notebooks/03-stock-price-prediction-machine-learning.ipynb

{
 "cells": [
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "# Stock Price Movement Prediction Using The Deutsche Börse Public Dataset & Machine Learning - Notebook 3 (Applying A Neural Network)\n",
    "\n",
    "Here we apply the neural network approach suggested in Notebook 1 to the data set product in Notebook 2.\n",
    "\n",
    "Note: this notebook has not yet been converted to Python 3 as it requires a lot of resources to recompute.\n",
    "As a simple example please use the notebook:\n",
    "```\n",
    "supported/adapted-03-stock-price-prediction-machine-learning-20-minutes-intervals.ipynb\n",
    "```\n",
    "which uses a smaller dataset."
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 1,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python2.7/dist-packages/h5py/__init__.py:36: FutureWarning: Conversion of the second argument of issubdtype from `float` to `np.floating` is deprecated. In future, it will be treated as `np.float64 == np.dtype(float).type`.\n",
      "  from ._conv import register_converters as _register_converters\n",
      "Using TensorFlow backend.\n"
     ]
    }
   ],
   "source": [
    "import pandas as pd\n",
    "import numpy as np\n",
    "import glob, os\n",
    "from datetime import datetime\n",
    "from sklearn.preprocessing import MinMaxScaler, StandardScaler, RobustScaler\n",
    "from pandas import DataFrame\n",
    "from pandas import concat\n",
    "\n",
    "import matplotlib as mpl\n",
    "import matplotlib.pyplot as plt\n",
    "%matplotlib inline\n",
    "mpl.rcParams['figure.figsize'] = (15, 10) # use bigger graphs\n",
    "\n",
    "from math import sqrt\n",
    "from numpy import concatenate\n",
    "from matplotlib import pyplot\n",
    "from pandas import read_csv\n",
    "from pandas import DataFrame\n",
    "from pandas import concat\n",
    "from sklearn.preprocessing import MinMaxScaler, StandardScaler\n",
    "from sklearn.preprocessing import LabelEncoder\n",
    "from sklearn.metrics import mean_squared_error\n",
    "from keras.models import Sequential\n",
    "from keras.layers import Dense, Flatten, Reshape, Conv1D, MaxPooling1D, BatchNormalization, LeakyReLU\n",
    "from keras.layers import LSTM\n",
    "from keras import regularizers\n",
    "\n",
    "import datetime\n"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "- Added L2 penalty\n",
    "- Added extra features\n",
    "- Added strategy computation"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We read the file we processed in the second notebook"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 2,
   "metadata": {},
   "outputs": [],
   "source": [
    "input_file = '/data/processed/cooked_v3.pkl'\n",
    "df = pd.read_pickle(input_file)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Checking what stocks are available in the file"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 3,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(array(['SNH', 'DBK', 'EOAN', 'DTE', 'CBK', 'RWE', 'IFX', 'SVAB', 'LHA',\n",
       "        'DAI', 'O2D', 'TKA', 'DPW', 'HDD', 'SIE', 'AIXA', 'BAYN', 'SAP',\n",
       "        'BAS', 'EVT', 'AT1', 'PSM', 'BMW', 'VOW3', 'FRE', 'GAZ', 'SDF',\n",
       "        'CEC', 'ALV', 'VNA', 'B4B', 'SHA', 'AB1', 'UN01', 'DLG', 'NDX1',\n",
       "        'NOA3', 'IGY', 'VODI', 'ADS', '1COV', 'TUI1', 'BPE5', 'HEI', 'KCO',\n",
       "        'ADV', 'SZU', 'EVK', 'HEN3', 'WDI', 'MUV2', 'DWNI', 'MRK', 'USE',\n",
       "        'PAH3', 'DEZ', 'FME', 'G1A', 'FNTN', 'RKET', 'QIA', 'DB1', 'ZAL',\n",
       "        'QSC', 'CON', 'SGL', 'BVB', 'TINA', 'PBB', 'PNE3', 'RIB', 'OSR',\n",
       "        'SHL', 'AOX', 'BEI', 'TEG', 'UTDI', 'ARL', 'MDG1', 'KGX', 'LXS',\n",
       "        'ARO', 'TTI', 'SANT', 'GYC', 'ANO', 'LINU', 'SOW', 'SZG', 'LLD',\n",
       "        'BOSS', 'BNR', 'WAF', 'LIN', 'DRI', 'NDA', 'ZIL2', 'SY1', 'CAP',\n",
       "        '3W9K'], dtype=object), 100)"
      ]
     },
     "execution_count": 3,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "all_mnemonics = df.Mnemonic.unique()\n",
    "df.Mnemonic.unique(), df.Mnemonic.unique().shape[0]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Check what dates are available in the file. We will split the original set\n",
    "into three parts, train, valid, test based on the dates.\n",
    "If the dates are ordered chronologically, we take the first dates for the test set,\n",
    "then we take the next dates for the validation set and finally we take what is\n",
    "left for the test set"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 4,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "(200, ['2017-07-03', '2017-07-04'], ['2018-04-26', '2018-04-27'])"
      ]
     },
     "execution_count": 4,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "def date_part(dt):\n",
    "    return str(dt).split(' ')[0]\n",
    "unique_days = sorted(list(set(map(date_part , list(df.index.unique())))))\n",
    "len(unique_days), unique_days[0:2], unique_days[-3:-1]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 5,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "('train', 120, 'valid', 10, 'test', 70)"
      ]
     },
     "execution_count": 5,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "percent_train = 60.0\n",
    "percent_valid = 5.0\n",
    "percent_test = 100.0 - percent_train - percent_valid\n",
    "\n",
    "offset_train = int(len(unique_days)*percent_train/100.0)\n",
    "offset_test = offset_train + int(len(unique_days)*percent_valid/100.0)\n",
    "\n",
    "train_valid_days = list(set(unique_days[0:offset_test]))\n",
    "\n",
    "np.random.seed(484811945)\n",
    "np.random.shuffle(train_valid_days)\n",
    "\n",
    "train_days = train_valid_days[0:offset_train]\n",
    "valid_days = train_valid_days[offset_train:]\n",
    "test_days = set(unique_days[offset_test:])\n",
    "'train', len(train_days), 'valid', len(valid_days), 'test', len(test_days)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 6,
   "metadata": {},
   "outputs": [],
   "source": [
    "df['CalcDateTime'] = df.index\n",
    "df['Date'] = df['CalcDateTime'].dt.strftime(\"%Y-%m-%d\")"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 7,
   "metadata": {},
   "outputs": [],
   "source": [
    "df_train = df[df.Date.isin(list(train_days))]\n",
    "df_valid = df[df.Date.isin(list(valid_days))]\n",
    "df_test = df[df.Date.isin(list(test_days))]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "We've prepared the train, test and valid sets. Make sure the days do not overlap"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 8,
   "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>CalcDateTime</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>8652000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>unique</th>\n",
       "      <td>86520</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>top</th>\n",
       "      <td>2017-08-09 08:10:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>freq</th>\n",
       "      <td>100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>first</th>\n",
       "      <td>2017-07-03 08:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>last</th>\n",
       "      <td>2018-01-09 20:00:00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "               CalcDateTime\n",
       "count               8652000\n",
       "unique                86520\n",
       "top     2017-08-09 08:10:00\n",
       "freq                    100\n",
       "first   2017-07-03 08:00:00\n",
       "last    2018-01-09 20:00:00"
      ]
     },
     "execution_count": 8,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_train[['CalcDateTime']].describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 9,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>CalcDateTime</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>721000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>unique</th>\n",
       "      <td>7210</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>top</th>\n",
       "      <td>2017-12-29 15:53:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>freq</th>\n",
       "      <td>100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>first</th>\n",
       "      <td>2017-07-05 08:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>last</th>\n",
       "      <td>2017-12-29 20:00:00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "               CalcDateTime\n",
       "count                721000\n",
       "unique                 7210\n",
       "top     2017-12-29 15:53:00\n",
       "freq                    100\n",
       "first   2017-07-05 08:00:00\n",
       "last    2017-12-29 20:00:00"
      ]
     },
     "execution_count": 9,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_valid[['CalcDateTime']].describe()"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 10,
   "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>CalcDateTime</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>count</th>\n",
       "      <td>5047000</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>unique</th>\n",
       "      <td>50470</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>top</th>\n",
       "      <td>2018-03-16 10:22:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>freq</th>\n",
       "      <td>100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>first</th>\n",
       "      <td>2018-01-10 08:00:00</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>last</th>\n",
       "      <td>2018-04-30 20:00:00</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "               CalcDateTime\n",
       "count               5047000\n",
       "unique                50470\n",
       "top     2018-03-16 10:22:00\n",
       "freq                    100\n",
       "first   2018-01-10 08:00:00\n",
       "last    2018-04-30 20:00:00"
      ]
     },
     "execution_count": 10,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "df_test[['CalcDateTime']].describe()"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "In the class below we create features from the raw features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 11,
   "metadata": {},
   "outputs": [],
   "source": [
    "def clipped(ind, limit):\n",
    "    return np.where(ind < -limit, -limit, np.where(ind > limit, limit, ind ))\n",
    "\n",
    "class NARemover:\n",
    "    def __init__(self, name):\n",
    "        self.name = name\n",
    "    def transform(self, single_stock):\n",
    "        before = single_stock.shape[0]\n",
    "        single_stock = single_stock.dropna()\n",
    "        after = single_stock.shape[0]\n",
    "        print(\"{}: Dropped {:2.2f} % of records due to NA\".format(self.name, 100.0*(before - after)/(0.0001 + before)))\n",
    "        return single_stock\n",
    "\n",
    "class Featurizer:\n",
    "    def transform(self, single_stock, max_offset = 10):\n",
    "        steps = 5    \n",
    "        limit = 5\n",
    "\n",
    "        # compute a smoothed standard deviation going back 5 and 10 steps\n",
    "        end_price_fixed = single_stock['EndPrice'].shift(1)\n",
    "        max_price_fixed = single_stock['MaxPrice'].shift(1)\n",
    "        min_price_fixed = single_stock['MinPrice'].shift(1)        \n",
    "\n",
    "        end_price_fixed = single_stock['EndPrice'].shift(1)\n",
    "        stds = 0.000001 + 0.9 * end_price_fixed.rolling(steps).std() + 0.1*end_price_fixed.rolling(10).std()\n",
    "\n",
    "        single_stock['x(NormalizingStd)'] = stds\n",
    "        \n",
    "        for offset in range(1, max_offset + 1):\n",
    "            # take the end price in the past at (t - offset)\n",
    "            end_price = single_stock['EndPrice'].shift(offset)\n",
    "\n",
    "            # take the max price in the past at (t - offset)        \n",
    "            min_price = single_stock['MinPrice'].shift(offset)\n",
    "\n",
    "            # take the min price in the past at (t - offset)\n",
    "            max_price = single_stock['MaxPrice'].shift(offset)\n",
    "        \n",
    "            # compute an indicator for time (t - offset)\n",
    "            ind = ((max_price - end_price) - (end_price - min_price)) / stds\n",
    "            single_stock['x((MaxP-EndP)-(EndP-MinP))(t - {})'.format(str(offset))] = clipped(ind, limit)   \n",
    "        \n",
    "            ind = (max_price - end_price) / stds\n",
    "            single_stock['x(MaxP - EndP)(t - {})'.format(str(offset))] = clipped(ind, limit)       \n",
    "        \n",
    "            ind = (end_price - min_price) / stds\n",
    "            single_stock['x(EndP-MinP)(t - {})'.format(str(offset))] = clipped(ind, limit)  \n",
    "                \n",
    "            ind = (end_price - end_price_fixed) / stds\n",
    "            single_stock['x(EndP-EndPBase)(t - {})'.format(str(offset))] = clipped(ind, limit)  \n",
    "                \n",
    "            ind = (max_price - max_price_fixed) / stds\n",
    "            single_stock['x(MaxP-MaxPBase)(t - {})'.format(str(offset))] = clipped(ind, limit) \n",
    "              \n",
    "            ind = (min_price - min_price_fixed) / stds\n",
    "            single_stock['x(MinP-MinPBase)(t - {})'.format(str(offset))] = clipped(ind, limit)              \n",
    "        \n",
    "        end_price_0 = single_stock['EndPrice'].shift(1)\n",
    "        for offset in range(2, max_offset):\n",
    "            end_price_1 = single_stock['EndPrice'].shift(offset + 1)             \n",
    "            ind = (end_price_0 - end_price_1) / stds\n",
    "            single_stock['x(EndP)(t - {}, t - {})'.format(str(1), str(offset + 1))] = \\\n",
