algo_trading_attention.ipynb

{
  "nbformat": 4,
  "nbformat_minor": 0,
  "metadata": {
    "colab": {
      "name": "algo trading attention.ipynb",
      "version": "0.3.2",
      "provenance": [],
      "include_colab_link": true
    },
    "kernelspec": {
      "name": "python3",
      "display_name": "Python 3"
    },
    "accelerator": "GPU"
  },
  "cells": [
    {
      "cell_type": "markdown",
      "metadata": {
        "id": "view-in-github",
        "colab_type": "text"
      },
      "source": [
        "<a href=\"https://colab.research.google.com/github/ShreshthSaxena/Algorithmic-Trading/blob/master/algo_trading_attention.ipynb\" target=\"_parent\"><img src=\"https://colab.research.google.com/assets/colab-badge.svg\" alt=\"Open In Colab\"/></a>"
      ]
    },
    {
      "metadata": {
        "id": "G-EcFexMRlvq",
        "colab_type": "code",
        "outputId": "f2f76243-8cc2-4a5e-e750-fcec42cf5ecf",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "cell_type": "code",
      "source": [
        "\n",
        "import pandas as pd\n",
        "import numpy as np\n",
        "import datetime\n",
        "from sklearn.metrics import mean_squared_error\n",
        "from math import sqrt\n",
        "from keras.models import Sequential\n",
        "from keras.layers.core import Dense, Dropout, Activation\n",
        "from keras.layers.recurrent import LSTM\n",
        "from keras.models import load_model\n",
        "import keras\n",
        "from sklearn import preprocessing\n",
        "import matplotlib.pyplot as plt\n",
        "import math, time\n",
        "import tensorflow as tf\n"
      ],
      "execution_count": 2,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Using TensorFlow backend.\n"
          ],
          "name": "stderr"
        }
      ]
    },
    {
      "metadata": {
        "id": "iqttlAhVRrJ-",
        "colab_type": "code",
        "outputId": "c8762597-f91f-4395-9def-27892c161e52",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 235
        }
      },
      "cell_type": "code",
      "source": [
        "from pandas_datareader import data as pdr\n",
        "import fix_yahoo_finance\n",
        "\n",
        "def get(tickers, startdate, enddate):\n",
        "  def data(ticker):\n",
        "    return (pdr.get_data_yahoo(ticker, start=startdate, end=enddate))\n",
        "  datas = map (data, tickers)\n",
        "  return(pd.concat(datas, keys=tickers, names=['Ticker', 'Date']))\n",
        "\n",
        "tickers = ['AAPL', 'MSFT', 'IBM', 'GOOG']\n",
        "all_data = get(tickers, datetime.datetime(2012, 10, 1), datetime.datetime(2018, 1, 1))\n",
        "\n",
        "all_data.head()"
      ],
      "execution_count": 3,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/html": [
              "<div>\n",
              "<style scoped>\n",
              "    .dataframe tbody tr th:only-of-type {\n",
              "        vertical-align: middle;\n",
              "    }\n",
              "\n",
              "    .dataframe tbody tr th {\n",
              "        vertical-align: top;\n",
              "    }\n",
              "\n",
              "    .dataframe thead th {\n",
              "        text-align: right;\n",
              "    }\n",
              "</style>\n",
              "<table border=\"1\" class=\"dataframe\">\n",
              "  <thead>\n",
              "    <tr style=\"text-align: right;\">\n",
              "      <th></th>\n",
              "      <th></th>\n",
              "      <th>High</th>\n",
              "      <th>Low</th>\n",
              "      <th>Open</th>\n",
              "      <th>Close</th>\n",
              "      <th>Volume</th>\n",
              "      <th>Adj Close</th>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Ticker</th>\n",
              "      <th>Date</th>\n",
              "      <th></th>\n",
              "      <th></th>\n",
              "      <th></th>\n",
              "      <th></th>\n",
              "      <th></th>\n",
              "      <th></th>\n",
              "    </tr>\n",
              "  </thead>\n",
              "  <tbody>\n",
              "    <tr>\n",
              "      <th rowspan=\"5\" valign=\"top\">AAPL</th>\n",
              "      <th>2012-10-01</th>\n",
              "      <td>96.678574</td>\n",
              "      <td>93.785713</td>\n",
              "      <td>95.879997</td>\n",
              "      <td>94.198570</td>\n",
              "      <td>135898700.0</td>\n",
              "      <td>64.749588</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2012-10-02</th>\n",
              "      <td>95.192856</td>\n",
              "      <td>92.949997</td>\n",
              "      <td>94.544289</td>\n",
              "      <td>94.472855</td>\n",
              "      <td>156998100.0</td>\n",
              "      <td>64.938141</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2012-10-03</th>\n",
              "      <td>95.980003</td>\n",
              "      <td>94.661430</td>\n",
