.836140       -4.257038      -3.393204      -0.685561   \n",
       "25%           -0.574930       -0.080457      -0.278461      -0.648587   \n",
       "50%           -0.166401        0.221903       0.119017      -0.497031   \n",
       "75%            0.400847        0.424382       0.459260      -0.087066   \n",
       "max            2.736179        1.081616       1.703814       2.910829   \n",
       "\n",
       "              P/Book            P/E          P/FCF         P/NCAV  \\\n",
       "count  219509.000000  219509.000000  219509.000000  219509.000000   \n",
       "mean        0.039496      -0.052480      -0.020469      -0.007880   \n",
       "std         1.002008       1.081207       1.015491       1.046830   \n",
       "min        -1.102352      -2.636409      -2.457002      -2.251178   \n",
       "25%        -0.582690      -0.383735      -0.363888      -0.399488   \n",
       "50%        -0.273378      -0.003164       0.009331      -0.098033   \n",
       "75%         0.313428       0.348195       0.392931       0.389160   \n",
       "max         3.171427       2.434734       2.342218       2.425320   \n",
       "\n",
       "            P/NetNet        P/Sales  \n",
       "count  219509.000000  219509.000000  \n",
       "mean       -0.072173       0.073102  \n",
       "std         0.996451       1.041276  \n",
       "min        -2.297130      -0.879493  \n",
       "25%        -0.399617      -0.634951  \n",
       "50%        -0.084065      -0.302029  \n",
       "75%         0.085461       0.338123  \n",
       "max         2.479116       3.322991  \n",
       "\n",
       "[8 rows x 24 columns]"
      ]
     },
     "execution_count": 30,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "X_test.describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Neural Network: Regression\n",
    "\n",
    "Let us now build a Neural Network to learn the mapping from signals X to stock-returns y. We will use Keras and TensorFlow. If you are new to this, you can watch [this tutorial](https://www.youtube.com/watch?v=3yfRJKA1BiQ) for a basic explanation."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [],
   "source": [
    "# This so-called \"activation function\" is applied after each layer.\n",
    "# It is important to use a non-linear function, otherwise\n",
    "# the Neural Network cannot learn non-linear mappings.\n",
    "activation = 'relu'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 32,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense (Dense)                (None, 128)               3200      \n",
      "_________________________________________________________________\n",
      "dense_1 (Dense)              (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "dense_2 (Dense)              (None, 32)                2080      \n",
      "_________________________________________________________________\n",
      "dense_3 (Dense)              (None, 16)                528       \n",
      "_________________________________________________________________\n",
      "dense_4 (Dense)              (None, 8)                 136       \n",
      "_________________________________________________________________\n",
      "dense_5 (Dense)              (None, 1)                 9         \n",
      "=================================================================\n",
      "Total params: 14,209\n",
      "Trainable params: 14,209\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# Create a new Keras model.\n",
    "model_regr = Sequential()\n",
    "\n",
    "# Add an input layer for the signals.\n",
    "# Note how we set the dimensionality of the input here.\n",
    "model_regr.add(InputLayer(input_shape=(num_signals,)))\n",
    "\n",
    "# Add several dense aka. fully-connected layers.\n",
    "# You can experiment with different designs.\n",
    "model_regr.add(Dense(128, activation=activation))\n",
    "model_regr.add(Dense(64, activation=activation))\n",
    "model_regr.add(Dense(32, activation=activation))\n",
    "model_regr.add(Dense(16, activation=activation))\n",
    "model_regr.add(Dense(8, activation=activation))\n",
    "\n",
    "# Add a layer for the output of the Neural Network.\n",
    "# This is 1-dimensional to match the stock-return data.\n",
    "model_regr.add(Dense(1))\n",
    "\n",
    "# Compile the model but don't train it yet.\n",
    "model_regr.compile(loss='mse', metrics=['mae'],\n",
    "                   optimizer=RMSprop(0.001))\n",
    "\n",
    "# Show the model.\n",
    "model_regr.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "These are common arguments for training both our regression and classification models:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 33,
   "metadata": {},
   "outputs": [],
   "source": [
    "fit_args = \\\n",
    "{\n",
    "    # For efficiency, the model is trained on batches of data.\n",
    "    'batch_size': 4096,\n",
    "    \n",
    "    # Number of iterations aka. epochs over the training-set.\n",
    "    'epochs': 40,\n",
    "    \n",
    "    # Fraction of the training-set used for validation after\n",
    "    # each training-epoch, to assess how well the model performs\n",
    "    # on unseen data.\n",
    "    'validation_split': 0.2,\n",
    "\n",
    "    # Show status during training.\n",
    "    'verbose': 1,\n",
    "}"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can now train the model. This takes a minute or two on an ordinary computer:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 34,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 712320 samples, validate on 178080 samples\n",
      "Epoch 1/40\n",
      "712320/712320 [==============================] - 2s 3us/sample - loss: 0.0374 - mae: 0.1462 - val_loss: 0.0668 - val_mae: 0.1940\n",
      "Epoch 2/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0273 - mae: 0.1252 - val_loss: 0.0747 - val_mae: 0.2033\n",
      "Epoch 3/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0225 - mae: 0.1141 - val_loss: 0.0770 - val_mae: 0.2102\n",
      "Epoch 4/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0194 - mae: 0.1063 - val_loss: 0.0794 - val_mae: 0.2115\n",
      "Epoch 5/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0173 - mae: 0.1005 - val_loss: 0.0815 - val_mae: 0.2145\n",
      "Epoch 6/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0157 - mae: 0.0961 - val_loss: 0.0847 - val_mae: 0.2185\n",
      "Epoch 7/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0145 - mae: 0.0925 - val_loss: 0.0841 - val_mae: 0.2169\n",
      "Epoch 8/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0136 - mae: 0.0895 - val_loss: 0.0871 - val_mae: 0.2219\n",
      "Epoch 9/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0128 - mae: 0.0870 - val_loss: 0.0883 - val_mae: 0.2232\n",
      "Epoch 10/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0121 - mae: 0.0847 - val_loss: 0.0897 - val_mae: 0.2239\n",
      "Epoch 11/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0115 - mae: 0.0828 - val_loss: 0.0885 - val_mae: 0.2237\n",
      "Epoch 12/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0110 - mae: 0.0809 - val_loss: 0.0893 - val_mae: 0.2241\n",
      "Epoch 13/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0105 - mae: 0.0793 - val_loss: 0.0932 - val_mae: 0.2296\n",
      "Epoch 14/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0102 - mae: 0.0778 - val_loss: 0.0919 - val_mae: 0.2274\n",
      "Epoch 15/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0098 - mae: 0.0765 - val_loss: 0.0915 - val_mae: 0.2279\n",
      "Epoch 16/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0095 - mae: 0.0750 - val_loss: 0.0912 - val_mae: 0.2263\n",
      "Epoch 17/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0092 - mae: 0.0740 - val_loss: 0.0916 - val_mae: 0.2276\n",
      "Epoch 18/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0089 - mae: 0.0728 - val_loss: 0.0922 - val_mae: 0.2281\n",
      "Epoch 19/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0087 - mae: 0.0720 - val_loss: 0.0921 - val_mae: 0.2281\n",