    "                clipped(ind, limit)  \n",
    "             \n",
    "        for offset in range(2, max_offset):\n",
    "            end_price_1 = single_stock['EndPrice'].shift(offset + 1)             \n",
    "            single_stock['x(CombinedP)(t - {}, t - {})'.format(str(1), str(offset + 1))] = \\\n",
    "                single_stock['x(EndP)(t - {}, t - {})'.format(str(1), str(offset + 1))] * \\\n",
    "                single_stock['x((MaxP-EndP)-(EndP-MinP))(t - {})'.format(str(offset))]\n",
    "              \n",
    "        for offset in range(1, max_offset):\n",
    "            # take the end price in the past at (t - offset)\n",
    "            end_price = single_stock['EndPrice'].shift(offset)\n",
    "            end_price_1 = single_stock['EndPrice'].shift(offset + 1)  \n",
    "\n",
    "            # take the max price in the past at (t - offset)        \n",
    "            min_price = single_stock['MinPrice'].shift(offset)\n",
    "            min_price_1 = single_stock['MinPrice'].shift(offset + 1)\n",
    "\n",
    "            # take the min price in the past at (t - offset)\n",
    "            max_price = single_stock['MaxPrice'].shift(offset)\n",
    "            max_price_1 = single_stock['MaxPrice'].shift(offset + 1)            \n",
    "            \n",
    "            ind = (max_price - end_price_1) - (end_price_1 - min_price) / stds\n",
    "            single_stock['x((MaxP-EndP[-1])-(EndP-MinP[-1]))(t - {})'.format(str(offset))] = clipped(ind, limit)  \n",
    "            \n",
    "            ind = (max_price - end_price_1) - (max_price - end_price_1) / stds\n",
    "            single_stock['x(f1)(t - {})'.format(str(offset))] = clipped(ind, limit)  \n",
    "\n",
    "            ind = (min_price - end_price_1) - (min_price - end_price_1) / stds\n",
    "            single_stock['x(f2)(t - {})'.format(str(offset))] = clipped(ind, limit)\n",
    "            \n",
    "           \n",
    "            ind = (max_price - end_price) - (max_price_1 - end_price) / stds\n",
    "            single_stock['x(f3)(t - {})'.format(str(offset))] = clipped(ind, limit)  \n",
    "\n",
    "            ind = (min_price - end_price) - (min_price_1 - end_price) / stds\n",
    "            single_stock['x(f4)(t - {})'.format(str(offset))] = clipped(ind, limit)\n",
    "            \n",
    "        single_stock['y(Return)'] = (single_stock['EndPrice'] - single_stock['EndPrice'].shift(1))\n",
    "        ret = single_stock['y(Return)']/stds\n",
    "        single_stock['pseudo_y(ClippedReturn)'] = clipped(ret, limit) \n",
    "        single_stock['pseudo_y(SignReturn)'] = np.sign(single_stock['y(Return)'])\n",
    "        \n",
    "        # for evaluation we should use pct change\n",
    "        prev = single_stock['EndPrice'].shift(1)\n",
    "        single_stock['pseudo_y(pctChange)'] = (single_stock['EndPrice'] - prev)/prev\n",
    "        return single_stock"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "The variable we use for predictions start with `x(`, while the variables that should be predicted start with `y(`"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 12,
   "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>Mnemonic</th>\n",
       "      <th>MinPrice</th>\n",
       "      <th>MaxPrice</th>\n",
       "      <th>StartPrice</th>\n",
       "      <th>EndPrice</th>\n",
       "      <th>HasTrade</th>\n",
       "      <th>TradedVolume</th>\n",
       "      <th>NumberOfTrades</th>\n",
       "      <th>CalcDateTime</th>\n",
       "      <th>Date</th>\n",
       "      <th>...</th>\n",
       "      <th>x(f4)(t - 8)</th>\n",
       "      <th>x((MaxP-EndP[-1])-(EndP-MinP[-1]))(t - 9)</th>\n",
       "      <th>x(f1)(t - 9)</th>\n",
       "      <th>x(f2)(t - 9)</th>\n",
       "      <th>x(f3)(t - 9)</th>\n",
       "      <th>x(f4)(t - 9)</th>\n",
       "      <th>y(Return)</th>\n",
       "      <th>pseudo_y(ClippedReturn)</th>\n",
       "      <th>pseudo_y(SignReturn)</th>\n",
       "      <th>pseudo_y(pctChange)</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CalcDateTime</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>2017-07-03 09:51:00</th>\n",
       "      <td>BMW</td>\n",
       "      <td>82.34</td>\n",
       "      <td>82.36</td>\n",
       "      <td>82.34</td>\n",
       "      <td>82.34</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2163.0</td>\n",
       "      <td>30.0</td>\n",
       "      <td>2017-07-03 09:51:00</td>\n",
       "      <td>2017-07-03</td>\n",
       "      <td>...</td>\n",
       "      <td>0.987222</td>\n",
       "      <td>-3.051665</td>\n",
       "      <td>0.0</td>\n",
       "      <td>3.021665</td>\n",
       "      <td>-3.031665</td>\n",
       "      <td>-1.027222</td>\n",
       "      <td>0.01</td>\n",
       "      <td>1.017222</td>\n",
       "      <td>1.0</td>\n",
       "      <td>0.000121</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1 rows × 136 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                    Mnemonic  MinPrice  MaxPrice  StartPrice  EndPrice  \\\n",
       "CalcDateTime                                                             \n",
       "2017-07-03 09:51:00      BMW     82.34     82.36       82.34     82.34   \n",
       "\n",
       "                     HasTrade  TradedVolume  NumberOfTrades  \\\n",
       "CalcDateTime                                                  \n",
       "2017-07-03 09:51:00       1.0        2163.0            30.0   \n",
       "\n",
       "                           CalcDateTime        Date         ...           \\\n",
       "CalcDateTime                                                ...            \n",
       "2017-07-03 09:51:00 2017-07-03 09:51:00  2017-07-03         ...            \n",
       "\n",
       "                     x(f4)(t - 8)  x((MaxP-EndP[-1])-(EndP-MinP[-1]))(t - 9)  \\\n",
       "CalcDateTime                                                                   \n",
       "2017-07-03 09:51:00      0.987222                                  -3.051665   \n",
       "\n",
       "                     x(f1)(t - 9)  x(f2)(t - 9)  x(f3)(t - 9)  x(f4)(t - 9)  \\\n",
       "CalcDateTime                                                                  \n",
       "2017-07-03 09:51:00           0.0      3.021665     -3.031665     -1.027222   \n",
       "\n",
       "                     y(Return)  pseudo_y(ClippedReturn)  pseudo_y(SignReturn)  \\\n",
       "CalcDateTime                                                                    \n",
       "2017-07-03 09:51:00       0.01                 1.017222                   1.0   \n",
       "\n",
       "                     pseudo_y(pctChange)  \n",
       "CalcDateTime                              \n",
       "2017-07-03 09:51:00             0.000121  \n",
       "\n",
       "[1 rows x 136 columns]"
      ]
     },
     "execution_count": 12,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dummy = df_train[df_train.Mnemonic == 'BMW'].copy()\n",
    "dummy = dummy[dummy.HasTrade == 1.0].iloc[100:111]\n",
    "Featurizer().transform(dummy)\n",
    "old = dummy.iloc[[-1]].copy()\n",
    "old"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 13,
   "metadata": {},
   "outputs": [
    {
     "name": "stderr",
     "output_type": "stream",
     "text": [
      "/usr/local/lib/python2.7/dist-packages/ipykernel_launcher.py:1: DeprecationWarning: \n",
      ".ix is deprecated. Please use\n",
      ".loc for label based indexing or\n",
      ".iloc for positional indexing\n",
      "\n",
      "See the documentation here:\n",
      "http://pandas.pydata.org/pandas-docs/stable/indexing.html#ix-indexer-is-deprecated\n",
      "  \"\"\"Entry point for launching an IPython kernel.\n"
     ]
    },
    {
     "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>Mnemonic</th>\n",
       "      <th>MinPrice</th>\n",
       "      <th>MaxPrice</th>\n",
       "      <th>StartPrice</th>\n",
       "      <th>EndPrice</th>\n",
       "      <th>HasTrade</th>\n",
       "      <th>TradedVolume</th>\n",
       "      <th>NumberOfTrades</th>\n",
       "      <th>CalcDateTime</th>\n",
       "      <th>Date</th>\n",
       "      <th>...</th>\n",
       "      <th>x(f4)(t - 8)</th>\n",
       "      <th>x((MaxP-EndP[-1])-(EndP-MinP[-1]))(t - 9)</th>\n",
       "      <th>x(f1)(t - 9)</th>\n",
       "      <th>x(f2)(t - 9)</th>\n",
       "      <th>x(f3)(t - 9)</th>\n",
       "      <th>x(f4)(t - 9)</th>\n",
       "      <th>y(Return)</th>\n",
       "      <th>pseudo_y(ClippedReturn)</th>\n",
       "      <th>pseudo_y(SignReturn)</th>\n",
       "      <th>pseudo_y(pctChange)</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CalcDateTime</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>2017-07-03 09:51:00</th>\n",
       "      <td>BMW</td>\n",
       "      <td>0.0</td>\n",
       "      <td>0.0</td>\n",
       "      <td>82.34</td>\n",
       "      <td>0.0</td>\n",
       "      <td>1.0</td>\n",
       "      <td>2163.0</td>\n",
       "      <td>30.0</td>\n",
       "      <td>2017-07-03 09:51:00</td>\n",
       "      <td>2017-07-03</td>\n",
       "      <td>...</td>\n",
       "      <td>0.987222</td>\n",
       "      <td>-3.051665</td>\n",
       "      <td>0.0</td>\n",
       "      <td>3.021665</td>\n",
       "      <td>-3.031665</td>\n",
       "      <td>-1.027222</td>\n",
       "      <td>-82.33</td>\n",
       "      <td>-5.0</td>\n",
       "      <td>-1.0</td>\n",
       "      <td>-1.0</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "<p>1 rows × 136 columns</p>\n",
       "</div>"
      ],
      "text/plain": [
       "                    Mnemonic  MinPrice  MaxPrice  StartPrice  EndPrice  \\\n",
       "CalcDateTime                                                             \n",
       "2017-07-03 09:51:00      BMW       0.0       0.0       82.34       0.0   \n",
       "\n",
       "                     HasTrade  TradedVolume  NumberOfTrades  \\\n",
       "CalcDateTime                                                  \n",
       "2017-07-03 09:51:00       1.0        2163.0            30.0   \n",
       "\n",
       "                           CalcDateTime        Date         ...           \\\n",
       "CalcDateTime                                                ...            \n",
       "2017-07-03 09:51:00 2017-07-03 09:51:00  2017-07-03         ...            \n",
       "\n",
       "                     x(f4)(t - 8)  x((MaxP-EndP[-1])-(EndP-MinP[-1]))(t - 9)  \\\n",
       "CalcDateTime                                                                   \n",
       "2017-07-03 09:51:00      0.987222                                  -3.051665   \n",
       "\n",
       "                     x(f1)(t - 9)  x(f2)(t - 9)  x(f3)(t - 9)  x(f4)(t - 9)  \\\n",
       "CalcDateTime                                                                  \n",
       "2017-07-03 09:51:00           0.0      3.021665     -3.031665     -1.027222   \n",
       "\n",
       "                     y(Return)  pseudo_y(ClippedReturn)  pseudo_y(SignReturn)  \\\n",
       "CalcDateTime                                                                    \n",
       "2017-07-03 09:51:00     -82.33                     -5.0                  -1.0   \n",
       "\n",
       "                     pseudo_y(pctChange)  \n",
       "CalcDateTime                              \n",
       "2017-07-03 09:51:00                 -1.0  \n",
       "\n",
       "[1 rows x 136 columns]"
      ]
     },
     "execution_count": 13,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "dummy.ix[ -1, 'MinPrice'] = 0.0\n",
    "dummy.ix[ -1, 'MaxPrice'] = 0.0\n",
    "dummy.ix[ -1, 'EndPrice'] = 0.0\n",
    "\n",
    "\n",
    "Featurizer().transform(dummy)\n",
    "dummy.iloc[[-1]]"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "Make sure old features and new features computed with different values of EndPrice, MaxPrice, MinPrice of the last\n",
    "record give the same values for the `x(` features"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 14,
   "metadata": {},
   "outputs": [],
   "source": [
    "for f in dummy.dtypes.index:\n",
    "    if f.startswith('x('):\n",
    "        assert dummy[f][-1] == old[f][0]"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 15,
   "metadata": {},
   "outputs": [],
   "source": [
    "class TrainingSet:\n",
    "    def __init__(self, X, y, orig_df):\n",
    "        self.X = X\n",
    "        self.y = y\n",
    "        self.original_df = orig_df\n",
    "        \n",
    "class TrainingSetBuilder:\n",
    "    def transform(self, single_stock):\n",
    "        x_features = filter(lambda name: name.startswith('x('), list(single_stock.dtypes.index))\n",
    "        X = single_stock[x_features].values\n",
    "        y = single_stock[['pseudo_y(SignReturn)']].values\n",
    "        return TrainingSet(X, y, single_stock)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 16,
   "metadata": {},
   "outputs": [],
   "source": [
    "class Predictions:\n",
    "    def __init__(self, predictions, training_set):\n",
    "\n",
    "        self.predictions = predictions\n",
    "        self.training_set = training_set\n",
    "        \n",
    "    def evaluate(self):\n",
    "        single_feature = 'x((MaxP-EndP)-(EndP-MinP))(t - 1)'\n",
    "        stats_df = pd.DataFrame({\n",
    "                      'predictions': self.predictions[:,0],\n",
    "                      'single_feature_pred': self.training_set.original_df[single_feature].values,\n",
    "                      'pseudo_y(SignReturn)': self.training_set.y[:,0],\n",