              "      <td>94.980003</td>\n",
              "      <td>95.921425</td>\n",
              "      <td>106070300.0</td>\n",
              "      <td>65.933838</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2012-10-04</th>\n",
              "      <td>96.321426</td>\n",
              "      <td>95.078575</td>\n",
              "      <td>95.892860</td>\n",
              "      <td>95.257141</td>\n",
              "      <td>92681400.0</td>\n",
              "      <td>65.477219</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2012-10-05</th>\n",
              "      <td>95.142860</td>\n",
              "      <td>93.040001</td>\n",
              "      <td>95.028572</td>\n",
              "      <td>93.227142</td>\n",
              "      <td>148501500.0</td>\n",
              "      <td>64.081856</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                        High        Low       Open      Close       Volume  \\\n",
              "Ticker Date                                                                  \n",
              "AAPL   2012-10-01  96.678574  93.785713  95.879997  94.198570  135898700.0   \n",
              "       2012-10-02  95.192856  92.949997  94.544289  94.472855  156998100.0   \n",
              "       2012-10-03  95.980003  94.661430  94.980003  95.921425  106070300.0   \n",
              "       2012-10-04  96.321426  95.078575  95.892860  95.257141   92681400.0   \n",
              "       2012-10-05  95.142860  93.040001  95.028572  93.227142  148501500.0   \n",
              "\n",
              "                   Adj Close  \n",
              "Ticker Date                   \n",
              "AAPL   2012-10-01  64.749588  \n",
              "       2012-10-02  64.938141  \n",
              "       2012-10-03  65.933838  \n",
              "       2012-10-04  65.477219  \n",
              "       2012-10-05  64.081856  "
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 3
        }
      ]
    },
    {
      "metadata": {
        "id": "Bqzik6JqRt4S",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "aapl = pdr.get_data_yahoo('AAPL', \n",
        "                          start=datetime.datetime(2000, 10, 1), \n",
        "                          end=datetime.datetime(2012, 1, 1))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "TitXaEEcRxJL",
        "colab_type": "code",
        "outputId": "9786f726-7a8c-4573-eb9a-3b2ef9092875",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 252
        }
      },
      "cell_type": "code",
      "source": [
        "df= aapl.drop(['Close'],1)\n",
        "print(df.shape)\n",
        "df.head()"
      ],
      "execution_count": 5,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "(2830, 5)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "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>High</th>\n",
              "      <th>Low</th>\n",
              "      <th>Open</th>\n",
              "      <th>Volume</th>\n",
              "      <th>Adj Close</th>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Date</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>2000-10-02</th>\n",
              "      <td>1.910714</td>\n",
              "      <td>1.678571</td>\n",
              "      <td>1.906250</td>\n",
              "      <td>606197200.0</td>\n",
              "      <td>1.154998</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2000-10-03</th>\n",
              "      <td>1.785714</td>\n",
              "      <td>1.584821</td>\n",
              "      <td>1.781250</td>\n",
              "      <td>509530000.0</td>\n",
              "      <td>1.062717</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2000-10-04</th>\n",
              "      <td>1.696429</td>\n",
              "      <td>1.562500</td>\n",
              "      <td>1.598214</td>\n",
              "      <td>366506000.0</td>\n",
              "      <td>1.125230</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2000-10-05</th>\n",
              "      <td>1.750000</td>\n",
              "      <td>1.571429</td>\n",
              "      <td>1.678571</td>\n",
              "      <td>218251600.0</td>\n",
              "      <td>1.050810</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2000-10-06</th>\n",
              "      <td>1.638393</td>\n",
              "      <td>1.500000</td>\n",
              "      <td>1.620536</td>\n",
              "      <td>153164200.0</td>\n",
              "      <td>1.056764</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                High       Low      Open       Volume  Adj Close\n",
              "Date                                                            \n",
              "2000-10-02  1.910714  1.678571  1.906250  606197200.0   1.154998\n",
              "2000-10-03  1.785714  1.584821  1.781250  509530000.0   1.062717\n",
              "2000-10-04  1.696429  1.562500  1.598214  366506000.0   1.125230\n",
              "2000-10-05  1.750000  1.571429  1.678571  218251600.0   1.050810\n",