      "Epoch 20/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0085 - mae: 0.0709 - val_loss: 0.0941 - val_mae: 0.2300\n",
      "Epoch 21/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0083 - mae: 0.0702 - val_loss: 0.0921 - val_mae: 0.2274\n",
      "Epoch 22/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0081 - mae: 0.0693 - val_loss: 0.0957 - val_mae: 0.2321\n",
      "Epoch 23/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0079 - mae: 0.0684 - val_loss: 0.0937 - val_mae: 0.2293\n",
      "Epoch 24/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0077 - mae: 0.0676 - val_loss: 0.0938 - val_mae: 0.2294\n",
      "Epoch 25/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0075 - mae: 0.0669 - val_loss: 0.0924 - val_mae: 0.2282\n",
      "Epoch 26/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0074 - mae: 0.0663 - val_loss: 0.0967 - val_mae: 0.2329\n",
      "Epoch 27/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0072 - mae: 0.0655 - val_loss: 0.0938 - val_mae: 0.2295\n",
      "Epoch 28/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0071 - mae: 0.0651 - val_loss: 0.0944 - val_mae: 0.2308\n",
      "Epoch 29/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0070 - mae: 0.0642 - val_loss: 0.0944 - val_mae: 0.2303\n",
      "Epoch 30/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0068 - mae: 0.0636 - val_loss: 0.0965 - val_mae: 0.2329\n",
      "Epoch 31/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0068 - mae: 0.0632 - val_loss: 0.0931 - val_mae: 0.2291\n",
      "Epoch 32/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0066 - mae: 0.0626 - val_loss: 0.0945 - val_mae: 0.2310\n",
      "Epoch 33/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0065 - mae: 0.0620 - val_loss: 0.0936 - val_mae: 0.2297\n",
      "Epoch 34/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0064 - mae: 0.0615 - val_loss: 0.0949 - val_mae: 0.2310\n",
      "Epoch 35/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0064 - mae: 0.0613 - val_loss: 0.0942 - val_mae: 0.2307\n",
      "Epoch 36/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0062 - mae: 0.0606 - val_loss: 0.0940 - val_mae: 0.2300\n",
      "Epoch 37/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0062 - mae: 0.0604 - val_loss: 0.0950 - val_mae: 0.2320\n",
      "Epoch 38/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0061 - mae: 0.0598 - val_loss: 0.0942 - val_mae: 0.2312\n",
      "Epoch 39/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0060 - mae: 0.0594 - val_loss: 0.0949 - val_mae: 0.2318\n",
      "Epoch 40/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.0059 - mae: 0.0589 - val_loss: 0.0964 - val_mae: 0.2337\n",
      "CPU times: user 3min 10s, sys: 7.33 s, total: 3min 17s\n",
      "Wall time: 46.3 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "history_regr = model_regr.fit(x=X_train.values,\n",
    "                              y=y_train.values, **fit_args)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can then plot the training-progress as the Mean Absolute Error (MAE) which is supposed to be minimized. After 40 epochs through the training-set, the model has learned to map from signals to stock-returns for the training-set, with an average absolute error of about 6%, which would be very good if it generalized to unseen data. Unfortunately, the performance on the validation-set gets worse during optimization and ends with a MAE about 24%, so the model is clearly not learning the underlying relation between signals and stock-returns, but it has instead learned peculiarities and noise in the training-data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 35,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f3751014470>"
      ]
     },
     "execution_count": 35,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "iVBORw0KGgoAAAANSUhEUgAAAYAAAAD4CAYAAADlwTGnAAAABHNCSVQICAgIfAhkiAAAAAlwSFlzAAALEgAACxIB0t1+/AAAADh0RVh0U29mdHdhcmUAbWF0cGxvdGxpYiB2ZXJzaW9uMy4xLjIsIGh0dHA6Ly9tYXRwbG90bGliLm9yZy8li6FKAAAgAElEQVR4nO3deXwTdf748VeOpuldWtq00FIEChRaBeQQQdByU8qNfvl64CrqKniAqOj+hF1WRV1WRDzxgGXV3a+igEsFlHoAuggIWIFyKYUWaAq9D9I0yfz+mDZQ29KCLUnI+/l45DHJZGbyninMe+bz+czno1EURUEIIYTX0bo6ACGEEK4hCUAIIbyUJAAhhPBSkgCEEMJLSQIQQggvpXd1ABdj165d+Pn5uTqMJqmsrMTX19fVYTSJxNoyPCVWT4kTJNZLVVlZSY8ePerM96gEoNFoSEhIcHUYTZKZmSmxtgCJtfl5SpwgsV6qzMzMeudLEZAQQngpSQBCCOGlJAEIIYSXkgQghBBeShKAEEJ4KUkAQgjhpSQBCCGEl/Ko5wCEEEI0UZUFcnbA6QMQNLDeRSQBCOHNTmUQkfEOBN0BMb1dF0fBr3DoCzB1g/Y3gEbjulg8lc0KJ3fB0S2QtRmyt4PNAlofmLKl3lUkAQhxuTkckPF/UJgF3SdAZNfLH0NlGXyzELa9QWvFDpn/gHb94foHofMo0F6G0uGi47BvNez9FE7tOTc/bgAMfgKuGnRlJgKbVb0yP7oZHFUQFA3Bbc5NAyJAq6t/XUUBSzGcLYCKQqjIB/NeyNoCx7dBVQWggahE6H03XHWD+nfNOlXv5iQBCHE55f4M62ZDznb187fPQ5uecM1USJwEAa1bPoaDG+DzOVCcDdfeyeE2E4ivOgDbXoN//y+EdYT+M6DH/4JPE/reqiyF/F9AbwS/VupLb6h/2eITsH+NetI/sVOd16YXDH8GuoyGI+mw9SVYORbaXQ83PgFXDW5aIjhbqJ4EHTaI6AqtrgLdRZziHHY1KZ05BBUF4GMEvZ96DGpeej91vo8/GEMaPlGfT1HUYphfvoZfv4as76CqHDRaQAOKvfbyGh0ERakJIaA1WEqqT/j56j46bHV/IyIBet6m3j21Hwj+Yb9ZQBKAEL+frRJO7gbfYLW4oqkqS+HrhfDDm+AXCuPfgI5DYO8n8NO/YP3jsPEp6DQMrvkf6DxSPdE0p5KTsP4JyPxMPWHctRHaXYctMxMSboQ+09Xvvn8F0mbD189Cn3ug7z3qichmhfzDYN4PefshLxPy9qknzd/yCVBPQn6h55JCWR4c/6/6fdTVMGS+egcUdtW59cI7Qq87YPc/YctLsHKcegU7+AnocGPt3zhbCMf+C1lb1Svg3J+B80a41fpAeCeI6KwmhNadIaILhMZBcQ6cOQinD52b5h9Wi0yaSqMF/9YQaILACHUaUD0NjCQk5zhkvgy/fgNlueo6YR2hx1R1X9rfAL5BUH5a/duUnjpvegpKT6oJ0xgCrePB/zrwCwP/cPXY+oerr9A49fcvgcaTxgTevXs3PXv2dHUYTeJOHUE15nfFWngMNjypXsV0Ggrxw6BV+2aN73yX/bjabWrxxNFv1bLV49vAdlb9LiJBvWpPnKieuOqLtWtX2L9WPUalJ+HaO9UT32+v0Mz7IePfkPGRegIwhkC38RASo16ZKnZQHOfeOxzqVGdQfzs8Xj3BBbSue7XssMOOdyF9gVrkMPhx6P+g8yq9zjFVFDj2PXy/FA6tV6/sW7WH/CPnrj61evX3IhPUV+vO6ndnC6tfRepVtPNzIeh9IWGsetJv3anxY19lOZcISk9CbD9ORg2lja6w+oS/F1BA5wuxfauvfgeoV+qnD557nTmoFrcpjnp+RAOhsdC6i5ocIrqo7wMj1GRfVaHGUXVW/btX1bwq1Cvysjz1VV49LTOD3Xpu835h6sm+403qNLRd4/vdAhr6fyMJoIV4RQLY+wn8ZxagqFd4RcfU+eHxaiLoNFQtz23GK9lLjtVmVW+ja26hNdpzLzTV76unliL1Nv3oZvVEaC1VtxHZXS1TbT8QSnPVYozj36vfRfeApMnqyS0kBoAjO76k08E34cgmMCXBmMUQ2+fCcTrsarL56d+Q+Z/qMl3UGLU6tXhAoz333mYBe+W59Y0h1ckgXr36DYmF7W/BiR+hYzKk/B3COjT9mJ4+BD+8oe5vZAJEdlNf4Z0aLuZpbrZK2LXyXCLQGyGmz7nijrbXNv5vrMqiJrAzB9U7luAY9c4gPB4M/s0Xa00ZfVkevx45RId+oy9PfUojJAFcZld0ArCWq0UWu99X/yNOeke9Dc0/op7sDn+p3pbbK9Wy0vY3QNz16hVrZan6spSce19ZrE6NIWp5eJte0LaXeoX9mzLceq9WS3PhdKZ6tVd4rLqstLrMtKJAfdWcxC9GeCe1IvKqQeo+1Fc+X5xTXZH5iVo0BGrZdVQijh//gVbnAzf9Cfree3Hl0VB9ta9Un+wbKAN3ONSy/PzDcKb6lX8YzhxRT5agFkuMfF69W6lnOx7zb9VWya/b19OhTwsUj7UAdzquDcXSpH+Rmzdv5tlnn8XhcDBlyhTuvffeWt8vX76cjz/+GJ1OR1hYGM899xxt27YlMzOTP//5z5SVlaHVarn//vsZPXo0AHPnzmX79u0EBQUB8Pzzz7vNwRIXcOonWHW3erK/YQ7cOBd0Pup3rauvPK+7H6wVahI48qWaEA5vVJfR+oAxWC379A1Sy9KDY8A3UC0L3bcaflyhLqv3g+irzyUEU3cCcndA0dfnTvinD6hXXDUMgefKRv3D1aIJZ7lpK/W9zkctDnC+lOpX9We9L7S7Tm2R0ZiQGLXlzPUPqhWhez+Fvatg+zLKYocQPOXVpm2nPk2pYNRqoVWc+uo0tPZ3lWVq88pW7dVj7un0vlS26uIRJ39P0WgCsNvtLFiwgOXLl2MymZg8eTLJycl06nSuDC8hIYFPPvkEPz8/PvzwQ/72t7/x8ssvYzQaeeGFF2jfvj1ms5lJkyYxcOBAgoPVf4yPP/44I0eObLm9E81HUWDbG7BpvnoynfaZemXcEIM/dB6uvqC6DNhPPbleqEWHoqgnrRO71DbNJ3apCeGHNwBwlqD6halFEomT1Qq+yK7qHUN9ZeCXS3hHGPwYDJoDlaWcOHqC4Es9+TcH30A1gQrRgEYTQEZGBnFxccTGxgKQkpJCenp6rQRw3XXXOd/36NGDzz77DICrrjpXu28ymQgLC6OgoMCZAISHKDsNax+Aw1+oTfXGvgoB4Re3Db9WTVtOo6mu1OwIV09R59lt6pV+XibHCiqJ6z3ikls9XBYaTfUV9wlXRyLEBTWaAMxmM1FRUc7PJpOJjIyMBpdftWoVgwbVvTLMyMigqqqKdu3O1YIvXryY1157jf79+zNnzhwMhgtXKjkcjgaHNnM3Fovl8seqONBZS9FVFqC3FKCzFKK1V6JofVB0PihaA4pWj0NrqP6svjQVJRwz70BXVY625mVTp7qqcoJObEZrLSWv1xwKO02C43lA3uXdN3Tgk4glxEJF9hngzGX+/Yvnkn8Dl8BT4gSJtbk163MAa9euZe/evbz//vu15ufl5fHYY4/xwgsvoK2uEZ89ezYRERFUVVXx9NNPs2zZMmbOnHnB7Wu1Wo+pJ2iRCiCHA4qyIO+A2g47/4ja7KzstFp+XnGm/odELpXeqJbTR8TDmJeIMnUnqvG1WpQ7Vaw1xlNi9ZQ4QWK9VA0lokYTgMlkIjc31/nZbDZjMpnqLPf999/z5ptv8v7779e6ki8rK+O+++5j1qxZtUalj4yMBMBgMDBx4kTee++9pu+NN6gogJyd6om+uviD0wfPtUEHCG6rPjEYEgNtelQ/hBKpTgNaQ0Ck2ibaXqW2yLFb1eaQ9kp1nk2dl23OJ7Zjt3MVs8YQtTL1cjXzE0K4RKMJICkpiaysLLKzszGZTKSlpfH3v/+91jL79+9n3rx5vPPOO4SHnysbtlqtzJgxg3HjxtWp7M3LyyMyMhJFUdi0aRPx8fHNtEseTlHUSs8vnj7XdDEoWq3o7H3XucrOiC7N1rKjTJcJ7d3jSkUIcfk0mgD0ej3z5s1j+vTp2O