    "                      'pseudo_y(pctChange)': self.training_set.original_df['pseudo_y(pctChange)'].values,\n",
    "                       'pseudo_y(ClippedReturn)': self.training_set.original_df['pseudo_y(ClippedReturn)'].values,\n",
    "                      'y(Return)': self.training_set.original_df['y(Return)'].values})\n",
    "        \n",
    "        corr = stats_df. \\\n",
    "            corr()[['predictions', 'single_feature_pred']]. \\\n",
    "            iloc[1:]\n",
    "            \n",
    "        pred_signs = np.sign(stats_df['predictions'])\n",
    "        y_signs = np.sign(stats_df['y(Return)'])\n",
    "        has_answer = np.absolute(pred_signs * y_signs).sum()\n",
    "        correct = np.where(pred_signs * y_signs == 1.0, 1.0, 0.0).sum()\n",
    "        \n",
    "        thresholds = []\n",
    "        accuracy = []\n",
    "        correct_lst = []\n",
    "        errors = []\n",
    "        percent_has_answer = []\n",
    "        abs_has_answer = []\n",
    "        achieved_returns = []\n",
    "\n",
    "        preds = stats_df['predictions']\n",
    "        \n",
    "        for d in range(5, 46, 5):\n",
    "            low = np.percentile(preds, d) \n",
    "            high = np.percentile(preds, 100 - d)\n",
    "            thresholded = np.where(preds > high, 1.0, np.where(preds < low, -1.0, 0.0))\n",
    "            c = np.where(np.sign(thresholded)*np.sign(y_signs) == 1.0, 1.0, 0.0).sum()\n",
    "            e = np.where(np.sign(thresholded)*np.sign(y_signs) == -1.0, 1.0, 0.0).sum()\n",
    "            achieved_ret = (stats_df['pseudo_y(pctChange)']*thresholded).sum()\n",
    "            correct_lst.append(c)\n",
    "            errors.append(e)\n",
    "            accuracy.append(c/(c + e))\n",
    "            percent_has_answer.append(100.0*(c + e)/pred_signs.shape[0])\n",
    "            abs_has_answer.append((c + e))\n",
    "            achieved_returns.append(achieved_ret)\n",
    "            thresholds.append(d)\n",
    "            \n",
    "        at_cutoff = DataFrame({\n",
    "                    'thresholds': thresholds,\n",
    "                    'accuracy': accuracy,\n",
    "                    'percent_with_answer': percent_has_answer,\n",
    "                    'absolute_has_answer': abs_has_answer,\n",
    "                    'achieved_returns': achieved_returns,\n",
    "                    'correct': correct_lst,\n",
    "                    'errors': errors\n",
    "        })\n",
    "        at_cutoff['achieved_norm_returns'] = at_cutoff['achieved_returns']/at_cutoff['absolute_has_answer']\n",
    "        \n",
    "        ret = stats_df['pseudo_y(pctChange)']\n",
    "        rand_feature = np.where(np.random.rand(ret.shape[0]) > 0.5, 1.0, -1.0)    \n",
    "        random_returns = (ret * rand_feature).sum()\n",
    "        always_up_returns = (ret*1.0).sum()\n",
    "        always_down_returns = (ret*-1.0).sum()\n",
    "        omnicient_returns = (np.absolute(ret)).sum()\n",
    "        achieved = (ret * pred_signs).sum()\n",
    "        return {\n",
    "            'corr': corr,\n",
    "            'accuracy_at_cutoff': at_cutoff,\n",
    "            'matches': {\n",
    "                'percent_correct': 100*correct/has_answer,\n",
    "                'percent_has_answer': has_answer/pred_signs.shape[0],\n",
    "                'absolute_with_answer': has_answer,\n",
    "                'size': pred_signs.shape[0]\n",
    "            },\n",
    "            'strategies': {\n",
    "                'omniscient': omnicient_returns,\n",
    "                'random': random_returns,\n",
    "                'always_up': always_up_returns,\n",
    "                'always_down': always_down_returns,\n",
    "                'achieved': achieved,\n",
    "                'num_trials': np.absolute(pred_signs).sum()\n",
    "            }\n",
    "        }\n",
    "    \n",
    "class MLModel:\n",
    "    def __init__(self):\n",
    "        self.model = None\n",
    "        \n",
    "    def fit(self, training_set, valid_set = None):\n",
    "        train_X, train_y = training_set.X, training_set.y\n",
    "        \n",
    "        if valid_set is None:\n",
    "            valid_X, valid_y = train_X, train_y\n",
    "        else:\n",
    "            valid_X, valid_y = valid_set.X, valid_set.y\n",
    "\n",
    "        model = Sequential()\n",
    "\n",
    "        model.add(Dense(100, activation='relu', input_shape =(train_X.shape[1],),\n",
    "                        kernel_regularizer=regularizers.l2(0.001))) \n",
    "        model.add(Dense(50, activation='relu', kernel_regularizer=regularizers.l2(0.001)))        \n",
    "\n",
    "        model.add(Dense(1))\n",
    "        \n",
    "        model.compile(loss='mean_squared_error', optimizer='adam')\n",
    "        self.model = model            \n",
    "\n",
    "        # fit network\n",
    "        # change the epochs back to 20\n",
    "        history = model.fit(train_X, train_y, epochs=50, batch_size=2500, validation_data=(valid_X, valid_y), verbose=2, shuffle=True)\n",
    "        # plot history\n",
    "        pyplot.plot(history.history['loss'], label='train')\n",
    "        pyplot.plot(history.history['val_loss'], label='valid')\n",
    "        pyplot.legend()\n",
    "        pyplot.show()\n",
    "        \n",
    "    def transform(self, input_set):\n",
    "        predictions = self.model.predict(input_set.X)\n",
    "        return Predictions(predictions, input_set)\n",
    "    \n",
    "    def fit_transform(self, training_set, valid_set):\n",
    "        self.fit(training_set, valid_set)\n",
    "        return self.transform(training_set), self.transform(valid_set)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 17,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "SNH: Dropped 0.02 % of records due to NA\n",
      "(43080, 136)\n",
      "SNH: Dropped 0.29 % of records due to NA\n",
      "SNH: Dropped 0.04 % of records due to NA\n",
      "DBK: Dropped 0.02 % of records due to NA\n",
      "(54629, 136)\n",
      "DBK: Dropped 0.23 % of records due to NA\n",
      "DBK: Dropped 0.03 % of records due to NA\n",
      "EOAN: Dropped 0.02 % of records due to NA\n",
      "(54570, 136)\n",
      "EOAN: Dropped 0.23 % of records due to NA\n",
      "EOAN: Dropped 0.03 % of records due to NA\n",
      "DTE: Dropped 0.02 % of records due to NA\n",
      "(51751, 136)\n",
      "DTE: Dropped 0.24 % of records due to NA\n",
      "DTE: Dropped 0.03 % of records due to NA\n",
      "CBK: Dropped 0.02 % of records due to NA\n",
      "(52433, 136)\n",
      "CBK: Dropped 0.25 % of records due to NA\n",
      "CBK: Dropped 0.03 % of records due to NA\n",
      "RWE: Dropped 0.02 % of records due to NA\n",
      "(52128, 136)\n",
      "RWE: Dropped 0.24 % of records due to NA\n",
      "RWE: Dropped 0.03 % of records due to NA\n",
      "IFX: Dropped 0.02 % of records due to NA\n",
      "(52194, 136)\n",
      "IFX: Dropped 0.24 % of records due to NA\n",
      "IFX: Dropped 0.03 % of records due to NA\n",
      "SVAB: Dropped 0.19 % of records due to NA\n",
      "(5224, 136)\n",
      "SVAB: Dropped 10.20 % of records due to NA\n",
      "SVAB: Dropped 0.33 % of records due to NA\n",
      "LHA: Dropped 0.02 % of records due to NA\n",
      "(53069, 136)\n",
      "LHA: Dropped 0.24 % of records due to NA\n",
      "LHA: Dropped 0.03 % of records due to NA\n",
      "DAI: Dropped 0.02 % of records due to NA\n",
      "(54899, 136)\n",
      "DAI: Dropped 0.23 % of records due to NA\n",
      "DAI: Dropped 0.03 % of records due to NA\n",
      "O2D: Dropped 0.03 % of records due to NA\n",
      "(35629, 136)\n",
      "O2D: Dropped 0.36 % of records due to NA\n",
      "O2D: Dropped 0.04 % of records due to NA\n",
      "TKA: Dropped 0.02 % of records due to NA\n",
      "(53135, 136)\n",
      "TKA: Dropped 0.24 % of records due to NA\n",
      "TKA: Dropped 0.03 % of records due to NA\n",
      "DPW: Dropped 0.02 % of records due to NA\n",
      "(52758, 136)\n",
      "DPW: Dropped 0.24 % of records due to NA\n",
      "DPW: Dropped 0.03 % of records due to NA\n",
      "HDD: Dropped 0.04 % of records due to NA\n",
      "(25694, 136)\n",
      "HDD: Dropped 0.45 % of records due to NA\n",
      "HDD: Dropped 0.07 % of records due to NA\n",
      "SIE: Dropped 0.02 % of records due to NA\n",
      "(50940, 136)\n",
      "SIE: Dropped 0.25 % of records due to NA\n",
      "SIE: Dropped 0.03 % of records due to NA\n",
      "AIXA: Dropped 0.03 % of records due to NA\n",
      "(34964, 136)\n",
      "AIXA: Dropped 0.34 % of records due to NA\n",
      "AIXA: Dropped 0.04 % of records due to NA\n",
      "BAYN: Dropped 0.02 % of records due to NA\n",
      "(49181, 136)\n",
      "BAYN: Dropped 0.26 % of records due to NA\n",
      "BAYN: Dropped 0.03 % of records due to NA\n",
      "SAP: Dropped 0.02 % of records due to NA\n",
      "(54520, 136)\n",
      "SAP: Dropped 0.23 % of records due to NA\n",
      "SAP: Dropped 0.03 % of records due to NA\n",
      "BAS: Dropped 0.02 % of records due to NA\n",
      "(54901, 136)\n",
      "BAS: Dropped 0.23 % of records due to NA\n",
      "BAS: Dropped 0.03 % of records due to NA\n",
      "EVT: Dropped 0.02 % of records due to NA\n",
      "(41726, 136)\n",
      "EVT: Dropped 0.31 % of records due to NA\n",
      "EVT: Dropped 0.04 % of records due to NA\n",
      "AT1: Dropped 0.06 % of records due to NA\n",
      "(16914, 136)\n",
      "AT1: Dropped 1.05 % of records due to NA\n",
      "AT1: Dropped 0.05 % of records due to NA\n",
      "PSM: Dropped 0.02 % of records due to NA\n",
      "(50618, 136)\n",
      "PSM: Dropped 0.24 % of records due to NA\n",
      "PSM: Dropped 0.04 % of records due to NA\n",
      "BMW: Dropped 0.02 % of records due to NA\n",
      "(54433, 136)\n",
      "BMW: Dropped 0.23 % of records due to NA\n",
      "BMW: Dropped 0.03 % of records due to NA\n",
      "VOW3: Dropped 0.02 % of records due to NA\n",
      "(52041, 136)\n",
      "VOW3: Dropped 0.24 % of records due to NA\n",
      "VOW3: Dropped 0.03 % of records due to NA\n",
      "FRE: Dropped 0.02 % of records due to NA\n",
      "(53542, 136)\n",
      "FRE: Dropped 0.24 % of records due to NA\n",
      "FRE: Dropped 0.03 % of records due to NA\n",
      "GAZ: Dropped 0.06 % of records due to NA\n",
      "(15783, 136)\n",
      "GAZ: Dropped 0.76 % of records due to NA\n",
      "GAZ: Dropped 0.11 % of records due to NA\n",
      "SDF: Dropped 0.02 % of records due to NA\n",
      "(41345, 136)\n",
      "SDF: Dropped 0.30 % of records due to NA\n",
      "SDF: Dropped 0.04 % of records due to NA\n",
      "CEC: Dropped 0.03 % of records due to NA\n",
      "(30171, 136)\n",
      "CEC: Dropped 0.55 % of records due to NA\n",
      "CEC: Dropped 0.05 % of records due to NA\n",
      "ALV: Dropped 0.02 % of records due to NA\n",
      "(52645, 136)\n",
      "ALV: Dropped 0.24 % of records due to NA\n",
      "ALV: Dropped 0.03 % of records due to NA\n",
      "VNA: Dropped 0.02 % of records due to NA\n",
      "(50631, 136)\n",
      "VNA: Dropped 0.25 % of records due to NA\n",
      "VNA: Dropped 0.03 % of records due to NA\n",
      "B4B: Dropped 0.03 % of records due to NA\n",
      "(33049, 136)\n",
      "B4B: Dropped 0.54 % of records due to NA\n",
      "B4B: Dropped 0.05 % of records due to NA\n",
      "SHA: Dropped 0.03 % of records due to NA\n",
      "(36181, 136)\n",
      "SHA: Dropped 0.36 % of records due to NA\n",
      "SHA: Dropped 0.04 % of records due to NA\n",
      "AB1: Dropped 0.15 % of records due to NA\n",
      "(6856, 136)\n",
      "AB1: Dropped 1.22 % of records due to NA\n",
      "AB1: Dropped 1.16 % of records due to NA\n",
      "UN01: Dropped 0.02 % of records due to NA\n",
      "(43030, 136)\n",
      "UN01: Dropped 0.29 % of records due to NA\n",
      "UN01: Dropped 0.05 % of records due to NA\n",
      "DLG: Dropped 0.02 % of records due to NA\n",
      "(42499, 136)\n",
      "DLG: Dropped 0.31 % of records due to NA\n",
      "DLG: Dropped 0.04 % of records due to NA\n",
      "NDX1: Dropped 0.04 % of records due to NA\n",
      "(27627, 136)\n",
      "NDX1: Dropped 0.39 % of records due to NA\n",
      "NDX1: Dropped 0.07 % of records due to NA\n",
      "NOA3: Dropped 0.07 % of records due to NA\n",
      "(13859, 136)\n",
      "NOA3: Dropped 1.05 % of records due to NA\n",
      "NOA3: Dropped 0.10 % of records due to NA\n",
      "IGY: Dropped 0.02 % of records due to NA\n",
      "(41778, 136)\n",
      "IGY: Dropped 0.30 % of records due to NA\n",
      "IGY: Dropped 0.04 % of records due to NA\n",
      "VODI: Dropped 0.12 % of records due to NA\n",
      "(8485, 136)\n",
      "VODI: Dropped 1.60 % of records due to NA\n",
      "VODI: Dropped 0.13 % of records due to NA\n",
      "ADS: Dropped 0.02 % of records due to NA\n",
      "(49531, 136)\n",
      "ADS: Dropped 0.26 % of records due to NA\n",
      "ADS: Dropped 0.03 % of records due to NA\n",
      "1COV: Dropped 0.02 % of records due to NA\n",
      "(45459, 136)\n",
      "1COV: Dropped 0.28 % of records due to NA\n",
      "1COV: Dropped 0.03 % of records due to NA\n",
      "TUI1: Dropped 0.04 % of records due to NA\n",
      "(28481, 136)\n",
      "TUI1: Dropped 0.47 % of records due to NA\n",
      "TUI1: Dropped 0.05 % of records due to NA\n",
      "BPE5: Dropped 0.08 % of records due to NA\n",
      "(13194, 136)\n",
      "BPE5: Dropped 0.91 % of records due to NA\n",
      "BPE5: Dropped 0.09 % of records due to NA\n",
      "HEI: Dropped 0.02 % of records due to NA\n",
      "(51655, 136)\n",