              "2000-10-06  1.638393  1.500000  1.620536  153164200.0   1.056764"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 5
        }
      ]
    },
    {
      "metadata": {
        "id": "XcHqz2bjR8yE",
        "colab_type": "code",
        "outputId": "0e4a6c04-7515-41a5-e7b0-2794477480ba",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 235
        }
      },
      "cell_type": "code",
      "source": [
        "from sklearn import preprocessing\n",
        "scaler = preprocessing.MinMaxScaler()\n",
        "\n",
        "def normalize_data(df):\n",
        "  smoothing_window_size = 500\n",
        "  for di in range(0,2830,smoothing_window_size):\n",
        "      if di +  smoothing_window_size > df.shape[0]:\n",
        "        dil  = df.shape[0]\n",
        "      else:\n",
        "        dil = di+smoothing_window_size\n",
        "      df['Open'][di:dil] = np.squeeze(scaler.fit_transform(df.Open.values.reshape(-1,1)[di:dil]))\n",
        "      df['High'][di:dil] = np.squeeze(scaler.fit_transform(df.High.values.reshape(-1,1)[di:dil]))\n",
        "      df['Low'][di:dil] = np.squeeze(scaler.fit_transform(df.Low.values.reshape(-1,1)[di:dil]))\n",
        "      df['Volume'][di:dil] = np.squeeze(scaler.fit_transform(df.Volume.values.reshape(-1,1)[di:dil]))\n",
        "      df['Adj Close'][di:dil] = np.squeeze(scaler.fit_transform(df['Adj Close'].values.reshape(-1,1)[di:dil]))\n",
        "  return df\n",
        "    \n",
        "dfw = normalize_data(df)\n",
        "dfw.tail()\n"
      ],
      "execution_count": 6,
      "outputs": [
        {
          "output_type": "execute_result",
          "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>High</th>\n",
              "      <th>Low</th>\n",
              "      <th>Open</th>\n",
              "      <th>Volume</th>\n",
              "      <th>Adj Close</th>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>Date</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>2011-12-23</th>\n",
              "      <td>0.854122</td>\n",
              "      <td>0.890344</td>\n",
              "      <td>0.858471</td>\n",
              "      <td>0.064929</td>\n",
              "      <td>0.878158</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2011-12-27</th>\n",
              "      <td>0.888840</td>\n",
              "      <td>0.913803</td>\n",
              "      <td>0.880339</td>\n",
              "      <td>0.062420</td>\n",
              "      <td>0.898777</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2011-12-28</th>\n",
              "      <td>0.883538</td>\n",
              "      <td>0.902638</td>\n",
              "      <td>0.904643</td>\n",
              "      <td>0.041292</td>\n",
              "      <td>0.873712</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2011-12-29</th>\n",
              "      <td>0.867125</td>\n",
              "      <td>0.897122</td>\n",
              "      <td>0.882262</td>\n",
              "      <td>0.033933</td>\n",
              "      <td>0.889691</td>\n",
              "    </tr>\n",
              "    <tr>\n",
              "      <th>2011-12-30</th>\n",
              "      <td>0.871102</td>\n",
              "      <td>0.916927</td>\n",
              "      <td>0.882968</td>\n",
              "      <td>0.012862</td>\n",
              "      <td>0.888918</td>\n",
              "    </tr>\n",
              "  </tbody>\n",
              "</table>\n",
              "</div>"
            ],
            "text/plain": [
              "                High       Low      Open    Volume  Adj Close\n",
              "Date                                                         \n",
              "2011-12-23  0.854122  0.890344  0.858471  0.064929   0.878158\n",
              "2011-12-27  0.888840  0.913803  0.880339  0.062420   0.898777\n",
              "2011-12-28  0.883538  0.902638  0.904643  0.041292   0.873712\n",
              "2011-12-29  0.867125  0.897122  0.882262  0.033933   0.889691\n",
              "2011-12-30  0.871102  0.916927  0.882968  0.012862   0.888918"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 6
        }
      ]
    },
    {
      "metadata": {
        "id": "YJAO-nfuSAhE",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def load_data(stock, seq_len):\n",
        "    amount_of_features = len(stock.columns) # 5\n",
        "    data = stock.as_matrix() \n",
        "    sequence_length = seq_len + 1 # index starting from 0\n",
        "    result = []\n",
        "    \n",
        "    for index in range(len(data) - sequence_length): # maxmimum date = lastest date - sequence length\n",
        "        result.append(data[index: index + sequence_length]) # index : index + 22days\n",
        "    \n",
        "    result = np.array(result)\n",
        "    row = round(0.9 * result.shape[0]) # 90% split\n",
        "    train = result[:int(row), :] # 90% date, all features \n",
        "    \n",
        "    x_train = train[:, :-1] \n",