12Jk2aRHx8PEuWLCExMZEhQ4bw4osvUlFRwcMPPwxAdHQ0b775JuvXr2fnzp0UFRWxevVq4Fxzzzlz5lBYWIiiKHTt2pW//OUvLbunnqAwCz57UH0Aqf0NahNLU/emV6AKIcRFaFIdwODBgxk8eHCteTUne4AVK1bUu964ceMYN25cvd+tXLmyiSF6AYcDdrwNm/6sPt055mW1y4ArsSdEIYTbkM7gWoJ5Hz6lJ0Dp2vhJPP8XWDtD7SSr4xBIXaL2TSKEEC1MEkBzOrlH7XDrl3Q6AXwVovYR06bnuVdoOzUpOOyw7XX46hn14ajxb6hdAstVvxDiMpEE0BzOHFZP5PvXqOX1Q//CqSIL0Uqu2j/Mf19Te2EE9QnWNj3VvmpO7lYfrEp5CYKjXbsPQgivIwng9yg+oQ7osfsDtc38oMfh+plgDKEoM5PomjbAtkow71NP+Cd3q3cKlcUw6d0GO+gSQoiWJgngUlQUwJa/w/a31c7D+t4DNzyqtsGvj95X7cysba/LG6cQQlyAJICLkf8L7PoH7FwO1jK4+n/Uppqt4lwdmRBCXDRJAI2psqgDc+z6hzoKkUYHCWPgxifV3iiFEMJDSQJoSF4m/PgPdZi+s4XqgCfJT6sDLwe5ukccIYT4/SQB1LBVqq15TvyojnSVs10dvCRhDPSaBlcNdouh3YQQorl4XwKwWdVeNE9nqr1q1kwLflWHLAR1nNDhz6jt8usbBlAIIa4A3pMArOWw5gE4sO5cl8karTpAdkRX6D6+emSpbmrZvjTNFEJc4bwjAVQUwAdT1CEG+94Lba9VT/Lh8TK+qBDCa135CaDoOPxzojq9eSUkpLo6IiGEcAtXdgIw74P3J4G1Am5fDe0HuDoiIYRwG1duAjj2PXz4P2Dwh7vWq/3qCyGEcLoy2zVmroOV49WuGe7+Qk7+QghRjysvAex8Dz66HaKS4K6NavfLQggh6rhyioAUBb59Eb55DuKHw5QVYAhwdVRCCOG2rpw7gC1/V0/+1/wv/M+HcvIXQohGNCkBbN68mREjRjBs2DCWLVtW5/vly5czevRoUlNTmTZtGidOnHB+t3r1aoYPH87w4cOdA8MD7N27l9TUVIYNG8YzzzyDoiiXvhcH16sDsiRNgfGvg87n0rclhBBeotEEYLfbWbBgAe+88w5paWmsW7eOI0eO1FomISGBTz75hP/85z+MGDGCv/3tbwAUFRXx6quv8tFHH/Hxxx/z6quvUlxcDMCf//xn/vrXv/LFF1+QlZXF5s2bL20P8g7AJ/dA9DUwdqk8wSuEEE3UaALIyMggLi6O2NhYDAYDKSkppKen11rmuuuuw8/PD4AePXqQm5sLwNatWxkwYAChoaGEhIQwYMAAtmzZQl5eHmVlZfTo0QONRsP48ePrbLNJzhbCv6eCj59a7OPjd/HbEEIIL9VoJbDZbCYq6lz3xyaTiYyMjAaXX7VqFYMGDWpwXbPZXGd+VFQUZrO50WAdDgeZmZnVH2zEbp5NQNFxjt30OmdPlsDJkka3cblYLJZzsbo5ibVleEqsnhInSKzNrVlbAa1du5a9e/fy/vvvN+dmnbRaLQk14+xu/BOYt8PYpbTvdUuL/N7vkZmZeS5WNyextgxPidVT4gSJ9VI1lIgaLQIymUzOIh1Qr+pNJlOd5b7//nvefPNN3njjDQwGwwXX/e383NzcerfZoD3/gv++Cn3vg153NH09IYQQTo0mgKSkJLKyssjOzsZqtZKWlkZycnKtZfbv38+8efN44403CA8Pd84fOHAgW7dupbi4mOLiYrZu3crAgQOJjIwkMDCQPXv2oCgKa9asYciQIU2LOGcn/OdhaH8DjHj24vZWCCGEU6NFQHq9nnnz5jF9+nTsdjuTJk0iPj6eJUuWkJiYyJAhQ3jxxRepqKjg4YcfBiA6Opo333yT0NBQHnjgASZPngzAjBkzCA0NBWD+/Pk8+eSTWCwWBg0a5Kw3uCBHFfz7dnVIxptXSnNPIYT4HZpUBzB48GAGDx5ca17NyR5gxYoVDa47efJkZwI4X1JSEuvWrWtimCpD2QmoLIXbPwX/sItaVwghRG0e9SSw1nYWJr4lnbsJIUQz8Ki+gKr8TZDQxLoCIYQQF+RRdwB2oxT7CCFEc/GoBCCEEKL5SAIQQggvJQlACCG8lCQAIYTwUpIAhBDCS0kCEEIILyUJQAghvJQkACGE8FKSAIQQwktJAhBCCC8lCUAIIbyUJAAhhPBSkgCEEMJLSQIQQggvJQlACCG8lCQAIYTwUk1KAJs3b2bEiBEMGzaMZcuW1fl+x44dTJgwgW7durFhwwbn/G3btjFu3DjnKykpiU2bNgEwd+5ckpOTnd9lZmY20y4JIYRoikaHhLTb7SxYsIDly5djMpmYPHkyycnJdOrUyblMdHQ0Cxcu5L333qu17nXXXcfatWsBKCoqYvjw4QwYMMD5/eOPP87IkSOba1+EEEJchEYTQEZGBnFxccTGxgKQkpJCenp6rQQQExMDgFbb8A3Fxo0bueGGG/Dz8/u9MQshhGgGjRYBmc1moqKinJ9NJhNms/mifygtLY0xY8bUmrd48WJSU1N57rnnsFqtF71NIYQQl67RO4DmkJeXx6FDhxg4cKBz3uzZs4mIiKCqqoqnn36aZcuWMXPmzAtux+FweExdgcVikVhbgMTa/DwlTpBYm1ujCcBkMpGbm+v8bDabMZlMF/Uj69evZ9iwYfj4+DjnRUZGAmAwGJg4cWKd+oP6aLVaEhISLuq3XSUzM1NibQESa/PzlDhBYr1UDSWiRouAkpKSyMrKIjs7G6vVSlpaGsnJyRf142lpaaSkpNSal5eXB4CiKGzatIn4+PiL2qYQQojfp9E7AL1ez7x585g+fTp2u51JkyYRHx/PkiVLSExMZMiQIWRkZDBz5kxKSkr4+uuvWbp0KWlpaQDk5ORw6tQp+vbtW2u7c+bMobCwEEVR6Nq1K3/5y19aZg+FEELUq0l1AIMHD2bw4MG15j388MPO91dffTWbN2+ud92YmBi2bNlSZ/7KlSsvJk4hhBDNTJ4EFkIILyUJQAghvJQkACGE8FKSAIQQwktJAhBCCC8lCUAIIbyUJAAhhPBSkgCEEMJLSQIQQggvJQlACCG8lCQAIYTwUpdlPAAhhOeoqqoiJycHi8Xi6lDqqKqqcvs+9mu4Ilaj0UhMTEytrvcvRBKAEKKWnJwcgoKCaN++PRqNxtXh1HL27FmPGVb2cseqKAr5+fnk5ORw1VVXNWkdKQISQtRisVgIDw93u5O/uDCNRkN4ePhF3blJAhBC1CEnf890sX83SQBCCLdSWFjIuHHjGDduHAMGDOCGG25wfq6qqmrSNp588kl+/fXXCy7zwQcf8NlnnzVHyEydOrXOSIkPPvggvXv3rjXv3Xff5ZprrqGsrMw57/vvv+faa6917uO4cePYtm1bs8TVGKkDEEK4lVatWrF27VoAli5dir+/P3fffTeglquDWt6tKApabf3XsAsXLmz0d2699dZmilgVEBDAnj176NGjB0VFReTn59dZJi0tje7du7Np0ybGjx/vnN+vXz9ef/31Zo2nKeQOQAjhEY4dO8bEiRN59NFHSUlJ4fTp0zz99NNMnDiRlJQUXn31VeeyU6dOJTMzE5vNRu/evVm0aBFjx47llltucZ6YFy9ezIoVK5zLL1q0iMmTJzNixAh27doFQEVFBQ8++CCjR4/moYceYuLEiQ227ElJSXEOhbtx40aGDRtW6/ujR49is9l48MEHncu5mtwBCCEa9MmPOXy0M7tZt3lz71gmXRtzSesePXqUF198kaSkJAAeffRRQkNDsdls3HHHHYwcOZJOnTrVWqe0tJQ+ffowZ84cFi5cyCeffMK9995bZ9uKorBq1SrS09N57bXXePfdd3n//fdp3bo1S5cu5cCBA0yYMKHB2K6//nr+9Kc/4XA4+Pzzz3n66ad57733nN+npaWRkpJC3759efLJJykoKCAsLAyAH374gXHjxjmXfe2114iJubRjdDEkAQghPEZMTIzz5A/qSXXVqlXYbDby8vI4cuRInQRgNBqdY5p3796dnTt31rvt4cOHA5CYmMiJEycA+PHHH7nnnnsA6Nq1a51tn8/Hx4cePXqQlpaGw+HAZDLV+j4tLY1ly5ah0+kYOnQoGzduZOrUqYDrioCalAA2b97Ms88+i8PhYMqUKXWy544dO3juuec4ePAgL730EiNHjnR+l5CQQOfOnQGIjo7mzTffBCA7O5vZs2dTVFRE9+7defHFFzEYDM21X0KIZjDp2phLvlpvCee3q8/KymLlypV8/PHHBAcHM2fOHCorK+usc/5DUTqdDrvdXu+2a84/Wq22wWUaM3r0aB555BEefvjhWvP3799PdnY206ZNA8BqtdK+fXtnAnCVRusA7HY7CxYs4J133iEtLY1169Zx5MiRWstER0ezcOFCxowZU2d9o9HI2rVrWbt2rfPkD7Bo0SLuvPNOvvzyS4KDg1m1alUz7I4QwluUlZUREBBAYGAgeXl5bN26tdl/o1evXqxfvx6AgwcP8ssvv1xw+X79+nHPPfcwevToWvPXrVvHI488wldffcVXX33Fli1byMnJITc3t9ljvhiNJoCMjAzi4uKIjY3FYDCQkpJCenp6rWViYmLo2rVrgzXyv6UoCtu2bWPEiBEATJgwoc42hRDiQrp3707Hjh0ZNWoUTzzxBL169Wr237jtttswm82MHj2a1157jY4dOxIYGNjg8lqtlunTpxMaGuqcpygK69evr1UprNFoGDp0KJ9//jlwrg6g5vXll182+77UR6MoinKhBTZs2MCWLVt49tlnAVizZg0ZGRnMmzevzrJz587lxhtvrFUE1K1bN7p27Yper+fee+9l6NChFBQUcMsttzh38tSpU9xzzz2sW7fugsH++OOP+Pv7X/ROuoLFYsFoNLo6jCaRWFuGp8T62zirqqqIj493YUQNUxTlsj6kZrPZsNvt+Pr6cuzYMe6//34+++wz9PrGS88vd6w1Dh8+XG9fQAkJCXXmtXgl8Ndff43JZHKWf3Xu3PmCGfRCtFptvTvhjjIzMyXWFiCxNr/fxpmZmem2/e1c7v51SkpKuOuuu7DZbCiKwl//+leCgoKatK6r+i3y8fGp8++uoaarjSYAk8lUq5zKbDbXqd1ubH2A2NhY+vbty/79+xkxYgQlJSXYbDb0ej25ubkXtU0hhLgcgoOD+fTTT10dRotptNA+KSmJrKwssrOzsVqtpKWl1XnkuSHFxcVYrVYACgoK2LVrF506dUKj0dCvXz82btwIwOrVq5u0zUqbo0m/K4QQonGN3gHo9XrmzZvH9OnTsdvtTJo0ifj4eJYsWUJiYiJDhgwhIyODmTNnUlJSwtdff83SpUtJS0vjl19+Yf78+Wg0GhRF4Z577nG2o33ssceYNWsWL7/8MgkJCUyZMqXRYE+X237/HgshhACaWAcwePBg54MUNc5v53r11VezefPmOuv16tWL//znP/VuMzY29qKbfp6tUjhZdJY2oe5ZPimEEJ7Eo/oCUoDPfjrp6jCEEOKK4FEJwM9Hw5rdJ1wdhhCiBd1+++1s2bKl1rwVK1Ywf/78C67Xs2dPQG2o8tBDDzW47Z9//vmC21mxYoWz11GAe+65h5KSkqaEfkFLly6lS5cuHDt2rNZvdenSpVZMmZmZdOnSpU6pSkJCQq1nBZYtW/a7Y/KoBBBk0HIgt5TMU7//jyGEcE9jxoxxPiBV4/PPP6+3p4H6mEwmXnnllUv+/ZUrV9ZKAG+//TbBwcGXvL3zde7cuVZPoBs2bKjzzMW6deu49tpr6/QYen6vCmvXrq23Q7uL5VEJINBXh16rYbXcBQhxxRoxYgTffPONswVhTk4OeXl59O7dm4qKCqZNm8aECRNITU1l06ZNddbPyclxJguLxcKsWbMYNWoUM2bMqDVc4vz5851dSdckjJUrV5KXl8e0adO4/fbbAUhOTqagoACA5cuXM2bMGMaMGePsSjonJ4dRo0bx//7f/yMlJYW77rqrwWEZhw4d6uz14Pjx4wQFBdGqVSvn94qisGHDBp5//nm+++67evs2ak4e1RuoXqvhxi6RrN1zgidGdkWnlWHrhGhRe/4Fu99v3m32vA16NNwJWmhoqLNhSU13CaNGjUKj0WAwGHjttdcIDAx09igwZMiQBp+4/de//oXRaGT9+vUcOHCAiRMnOr+bNWsWoaGh2O127rzzTg4cOMAdd9zBihUr+Mc//uHsqrnG3r17+fTTT/noo49QFIWbb76Zvn37EhwczLFjx3jppZd45plnePjhh9m4caOzd9HzBQYGEh0dzaFDh0hPT2f06NG1njPYtWsXMTExtGvXjn79+vHNN984u8yxWCy1uoy+77776vQ5dLE86g4AYELPtphLKtn2a93RdoQQV4aUlBRnMVBNP/qgXiG/9NJLpKam8oc//AGz2cyZM2ca3M6OHTsYO3YsoHbn3KVLF+d369evZ8KECYwfP57Dhw832tHbjz/+yNChQ/H39ycgIIBhw4Y5u5aOiYlxPn3bvXt3Z3fS9Rk9ejRpaWls2rSpzqAx5+9rzXI1flsE9HtP/uBhdwAAQxIiCfLVs3r3CQZ0au3qcIS4svWYesGr9ZYyZMgQFi5cyL59+7BYLCQmJgJqXUBBQQGffvopPj4+JCcnX1IxSXZ2Nu+99x6rVq0iJCSEuXPn/q7ilvO7stfpdBfc1k033cTf/vY3EhMTa3WLY7fb+eKLL0hPT+fNN99EURSKioooKyu75O5zGuNxdwBGHx2jkqLYsDeXs9ZL67NbCOHeAgIC6NevH0899ZTzihjULqDDw8Px8fFh27ZtF7zSBujTp4+zk8lDhw5x8OBBAMrLy/Hz8yMoKIgzZ87UanETEBBAeXl5nW317t2bTZs2cfbsWSoqKti0aXvRqbEAAB1SSURBVFOdQd+bws/Pjzlz5vDHP/6x1vz//ve/dOnShW+//ZavvvqKr7/+muHDh9dbz9FcPC4BAIzv2ZayShubMs2uDkUI0ULGjBnDgQMHaiWA0aNHs3fvXlJTU1m7di0dOnS44DamTp1KRUUFo0aN4pVXXqF79+6AWhzUrVs3Ro0axaOPPlqrK+mbb76Z6dOnOyuBa3Tv3p2JEycyZcoUbr75ZiZPnky3bt0uad9SUlKcsdRIS0tj6NChteYNHz7cmcBq6gBqXosWLbqk3z5fo91Bu5Pdu3fTs2dPHA6FAS98RbfoYN69s4+rw6qXp/QECRJrS/GUWOvrDdRd43ZVD5uXwlWx1vf3a+hv6pF3AFqthrE92vDtodPkl7VsMykhhLhSeWQCALU1kM2hsC7jlKtDEUIIj+SxCaBrVDAJ0cHyUJgQQlwij00AABN6tmFPdhFHz9StsRdCXDoPqhoU57nYv5tHJ4Cx17RFo0E6iBOiGRmNRvLz8yUJeBhFUcjPz7+ocag97kGw80WFGLm+Yzhr9pzgkaHxLhmAWYgrTUxMDDk5OZw+fdrVodRRVVVV74Dn7sgVsRqNRmJiYpq8vEcnAIDxPdry2KoMdmcX0atdq8ZXEEJckI+PD1dddZWrw6iXOzdR/S1PiNWji4AARiZG4avXsnqXFAMJIcTF8PgEEGT0YVg3E+syTmKVQeOFEKLJmpQANm/ezIgRIxg2bFi9o9Ds2LGDCRMm0K1bNzZs2OCcn5mZyS233EJKSgqpqam1BnmYO3cuycnJzseaMzMzL3knJvZqS2FFFZsPuV+ZpRBCuKtG6wDsdjsLFixg+fLlmEwmJk+eTHJyMp06dXIuEx0dzcKFC3nvvfdqrWs0GnnhhRdo3749ZrOZSZMmMXDgQOfoOo8//jgjR4783TtxQ3wEYQEGVu85wdBupt+9PSGE8AaNJoCMjAzi4uKIjY0F1E6M0tPTayWAmlpnrbb2DcX5FUkmk4mwsDAKCgqabXi1Gj46LalXR/PvHdkUlltpFWBofCUhhPByjSYAs9lMVFSU87PJZCIjI+OifygjI4OqqiratWvnnLd48WJee