      "HEI: Dropped 0.24 % of records due to NA\n",
      "HEI: Dropped 0.03 % of records due to NA\n",
      "KCO: Dropped 0.04 % of records due to NA\n",
      "(24124, 136)\n",
      "KCO: Dropped 0.49 % of records due to NA\n",
      "KCO: Dropped 0.07 % of records due to NA\n",
      "ADV: Dropped 0.05 % of records due to NA\n",
      "(21386, 136)\n",
      "ADV: Dropped 0.52 % of records due to NA\n",
      "ADV: Dropped 0.12 % of records due to NA\n",
      "SZU: Dropped 0.03 % of records due to NA\n",
      "(32356, 136)\n",
      "SZU: Dropped 0.40 % of records due to NA\n",
      "SZU: Dropped 0.06 % of records due to NA\n",
      "EVK: Dropped 0.02 % of records due to NA\n",
      "(41859, 136)\n",
      "EVK: Dropped 0.31 % of records due to NA\n",
      "EVK: Dropped 0.04 % of records due to NA\n",
      "HEN3: Dropped 0.02 % of records due to NA\n",
      "(42509, 136)\n",
      "HEN3: Dropped 0.30 % of records due to NA\n",
      "HEN3: Dropped 0.04 % of records due to NA\n",
      "WDI: Dropped 0.02 % of records due to NA\n",
      "(45826, 136)\n",
      "WDI: Dropped 0.27 % of records due to NA\n",
      "WDI: Dropped 0.03 % of records due to NA\n",
      "MUV2: Dropped 0.02 % of records due to NA\n",
      "(49206, 136)\n",
      "MUV2: Dropped 0.25 % of records due to NA\n",
      "MUV2: Dropped 0.03 % of records due to NA\n",
      "DWNI: Dropped 0.02 % of records due to NA\n",
      "(41801, 136)\n",
      "DWNI: Dropped 0.30 % of records due to NA\n",
      "DWNI: Dropped 0.04 % of records due to NA\n",
      "MRK: Dropped 0.02 % of records due to NA\n",
      "(49021, 136)\n",
      "MRK: Dropped 0.26 % of records due to NA\n",
      "MRK: Dropped 0.03 % of records due to NA\n",
      "USE: Dropped 1.00 % of records due to NA\n",
      "(994, 136)\n",
      "USE: Dropped 16.95 % of records due to NA\n",
      "USE: Dropped 0.64 % of records due to NA\n",
      "PAH3: Dropped 0.02 % of records due to NA\n",
      "(46348, 136)\n",
      "PAH3: Dropped 0.29 % of records due to NA\n",
      "PAH3: Dropped 0.03 % of records due to NA\n",
      "DEZ: Dropped 0.04 % of records due to NA\n",
      "(22384, 136)\n",
      "DEZ: Dropped 0.53 % of records due to NA\n",
      "DEZ: Dropped 0.08 % of records due to NA\n",
      "FME: Dropped 0.02 % of records due to NA\n",
      "(50673, 136)\n",
      "FME: Dropped 0.25 % of records due to NA\n",
      "FME: Dropped 0.03 % of records due to NA\n",
      "G1A: Dropped 0.02 % of records due to NA\n",
      "(42799, 136)\n",
      "G1A: Dropped 0.29 % of records due to NA\n",
      "G1A: Dropped 0.04 % of records due to NA\n",
      "FNTN: Dropped 0.03 % of records due to NA\n",
      "(35417, 136)\n",
      "FNTN: Dropped 0.35 % of records due to NA\n",
      "FNTN: Dropped 0.04 % of records due to NA\n",
      "RKET: Dropped 0.04 % of records due to NA\n",
      "(27875, 136)\n",
      "RKET: Dropped 0.46 % of records due to NA\n",
      "RKET: Dropped 0.07 % of records due to NA\n"
     ]
    },
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "QIA: Dropped 0.03 % of records due to NA\n",
      "(33737, 136)\n",
      "QIA: Dropped 0.40 % of records due to NA\n",
      "QIA: Dropped 0.05 % of records due to NA\n",
      "DB1: Dropped 0.02 % of records due to NA\n",
      "(49103, 136)\n",
      "DB1: Dropped 0.25 % of records due to NA\n",
      "DB1: Dropped 0.04 % of records due to NA\n",
      "ZAL: Dropped 0.02 % of records due to NA\n",
      "(41253, 136)\n",
      "ZAL: Dropped 0.30 % of records due to NA\n",
      "ZAL: Dropped 0.04 % of records due to NA\n",
      "QSC: Dropped 0.10 % of records due to NA\n",
      "(9595, 136)\n",
      "QSC: Dropped 1.32 % of records due to NA\n",
      "QSC: Dropped 0.29 % of records due to NA\n",
      "CON: Dropped 0.02 % of records due to NA\n",
      "(49654, 136)\n",
      "CON: Dropped 0.26 % of records due to NA\n",
      "CON: Dropped 0.03 % of records due to NA\n",
      "SGL: Dropped 0.04 % of records due to NA\n",
      "(23599, 136)\n",
      "SGL: Dropped 0.55 % of records due to NA\n",
      "SGL: Dropped 0.09 % of records due to NA\n",
      "BVB: Dropped 0.05 % of records due to NA\n",
      "(18209, 136)\n",
      "BVB: Dropped 0.81 % of records due to NA\n",
      "BVB: Dropped 0.10 % of records due to NA\n",
      "TINA: Dropped 0.13 % of records due to NA\n",
      "(7575, 136)\n",
      "TINA: Dropped 1.49 % of records due to NA\n",
      "TINA: Dropped 0.26 % of records due to NA\n",
      "PBB: Dropped 0.05 % of records due to NA\n",
      "(21457, 136)\n",
      "PBB: Dropped 0.54 % of records due to NA\n",
      "PBB: Dropped 0.07 % of records due to NA\n",
      "PNE3: Dropped 0.11 % of records due to NA\n",
      "(8856, 136)\n",
      "PNE3: Dropped 1.54 % of records due to NA\n",
      "PNE3: Dropped 0.17 % of records due to NA\n",
      "RIB: Dropped 0.05 % of records due to NA\n",
      "(20077, 136)\n",
      "RIB: Dropped 0.74 % of records due to NA\n",
      "RIB: Dropped 0.06 % of records due to NA\n",
      "OSR: Dropped 0.02 % of records due to NA\n",
      "(42827, 136)\n",
      "OSR: Dropped 0.31 % of records due to NA\n",
      "OSR: Dropped 0.04 % of records due to NA\n",
      "SHL: Dropped 0.00 % of records due to NA\n",
      "(0, 136)\n",
      "SHL: Dropped 0.00 % of records due to NA\n",
      "SHL: Dropped 0.12 % of records due to NA\n",
      "AOX: Dropped 0.05 % of records due to NA\n",
      "(18333, 136)\n",
      "AOX: Dropped 0.80 % of records due to NA\n",
      "AOX: Dropped 0.09 % of records due to NA\n",
      "BEI: Dropped 0.02 % of records due to NA\n",
      "(47736, 136)\n",
      "BEI: Dropped 0.26 % of records due to NA\n",
      "BEI: Dropped 0.03 % of records due to NA\n",
      "TEG: Dropped 0.04 % of records due to NA\n",
      "(22816, 136)\n",
      "TEG: Dropped 0.56 % of records due to NA\n",
      "TEG: Dropped 0.07 % of records due to NA\n",
      "UTDI: Dropped 0.03 % of records due to NA\n",
      "(37004, 136)\n",
      "UTDI: Dropped 0.32 % of records due to NA\n",
      "UTDI: Dropped 0.04 % of records due to NA\n",
      "ARL: Dropped 0.03 % of records due to NA\n",
      "(35840, 136)\n",
      "ARL: Dropped 0.33 % of records due to NA\n",
      "ARL: Dropped 0.04 % of records due to NA\n",
      "MDG1: Dropped 0.06 % of records due to NA\n",
      "(16017, 136)\n",
      "MDG1: Dropped 0.66 % of records due to NA\n",
      "MDG1: Dropped 0.07 % of records due to NA\n",
      "KGX: Dropped 0.02 % of records due to NA\n",
      "(45061, 136)\n",
      "KGX: Dropped 0.29 % of records due to NA\n",
      "KGX: Dropped 0.04 % of records due to NA\n",
      "LXS: Dropped 0.02 % of records due to NA\n",
      "(42922, 136)\n",
      "LXS: Dropped 0.30 % of records due to NA\n",
      "LXS: Dropped 0.04 % of records due to NA\n",
      "ARO: Dropped 2.17 % of records due to NA\n",
      "(451, 136)\n",
      "ARO: Dropped 18.87 % of records due to NA\n",
      "ARO: Dropped 2.75 % of records due to NA\n",
      "TTI: Dropped 0.07 % of records due to NA\n",
      "(14958, 136)\n",
      "TTI: Dropped 1.00 % of records due to NA\n",
      "TTI: Dropped 0.10 % of records due to NA\n",
      "SANT: Dropped 0.05 % of records due to NA\n",
      "(19809, 136)\n",
      "SANT: Dropped 0.55 % of records due to NA\n",
      "SANT: Dropped 0.07 % of records due to NA\n",
      "GYC: Dropped 0.04 % of records due to NA\n",
      "(22858, 136)\n",
      "GYC: Dropped 0.56 % of records due to NA\n",
      "GYC: Dropped 0.07 % of records due to NA\n",
      "ANO: Dropped 0.64 % of records due to NA\n",
      "(1548, 136)\n",
      "ANO: Dropped 10.87 % of records due to NA\n",
      "ANO: Dropped 0.00 % of records due to NA\n",
      "LINU: Dropped 0.05 % of records due to NA\n",
      "(18534, 136)\n",
      "LINU: Dropped 1.16 % of records due to NA\n",
      "LINU: Dropped 0.03 % of records due to NA\n",
      "SOW: Dropped 0.03 % of records due to NA\n",
      "(32190, 136)\n",
      "SOW: Dropped 0.46 % of records due to NA\n",
      "SOW: Dropped 0.05 % of records due to NA\n",
      "SZG: Dropped 0.03 % of records due to NA\n",
      "(35554, 136)\n",
      "SZG: Dropped 0.35 % of records due to NA\n",
      "SZG: Dropped 0.04 % of records due to NA\n",
      "LLD: Dropped 0.86 % of records due to NA\n",
      "(1155, 136)\n",
      "LLD: Dropped 9.90 % of records due to NA\n",
      "LLD: Dropped 1.67 % of records due to NA\n",
      "BOSS: Dropped 0.03 % of records due to NA\n",
      "(39917, 136)\n",
      "BOSS: Dropped 0.31 % of records due to NA\n",
      "BOSS: Dropped 0.04 % of records due to NA\n",
      "BNR: Dropped 0.03 % of records due to NA\n",
      "(38167, 136)\n",
      "BNR: Dropped 0.35 % of records due to NA\n",
      "BNR: Dropped 0.05 % of records due to NA\n",
      "WAF: Dropped 0.03 % of records due to NA\n",
      "(37339, 136)\n",
      "WAF: Dropped 0.36 % of records due to NA\n",
      "WAF: Dropped 0.04 % of records due to NA\n",
      "LIN: Dropped 0.02 % of records due to NA\n",
      "(43474, 136)\n",
      "LIN: Dropped 0.31 % of records due to NA\n",
      "LIN: Dropped 0.06 % of records due to NA\n",
      "DRI: Dropped 0.03 % of records due to NA\n",
      "(30972, 136)\n",
      "DRI: Dropped 0.42 % of records due to NA\n",
      "DRI: Dropped 0.07 % of records due to NA\n",
      "NDA: Dropped 0.03 % of records due to NA\n",
      "(36381, 136)\n",
      "NDA: Dropped 0.35 % of records due to NA\n",
      "NDA: Dropped 0.04 % of records due to NA\n",
      "ZIL2: Dropped 0.05 % of records due to NA\n",
      "(19313, 136)\n",
      "ZIL2: Dropped 0.65 % of records due to NA\n",
      "ZIL2: Dropped 0.12 % of records due to NA\n",
      "SY1: Dropped 0.03 % of records due to NA\n",
      "(38471, 136)\n",
      "SY1: Dropped 0.35 % of records due to NA\n",
      "SY1: Dropped 0.04 % of records due to NA\n",
      "CAP: Dropped 0.08 % of records due to NA\n",
      "(12263, 136)\n",
      "CAP: Dropped 1.23 % of records due to NA\n",
      "CAP: Dropped 0.15 % of records due to NA\n",
      "3W9K: Dropped 1.91 % of records due to NA\n",
      "(514, 136)\n",
      "3W9K: Dropped 20.83 % of records due to NA\n",
      "3W9K: Dropped 1.83 % of records due to NA\n",
      "Train on 3397349 samples, validate on 264794 samples\n",
      "Epoch 1/50\n",
      " - 21s - loss: 0.9017 - val_loss: 0.8270\n",
      "Epoch 2/50\n",
      " - 21s - loss: 0.8108 - val_loss: 0.7943\n",
      "Epoch 3/50\n",
      " - 21s - loss: 0.7941 - val_loss: 0.7870\n",
      "Epoch 4/50\n",
      " - 21s - loss: 0.7906 - val_loss: 0.7862\n",
      "Epoch 5/50\n",
      " - 21s - loss: 0.7899 - val_loss: 0.7856\n",
      "Epoch 6/50\n",
      " - 21s - loss: 0.7897 - val_loss: 0.7855\n",
      "Epoch 7/50\n",
      " - 22s - loss: 0.7897 - val_loss: 0.7852\n",
      "Epoch 8/50\n",
      " - 21s - loss: 0.7896 - val_loss: 0.7853\n",
      "Epoch 9/50\n",
      " - 21s - loss: 0.7896 - val_loss: 0.7852\n",
      "Epoch 10/50\n",
      " - 20s - loss: 0.7896 - val_loss: 0.7852\n",
      "Epoch 11/50\n",
      " - 20s - loss: 0.7896 - val_loss: 0.7850\n",
      "Epoch 12/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 13/50\n",
      " - 21s - loss: 0.7896 - val_loss: 0.7853\n",
      "Epoch 14/50\n",
      " - 20s - loss: 0.7896 - val_loss: 0.7856\n",
      "Epoch 15/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7854\n",
      "Epoch 16/50\n",
      " - 22s - loss: 0.7895 - val_loss: 0.7855\n",
      "Epoch 17/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 18/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7852\n",
      "Epoch 19/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 20/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 21/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7853\n",
      "Epoch 22/50\n",
      " - 22s - loss: 0.7895 - val_loss: 0.7857\n",
      "Epoch 23/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7852\n",
      "Epoch 24/50\n",
      " - 22s - loss: 0.7895 - val_loss: 0.7852\n",
      "Epoch 25/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7850\n",
      "Epoch 26/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7852\n",
      "Epoch 27/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 28/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 29/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 30/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7852\n",
      "Epoch 31/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 32/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 33/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 34/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7852\n",
      "Epoch 35/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7852\n",
      "Epoch 36/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7855\n",
      "Epoch 37/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7854\n",
      "Epoch 38/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7850\n",
      "Epoch 39/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7853\n",
      "Epoch 40/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7850\n",
      "Epoch 41/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7850\n",
      "Epoch 42/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7852\n",
      "Epoch 43/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7852\n",
      "Epoch 44/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7852\n",
      "Epoch 45/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 46/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7853\n",
      "Epoch 47/50\n",
      " - 20s - loss: 0.7895 - val_loss: 0.7855\n",
      "Epoch 48/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7850\n",