        "    y_train = train[:, -1][:,-1]\n",
        "    \n",
        "    x_test = result[int(row):, :-1] \n",
        "    y_test = result[int(row):, -1][:,-1]\n",
        "\n",
        "    x_train = np.reshape(x_train, (x_train.shape[0], x_train.shape[1], amount_of_features))\n",
        "    x_test = np.reshape(x_test, (x_test.shape[0], x_test.shape[1], amount_of_features))  \n",
        "\n",
        "    return [x_train, y_train, x_test, y_test]\n"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "aBBlB06VSFLB",
        "colab_type": "code",
        "outputId": "9522e32d-9181-4594-c977-66948e5e6253",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        }
      },
      "cell_type": "code",
      "source": [
        "window = 22\n",
        "X_train, y_train, X_test, y_test = load_data(df, window)\n",
        "X_train.shape"
      ],
      "execution_count": 8,
      "outputs": [
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(2526, 22, 5)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 8
        }
      ]
    },
    {
      "metadata": {
        "id": "__yl9IU3XEAx",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def _time_distributed_dense(x, w, b=None, dropout=None,\n",
        "                           input_dim=None, output_dim=None, timesteps=None):\n",
        "    '''Apply y.w + b for every temporal slice y of x.\n",
        "    '''\n",
        "    if not input_dim:\n",
        "        # won't work with TensorFlow\n",
        "        input_dim = K.shape(x)[2]\n",
        "    if not timesteps:\n",
        "        # won't work with TensorFlow\n",
        "        timesteps = K.shape(x)[1]\n",
        "    if not output_dim:\n",
        "        # won't work with TensorFlow\n",
        "        output_dim = K.shape(w)[1]\n",
        "\n",
        "    if dropout:\n",
        "        # apply the same dropout pattern at every timestep\n",
        "        ones = K.ones_like(K.reshape(x[:, 0, :], (-1, input_dim)))\n",
        "        dropout_matrix = K.dropout(ones, dropout)\n",
        "        expanded_dropout_matrix = K.repeat(dropout_matrix, timesteps)\n",
        "        x *= expanded_dropout_matrix\n",
        "\n",
        "    # collapse time dimension and batch dimension together\n",
        "    x = K.reshape(x, (-1, input_dim))\n",
        "\n",
        "    x = K.dot(x, w)\n",
        "    if b:\n",
        "        x = x + b\n",
        "    # reshape to 3D tensor\n",
        "    x = K.reshape(x, (-1, timesteps, output_dim))\n",
        "    return x"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "g3oFC-FmSQMv",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "import tensorflow as tf\n",
        "from keras import backend as K\n",
        "from keras import regularizers, constraints, initializers, activations\n",
        "from keras.layers.recurrent import Recurrent\n",
        "from keras.engine import InputSpec\n",
        "\n",
        "tfPrint = lambda d, T: tf.Print(input_=T, data=[T, tf.shape(T)], message=d)\n",
        "\n",
        "class AttentionDecoder(Recurrent):\n",
        "\n",
        "    def __init__(self, units, output_dim,\n",
        "                 activation='tanh',\n",
        "                 return_probabilities=False,\n",
        "                 name='AttentionDecoder',\n",
        "                 kernel_initializer='glorot_uniform',\n",
        "                 recurrent_initializer='orthogonal',\n",
        "                 bias_initializer='zeros',\n",
        "                 kernel_regularizer=None,\n",
        "                 bias_regularizer=None,\n",
        "                 activity_regularizer=None,\n",
        "                 kernel_constraint=None,\n",
        "                 bias_constraint=None,\n",
        "                 **kwargs):\n",
        "        \"\"\"\n",
        "        Implements an AttentionDecoder that takes in a sequence encoded by an\n",
        "        encoder and outputs the decoded states\n",
        "        :param units: dimension of the hidden state and the attention matrices\n",
        "        :param output_dim: the number of labels in the output space\n",
        "\n",
        "        references:\n",
        "            Bahdanau, Dzmitry, Kyunghyun Cho, and Yoshua Bengio.\n",
        "            \"Neural machine translation by jointly learning to align and translate.\"\n",
        "            arXiv preprint arXiv:1409.0473 (2014).\n",
        "        \"\"\"\n",
        "        self.units = units\n",
        "        self.output_dim = output_dim\n",
        "        self.return_probabilities = return_probabilities\n",
        "        self.activation = activations.get(activation)\n",
        "        self.kernel_initializer = initializers.get(kernel_initializer)\n",
        "        self.recurrent_initializer = initializers.get(recurrent_initializer)\n",
        "        self.bias_initializer = initializers.get(bias_initializer)\n",
        "\n",
        "        self.kernel_regularizer = regularizers.get(kernel_regularizer)\n",