+01+vfvz5w5c2r1qV0fh8PRYFFR/wg779sdPPLP75k72PV3ARaL5XcVa11OEmvL8JRYPSVOkFib22VpBpqXl8djjz3GCy+84LxLmD17NhEREVRVVfH000+zbNkyZs6cecHtaLXaBptVJQCzLP4s+uIQk/qHMPaaNs29GxfFE5qA1ZBYW4anxOopcYLEeqkaSkSNVgKbTCZyc3Odn81mMyZT06+wy8rKuO+++5g1axY9evRwzo+MjHSO8Tlx4kR+/vnnJm+zIX8c3JEesaE8vWYv5pL6B2UWQgihajQBJCUlkZWVRXZ2NlarlbS0NJKTk5u0cavVyowZMxg3blydyt68vDxAfXx506ZNxMfHX0L4tel1Wl66+RoqbXae+CRDHmUXQogLaDQB6PV65s2bx/Tp0xk9ejSjRo0iPj6eJUuWkJ6eDqjl+4MGDWLDhg3Mnz/fOYLP+vXr2blzJ6tXr67T3HPOnDmkpqaSmppKYWEh999/f7PsUIeIQJ4ancA3B0/zr+3ZzbJNIYS4EjWpDmDw4MEMHjy41ryHH37Y+f7qq6+uNaZmjZqTfn1Wrlx5MXFelNv6xfHFPjPPpO1nQKdw4sIDWuy3hBDCU3n8k8D10Wo1vDj5anRaDY9+9BN2hxQFCSHEb12RCQCgTagfC8Z1Z+exQt7e8qurwxFCCLdzxSYAUHsKHZUYxUtfHOJAbomrwxFCCLdyRScAjUbDM+MTCfbzYdb//SSdxQkhxHmu6AQAEB7oy/MTk8g8VcKS9EOuDkcIIdzGFZ8AAIZ2M3Fz7xje+OYXfjxW6OpwhBDCLXhFAgB4ekw32oT68cAHP5Ilg8gLIYT3JIAgow/vTutDlV3hf9/eRnZBhatDEkIIl/KaBADQJSqI9+/uR7nVztS3t3Gi6KyrQxJCCJfxqgQA0K1NMO/f3Y/is1VMXbaN3GLpNE4I4Z28LgEAJMWE8M+7+1FQbmXq29vIk55DhRBeyCsTAECP2FD+cVcfzCUWpr69jdOlla4OSQghLiuvTQAA18aFsfzOPpwssnDbOz9QUG51dUhCCHHZeHUCAOjXIZx3p/UmK7+c2975gaIKSQJCCO/g9QkA4PpOrXn7jt4cySvj9ne3Uyh3AkIILyAJoNqgzhG8dfu1HMwtZczSrfyUXeTqkIQQokVJAjjPTV0jWXV/fwCmvPlfPvjhmAwrKYS4YkkC+I2rY0JZ9+BA+ncM50+r9/Loxz9x1mp3dVhCCNHsJAHUo1WAgeV39mHW0M6s3n2CCa9/x1HpP0gIcYVpUgLYvHkzI0aMYNiwYSxbtqzO9zt27GDChAl069aNDRs21Ppu9erVDB8+nOHDh7N69Wrn/L1795KamsqwYcN45pln3K6oRavV8PDQeFb8oS+5JRbGLt3Kxn25rg5LCCGaTaMJwG63s2DBAt555x3S0tJYt24dR44cqbVMdHQ0CxcuZMyYMbXmFxUV8eqrr/LRRx/x8ccf8+qrr1JcXAzAn//8Z/7617/yxRdfkJWVVe+g8u5gcOcI1j04kA4RAdz3zx9ZuD4Tm10GlhFCeL5GE0BGRgZxcXHExsZiMBhISUkhPT291jIxMTF07doVrbb25rZu3cqAAQMIDQ0lJCSEAQMGsGXLFvLy8igrK6NHjx5oNBrGjx9fZ5vuJKaVPx/9sT+3XdeOt779lVuWbWP/SRliUgjh2fSNLWA2m4mKinJ+NplMZGRkNGnj9a1rNpvrzI+KisJsNje6PYfDQWZmZpN+uyXc2kVPtE8Eb23PZ8zSLYyMD+L2nmGEGnV1lrVYLC6N9WJIrC3DU2L1lDhBYm1ujSYAd6LVaklISHBpDAkJcNtNVbycfoiV/z3G1uNneWRoZ27vH4eP7twdUGZmpstjbSqJtWV4SqyeEidIrJeqoUTUaBGQyWQiN/dc5afZbMZkMjXpRxta97fzc3Nzm7xNdxDi78P81O5sePgGrokNZcG6/Yx8eTPfHjrt6tCEEKLJGk0ASUlJZGVlkZ2djdVqJS0tjeTk5CZtfODAgWzdupXi4mKKi4vZunUrAwcOJDIyksDAQPbs2YOiKKxZs4YhQ4b87p253OJNQay8qy/vTuuN3aEw7b3t3L1iB7+eLnN1aEII0ahGi4D0ej3z5s1j+vTp2O12Jk2aRHx8PEuWLCExMZEhQ4aQkZHBzJkzKSkp4euvv2bp0qWkpaURGhrKAw88wOTJkwGYMWMGoaGhAMyfP58nn3wSi8XCoEGDGDRoUMvuaQvRaDQMSTAxML41//g+i1fSjzDi5c0M6xjIE5HtiAsPcHWIQghRL43ibg3wL2D37t307NnT1WFc0OnSShZvOsRHO47jUGBUUjT3D+5IYtsQV4fWIHcqq2yMxNr8PCVOkFgvVUOxeFQlsCeICPLluQlJpLSDLXk+fLDtGGkZp7ghvjV/HNyR6zuGo9FoXB2mEEJIVxAtJcxfz9xRXfnuyWSeGNmVA7ml3PrOD4x99TvSMk5hd3jMjZcQ4goldwAtLNjow/03duQPA9qzevcJ3vr2F2Z8uIu4cH9u6xfHpGtjCAswuDpMIYQXkjuAy8Too2Nq33akP3ojr9/ai9aBvjz7eSbXPZfOI//ezY6sArfrD0kIcWWTO4DLTKfVMDopmtFJ0RzILeHDH46zetcJ1uw5SWdTILf2i2NCr7YEG31cHaoQ4gondwAu1DUqmAXjEvnhT0N4YVISRh8d8z/bR79n03l81U/syCrAIXUFQogWIncAbsDfoOeWPu24pU87fs4p5sPtx1iz+yQf7cwhIsiXEd1NjEqMpt9VYeh1krOFEM1DEoCbSYoJYWHM1fwppRvpmWY27M3lkx9P8P6247Ty92FYNzUZXN8pHF993U7ohBCiqSQBuKlAXz3jerRlXI+2nLXa+fZQHuv35rL+51w+2plDkK+e5IRIhneLYlDn1gRJnYEQ4iJJAvAAfgYdIxOjGZkYTaXNzndHzrBhby5f7jezds9JfHQarusQztAEE0MSIolp5e/qkIUQHkASgIfx1etI7moiuasJm93BruNFpGea+TLTzPzP9jH/s310jQpiWDcTQxNMJLUNQauVJ4+FEHVJAvBgep2WvleF0feqMJ4cncAvp8tIzzSzKTOP174+wtKvjhAWYKBv+zD6dQij31XhdI0KkoQghAAkAVxROkYE0jEikHsHdaSw3MrXB/PYeuQMP/xawIbqAe1D/Hzo0z6MflepSaFbdLCLoxZCuIokgCtUqwADE3vFMLFXDAA5hRVsP1rAD78W8MPRfDZlqkNwBvrq6RZhYPgZXwZ0ak0Xk9whCOEtJAF4iZhW/sS08ncmBHOJhR+OFrDt13y+zTzFM2nqkHHhAQb6dwxnQKfWDOjYmnbhUqEsxJVKEoCXMgUbGXtNG8Ze04bMrnpCotvz3ZEzfP9LPt8dOcO6jFMAxLTyo3+HcHq0C6VHbChdTEHyMJoQVwhJAAKANqF+TOkdy5TesSiKwi+ny/juiJoMvsw08/GPOQAYfbQktgnhmlg1IfSIDSWmlZ+McSCEB5IEIOrQaDR0igyiU2QQ065vj6IoZBecZXd2IT9lF/NTThHvbzvGu1uPAhAWYCCpbQjd2wSTWD1tF+YvSUEINycJQDRKo9HQLtyfduH+jOvRFoAqu4ODuaX8lFPEnuNF7D1Zwnebf8VW3XldkK+ebm2C6d4mhMS26vSq1gEY9FJ8JIS7kAQgLomPTkti2xAS24Zwa784ACptdg7llrHvZDF7Txaz72QJH24/hqXKAYBeq6FDRADxpiC6mILobAok3hREXJi/1CsI4QJNSgCbN2/m2WefxeFwMGXKFO69995a31utVh5//HH27dtHaGgoixcvJiYmhs8++4x3333XudzBgwdZvXo1CQkJ3H777eTl5WE0GgF47733CA8Pb8ZdE5ebr15HUkwISTEhznk2u4OjZ8rZd7KEQ+ZSDplL+TmnmLTqSmYAg15Lx4hAEqKC6NYmWL1ziA4hxF/6NxKiJTWaAOx2OwsWLGD58uWYTCYmT55McnIynTp1ci7z8ccfExwczJdffklaWhqLFi3i5ZdfZuzYsYwdOxZQT/4zZsyoNTL9okWLSEpKaoHdEu5Cr9MSbwoi3hRUa36F1caRvDIO5pZyuHq69cgZPt19wrlMTCs/ulcXI3WvTgwyapoQzafRBJCRkUFcXByxsbEApKSkkJ6eXisBfPXVV8ycOROAESNGsGDBAhRFqVUJmJaWRkpKSnPHLzyUv0HP1TGhXB0TWmv+6dJK9p8qYV91EdL+kyVs3Gd2fu+r19A+/IxaJxHmT1z1tF2Y+pyD1DEI0XSNJgCz2UxUVJTzs8lkIiMjo84y0dHR6gb1eoKCgigsLCQsLMy5zOeff87rr79ea72nnnoKrVbL8OHDeeCBBxptNeJwOMjMzGx8r9yAxWKRWC9RJBBpgptMftDTj4oqB0cLrfxaUEl2oYXTZ+0cOlnA5oN5VNrP3RFoNRARoKddiA+xIQbahfrQLtRAuxADAYbLnxjc7bg2xFPiBIm1uV2WSuCffvoJPz8/Onfu7Jy3aNEiTCYTZWVlPPTQQ6xdu5bx48dfcDtarbZWEZI7y8zMlFib0bXV0/NjVRSF06WVHCuo4Fh+BccLKjh6ppwjeWWsO1SK1eZwrm8K9iU+MohOkYHEtPIjOsSPqBBfokL8iAzyxacFKqE94biC58QJEuulaigRNZoATCYTubm5zs9msxmTyVRnmVOnThEVFYXNZqO0tJRWrVo5v6+v+KdmG4GBgYwZM4aMjIxGE4AQ59NoNEQGG4kMNtKnfVit7+wOhZzCCg6byzhyusw5/XhnNuVW+2+2A60DfYkOMWIKNtImxEhsWE0RUwCxYX74G6TBnLjyNPqvOikpiaysLLKzszGZTKSlpfH3v/+91jLJycmsXr2anj17snHjRq677jpncY7D4WD9+vV8+OGHzuVtNhslJSWEhYVRVVXFN998Q//+/Zt514Q302k1xIUHEBcewFDOXbAoikLJWRunSs6SW2wht9jCqWIL5hJ1ejy/gm2/5FNaaau1vdaBvs76htgwf9qGGokI8iUiUJ2GBxpa5C5CiJbUaALQ6/XMmzeP6dOnY7fbmTRpEvHx8SxZsoTExESGDBnC5MmTeeyxxxg2bBghISEsXrzYuf6OHTuIjo52ViKD2mx0+vTpVFVV4XA46N+/PzfffHPL7KEQ59FoNIT4+xDi70PXqPq7wlYUhaKKKo4XVJx75VdwrKCc7UcLWLPnBPU1RgoLMBAR6KsmhiBffG3l9Cw7Xt0Rn1rsJJXUwp1oFA9qV7d792569uzp6jCaxJ3K/xojsV6cSpud06WV515llXU+55VUcqr4LI7z/ndpNRAVbHQmhLbVSSE61Eh0iJHoED+CjfrL3oWGOxzTppJYL01DsUjBphAXyVevc3avfSE/79tPaHR7cgrPklNYQXb1NKfwLD8cLeDUntoJAiDAoCM61K86IRiJCvEjKtiIKdgXU7CRqBAjYf4GGbNBNAtJAEK0EL1WQ2x1nQHUfcrdZneQV6reKZwsUusjThaf5VSRhVPFZzmQW8qZsso6xU0+Og2RQWpSiAoxEhmk1kNEBvmqleLV71tJohCNkAQghIvodVrahPrRJtSPa+PqX6bK7uB0aSXmErWiOrfYQm7Juc8HckvZcuhMnUprUBNF60BfWgf60irAQHiAgVb+BsID1WlYgA9hAb4UFVuJrbQR6CunA28jf3Eh3JjPeUniQs5a7eSVWsgrVesfzn9fUF5JQbmVo2fKKCyvoqyeZMGaHIJ89Zhqip6C1WnN57AAXwJ9dQT46tWXQY9O7i48niQAIa4Afgads9lrYyxVdooqqsgvr6SwvIqMQ0fRBYZxqrjmDsPCYfMZ8kotdeoozmf00RLoq8ffoCfQV48p2NeZrNqEGmkTor6PCjFKE1k3JQlACC9j9NERFaIjKkTtiTfcdpqEhI51lrPZHZwuq+RUsYXCcivlVjvllbbql50Kq42y6s+lFhvmUgs/5RRTUG6ttR2NBkxB6t1EK38fwvwNhPobaOXvQ6vqYqma98F+PgT6qglF7jBaniQAIUS99Dqt2kw15MLFT7911mp3VmafLDrLiaKznCw6S26JhfwyK0fyyiiqaKAo6jz+Bp2aDIx6gqqnWM/SLtNGeICBsAC1PiM8wNf5vpW/QZ61uAiSAIQQzcrPoKNjRCAdIwIvuFylzU5xRRUFFVYKy6soqrBSarFRWmmj1FJFmUW9wyittDnf5xVXcSA/l8IKa4PFU0FGvVrhfV7Fd1h1wmgVYCDMX73TCDLqndNAg94rW0xJAhBCuISvXkdksI7IYGOT16l5oMnuUCg+W0VBeSVnyqwUlFvJL7dSUGalsEJ9X1hu5WSRhb0nSigot2K1OxrcrkYDgb56go1qQgjw1WPQafH10VZPdRh0Wgx6Lb56dX6Yv4HIYF8ig2qa3hoJ9rv8D/L9HpIAhBAeR6fVOK/qO0U2vryiKFRY7RSUq8mi1GKjxFJFqaWq+r2NkrPq+1JLFRVWO1abg8JyK5U2B1abg0rny05llaPehOKr1zqfyTAoVsJ2VaDTavHRatDrNOh16nudVouPTkOQUU/rQF/CA9X+pFoHqFN/g+6yJBJJAEKIK55Go3E2YVUfzPt9FEVRi6TOa3Z7urSy+rPaBPdUgY0zlWXY7A5sDgWbXcHmcFBlV7DZHVQ5lFpdlp/P6KMlvDoZBBn11RXjPgT66gg0Vr836tXPvupdS5BRT1D1+0CjvkktryQBCCHERdJoNAQZfQgy+jRY19GUvoAsVepdSX6ZlTPlleq0rJL8MvV9frmVskob+WUVlFbXg5RV2rBfqH1uNaOPliCjD60DfXlpeOt6l5EEIIQQLmL00TXpQb/zKYpCpc1xLiFUF1uV1FSaVxdr1by/ULKQBCCEEB5Eo9Fg9NFh9NEREeTbpHUaGhFMGswKIYSXkgQghBBeShKAEEJ4KUkAQgjhpSQBCCGEl5IEIIQQXkoSgBBCeClJAEII4aU0ivLbIafd1549e/D1bdqDD0IIIVSVlZX06NGjznyPSgBCCCGajxQBCSGEl5IEIIQQXkoSgBBCeClJAEII4aUkAQghhJeSBCCEEF7KIwaE2bx5M88++ywOh4MpU6Zw7733ujqkBiUnJxMQEIBWq0Wn0/Hpp5+6OiSnJ598km+++Ybw8HDWrVsHQFFREbNmzeLEiRO0bduWl19+mZCQEBdHWn+sS5cu5aOPPiIsLAyA2bNnM3jwYFeGCcCpU6d4/PHHyc/PR6PRcPPNNzNt2jS3PLYNxeqOx7ayspJbb70Vq9WK3W5nxIgRPPTQQ2RnZzN79myKioro3r07L774IgaDwS1jnTt3Ltu3bycoKAiA559/vtFhIi8rxc3ZbDZlyJAhyvHjx5XKykolNTVVOXz4sKvDatBNN92k5OfnuzqMem3fvl3Zu3evkpKS4pz3wgsvKG+99ZaiKIry1ltvKS+++KKrwqulvlhfeeUV5Z133nFhVPUzm83K3r17FUVRlNLSUmX48OHK4cOH3fLYNhSrOx5bh8OhlJWVKYqiKFarVZk8ebKye/du5aGHHlLWrVunKIqiPP3008oHH3zgyjAVRWk41ieeeEJZv369i6NrmNsXAWVkZBAXF0dsbCwGg4GUlBTS09NdHZZH6tOnT50r0PT0dMaPHw/A+PHj2bRpkytCq6O+WN1VZGQk3bt3ByAwMJAOHTpgNpvd8tg2FKs70mg0BAQEAGCz2bDZbGg0GrZt28aIESMAmDBhglucDxqK1d25fQIwm81ERUU5P5tMJrf9B1vj7rvvZuLEifzf//2fq0NpVH5+PpGRkQBERESQn5/v4ogu7IMPPiA1NZUnn3yS4uJiV4dTR05ODpmZmVxzzTVuf2zPjxXc89ja7XbGjRvH9ddfz/XXX09sbCzBwcHo9WrpdVRUlNucD34ba81xXbx4MampqTz33HNYrVYXR1mb2ycAT/Ovf/2L1atX8/bbb/PBBx+wY8cOV4fUZBqNxq2vWqZOncqXX37J2rVriYyM5Pnnn3d1SLWUl5fz0EMP8dRTTxEYGFjrO3c7tr+N1V2PrU6nY+3atXz77bdkZGTw66+/ujqkBv021kOHDjF79mw2bNjAJ598QnFxMcuWLXN1mLW4fQIwmUzk5uY6P5vNZkwmkwsjurCa2MLDwxk2bBgZGRkujujCwsPDycvLAyAvL89ZCeiOWrdujU6nQ6vVMmXKFH7++WdXh+RUVVXFQw89RGpqKsOHDwfc99jWF6s7H1uA4OBg+vXrx549eygpKcFmswGQm5vrdueDmli3bNlCZGQkGo0Gg8HAxIkT3e64un0CSEpKIisri+zsbKxWK2lpaSQnJ7s6rHpVVFRQVlbmfP/dd98RHx/v4qguLDk5mTVr1gCwZs0ahgwZ4uKIGlZzMgXYtGmT2xxbRVH405/+RIcOHfjDH/7gnO+Ox7ahWN3x2BYUFFBSUgKAxWLh+++/p2PHjvTr14+NGzcCsHr1arc4H9QXa4cOHZzHVVEUtzmu5/OI3kC//fZbnnvuOex2O5MmTeL+++93dUj1ys7OZsaMGYBaHjhmzBi3inX27Nls376dwsJCwsPDefDBBxk6dCiPPPIIp06dok2bNrz88suEhoa6OtR6Y92+fTsHDhwAoG3btixYsMBZxu5KO3fu5NZbb6Vz585oteo11ezZs7n66qvd7tg2FOu6devc7tgeOHCAuXPnYrfbURSFkSNHMnPmTLKzs5k1axbFxcUkJCSwaNEilzcDbSjWO+64g8LCQhRFoWvXrvzlL39xVha7A49IAEIIIZqf2xcBCSGEaBmSAIQQwktJAhBCCC8lCUAIIbyUJAAhhPBSkgCEEMJLSQIQQggv9f8BEHmAnnjZFKQAAAAASUVORK5CYII=\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Convert training history to a DataFrame.\n",