      "Epoch 49/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7851\n",
      "Epoch 50/50\n",
      " - 21s - loss: 0.7895 - val_loss: 0.7852\n"
     ]
    },
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fd1dee6c650>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "combined_training_set = []\n",
    "combined_valid_set = []\n",
    "combined_test_set = []\n",
    "\n",
    "for mnemonic in all_mnemonics: #['SIE', 'BMW', 'SNH', 'DBK', 'EOAN',  'DTE', 'CBK', 'RWE', 'IFX']:\n",
    "\n",
    "    single_stock = df_train[df_train.Mnemonic == mnemonic].copy()\n",
    "    single_stock = single_stock[single_stock.HasTrade == 1.0]\n",
    "    single_stock = Featurizer().transform(single_stock)\n",
    "    single_stock = NARemover(mnemonic).transform(single_stock)\n",
    "    combined_training_set.append(single_stock)\n",
    "    print(single_stock.shape)\n",
    "\n",
    "    single_stock = df_valid[df_valid.Mnemonic == mnemonic].copy()\n",
    "    single_stock = single_stock[single_stock.HasTrade == 1.0] \n",
    "    single_stock = Featurizer().transform(single_stock)\n",
    "    single_stock = NARemover(mnemonic).transform(single_stock)\n",
    "    combined_valid_set.append(single_stock)\n",
    "    \n",
    "    single_stock = df_test[df_test.Mnemonic == mnemonic].copy()\n",
    "    single_stock = single_stock[single_stock.HasTrade == 1.0] \n",
    "    single_stock = Featurizer().transform(single_stock)\n",
    "    single_stock = NARemover(mnemonic).transform(single_stock)\n",
    "    combined_test_set.append(single_stock)\n",
    "    \n",
    "combined_training_set_df = pd.concat(combined_training_set, axis=0)\n",
    "training_set = TrainingSetBuilder().transform(combined_training_set_df)\n",
    "    \n",
    "combined_valid_set_df = pd.concat(combined_valid_set, axis=0)\n",
    "valid_set = TrainingSetBuilder().transform(combined_valid_set_df) \n",
    "\n",
    "combined_test_set_df = pd.concat(combined_test_set, axis=0)\n",
    "test_set = TrainingSetBuilder().transform(combined_test_set_df) \n",
    "    \n",
    "model = MLModel()\n",
    "train_predictions, valid_predictions = model.fit_transform(training_set, valid_set)\n"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 18,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'absolute_with_answer': 2700262.0,\n",
       " 'percent_correct': 53.93484039696889,\n",
       " 'percent_has_answer': 0.7948144273667498,\n",
       " 'size': 3397349}"
      ]
     },
     "execution_count": 18,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_predictions.evaluate()['matches']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 19,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'absolute_with_answer': 209181.0,\n",
       " 'percent_correct': 53.83471730224064,\n",
       " 'percent_has_answer': 0.7899763589809436,\n",
       " 'size': 264794}"
      ]
     },
     "execution_count": 19,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "valid_predictions.evaluate()['matches']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 20,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'absolute_with_answer': 1662337.0,\n",
       " 'percent_correct': 53.5093064763643,\n",
       " 'percent_has_answer': 0.7810295283951453,\n",
       " 'size': 2128392}"
      ]
     },
     "execution_count": 20,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "test_predictions = model.transform(test_set)\n",
    "test_predictions.evaluate()['matches']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 21,
   "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>absolute_has_answer</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>achieved_returns</th>\n",
       "      <th>correct</th>\n",
       "      <th>errors</th>\n",
       "      <th>percent_with_answer</th>\n",
       "      <th>thresholds</th>\n",
       "      <th>achieved_norm_returns</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>267937.0</td>\n",
       "      <td>0.597517</td>\n",
       "      <td>115.067683</td>\n",
       "      <td>160097.0</td>\n",
       "      <td>107840.0</td>\n",
       "      <td>7.886649</td>\n",
       "      <td>5</td>\n",
       "      <td>0.000429</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>537230.0</td>\n",
       "      <td>0.584169</td>\n",
       "      <td>178.481673</td>\n",
       "      <td>313833.0</td>\n",
       "      <td>223397.0</td>\n",
       "      <td>15.813212</td>\n",
       "      <td>10</td>\n",
       "      <td>0.000332</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>807813.0</td>\n",
       "      <td>0.575225</td>\n",
       "      <td>227.250309</td>\n",
       "      <td>464674.0</td>\n",
       "      <td>343139.0</td>\n",
       "      <td>23.777746</td>\n",
       "      <td>15</td>\n",
       "      <td>0.000281</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>1079270.0</td>\n",
       "      <td>0.568268</td>\n",
       "      <td>262.329637</td>\n",
       "      <td>613315.0</td>\n",
       "      <td>465955.0</td>\n",
       "      <td>31.768005</td>\n",
       "      <td>20</td>\n",
       "      <td>0.000243</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>1351000.0</td>\n",
       "      <td>0.561947</td>\n",
       "      <td>289.974632</td>\n",
       "      <td>759191.0</td>\n",
       "      <td>591809.0</td>\n",
       "      <td>39.766300</td>\n",
       "      <td>25</td>\n",
       "      <td>0.000215</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>1622200.0</td>\n",
       "      <td>0.556761</td>\n",
       "      <td>314.370359</td>\n",
       "      <td>903178.0</td>\n",
       "      <td>719022.0</td>\n",
       "      <td>47.748995</td>\n",
       "      <td>30</td>\n",
       "      <td>0.000194</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>1892992.0</td>\n",
       "      <td>0.551755</td>\n",
       "      <td>337.534825</td>\n",
       "      <td>1044468.0</td>\n",
       "      <td>848524.0</td>\n",
       "      <td>55.719680</td>\n",
       "      <td>35</td>\n",
       "      <td>0.000178</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>2162466.0</td>\n",
       "      <td>0.547523</td>\n",
       "      <td>352.966457</td>\n",
       "      <td>1184000.0</td>\n",
       "      <td>978466.0</td>\n",
       "      <td>63.651571</td>\n",
       "      <td>40</td>\n",
       "      <td>0.000163</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>2431481.0</td>\n",
       "      <td>0.543373</td>\n",
       "      <td>364.691663</td>\n",
       "      <td>1321202.0</td>\n",
       "      <td>1110279.0</td>\n",
       "      <td>71.569951</td>\n",
       "      <td>45</td>\n",
       "      <td>0.000150</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   absolute_has_answer  accuracy  achieved_returns    correct     errors  \\\n",
       "0             267937.0  0.597517        115.067683   160097.0   107840.0   \n",
       "1             537230.0  0.584169        178.481673   313833.0   223397.0   \n",
       "2             807813.0  0.575225        227.250309   464674.0   343139.0   \n",
       "3            1079270.0  0.568268        262.329637   613315.0   465955.0   \n",
       "4            1351000.0  0.561947        289.974632   759191.0   591809.0   \n",
       "5            1622200.0  0.556761        314.370359   903178.0   719022.0   \n",
       "6            1892992.0  0.551755        337.534825  1044468.0   848524.0   \n",
       "7            2162466.0  0.547523        352.966457  1184000.0   978466.0   \n",
       "8            2431481.0  0.543373        364.691663  1321202.0  1110279.0   \n",
       "\n",
       "   percent_with_answer  thresholds  achieved_norm_returns  \n",
       "0             7.886649           5               0.000429  \n",
       "1            15.813212          10               0.000332  \n",
       "2            23.777746          15               0.000281  \n",
       "3            31.768005          20               0.000243  \n",
       "4            39.766300          25               0.000215  \n",
       "5            47.748995          30               0.000194  \n",
       "6            55.719680          35               0.000178  \n",
       "7            63.651571          40               0.000163  \n",
       "8            71.569951          45               0.000150  "
      ]
     },
     "execution_count": 21,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_predictions.evaluate()['accuracy_at_cutoff']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 22,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "So if you play 3397349 times on the Training Set with 1 EUR and you always guess the movement,\n",
      "ignoring all transactions cost, you will make 2191.20627456. \n",
      "Instead you make 371.468652327 or 16.9527011966 percent of the ideally achievable.\n",
      "If you use the baseline you will make 156.787598049 or 7.15530983409 percent of ideal\n"
     ]
    }
   ],
   "source": [
    "def pred_baseline(d):\n",
    "    single_feature = 'x((MaxP-EndP)-(EndP-MinP))(t - 1)'\n",
    "    preds = d.training_set.original_df[single_feature].values\n",
    "    preds = preds.reshape((preds.shape[0], 1))\n",
    "    return Predictions(preds, d.training_set).evaluate()\n",
    "\n",
    "def readable_summary(which_set, p):\n",
    "    achieved = p.evaluate()['strategies']['achieved']\n",
    "    achieved_baseline = pred_baseline(p)['strategies']['achieved']\n",
    "    per_change = np.mean(np.absolute(p.training_set.original_df['pseudo_y(pctChange)']))\n",
    "    n = p.training_set.original_df.shape[0]\n",
    "    print (\"\"\"So if you play {} times on the {} with 1 EUR and you always guess the movement,\n",
    "ignoring all transactions cost, you will make {}. \n",
    "Instead you make {} or {} percent of the ideally achievable.\n",
    "If you use the baseline you will make {} or {} percent of ideal\"\"\".format(\n",
    "        n, which_set, n * per_change, achieved, 100.0*achieved/(n*per_change),\n",
    "          achieved_baseline, 100.0*achieved_baseline/(n*per_change)))\n",
    "readable_summary('Training Set', train_predictions)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 23,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'achieved': 371.4686523273236,\n",
       " 'always_down': -36.53431596652022,\n",
       " 'always_up': 36.53431596652022,\n",
       " 'num_trials': 3397349.0,\n",
       " 'omniscient': 2191.206274564835,\n",
       " 'random': 0.37983003012628735}"
      ]
     },
     "execution_count": 23,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "train_predictions.evaluate()['strategies']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 24,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'achieved': 10.300250300757304,\n",
       " 'always_down': -16.120621322692667,\n",
       " 'always_up': 16.120621322692667,\n",
       " 'num_trials': 175315.0,\n",
       " 'omniscient': 201.38785391708964,\n",
       " 'random': 7.49194215978099}"
      ]
     },
     "execution_count": 24,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "pred_baseline(valid_predictions)['strategies'] "
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 25,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>absolute_has_answer</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>achieved_returns</th>\n",
       "      <th>correct</th>\n",
       "      <th>errors</th>\n",
       "      <th>percent_with_answer</th>\n",
       "      <th>thresholds</th>\n",
       "      <th>achieved_norm_returns</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>20696.0</td>\n",
       "      <td>0.593158</td>\n",
       "      <td>4.859130</td>\n",
       "      <td>12276.0</td>\n",
       "      <td>8420.0</td>\n",
       "      <td>7.815887</td>\n",
       "      <td>5</td>\n",
       "      <td>0.000235</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>41537.0</td>\n",
       "      <td>0.579170</td>\n",
       "      <td>8.765475</td>\n",
       "      <td>24057.0</td>\n",
       "      <td>17480.0</td>\n",
       "      <td>15.686534</td>\n",
       "      <td>10</td>\n",
       "      <td>0.000211</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>62395.0</td>\n",
       "      <td>0.572209</td>\n",
       "      <td>11.851657</td>\n",
       "      <td>35703.0</td>\n",
       "      <td>26692.0</td>\n",
       "      <td>23.563600</td>\n",
       "      <td>15</td>\n",
       "      <td>0.000190</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>83407.0</td>\n",
       "      <td>0.564605</td>\n",
       "      <td>13.683131</td>\n",
       "      <td>47092.0</td>\n",
       "      <td>36315.0</td>\n",
       "      <td>31.498826</td>\n",
       "      <td>20</td>\n",
       "      <td>0.000164</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>104389.0</td>\n",
       "      <td>0.558814</td>\n",
       "      <td>16.520182</td>\n",
       "      <td>58334.0</td>\n",
       "      <td>46055.0</td>\n",
       "      <td>39.422721</td>\n",
       "      <td>25</td>\n",
       "      <td>0.000158</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>125342.0</td>\n",
       "      <td>0.553980</td>\n",
       "      <td>17.181956</td>\n",
       "      <td>69437.0</td>\n",
       "      <td>55905.0</td>\n",
       "      <td>47.335665</td>\n",
       "      <td>30</td>\n",
       "      <td>0.000137</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>146367.0</td>\n",
       "      <td>0.549632</td>\n",
       "      <td>30.192079</td>\n",
       "      <td>80448.0</td>\n",
       "      <td>65919.0</td>\n",
       "      <td>55.275799</td>\n",