        "        self.recurrent_regularizer = regularizers.get(kernel_regularizer)\n",
        "        self.bias_regularizer = regularizers.get(bias_regularizer)\n",
        "        self.activity_regularizer = regularizers.get(activity_regularizer)\n",
        "\n",
        "        self.kernel_constraint = constraints.get(kernel_constraint)\n",
        "        self.recurrent_constraint = constraints.get(kernel_constraint)\n",
        "        self.bias_constraint = constraints.get(bias_constraint)\n",
        "\n",
        "        super(AttentionDecoder, self).__init__(**kwargs)\n",
        "        self.name = name\n",
        "        self.return_sequences = True  # must return sequences\n",
        "\n",
        "    def build(self, input_shape):\n",
        "        \"\"\"\n",
        "          See Appendix 2 of Bahdanau 2014, arXiv:1409.0473\n",
        "          for model details that correspond to the matrices here.\n",
        "        \"\"\"\n",
        "\n",
        "        self.batch_size, self.timesteps, self.input_dim = input_shape\n",
        "\n",
        "        if self.stateful:\n",
        "            super(AttentionDecoder, self).reset_states()\n",
        "\n",
        "        self.states = [None, None]  # y, s\n",
        "\n",
        "        \"\"\"\n",
        "            Matrices for creating the context vector\n",
        "        \"\"\"\n",
        "\n",
        "        self.V_a = self.add_weight(shape=(self.units,),\n",
        "                                   name='V_a',\n",
        "                                   initializer=self.kernel_initializer,\n",
        "                                   regularizer=self.kernel_regularizer,\n",
        "                                   constraint=self.kernel_constraint)\n",
        "        self.W_a = self.add_weight(shape=(self.units, self.units),\n",
        "                                   name='W_a',\n",
        "                                   initializer=self.kernel_initializer,\n",
        "                                   regularizer=self.kernel_regularizer,\n",
        "                                   constraint=self.kernel_constraint)\n",
        "        self.U_a = self.add_weight(shape=(self.input_dim, self.units),\n",
        "                                   name='U_a',\n",
        "                                   initializer=self.kernel_initializer,\n",
        "                                   regularizer=self.kernel_regularizer,\n",
        "                                   constraint=self.kernel_constraint)\n",
        "        self.b_a = self.add_weight(shape=(self.units,),\n",
        "                                   name='b_a',\n",
        "                                   initializer=self.bias_initializer,\n",
        "                                   regularizer=self.bias_regularizer,\n",
        "                                   constraint=self.bias_constraint)\n",
        "        \"\"\"\n",
        "            Matrices for the r (reset) gate\n",
        "        \"\"\"\n",
        "        self.C_r = self.add_weight(shape=(self.input_dim, self.units),\n",
        "                                   name='C_r',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.U_r = self.add_weight(shape=(self.units, self.units),\n",
        "                                   name='U_r',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.W_r = self.add_weight(shape=(self.output_dim, self.units),\n",
        "                                   name='W_r',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.b_r = self.add_weight(shape=(self.units, ),\n",
        "                                   name='b_r',\n",
        "                                   initializer=self.bias_initializer,\n",
        "                                   regularizer=self.bias_regularizer,\n",
        "                                   constraint=self.bias_constraint)\n",
        "\n",
        "        \"\"\"\n",
        "            Matrices for the z (update) gate\n",
        "        \"\"\"\n",
        "        self.C_z = self.add_weight(shape=(self.input_dim, self.units),\n",
        "                                   name='C_z',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.U_z = self.add_weight(shape=(self.units, self.units),\n",
        "                                   name='U_z',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.W_z = self.add_weight(shape=(self.output_dim, self.units),\n",
        "                                   name='W_z',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.b_z = self.add_weight(shape=(self.units, ),\n",
        "                                   name='b_z',\n",
        "                                   initializer=self.bias_initializer,\n",
        "                                   regularizer=self.bias_regularizer,\n",
        "                                   constraint=self.bias_constraint)\n",
        "        \"\"\"\n",