    "df = pd.DataFrame(history_regr.history)\n",
    "\n",
    "# Rename the relevant columns.\n",
    "names = {'mae': 'Training MAE', 'val_mae': 'Validation MAE'}\n",
    "df = df[['mae', 'val_mae']].rename(columns=names)\n",
    "\n",
    "# Plot the results.\n",
    "df.plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Regression Results\n",
    "\n",
    "Let us now compute some performance measures for the Neural Network. First we use the fitted model to predict the stock-returns for the training- and test-sets:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 36,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 29.6 s, sys: 2.01 s, total: 31.6 s\n",
      "Wall time: 21.2 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "y_train_pred = model_regr.predict(X_train.values)\n",
    "y_test_pred = model_regr.predict(X_test.values)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can then calculate the $R^2$ value (aka. the coefficient of determination) between the actual stock-returns and those predicted by the model. A value of 1 means a perfect prediction, while a value of 0 means the model basically predicted the average stock-return, and a negative value means the model's prediction was worse than just using the average stock-return.\n",
    "\n",
    "The $R^2$ value is about 0.5 for the training-set, which is OK but could be much higher if we trained it more, and if we used a bigger Neural Network with more parameters to optimize."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 37,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.4976140337851359"
      ]
     },
     "execution_count": 37,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "r2_score(y_true=y_train, y_pred=y_train_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "But the $R^2$ value is very negative for the test-set, which is so bad, that it would be much better to just use the average stock-return. This shows the model has not learned the underlying relation between the signals X and stock-returns y. The model has merely learned noise or peculiarities in the training-data, which do not generalize to the unseen data in the test-set:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 38,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "-0.7089485805293441"
      ]
     },
     "execution_count": 38,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "r2_score(y_true=y_test, y_pred=y_test_pred)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Regression Examples\n",
    "\n",
    "We can also try and plot the model's predicted versus actual stock-returns for a stock in the training-set. As we can see, the model has learned this mapping reasonably well:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 39,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Column-name for the models' predicted stock-returns.\n",
    "TOTAL_RETURN_PRED = 'Total Return Predicted'\n",
    "\n",
    "# Create a DataFrame with actual and predicted stock-returns.\n",
    "# This is for the training-set.\n",
    "df_y_train = pd.DataFrame(y_train)\n",
    "df_y_train[TOTAL_RETURN_PRED] = y_train_pred\n",
    "\n",
    "# Plot the actual and predicted stock-returns for the first stock.\n",
    "ticker = tickers_train[0]\n",
    "_ = df_y_train.loc[ticker].plot(title=ticker)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now let us try and plot the model's predicted versus actual stock-returns for the first stock in the test-set. As we can see, the model is mostly unable to predict the stock-returns, for this data that it has not seen during its training:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 40,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Create a DataFrame with actual and predicted stock-returns.\n",
    "# This is for the test-set.\n",
    "df_y_test = pd.DataFrame(y_test)\n",
    "df_y_test[TOTAL_RETURN_PRED] = y_test_pred\n",
    "\n",
    "# Plot the actual and predicted stock-returns for the first stock.\n",
    "ticker = tickers_test[0]\n",
    "_ = df_y_test.loc[ticker].plot(title=ticker)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Neural Network: Classification\n",
    "\n",
    "Instead of mapping from signals X to a continuous variable y for the stock-return, we can try and map to discrete output-classes that indicate if the stock-return was a gain or loss. This is a so-called binary classification problem, and perhaps that is easier for the Neural Network.\n",
    "\n",
    "### Convert Output Data\n",
    "\n",
    "First we need to convert the stock-returns into 1 for gains and 0 for losses, and then split into training- and test-sets:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 41,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 2.98 s, sys: 9.1 ms, total: 2.99 s\n",
      "Wall time: 2.99 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "# Convert stock-returns to binary classes where loss=0 and gain=1\n",
    "# In the previous tutorial we just used the sign directly, but it\n",
    "# is easier in TensorFlow / Keras to use class-values of 0 and 1.\n",
    "df_cls = (np.sign(df_sig_rets[TOTAL_RETURN_1_3Y]) + 1) * 0.5\n",
    "\n",
    "# Split the classes into training- and test-sets.\n",
    "y_train_cls = df_cls.loc[tickers_train]\n",
    "y_test_cls = df_cls.loc[tickers_test]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Because this is a binary classification problem (the result is either a gain or a loss), we could build a Neural Network to use `y_train_cls` directly. But instead we want to build a more general type of Neural Network which supports any number of output classes. For this we must convert the output to so-called one-hot encoded arrays, so a class of 0 (loss) becomes the array `[1,0]` and a class of 1 (gain) becomes the array `[0,1]`."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 42,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0., 1.],\n",
       "       [0., 1.],\n",
       "       [0., 1.],\n",
       "       ...,\n",
       "       [0., 1.],\n",
       "       [0., 1.],\n",
       "       [0., 1.]], dtype=float32)"
      ]
     },
     "execution_count": 42,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Convert classes to one-hot-encoded arrays.\n",
    "y_train_one_hot = to_categorical(y=y_train_cls.values,\n",
    "                                 num_classes=2)\n",
    "\n",
    "# Show the result.\n",
    "y_train_one_hot"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The training-set is also biased because it contains many more samples with gains compared to losses:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 43,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.7068901617250674"
      ]
     },
     "execution_count": 43,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Fraction of the training-set that is classified as gains.\n",
    "gain_fraction = (y_train_cls == 1.0).sum() / len(y_train_cls)\n",
    "loss_fraction = 1.0 - gain_fraction\n",
    "\n",
    "# Show it.\n",