       "      <td>35</td>\n",
       "      <td>0.000206</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>167318.0</td>\n",
       "      <td>0.545393</td>\n",
       "      <td>30.687667</td>\n",
       "      <td>91254.0</td>\n",
       "      <td>76064.0</td>\n",
       "      <td>63.187988</td>\n",
       "      <td>40</td>\n",
       "      <td>0.000183</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>188275.0</td>\n",
       "      <td>0.541968</td>\n",
       "      <td>31.225679</td>\n",
       "      <td>102039.0</td>\n",
       "      <td>86236.0</td>\n",
       "      <td>71.102442</td>\n",
       "      <td>45</td>\n",
       "      <td>0.000166</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   absolute_has_answer  accuracy  achieved_returns   correct   errors  \\\n",
       "0              20696.0  0.593158          4.859130   12276.0   8420.0   \n",
       "1              41537.0  0.579170          8.765475   24057.0  17480.0   \n",
       "2              62395.0  0.572209         11.851657   35703.0  26692.0   \n",
       "3              83407.0  0.564605         13.683131   47092.0  36315.0   \n",
       "4             104389.0  0.558814         16.520182   58334.0  46055.0   \n",
       "5             125342.0  0.553980         17.181956   69437.0  55905.0   \n",
       "6             146367.0  0.549632         30.192079   80448.0  65919.0   \n",
       "7             167318.0  0.545393         30.687667   91254.0  76064.0   \n",
       "8             188275.0  0.541968         31.225679  102039.0  86236.0   \n",
       "\n",
       "   percent_with_answer  thresholds  achieved_norm_returns  \n",
       "0             7.815887           5               0.000235  \n",
       "1            15.686534          10               0.000211  \n",
       "2            23.563600          15               0.000190  \n",
       "3            31.498826          20               0.000164  \n",
       "4            39.422721          25               0.000158  \n",
       "5            47.335665          30               0.000137  \n",
       "6            55.275799          35               0.000206  \n",
       "7            63.187988          40               0.000183  \n",
       "8            71.102442          45               0.000166  "
      ]
     },
     "execution_count": 25,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "valid_predictions.evaluate()['accuracy_at_cutoff']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Baseline on Testset: Accuracy at different levels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 26,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>absolute_has_answer</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>achieved_returns</th>\n",
       "      <th>correct</th>\n",
       "      <th>errors</th>\n",
       "      <th>percent_with_answer</th>\n",
       "      <th>thresholds</th>\n",
       "      <th>achieved_norm_returns</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>168159.0</td>\n",
       "      <td>0.567487</td>\n",
       "      <td>34.145246</td>\n",
       "      <td>95428.0</td>\n",
       "      <td>72731.0</td>\n",
       "      <td>7.900753</td>\n",
       "      <td>5</td>\n",
       "      <td>0.000203</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>338247.0</td>\n",
       "      <td>0.560859</td>\n",
       "      <td>55.956382</td>\n",
       "      <td>189709.0</td>\n",
       "      <td>148538.0</td>\n",
       "      <td>15.892138</td>\n",
       "      <td>10</td>\n",
       "      <td>0.000165</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>507052.0</td>\n",
       "      <td>0.555728</td>\n",
       "      <td>73.317496</td>\n",
       "      <td>281783.0</td>\n",
       "      <td>225269.0</td>\n",
       "      <td>23.823243</td>\n",
       "      <td>15</td>\n",
       "      <td>0.000145</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>676940.0</td>\n",
       "      <td>0.550663</td>\n",
       "      <td>87.090783</td>\n",
       "      <td>372766.0</td>\n",
       "      <td>304174.0</td>\n",
       "      <td>31.805231</td>\n",
       "      <td>20</td>\n",
       "      <td>0.000129</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>848942.0</td>\n",
       "      <td>0.545472</td>\n",
       "      <td>97.624355</td>\n",
       "      <td>463074.0</td>\n",
       "      <td>385868.0</td>\n",
       "      <td>39.886543</td>\n",
       "      <td>25</td>\n",
       "      <td>0.000115</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>1026028.0</td>\n",
       "      <td>0.540843</td>\n",
       "      <td>102.612630</td>\n",
       "      <td>554920.0</td>\n",
       "      <td>471108.0</td>\n",
       "      <td>48.206721</td>\n",
       "      <td>30</td>\n",
       "      <td>0.000100</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>1160196.0</td>\n",
       "      <td>0.536681</td>\n",
       "      <td>99.225102</td>\n",
       "      <td>622655.0</td>\n",
       "      <td>537541.0</td>\n",
       "      <td>54.510447</td>\n",
       "      <td>35</td>\n",
       "      <td>0.000086</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>1160196.0</td>\n",
       "      <td>0.536681</td>\n",
       "      <td>99.225102</td>\n",
       "      <td>622655.0</td>\n",
       "      <td>537541.0</td>\n",
       "      <td>54.510447</td>\n",
       "      <td>40</td>\n",
       "      <td>0.000086</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>1160196.0</td>\n",
       "      <td>0.536681</td>\n",
       "      <td>99.225102</td>\n",
       "      <td>622655.0</td>\n",
       "      <td>537541.0</td>\n",
       "      <td>54.510447</td>\n",
       "      <td>45</td>\n",
       "      <td>0.000086</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   absolute_has_answer  accuracy  achieved_returns   correct    errors  \\\n",
       "0             168159.0  0.567487         34.145246   95428.0   72731.0   \n",
       "1             338247.0  0.560859         55.956382  189709.0  148538.0   \n",
       "2             507052.0  0.555728         73.317496  281783.0  225269.0   \n",
       "3             676940.0  0.550663         87.090783  372766.0  304174.0   \n",
       "4             848942.0  0.545472         97.624355  463074.0  385868.0   \n",
       "5            1026028.0  0.540843        102.612630  554920.0  471108.0   \n",
       "6            1160196.0  0.536681         99.225102  622655.0  537541.0   \n",
       "7            1160196.0  0.536681         99.225102  622655.0  537541.0   \n",
       "8            1160196.0  0.536681         99.225102  622655.0  537541.0   \n",
       "\n",
       "   percent_with_answer  thresholds  achieved_norm_returns  \n",
       "0             7.900753           5               0.000203  \n",
       "1            15.892138          10               0.000165  \n",
       "2            23.823243          15               0.000145  \n",
       "3            31.805231          20               0.000129  \n",
       "4            39.886543          25               0.000115  \n",
       "5            48.206721          30               0.000100  \n",
       "6            54.510447          35               0.000086  \n",
       "7            54.510447          40               0.000086  \n",
       "8            54.510447          45               0.000086  "
      ]
     },
     "execution_count": 26,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# this is for the baseline\n",
    "pred_baseline(test_predictions)['accuracy_at_cutoff'] "
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Baseline on Testset: Comparison of Returns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 27,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'achieved': 99.22510172088424,\n",
       " 'always_down': -2.8184952097059814,\n",
       " 'always_up': 2.8184952097059814,\n",
       " 'num_trials': 1436174.0,\n",
       " 'omniscient': 1638.5998374293326,\n",
       " 'random': -6.32069528321988}"
      ]
     },
     "execution_count": 27,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# this is for the baseline\n",
    "pred_baseline(test_predictions)['strategies']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### ML Model on Testset: Accuracy at different levels"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 28,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/html": [
       "<div>\n",
       "<style scoped>\n",
       "    .dataframe tbody tr th:only-of-type {\n",
       "        vertical-align: middle;\n",
       "    }\n",
       "\n",
       "    .dataframe tbody tr th {\n",
       "        vertical-align: top;\n",
       "    }\n",
       "\n",
       "    .dataframe thead th {\n",
       "        text-align: right;\n",
       "    }\n",
       "</style>\n",
       "<table border=\"1\" class=\"dataframe\">\n",
       "  <thead>\n",
       "    <tr style=\"text-align: right;\">\n",
       "      <th></th>\n",
       "      <th>absolute_has_answer</th>\n",
       "      <th>accuracy</th>\n",
       "      <th>achieved_returns</th>\n",
       "      <th>correct</th>\n",
       "      <th>errors</th>\n",
       "      <th>percent_with_answer</th>\n",
       "      <th>thresholds</th>\n",
       "      <th>achieved_norm_returns</th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>0</th>\n",
       "      <td>164730.0</td>\n",
       "      <td>0.588035</td>\n",
       "      <td>72.068937</td>\n",
       "      <td>96867.0</td>\n",
       "      <td>67863.0</td>\n",
       "      <td>7.739646</td>\n",
       "      <td>5</td>\n",
       "      <td>0.000437</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1</th>\n",
       "      <td>330183.0</td>\n",
       "      <td>0.575554</td>\n",
       "      <td>115.657145</td>\n",
       "      <td>190038.0</td>\n",
       "      <td>140145.0</td>\n",
       "      <td>15.513261</td>\n",
       "      <td>10</td>\n",
       "      <td>0.000350</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2</th>\n",
       "      <td>496976.0</td>\n",
       "      <td>0.566679</td>\n",
       "      <td>148.828014</td>\n",
       "      <td>281626.0</td>\n",
       "      <td>215350.0</td>\n",
       "      <td>23.349834</td>\n",
       "      <td>15</td>\n",
       "      <td>0.000299</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3</th>\n",
       "      <td>664073.0</td>\n",
       "      <td>0.560366</td>\n",
       "      <td>173.802069</td>\n",
       "      <td>372124.0</td>\n",
       "      <td>291949.0</td>\n",
       "      <td>31.200690</td>\n",
       "      <td>20</td>\n",
       "      <td>0.000262</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>4</th>\n",
       "      <td>831629.0</td>\n",
       "      <td>0.555227</td>\n",
       "      <td>193.864521</td>\n",
       "      <td>461743.0</td>\n",
       "      <td>369886.0</td>\n",
       "      <td>39.073112</td>\n",
       "      <td>25</td>\n",
       "      <td>0.000233</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>5</th>\n",
       "      <td>999001.0</td>\n",
       "      <td>0.550809</td>\n",
       "      <td>211.609139</td>\n",
       "      <td>550259.0</td>\n",
       "      <td>448742.0</td>\n",
       "      <td>46.936889</td>\n",
       "      <td>30</td>\n",
       "      <td>0.000212</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>6</th>\n",
       "      <td>1165834.0</td>\n",
       "      <td>0.546665</td>\n",
       "      <td>223.401865</td>\n",
       "      <td>637321.0</td>\n",
       "      <td>528513.0</td>\n",
       "      <td>54.775342</td>\n",
       "      <td>35</td>\n",
       "      <td>0.000192</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>7</th>\n",
       "      <td>1331767.0</td>\n",
       "      <td>0.542870</td>\n",
       "      <td>233.593762</td>\n",
       "      <td>722976.0</td>\n",
       "      <td>608791.0</td>\n",
       "      <td>62.571509</td>\n",
       "      <td>40</td>\n",
       "      <td>0.000175</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>8</th>\n",
       "      <td>1497333.0</td>\n",
       "      <td>0.539412</td>\n",
       "      <td>239.141232</td>\n",
       "      <td>807680.0</td>\n",
       "      <td>689653.0</td>\n",
       "      <td>70.350434</td>\n",
       "      <td>45</td>\n",
       "      <td>0.000160</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "   absolute_has_answer  accuracy  achieved_returns   correct    errors  \\\n",
       "0             164730.0  0.588035         72.068937   96867.0   67863.0   \n",
       "1             330183.0  0.575554        115.657145  190038.0  140145.0   \n",
       "2             496976.0  0.566679        148.828014  281626.0  215350.0   \n",
       "3             664073.0  0.560366        173.802069  372124.0  291949.0   \n",
       "4             831629.0  0.555227        193.864521  461743.0  369886.0   \n",
       "5             999001.0  0.550809        211.609139  550259.0  448742.0   \n",
       "6            1165834.0  0.546665        223.401865  637321.0  528513.0   \n",
       "7            1331767.0  0.542870        233.593762  722976.0  608791.0   \n",
       "8            1497333.0  0.539412        239.141232  807680.0  689653.0   \n",
       "\n",
       "   percent_with_answer  thresholds  achieved_norm_returns  \n",
       "0             7.739646           5               0.000437  \n",
       "1            15.513261          10               0.000350  \n",
       "2            23.349834          15               0.000299  \n",
       "3            31.200690          20               0.000262  \n",
       "4            39.073112          25               0.000233  \n",