        "            Matrices for the proposal\n",
        "        \"\"\"\n",
        "        self.C_p = self.add_weight(shape=(self.input_dim, self.units),\n",
        "                                   name='C_p',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.U_p = self.add_weight(shape=(self.units, self.units),\n",
        "                                   name='U_p',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.W_p = self.add_weight(shape=(self.output_dim, self.units),\n",
        "                                   name='W_p',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.b_p = self.add_weight(shape=(self.units, ),\n",
        "                                   name='b_p',\n",
        "                                   initializer=self.bias_initializer,\n",
        "                                   regularizer=self.bias_regularizer,\n",
        "                                   constraint=self.bias_constraint)\n",
        "        \"\"\"\n",
        "            Matrices for making the final prediction vector\n",
        "        \"\"\"\n",
        "        self.C_o = self.add_weight(shape=(self.input_dim, self.output_dim),\n",
        "                                   name='C_o',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.U_o = self.add_weight(shape=(self.units, self.output_dim),\n",
        "                                   name='U_o',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.W_o = self.add_weight(shape=(self.output_dim, self.output_dim),\n",
        "                                   name='W_o',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "        self.b_o = self.add_weight(shape=(self.output_dim, ),\n",
        "                                   name='b_o',\n",
        "                                   initializer=self.bias_initializer,\n",
        "                                   regularizer=self.bias_regularizer,\n",
        "                                   constraint=self.bias_constraint)\n",
        "\n",
        "        # For creating the initial state:\n",
        "        self.W_s = self.add_weight(shape=(self.input_dim, self.units),\n",
        "                                   name='W_s',\n",
        "                                   initializer=self.recurrent_initializer,\n",
        "                                   regularizer=self.recurrent_regularizer,\n",
        "                                   constraint=self.recurrent_constraint)\n",
        "\n",
        "        self.input_spec = [\n",
        "            InputSpec(shape=(self.batch_size, self.timesteps, self.input_dim))]\n",
        "        self.built = True\n",
        "\n",
        "    def call(self, x):\n",
        "        # store the whole sequence so we can \"attend\" to it at each timestep\n",
        "        self.x_seq = x\n",
        "\n",
        "        # apply the a dense layer over the time dimension of the sequence\n",
        "        # do it here because it doesn't depend on any previous steps\n",
        "        # thefore we can save computation time:\n",
        "        self._uxpb = _time_distributed_dense(self.x_seq, self.U_a, b=self.b_a,\n",
        "                                             input_dim=self.input_dim,\n",
        "                                             timesteps=self.timesteps,\n",
        "                                             output_dim=self.units)\n",
        "\n",
        "        return super(AttentionDecoder, self).call(x)\n",
        "\n",
        "    def get_initial_state(self, inputs):\n",
        "        # apply the matrix on the first time step to get the initial s0.\n",
        "        s0 = activations.tanh(K.dot(inputs[:, 0], self.W_s))\n",
        "\n",
        "        # from keras.layers.recurrent to initialize a vector of (batchsize,\n",
        "        # output_dim)\n",
        "        y0 = K.zeros_like(inputs)  # (samples, timesteps, input_dims)\n",
        "        y0 = K.sum(y0, axis=(1, 2))  # (samples, )\n",
        "        y0 = K.expand_dims(y0)  # (samples, 1)\n",
        "        y0 = K.tile(y0, [1, self.output_dim])\n",
        "\n",
        "        return [y0, s0]\n",
        "\n",
        "    def step(self, x, states):\n",
        "\n",
        "        ytm, stm = states\n",
        "\n",
        "        # repeat the hidden state to the length of the sequence\n",
        "        _stm = K.repeat(stm, self.timesteps)\n",
        "\n",
        "        # now multiplty the weight matrix with the repeated hidden state\n",
        "        _Wxstm = K.dot(_stm, self.W_a)\n",
        "\n",
        "        # calculate the attention probabilities\n",
        "        # this relates how much other timesteps contributed to this one.\n",
        "        et = K.dot(activations.tanh(_Wxstm + self._uxpb),\n",
        "                   K.expand_dims(self.V_a))\n",
        "        at = K.exp(et)\n",
        "        at_sum = K.sum(at, axis=1)\n",
        "        at_sum_repeated = K.repeat(at_sum, self.timesteps)\n",