    "gain_fraction"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "When we train the Neural Network further below, we need to provide new class-weights to counter this imbalance between the two classes:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 44,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([1.70584516, 0.70732347])"
      ]
     },
     "execution_count": 44,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Calculate the class-weights.\n",
    "class_weight = compute_class_weight('balanced',\n",
    "                                    y=y_train_cls,\n",
    "                                    classes=[0.0, 1.0])\n",
    "\n",
    "# Show them.\n",
    "class_weight"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can check that when we multiply these weights by the class-fractions for losses and gains, they both end up with the same weights:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 45,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([0.5, 0.5])"
      ]
     },
     "execution_count": 45,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "class_weight * [loss_fraction, gain_fraction]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Build and Train Neural Network: Classification\n",
    "\n",
    "Now that we have prepared the classification data, we can proceed to build the Neural Network. We will use the same overall design as for the regression model above, with minor changes for the classification task."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 46,
   "metadata": {},
   "outputs": [],
   "source": [
    "# The activation function applied after each layer.\n",
    "# It is important to use a non-linear function, otherwise\n",
    "# the Neural Network cannot learn non-linear mappings.\n",
    "activation = 'relu'"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 47,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Model: \"sequential_1\"\n",
      "_________________________________________________________________\n",
      "Layer (type)                 Output Shape              Param #   \n",
      "=================================================================\n",
      "dense_6 (Dense)              (None, 128)               3200      \n",
      "_________________________________________________________________\n",
      "dense_7 (Dense)              (None, 64)                8256      \n",
      "_________________________________________________________________\n",
      "dense_8 (Dense)              (None, 32)                2080      \n",
      "_________________________________________________________________\n",
      "dense_9 (Dense)              (None, 16)                528       \n",
      "_________________________________________________________________\n",
      "dense_10 (Dense)             (None, 8)                 136       \n",
      "_________________________________________________________________\n",
      "dense_11 (Dense)             (None, 2)                 18        \n",
      "=================================================================\n",
      "Total params: 14,218\n",
      "Trainable params: 14,218\n",
      "Non-trainable params: 0\n",
      "_________________________________________________________________\n"
     ]
    }
   ],
   "source": [
    "# Create a new Keras model.\n",
    "model_clf = keras.Sequential()\n",
    "\n",
    "# Add an input layer for the signals.\n",
    "# Note how we set the dimensionality of the input here.\n",
    "model_clf.add(InputLayer(input_shape=(num_signals,)))\n",
    "\n",
    "# Add several dense aka. fully-connected layers.\n",
    "# You can experiment with different designs.\n",
    "model_clf.add(Dense(128, activation=activation))\n",
    "model_clf.add(Dense(64, activation=activation))\n",
    "model_clf.add(Dense(32, activation=activation))\n",
    "model_clf.add(Dense(16, activation=activation))\n",
    "model_clf.add(Dense(8, activation=activation))\n",
    "\n",
    "# Add a layer for the output of the Neural Network.\n",
    "# This has 2 outputs, one for each possible class, and it uses\n",
    "# the softmax activation function for classification.\n",
    "model_clf.add(Dense(2, activation='softmax'))\n",
    "\n",
    "# Compile the model but don't train it yet.\n",
    "# Note that it uses a different loss-function and metric\n",
    "# than the regression-model above.\n",
    "model_clf.compile(loss='categorical_crossentropy',\n",
    "                  metrics=['accuracy'],\n",
    "                  optimizer=RMSprop(0.001))\n",
    "\n",
    "# Show the model.\n",
    "model_clf.summary()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can now train the model. This takes a minute or two on an ordinary computer:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 48,
   "metadata": {
    "scrolled": true
   },
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "Train on 712320 samples, validate on 178080 samples\n",
      "Epoch 1/40\n",
      "712320/712320 [==============================] - 2s 3us/sample - loss: 0.5485 - accuracy: 0.7389 - val_loss: 0.6236 - val_accuracy: 0.6729\n",
      "Epoch 2/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.4801 - accuracy: 0.7778 - val_loss: 0.6679 - val_accuracy: 0.6533\n",
      "Epoch 3/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.4369 - accuracy: 0.8045 - val_loss: 0.7097 - val_accuracy: 0.6524\n",
      "Epoch 4/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.4043 - accuracy: 0.8239 - val_loss: 0.7599 - val_accuracy: 0.6606\n",
      "Epoch 5/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.3756 - accuracy: 0.8403 - val_loss: 0.8002 - val_accuracy: 0.6617\n",
      "Epoch 6/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.3505 - accuracy: 0.8540 - val_loss: 0.8559 - val_accuracy: 0.6635\n",
      "Epoch 7/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.3281 - accuracy: 0.8656 - val_loss: 0.9290 - val_accuracy: 0.6155\n",
      "Epoch 8/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.3086 - accuracy: 0.8754 - val_loss: 0.9807 - val_accuracy: 0.6066\n",
      "Epoch 9/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.2912 - accuracy: 0.8845 - val_loss: 1.0285 - val_accuracy: 0.6112\n",
      "Epoch 10/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.2737 - accuracy: 0.8937 - val_loss: 1.0722 - val_accuracy: 0.6119\n",
      "Epoch 11/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.2606 - accuracy: 0.8997 - val_loss: 1.0686 - val_accuracy: 0.6597\n",
      "Epoch 12/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.2465 - accuracy: 0.9065 - val_loss: 1.1256 - val_accuracy: 0.6408\n",
      "Epoch 13/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.2355 - accuracy: 0.9117 - val_loss: 1.1427 - val_accuracy: 0.6448\n",
      "Epoch 14/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.2244 - accuracy: 0.9161 - val_loss: 1.1987 - val_accuracy: 0.6523\n",
      "Epoch 15/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.2149 - accuracy: 0.9204 - val_loss: 1.3874 - val_accuracy: 0.5823\n",
      "Epoch 16/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.2062 - accuracy: 0.9243 - val_loss: 1.3780 - val_accuracy: 0.5971\n",
      "Epoch 17/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1989 - accuracy: 0.9272 - val_loss: 1.3306 - val_accuracy: 0.6281\n",
      "Epoch 18/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1919 - accuracy: 0.9302 - val_loss: 1.3303 - val_accuracy: 0.6639\n",
      "Epoch 19/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1870 - accuracy: 0.9316 - val_loss: 1.3558 - val_accuracy: 0.6296\n",
      "Epoch 20/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1800 - accuracy: 0.9343 - val_loss: 1.3655 - val_accuracy: 0.6415\n",
      "Epoch 21/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1735 - accuracy: 0.9372 - val_loss: 1.4403 - val_accuracy: 0.6323\n",
      "Epoch 22/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1690 - accuracy: 0.9389 - val_loss: 1.4594 - val_accuracy: 0.6650\n",