       "5            46.936889          30               0.000212  \n",
       "6            54.775342          35               0.000192  \n",
       "7            62.571509          40               0.000175  \n",
       "8            70.350434          45               0.000160  "
      ]
     },
     "execution_count": 28,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# this is for ML\n",
    "test_predictions.evaluate()['accuracy_at_cutoff']"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### ML Model on Testset: Comparison of Returns"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 29,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "text/plain": [
       "{'achieved': 238.0119688885496,\n",
       " 'always_down': -2.8184952097059814,\n",
       " 'always_up': 2.8184952097059814,\n",
       " 'num_trials': 2128392.0,\n",
       " 'omniscient': 1638.5998374293326,\n",
       " 'random': 1.3535557596882493}"
      ]
     },
     "execution_count": 29,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "# for ML\n",
    "Predictions(test_predictions.predictions, test_predictions.training_set).evaluate()['strategies']"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 31,
   "metadata": {},
   "outputs": [
    {
     "name": "stdout",
     "output_type": "stream",
     "text": [
      "So if you play 2128392 times on the Test set with 1 EUR and you always guess the movement,\n",
      "ignoring all transactions cost, you will make 1638.59983743. \n",
      "Instead you make 238.011968889 or 14.5253260407 percent of the ideally achievable.\n",
      "If you use the baseline you will make 99.2251017209 or 6.05548099386 percent of ideal\n"
     ]
    }
   ],
   "source": [
    "readable_summary('Test set', test_predictions)"
   ]
  },
  {
   "cell_type": "markdown",
   "metadata": {},
   "source": [
    "### Analysis by Mnemonic and by Date\n",
    "\n",
    "It is known that when it comes to stock predictions, different stocks and different days will exibit different performance"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 153,
   "metadata": {},
   "outputs": [],
   "source": [
    "def plot_accuracy_by(grouping_feature, predictions):\n",
    "    df = predictions.training_set.original_df\n",
    "    s = df[['Date', 'Mnemonic', 'pseudo_y(SignReturn)']].copy()\n",
    "    s['Predictions'] = predictions.predictions\n",
    "    s['Baseline'] = df['x((MaxP-EndP)-(EndP-MinP))(t - 1)']\n",
    "\n",
    "    def agg(group):\n",
    "        pred = group['Predictions']\n",
    "        baseline = group['Baseline']\n",
    "        rets = group['pseudo_y(SignReturn)']\n",
    "        c = pred.corr(rets)\n",
    "        c = np.where(np.sign(pred)*np.sign(rets) == 1.0, 1.0, 0.0).sum()\n",
    "        e = np.where(np.sign(pred)*np.sign(rets) == -1.0, 1.0, 0.0).sum()\n",
    "        acc = c/(c + e)\n",
    "\n",
    "        c_baseline = np.where(np.sign(baseline)*np.sign(rets) == 1.0, 1.0, 0.0).sum()\n",
    "        e_baseline = np.where(np.sign(baseline)*np.sign(rets) == -1.0, 1.0, 0.0).sum()\n",
    "        acc_baseline = c_baseline/(c_baseline + e_baseline)\n",
    "\n",
    "        l = group.shape[0]\n",
    "        return {\"corr\": c, 'size': l, 'accuracy': acc, 'acc_baseline': acc_baseline}\n",
    "    f = s.groupby(grouping_feature).apply(agg).to_frame(\"agg\")\n",
    "\n",
    "    f['AccuracyPred'] = f['agg'].map(lambda i: i['accuracy'])\n",
    "    f['AccuracyBaseline'] = f['agg'].map(lambda i: i['acc_baseline'])\n",
    "    f['AccPred - AccBaseline'] = f['AccuracyPred'] - f['AccuracyBaseline']\n",
    "    f = f.drop(columns=['agg'])\n",
    "\n",
    "    f = f[f.index != '2017-10-14'] # remove this date which has one data point\n",
    "    f[['AccuracyPred', 'AccuracyBaseline']].plot()\n",
    "    return f"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 154,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fce244f82d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "perf = plot_accuracy_by('Mnemonic', test_predictions)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 155,
   "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>AccuracyPred</th>\n",
       "      <th>AccuracyBaseline</th>\n",
       "      <th>AccPred - AccBaseline</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mnemonic</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>ARO</th>\n",
       "      <td>0.745455</td>\n",
       "      <td>0.517241</td>\n",
       "      <td>0.228213</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>AT1</th>\n",
       "      <td>0.652975</td>\n",
       "      <td>0.678565</td>\n",
       "      <td>-0.025589</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>GAZ</th>\n",
       "      <td>0.642234</td>\n",
       "      <td>0.641590</td>\n",
       "      <td>0.000644</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>AB1</th>\n",
       "      <td>0.635802</td>\n",
       "      <td>0.547945</td>\n",
       "      <td>0.087857</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SVAB</th>\n",
       "      <td>0.623996</td>\n",
       "      <td>0.581560</td>\n",
       "      <td>0.042436</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>3W9K</th>\n",
       "      <td>0.620773</td>\n",
       "      <td>0.579365</td>\n",
       "      <td>0.041408</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>USE</th>\n",
       "      <td>0.619565</td>\n",
       "      <td>0.598214</td>\n",
       "      <td>0.021351</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>MDG1</th>\n",
       "      <td>0.609740</td>\n",
       "      <td>0.580709</td>\n",
       "      <td>0.029031</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>QSC</th>\n",
       "      <td>0.595390</td>\n",
       "      <td>0.569677</td>\n",
       "      <td>0.025713</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TINA</th>\n",
       "      <td>0.595362</td>\n",
       "      <td>0.586592</td>\n",
       "      <td>0.008770</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CAP</th>\n",
       "      <td>0.592759</td>\n",
       "      <td>0.599061</td>\n",
       "      <td>-0.006302</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SNH</th>\n",
       "      <td>0.588323</td>\n",
       "      <td>0.586374</td>\n",
       "      <td>0.001949</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>RIB</th>\n",
       "      <td>0.584391</td>\n",
       "      <td>0.589032</td>\n",
       "      <td>-0.004641</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BVB</th>\n",
       "      <td>0.583552</td>\n",
       "      <td>0.595956</td>\n",
       "      <td>-0.012403</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SHL</th>\n",
       "      <td>0.581003</td>\n",
       "      <td>0.585011</td>\n",
       "      <td>-0.004008</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>ADV</th>\n",
       "      <td>0.572709</td>\n",
       "      <td>0.558814</td>\n",
       "      <td>0.013895</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>AOX</th>\n",
       "      <td>0.566464</td>\n",
       "      <td>0.594437</td>\n",
       "      <td>-0.027973</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>EVT</th>\n",
       "      <td>0.561056</td>\n",
       "      <td>0.561461</td>\n",
       "      <td>-0.000405</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PNE3</th>\n",
       "      <td>0.557648</td>\n",
       "      <td>0.564229</td>\n",
       "      <td>-0.006582</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>UN01</th>\n",
       "      <td>0.557487</td>\n",
       "      <td>0.552025</td>\n",
       "      <td>0.005463</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          AccuracyPred  AccuracyBaseline  AccPred - AccBaseline\n",
       "Mnemonic                                                       \n",
       "ARO           0.745455          0.517241               0.228213\n",
       "AT1           0.652975          0.678565              -0.025589\n",
       "GAZ           0.642234          0.641590               0.000644\n",
       "AB1           0.635802          0.547945               0.087857\n",
       "SVAB          0.623996          0.581560               0.042436\n",
       "3W9K          0.620773          0.579365               0.041408\n",
       "USE           0.619565          0.598214               0.021351\n",
       "MDG1          0.609740          0.580709               0.029031\n",
       "QSC           0.595390          0.569677               0.025713\n",
       "TINA          0.595362          0.586592               0.008770\n",
       "CAP           0.592759          0.599061              -0.006302\n",
       "SNH           0.588323          0.586374               0.001949\n",
       "RIB           0.584391          0.589032              -0.004641\n",
       "BVB           0.583552          0.595956              -0.012403\n",
       "SHL           0.581003          0.585011              -0.004008\n",
       "ADV           0.572709          0.558814               0.013895\n",
       "AOX           0.566464          0.594437              -0.027973\n",
       "EVT           0.561056          0.561461              -0.000405\n",
       "PNE3          0.557648          0.564229              -0.006582\n",
       "UN01          0.557487          0.552025               0.005463"
      ]
     },
     "execution_count": 155,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "perf.sort_values(by='AccuracyPred', ascending=False).head(20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 156,
   "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>AccuracyPred</th>\n",
       "      <th>AccuracyBaseline</th>\n",
       "      <th>AccPred - AccBaseline</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Mnemonic</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>DB1</th>\n",
       "      <td>0.523620</td>\n",
       "      <td>0.536047</td>\n",
       "      <td>-0.012426</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>PAH3</th>\n",
       "      <td>0.523028</td>\n",
       "      <td>0.525957</td>\n",
       "      <td>-0.002929</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BMW</th>\n",
       "      <td>0.522967</td>\n",
       "      <td>0.521218</td>\n",
       "      <td>0.001749</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>QIA</th>\n",
       "      <td>0.522816</td>\n",
       "      <td>0.536484</td>\n",
       "      <td>-0.013669</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LIN</th>\n",
       "      <td>0.522549</td>\n",
       "      <td>0.526406</td>\n",
       "      <td>-0.003857</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TKA</th>\n",
       "      <td>0.521335</td>\n",
       "      <td>0.533668</td>\n",
       "      <td>-0.012333</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SAP</th>\n",
       "      <td>0.519582</td>\n",
       "      <td>0.519968</td>\n",
       "      <td>-0.000386</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>SZG</th>\n",
       "      <td>0.518228</td>\n",
       "      <td>0.515717</td>\n",
       "      <td>0.002511</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>LHA</th>\n",
       "      <td>0.517990</td>\n",
       "      <td>0.523647</td>\n",
       "      <td>-0.005657</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>HEI</th>\n",
       "      <td>0.517751</td>\n",
       "      <td>0.518109</td>\n",
       "      <td>-0.000357</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>CBK</th>\n",
       "      <td>0.517694</td>\n",
       "      <td>0.518475</td>\n",
       "      <td>-0.000781</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>FME</th>\n",
       "      <td>0.517620</td>\n",
       "      <td>0.519113</td>\n",
       "      <td>-0.001494</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BAS</th>\n",
       "      <td>0.517197</td>\n",
       "      <td>0.515772</td>\n",
       "      <td>0.001425</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>BAYN</th>\n",
       "      <td>0.517119</td>\n",
       "      <td>0.520142</td>\n",
       "      <td>-0.003023</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>TUI1</th>\n",
       "      <td>0.516325</td>\n",
       "      <td>0.529907</td>\n",
       "      <td>-0.013582</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>DBK</th>\n",
       "      <td>0.516079</td>\n",
       "      <td>0.519342</td>\n",
       "      <td>-0.003263</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>1COV</th>\n",
       "      <td>0.515834</td>\n",
       "      <td>0.521757</td>\n",
       "      <td>-0.005923</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VOW3</th>\n",
       "      <td>0.515620</td>\n",
       "      <td>0.517962</td>\n",
       "      <td>-0.002342</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>EOAN</th>\n",
       "      <td>0.513831</td>\n",
       "      <td>0.519081</td>\n",
       "      <td>-0.005250</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>VNA</th>\n",
       "      <td>0.513023</td>\n",
       "      <td>0.518408</td>\n",
       "      <td>-0.005385</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "          AccuracyPred  AccuracyBaseline  AccPred - AccBaseline\n",
       "Mnemonic                                                       \n",
       "DB1           0.523620          0.536047              -0.012426\n",
       "PAH3          0.523028          0.525957              -0.002929\n",
       "BMW           0.522967          0.521218               0.001749\n",
       "QIA           0.522816          0.536484              -0.013669\n",