        "        at /= at_sum_repeated  # vector of size (batchsize, timesteps, 1)\n",
        "\n",
        "        # calculate the context vector\n",
        "        context = K.squeeze(K.batch_dot(at, self.x_seq, axes=1), axis=1)\n",
        "        # ~~~> calculate new hidden state\n",
        "        # first calculate the \"r\" gate:\n",
        "\n",
        "        rt = activations.sigmoid(\n",
        "            K.dot(ytm, self.W_r)\n",
        "            + K.dot(stm, self.U_r)\n",
        "            + K.dot(context, self.C_r)\n",
        "            + self.b_r)\n",
        "\n",
        "        # now calculate the \"z\" gate\n",
        "        zt = activations.sigmoid(\n",
        "            K.dot(ytm, self.W_z)\n",
        "            + K.dot(stm, self.U_z)\n",
        "            + K.dot(context, self.C_z)\n",
        "            + self.b_z)\n",
        "\n",
        "        # calculate the proposal hidden state:\n",
        "        s_tp = activations.tanh(\n",
        "            K.dot(ytm, self.W_p)\n",
        "            + K.dot((rt * stm), self.U_p)\n",
        "            + K.dot(context, self.C_p)\n",
        "            + self.b_p)\n",
        "\n",
        "        # new hidden state:\n",
        "        st = (1-zt)*stm + zt * s_tp\n",
        "\n",
        "        yt = activations.softmax(\n",
        "            K.dot(ytm, self.W_o)\n",
        "            + K.dot(stm, self.U_o)\n",
        "            + K.dot(context, self.C_o)\n",
        "            + self.b_o)\n",
        "\n",
        "        if self.return_probabilities:\n",
        "            return at, [yt, st]\n",
        "        else:\n",
        "            return yt, [yt, st]\n",
        "\n",
        "    def compute_output_shape(self, input_shape):\n",
        "        \"\"\"\n",
        "            For Keras internal compatability checking\n",
        "        \"\"\"\n",
        "        if self.return_probabilities:\n",
        "            return (None, self.timesteps, self.timesteps)\n",
        "        else:\n",
        "            return (None, self.timesteps, self.output_dim)\n",
        "\n",
        "    def get_config(self):\n",
        "        \"\"\"\n",
        "            For rebuilding models on load time.\n",
        "        \"\"\"\n",
        "        config = {\n",
        "            'output_dim': self.output_dim,\n",
        "            'units': self.units,\n",
        "            'return_probabilities': self.return_probabilities\n",
        "        }\n",
        "        base_config = super(AttentionDecoder, self).get_config()\n",
        "        return dict(list(base_config.items()) + list(config.items()))"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "kNyXMcXaSMKh",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "def build_model(layers):\n",
        "    d = 0.3\n",
        "    model = Sequential()\n",
        "    \n",
        "    \n",
        "    \n",
        "    model.add(LSTM(256, input_shape=(layers[1], layers[0]), return_sequences=True))\n",
        "    model.add(Dropout(d))\n",
        "    \n",
        "    model.add(AttentionDecoder(256, 5))\n",
        "        \n",
        "    model.add(LSTM(256, input_shape=(layers[1], layers[0]), return_sequences=False))\n",
        "    model.add(Dropout(d))\n",
        "    \n",
        "    #model.add(AttentionDecoder(256, 5))\n",
        "        \n",
        "    model.add(Dense(32,kernel_initializer=\"uniform\",activation='relu'))        \n",
        "    model.add(Dense(1,kernel_initializer=\"uniform\",activation='linear'))\n",
        "    \n",
        "    # adam = keras.optimizers.Adam(decay=0.2)\n",
        "        \n",
        "    start = time.time()\n",
        "    model.compile(loss='mse',optimizer='adam', metrics=['accuracy'])\n",
        "    print(\"Compilation Time : \", time.time() - start)\n",
        "    return model"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "wm48pcUpTPxU",
        "colab_type": "code",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 34
        },
        "outputId": "10c9386a-c9d6-4816-90d8-12e1bce02653"
      },
      "cell_type": "code",
      "source": [
        "model = build_model([5,window,1])"
      ],
      "execution_count": 27,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Compilation Time :  0.03598737716674805\n"
          ],
          "name": "stdout"
        }
      ]
    },
    {
      "metadata": {
        "id": "EX33FmLLT2bn",
        "colab_type": "code",
        "outputId": "50578fe2-a51b-44bf-98d0-fae0ebe12d32",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 3451
        }
      },
      "cell_type": "code",
      "source": [
        "model.fit(X_train,y_train,batch_size=512,epochs=100,validation_split=0.1,verbose=2)"
      ],
      "execution_count": 28,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train on 2273 samples, validate on 253 samples\n",
            "Epoch 1/100\n",
            " - 6s - loss: 0.2447 - acc: 0.0022 - val_loss: 0.4819 - val_acc: 0.0040\n",