      "Epoch 23/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1655 - accuracy: 0.9403 - val_loss: 1.4784 - val_accuracy: 0.6461\n",
      "Epoch 24/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1597 - accuracy: 0.9425 - val_loss: 1.6489 - val_accuracy: 0.6042\n",
      "Epoch 25/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1569 - accuracy: 0.9435 - val_loss: 1.7294 - val_accuracy: 0.5939\n",
      "Epoch 26/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1534 - accuracy: 0.9450 - val_loss: 1.6694 - val_accuracy: 0.6080\n",
      "Epoch 27/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1485 - accuracy: 0.9465 - val_loss: 1.7625 - val_accuracy: 0.5981\n",
      "Epoch 28/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1469 - accuracy: 0.9474 - val_loss: 1.7854 - val_accuracy: 0.6033\n",
      "Epoch 29/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1447 - accuracy: 0.9484 - val_loss: 1.7102 - val_accuracy: 0.6137\n",
      "Epoch 30/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1412 - accuracy: 0.9494 - val_loss: 1.6183 - val_accuracy: 0.6378\n",
      "Epoch 31/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1387 - accuracy: 0.9503 - val_loss: 1.8300 - val_accuracy: 0.6053\n",
      "Epoch 32/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1368 - accuracy: 0.9510 - val_loss: 1.8711 - val_accuracy: 0.6003\n",
      "Epoch 33/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1331 - accuracy: 0.9522 - val_loss: 1.8598 - val_accuracy: 0.6160\n",
      "Epoch 34/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1341 - accuracy: 0.9520 - val_loss: 1.7861 - val_accuracy: 0.6177\n",
      "Epoch 35/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1293 - accuracy: 0.9538 - val_loss: 1.7999 - val_accuracy: 0.6243\n",
      "Epoch 36/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1281 - accuracy: 0.9540 - val_loss: 1.8664 - val_accuracy: 0.6126\n",
      "Epoch 37/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1274 - accuracy: 0.9542 - val_loss: 2.2020 - val_accuracy: 0.5787\n",
      "Epoch 38/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1264 - accuracy: 0.9546 - val_loss: 1.9973 - val_accuracy: 0.5971\n",
      "Epoch 39/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1224 - accuracy: 0.9562 - val_loss: 1.9357 - val_accuracy: 0.6076\n",
      "Epoch 40/40\n",
      "712320/712320 [==============================] - 1s 2us/sample - loss: 0.1224 - accuracy: 0.9559 - val_loss: 1.9765 - val_accuracy: 0.6109\n",
      "CPU times: user 3min 21s, sys: 7.42 s, total: 3min 29s\n",
      "Wall time: 51.5 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "history_clf = model_clf.fit(x=X_train.values, y=y_train_one_hot,\n",
    "                            class_weight=class_weight, **fit_args)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can then plot the training-progress as the classification accuracy which is supposed to be maximized. After 40 epochs through the training-set, the model has learned to map from signals to stock gain/loss for the training-set, with an accuracy of about 96% (this is not for the entire training-set, but only for one or more batches of data).\n",
    "\n",
    "Unfortunately, the performance on the validation-set gets worse during optimization and ends with an accuracy of about 60%, so the model is clearly not learning the underlying relation between signals and stock gain/loss, but it has instead learned peculiarities and noise in the training-data."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 49,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "<matplotlib.axes._subplots.AxesSubplot at 0x7f36e0232b38>"
      ]
     },
     "execution_count": 49,
     "metadata": {},
     "output_type": "execute_result"
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Convert training history to a DataFrame.\n",
    "df = pd.DataFrame(history_clf.history)\n",
    "\n",
    "# Rename the relevant columns.\n",
    "names = {'accuracy': 'Training Accuracy',\n",
    "         'val_accuracy': 'Validation Accuracy'}\n",
    "df = df[['accuracy', 'val_accuracy']].rename(columns=names)\n",
    "\n",
    "# Plot the results.\n",
    "df.plot()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Classification Results\n",
    "\n",
    "We can then use the fitted model to predict the classes for the training- and test-sets:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 50,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "CPU times: user 32 s, sys: 2.25 s, total: 34.2 s\n",
      "Wall time: 23 s\n"
     ]
    }
   ],
   "source": [
    "%%time\n",
    "y_train_pred_cls = model_clf.predict(X_train.values)\n",
    "y_test_pred_cls = model_clf.predict(X_test.values)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The Neural Network has 2 outputs which is a likelihood score for each of the two classes:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 51,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([[0.00180785, 0.99819213],\n",
       "       [0.00175658, 0.99824345],\n",
       "       [0.00175229, 0.99824774],\n",
       "       ...,\n",
       "       [0.51388144, 0.48611856],\n",
       "       [0.5675287 , 0.4324713 ],\n",
       "       [0.55133164, 0.44866836]], dtype=float32)"
      ]
     },
     "execution_count": 51,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "y_test_pred_cls"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Let us convert this to a single integer class-number, by taking the so-called argmax, which is the index of the largest element in each row:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 52,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "array([1, 1, 1, ..., 0, 0, 0])"
      ]
     },
     "execution_count": 52,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Convert one-hot-encoded arrays to integer class-numbers.\n",
    "y_train_pred_cls = y_train_pred_cls.argmax(axis=1)\n",
    "y_test_pred_cls = y_test_pred_cls.argmax(axis=1)\n",
    "\n",
    "# Show the result.\n",
    "y_test_pred_cls"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We can now calculate the classification accuracy for the entire training-set, which is almost 90%, so the model has learned to map financial signals to gain/loss classifications for 1-3 year investment periods:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 53,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.8960455974842767"
      ]
     },
     "execution_count": 53,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(y_true=y_train_cls, y_pred=y_train_pred_cls)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Unfortunately this performance does not generalize to the test-set, which the model has not seen during training. Here the classification accuracy is much lower:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 54,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.6093554250622981"
      ]
     },
     "execution_count": 54,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "accuracy_score(y_true=y_test_cls, y_pred=y_test_pred_cls)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "This is even worse than simply guessing that the stock-returns are always gains, which would give a classification accuracy of about 70%:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 55,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "0.7024723359862238"
      ]
     },
     "execution_count": 55,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# Fraction of the test-set that is classified as gains.\n",
    "(y_test_cls == 1.0).sum() / len(y_test_cls)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Confusion Matrix\n",
    "\n",
    "We can also plot a so-called Confusion Matrix, which shows how well the classes were predicted. First we need a small helper-function for making the plot:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 56,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_confusion_matrix(y_true, y_pred):\n",
    "    \"\"\"\n",