       "LIN           0.522549          0.526406              -0.003857\n",
       "TKA           0.521335          0.533668              -0.012333\n",
       "SAP           0.519582          0.519968              -0.000386\n",
       "SZG           0.518228          0.515717               0.002511\n",
       "LHA           0.517990          0.523647              -0.005657\n",
       "HEI           0.517751          0.518109              -0.000357\n",
       "CBK           0.517694          0.518475              -0.000781\n",
       "FME           0.517620          0.519113              -0.001494\n",
       "BAS           0.517197          0.515772               0.001425\n",
       "BAYN          0.517119          0.520142              -0.003023\n",
       "TUI1          0.516325          0.529907              -0.013582\n",
       "DBK           0.516079          0.519342              -0.003263\n",
       "1COV          0.515834          0.521757              -0.005923\n",
       "VOW3          0.515620          0.517962              -0.002342\n",
       "EOAN          0.513831          0.519081              -0.005250\n",
       "VNA           0.513023          0.518408              -0.005385"
      ]
     },
     "execution_count": 156,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "perf.sort_values(by='AccuracyPred', ascending=False).tail(20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 158,
   "metadata": {},
   "outputs": [
    {
     "data": {
      "image/png": "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\n",
      "text/plain": [
       "<matplotlib.figure.Figure at 0x7fce246488d0>"
      ]
     },
     "metadata": {},
     "output_type": "display_data"
    }
   ],
   "source": [
    "perf = plot_accuracy_by('Date', test_predictions)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 159,
   "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>AccuracyPred</th>\n",
       "      <th>AccuracyBaseline</th>\n",
       "      <th>AccPred - AccBaseline</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Date</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>2018-01-26</th>\n",
       "      <td>0.552227</td>\n",
       "      <td>0.544215</td>\n",
       "      <td>0.008011</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-01-22</th>\n",
       "      <td>0.551807</td>\n",
       "      <td>0.547880</td>\n",
       "      <td>0.003927</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-01-23</th>\n",
       "      <td>0.551116</td>\n",
       "      <td>0.555849</td>\n",
       "      <td>-0.004733</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-01-15</th>\n",
       "      <td>0.550708</td>\n",
       "      <td>0.551314</td>\n",
       "      <td>-0.000606</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-01-18</th>\n",
       "      <td>0.549494</td>\n",
       "      <td>0.549597</td>\n",
       "      <td>-0.000102</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-01-12</th>\n",
       "      <td>0.547754</td>\n",
       "      <td>0.547022</td>\n",
       "      <td>0.000732</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-04-10</th>\n",
       "      <td>0.547308</td>\n",
       "      <td>0.540861</td>\n",
       "      <td>0.006448</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-23</th>\n",
       "      <td>0.546640</td>\n",
       "      <td>0.546975</td>\n",
       "      <td>-0.000335</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-05</th>\n",
       "      <td>0.543933</td>\n",
       "      <td>0.543524</td>\n",
       "      <td>0.000408</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-07</th>\n",
       "      <td>0.543679</td>\n",
       "      <td>0.544006</td>\n",
       "      <td>-0.000327</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-01</th>\n",
       "      <td>0.543029</td>\n",
       "      <td>0.546152</td>\n",
       "      <td>-0.003123</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-04-12</th>\n",
       "      <td>0.543026</td>\n",
       "      <td>0.542455</td>\n",
       "      <td>0.000571</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-04-13</th>\n",
       "      <td>0.542929</td>\n",
       "      <td>0.545968</td>\n",
       "      <td>-0.003039</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-13</th>\n",
       "      <td>0.542851</td>\n",
       "      <td>0.543878</td>\n",
       "      <td>-0.001027</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-01-30</th>\n",
       "      <td>0.542710</td>\n",
       "      <td>0.542083</td>\n",
       "      <td>0.000627</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-06</th>\n",
       "      <td>0.542633</td>\n",
       "      <td>0.542272</td>\n",
       "      <td>0.000361</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-14</th>\n",
       "      <td>0.541659</td>\n",
       "      <td>0.539073</td>\n",
       "      <td>0.002586</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-01-17</th>\n",
       "      <td>0.541635</td>\n",
       "      <td>0.545702</td>\n",
       "      <td>-0.004067</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-04-23</th>\n",
       "      <td>0.541543</td>\n",
       "      <td>0.542864</td>\n",
       "      <td>-0.001321</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-01-31</th>\n",
       "      <td>0.541287</td>\n",
       "      <td>0.540160</td>\n",
       "      <td>0.001127</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            AccuracyPred  AccuracyBaseline  AccPred - AccBaseline\n",
       "Date                                                             \n",
       "2018-01-26      0.552227          0.544215               0.008011\n",
       "2018-01-22      0.551807          0.547880               0.003927\n",
       "2018-01-23      0.551116          0.555849              -0.004733\n",
       "2018-01-15      0.550708          0.551314              -0.000606\n",
       "2018-01-18      0.549494          0.549597              -0.000102\n",
       "2018-01-12      0.547754          0.547022               0.000732\n",
       "2018-04-10      0.547308          0.540861               0.006448\n",
       "2018-02-23      0.546640          0.546975              -0.000335\n",
       "2018-02-05      0.543933          0.543524               0.000408\n",
       "2018-02-07      0.543679          0.544006              -0.000327\n",
       "2018-02-01      0.543029          0.546152              -0.003123\n",
       "2018-04-12      0.543026          0.542455               0.000571\n",
       "2018-04-13      0.542929          0.545968              -0.003039\n",
       "2018-02-13      0.542851          0.543878              -0.001027\n",
       "2018-01-30      0.542710          0.542083               0.000627\n",
       "2018-02-06      0.542633          0.542272               0.000361\n",
       "2018-02-14      0.541659          0.539073               0.002586\n",
       "2018-01-17      0.541635          0.545702              -0.004067\n",
       "2018-04-23      0.541543          0.542864              -0.001321\n",
       "2018-01-31      0.541287          0.540160               0.001127"
      ]
     },
     "execution_count": 159,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "perf.sort_values(by='AccuracyPred', ascending=False).head(20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": 160,
   "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>AccuracyPred</th>\n",
       "      <th>AccuracyBaseline</th>\n",
       "      <th>AccPred - AccBaseline</th>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>Date</th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "      <th></th>\n",
       "    </tr>\n",
       "  </thead>\n",
       "  <tbody>\n",
       "    <tr>\n",
       "      <th>2018-04-19</th>\n",
       "      <td>0.531091</td>\n",
       "      <td>0.530912</td>\n",
       "      <td>0.000179</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-04-26</th>\n",
       "      <td>0.530988</td>\n",
       "      <td>0.529704</td>\n",
       "      <td>0.001284</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-26</th>\n",
       "      <td>0.529216</td>\n",
       "      <td>0.527383</td>\n",
       "      <td>0.001833</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-03-12</th>\n",
       "      <td>0.529037</td>\n",
       "      <td>0.526849</td>\n",
       "      <td>0.002188</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-27</th>\n",
       "      <td>0.528221</td>\n",
       "      <td>0.528449</td>\n",
       "      <td>-0.000228</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-03-23</th>\n",
       "      <td>0.527086</td>\n",
       "      <td>0.533920</td>\n",
       "      <td>-0.006833</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-03-13</th>\n",
       "      <td>0.526939</td>\n",
       "      <td>0.535091</td>\n",
       "      <td>-0.008153</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-04-24</th>\n",
       "      <td>0.526151</td>\n",
       "      <td>0.516694</td>\n",
       "      <td>0.009457</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-03-09</th>\n",
       "      <td>0.525953</td>\n",
       "      <td>0.525134</td>\n",
       "      <td>0.000818</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-03-26</th>\n",
       "      <td>0.525815</td>\n",
       "      <td>0.524914</td>\n",
       "      <td>0.000901</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-04-27</th>\n",
       "      <td>0.524572</td>\n",
       "      <td>0.521916</td>\n",
       "      <td>0.002656</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-03-22</th>\n",
       "      <td>0.523474</td>\n",
       "      <td>0.526847</td>\n",
       "      <td>-0.003373</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-22</th>\n",
       "      <td>0.522100</td>\n",
       "      <td>0.530654</td>\n",
       "      <td>-0.008554</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-04-20</th>\n",
       "      <td>0.522072</td>\n",
       "      <td>0.521685</td>\n",
       "      <td>0.000386</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-03-14</th>\n",
       "      <td>0.521883</td>\n",
       "      <td>0.518819</td>\n",
       "      <td>0.003064</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-03-07</th>\n",
       "      <td>0.520974</td>\n",
       "      <td>0.515606</td>\n",
       "      <td>0.005369</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-09</th>\n",
       "      <td>0.520462</td>\n",
       "      <td>0.520975</td>\n",
       "      <td>-0.000513</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-02-08</th>\n",
       "      <td>0.518119</td>\n",
       "      <td>0.522615</td>\n",
       "      <td>-0.004495</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-03-01</th>\n",
       "      <td>0.518090</td>\n",
       "      <td>0.527935</td>\n",
       "      <td>-0.009845</td>\n",
       "    </tr>\n",
       "    <tr>\n",
       "      <th>2018-04-25</th>\n",
       "      <td>0.517547</td>\n",
       "      <td>0.526898</td>\n",
       "      <td>-0.009351</td>\n",
       "    </tr>\n",
       "  </tbody>\n",
       "</table>\n",
       "</div>"
      ],
      "text/plain": [
       "            AccuracyPred  AccuracyBaseline  AccPred - AccBaseline\n",
       "Date                                                             \n",
       "2018-04-19      0.531091          0.530912               0.000179\n",
       "2018-04-26      0.530988          0.529704               0.001284\n",
       "2018-02-26      0.529216          0.527383               0.001833\n",
       "2018-03-12      0.529037          0.526849               0.002188\n",
       "2018-02-27      0.528221          0.528449              -0.000228\n",
       "2018-03-23      0.527086          0.533920              -0.006833\n",
       "2018-03-13      0.526939          0.535091              -0.008153\n",
       "2018-04-24      0.526151          0.516694               0.009457\n",
       "2018-03-09      0.525953          0.525134               0.000818\n",
       "2018-03-26      0.525815          0.524914               0.000901\n",
       "2018-04-27      0.524572          0.521916               0.002656\n",
       "2018-03-22      0.523474          0.526847              -0.003373\n",
       "2018-02-22      0.522100          0.530654              -0.008554\n",
       "2018-04-20      0.522072          0.521685               0.000386\n",
       "2018-03-14      0.521883          0.518819               0.003064\n",
       "2018-03-07      0.520974          0.515606               0.005369\n",
       "2018-02-09      0.520462          0.520975              -0.000513\n",
       "2018-02-08      0.518119          0.522615              -0.004495\n",
       "2018-03-01      0.518090          0.527935              -0.009845\n",
       "2018-04-25      0.517547          0.526898              -0.009351"
      ]
     },
     "execution_count": 160,
     "metadata": {},
     "output_type": "execute_result"
    }
   ],
   "source": [
    "perf.sort_values(by='AccuracyPred', ascending=False).tail(20)"
   ]
  },
  {
   "cell_type": "code",
   "execution_count": null,
   "metadata": {},
   "outputs": [],
   "source": []
  }
 ],
 "metadata": {
  "kernelspec": {
   "display_name": "Python 3",
   "language": "python",
   "name": "python3"
  },
  "language_info": {
   "codemirror_mode": {
    "name": "ipython",
    "version": 3
   },
   "file_extension": ".py",
   "mimetype": "text/x-python",
   "name": "python",
   "nbconvert_exporter": "python",
   "pygments_lexer": "ipython3",
   "version": "3.6.9"
  }
 },
 "nbformat": 4,
 "nbformat_minor": 2
}