    "    Plot a classification confusion matrix.\n",
    "    \n",
    "    :param y_true: Array of true classes.\n",
    "    :param y_pred: Array of predicted classes.\n",
    "    \"\"\"\n",
    "    \n",
    "    # Class labels.\n",
    "    labels = [1.0, 0.0]\n",
    "    labels_text = ['Gain', 'Loss']\n",
    "\n",
    "    # Create confusion matrix.\n",
    "    mat = confusion_matrix(y_true=y_true, y_pred=y_pred,\n",
    "                           labels=labels)\n",
    "    \n",
    "    # Normalize so all matrix entries sum to 1.0\n",
    "    mat = mat / len(y_true)\n",
    "    \n",
    "    # Plot the matrix as a heatmap.\n",
    "    sns.heatmap(mat, robust=True,\n",
    "                xticklabels=labels_text,\n",
    "                yticklabels=labels_text)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "For the training-set, we again see that the model has learned to accurately classify gains and losses. Note how the colours are different, because the training-set contains many more gains than losses."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 57,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_confusion_matrix(y_true=y_train_cls, y_pred=y_train_pred_cls)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The confusion matrix for the test-set, shows that many classes were correctly predicted as gains, and there were a lot of errors in the rest of the predictions:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 58,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 2 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "plot_confusion_matrix(y_true=y_test_cls, y_pred=y_test_pred_cls)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Classification Examples\n",
    "\n",
    "We can also try and plot the model's predicted versus actual gain/loss classification for a stock in the training-set. As we can see, the model has learned this mapping quite well:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 59,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Create a DataFrame with actual and predicted classifications.\n",
    "# This is for the training-set.\n",
    "df_y_train_cls = pd.DataFrame(y_train_cls)\n",
    "df_y_train_cls[TOTAL_RETURN_PRED] = y_train_pred_cls\n",
    "\n",
    "# Plot the actual and predicted stock-returns for the first stock.\n",
    "ticker = tickers_train[0]\n",
    "_ = df_y_train_cls.loc[ticker].plot(title=ticker)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Now let us try and plot the model's predicted versus actual gain/loss classifications for the first stock in the test-set. As we can see, the model is quite bad at predicting this for data that it has not seen during its training:"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 60,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<Figure size 432x288 with 1 Axes>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "# Create a DataFrame with actual and predicted classifications.\n",
    "# This is for the test-set.\n",
    "df_y_test_cls = pd.DataFrame(y_test_cls)\n",
    "df_y_test_cls[TOTAL_RETURN_PRED] = y_test_pred_cls\n",
    "\n",
    "# Plot the actual and predicted stock-returns for the first stock.\n",
    "ticker = tickers_test[0]\n",
    "_ = df_y_test_cls.loc[ticker].plot(title=ticker)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## Discussion\n",
    "\n",
    "We used a Neural Network to learn the relation between signals and future stock-returns. This worked reasonably well on the training-data, but it did not generalize to the test-data, where the Neural Network performed worse than just guessing. In fact, the result was even worse than the previous tutorial which used a Machine Learning (ML) model called a Random Forest.\n",
    "\n",
    "The problem might be the design of the Neural Network, that is, the number and size of the layers, their activation functions, etc. You can try and use so-called Hyper-Parameter Optimization such as the [Keras Tuner](https://keras-team.github.io/keras-tuner/), to try and find a better design for the Neural Network.\n",
    "\n",
    "But we have now seen two very different ML models having the same problem with this dataset, so perhaps it means that these financial signals simply cannot be used to predict future stock-returns for 1-3 year investment periods.\n",
    "\n",
    "\n",
    "### Machine Learning vs. Good Science\n",
    "\n",
    "In the previous tutorial, we had a brief discussion about Machine Learning versus Good Science. We made an important point that is worth repeating here.\n",
    "\n",
    "In general, these experiments show that Machine Learning cannot replace good science. Even if the ML model had performed well on the test-data, it would not have provided an explanation for the underlying relation between signals and stock-returns, so we would not know whether it was truly something that would work in the future. To explain this we need proper scientific reasoning.\n",
    "\n",
    "A good example of scientific reasoning is the discovery and formulation of the Law of Gravity. A simple experiment would be to drop various items from different heights and take lots of measurements of how fast the items fall to the ground.\n",
    "\n",
    "You can then make a noisy scatter-plot that shows the relation between item-size, item-weight, drop-height, and drop-speed. Statistical analysis would be able to give you a rough idea whether there is some relation in the data and if it is statistically significant, so the observed data is probably not just due to random chance. \n",
    "\n",
    "Machine Learning takes this one step further, in that it might be able to create a mathematical or computational model for the non-linear relation in the data, so you could use the ML model to predict the drop-speed if you input the other data. But the ML model would not be able to explain *why* this relation exists in the data. For this we need a proper theoretical framework such as Classical Mechanics in the case of the Law of Gravity.\n",
    "\n",
    "\n",
    "### Predicting Long-Term Stock Returns\n",
    "\n",
    "When it comes to predicting long-term stock-returns, there are actually only 3 factors that matter: The future dividends, sales-growth, and change in the P/Sales valuation ratio. (You could instead use earnings-growth or book-value growth, and the P/E or P/Book valuation ratios.) If we can predict the future dividend, sales-growth and P/Sales ratio, then we can predict the future stock-return.\n",
    "\n",
    "When making scatter-plots of historical P/Sales ratios versus long-term stock-returns, we often see a particular downwards-sloping curve, and it turns out there is a certain formula for this curve, which is derived from the mathematical definition of Annualized Return, see [here](https://github.com/Hvass-Labs/FinanceOps/blob/master/01C_Theory_of_Long-Term_Stock_Forecasting.ipynb). The formula takes as input the average dividend yield, sales-growth and P/Sales ratios that you predict for the future, and produces a curve that can be used to predict future stock-returns from these 3 factors and the P/Sales ratio today. The accuracy of the prediction depends on how well you have predicted these 3 factors.\n",
    "\n",
    "Machine Learning cannot uncover and explain such relations, whether it is the Law of Gravity, or the formula for long-term stock-returns. Human reasoning is still needed for this."
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "## License (MIT)\n",
    "\n",
    "This is published under the\n",
    "[MIT License](https://github.com/simfin/simfin-tutorials/blob/master/LICENSE.txt)\n",
    "which allows very broad use for both academic and commercial purposes.\n",
    "\n",
    "You are very welcome to modify and use this source-code in your own project. Please keep a link to the [original repository](https://github.com/simfin/simfin-tutorials).\n"
   ]
  }
 ],
 "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": "3.6.8"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}