            "Epoch 2/100\n",
            " - 2s - loss: 0.1698 - acc: 0.0022 - val_loss: 0.0813 - val_acc: 0.0040\n",
            "Epoch 3/100\n",
            " - 2s - loss: 0.0905 - acc: 8.7989e-04 - val_loss: 0.1483 - val_acc: 0.0040\n",
            "Epoch 4/100\n",
            " - 2s - loss: 0.0776 - acc: 0.0022 - val_loss: 0.1893 - val_acc: 0.0040\n",
            "Epoch 5/100\n",
            " - 2s - loss: 0.0756 - acc: 0.0022 - val_loss: 0.1314 - val_acc: 0.0040\n",
            "Epoch 6/100\n",
            " - 2s - loss: 0.0728 - acc: 0.0022 - val_loss: 0.1077 - val_acc: 0.0040\n",
            "Epoch 7/100\n",
            " - 2s - loss: 0.0719 - acc: 0.0026 - val_loss: 0.1302 - val_acc: 0.0040\n",
            "Epoch 8/100\n",
            " - 2s - loss: 0.0669 - acc: 0.0022 - val_loss: 0.1470 - val_acc: 0.0040\n",
            "Epoch 9/100\n",
            " - 2s - loss: 0.0533 - acc: 0.0022 - val_loss: 0.0924 - val_acc: 0.0040\n",
            "Epoch 10/100\n",
            " - 2s - loss: 0.0322 - acc: 0.0031 - val_loss: 0.0434 - val_acc: 0.0040\n",
            "Epoch 11/100\n",
            " - 2s - loss: 0.0297 - acc: 0.0031 - val_loss: 0.0293 - val_acc: 0.0040\n",
            "Epoch 12/100\n",
            " - 2s - loss: 0.0186 - acc: 0.0031 - val_loss: 0.0170 - val_acc: 0.0040\n",
            "Epoch 13/100\n",
            " - 2s - loss: 0.0118 - acc: 0.0031 - val_loss: 0.0184 - val_acc: 0.0040\n",
            "Epoch 14/100\n",
            " - 2s - loss: 0.0094 - acc: 0.0031 - val_loss: 0.0172 - val_acc: 0.0040\n",
            "Epoch 15/100\n",
            " - 2s - loss: 0.0087 - acc: 0.0031 - val_loss: 0.0153 - val_acc: 0.0040\n",
            "Epoch 16/100\n",
            " - 2s - loss: 0.0075 - acc: 0.0031 - val_loss: 0.0145 - val_acc: 0.0040\n",
            "Epoch 17/100\n",
            " - 2s - loss: 0.0075 - acc: 0.0031 - val_loss: 0.0101 - val_acc: 0.0040\n",
            "Epoch 18/100\n",
            " - 2s - loss: 0.0064 - acc: 0.0031 - val_loss: 0.0091 - val_acc: 0.0040\n",
            "Epoch 19/100\n",
            " - 2s - loss: 0.0062 - acc: 0.0031 - val_loss: 0.0097 - val_acc: 0.0040\n",
            "Epoch 20/100\n",
            " - 2s - loss: 0.0055 - acc: 0.0031 - val_loss: 0.0080 - val_acc: 0.0040\n",
            "Epoch 21/100\n",
            " - 2s - loss: 0.0048 - acc: 0.0031 - val_loss: 0.0074 - val_acc: 0.0040\n",
            "Epoch 22/100\n",
            " - 2s - loss: 0.0049 - acc: 0.0031 - val_loss: 0.0075 - val_acc: 0.0040\n",
            "Epoch 23/100\n",
            " - 2s - loss: 0.0047 - acc: 0.0035 - val_loss: 0.0077 - val_acc: 0.0040\n",
            "Epoch 24/100\n",
            " - 2s - loss: 0.0051 - acc: 0.0035 - val_loss: 0.0066 - val_acc: 0.0040\n",
            "Epoch 25/100\n",
            " - 2s - loss: 0.0050 - acc: 0.0035 - val_loss: 0.0094 - val_acc: 0.0040\n",
            "Epoch 26/100\n",
            " - 2s - loss: 0.0053 - acc: 0.0035 - val_loss: 0.0064 - val_acc: 0.0040\n",
            "Epoch 27/100\n",
            " - 2s - loss: 0.0047 - acc: 0.0035 - val_loss: 0.0064 - val_acc: 0.0040\n",
            "Epoch 28/100\n",
            " - 2s - loss: 0.0047 - acc: 0.0035 - val_loss: 0.0070 - val_acc: 0.0040\n",
            "Epoch 29/100\n",
            " - 2s - loss: 0.0044 - acc: 0.0035 - val_loss: 0.0075 - val_acc: 0.0040\n",
            "Epoch 30/100\n",
            " - 2s - loss: 0.0044 - acc: 0.0035 - val_loss: 0.0060 - val_acc: 0.0040\n",
            "Epoch 31/100\n",
            " - 2s - loss: 0.0042 - acc: 0.0035 - val_loss: 0.0065 - val_acc: 0.0040\n",
            "Epoch 32/100\n",
            " - 2s - loss: 0.0039 - acc: 0.0035 - val_loss: 0.0070 - val_acc: 0.0040\n",
            "Epoch 33/100\n",
            " - 2s - loss: 0.0041 - acc: 0.0035 - val_loss: 0.0056 - val_acc: 0.0040\n",
            "Epoch 34/100\n",
            " - 2s - loss: 0.0039 - acc: 0.0035 - val_loss: 0.0056 - val_acc: 0.0040\n",
            "Epoch 35/100\n",
            " - 2s - loss: 0.0038 - acc: 0.0035 - val_loss: 0.0060 - val_acc: 0.0040\n",
            "Epoch 36/100\n",
            " - 2s - loss: 0.0038 - acc: 0.0035 - val_loss: 0.0067 - val_acc: 0.0040\n",
            "Epoch 37/100\n",
            " - 2s - loss: 0.0037 - acc: 0.0035 - val_loss: 0.0054 - val_acc: 0.0040\n",
            "Epoch 38/100\n",
            " - 2s - loss: 0.0042 - acc: 0.0035 - val_loss: 0.0053 - val_acc: 0.0040\n",
            "Epoch 39/100\n",
            " - 2s - loss: 0.0040 - acc: 0.0035 - val_loss: 0.0059 - val_acc: 0.0040\n",
            "Epoch 40/100\n",
            " - 2s - loss: 0.0039 - acc: 0.0035 - val_loss: 0.0056 - val_acc: 0.0040\n",
            "Epoch 41/100\n",
            " - 2s - loss: 0.0037 - acc: 0.0035 - val_loss: 0.0052 - val_acc: 0.0040\n",
            "Epoch 42/100\n",
            " - 2s - loss: 0.0036 - acc: 0.0035 - val_loss: 0.0056 - val_acc: 0.0040\n",
            "Epoch 43/100\n",
            " - 2s - loss: 0.0035 - acc: 0.0035 - val_loss: 0.0058 - val_acc: 0.0040\n",
            "Epoch 44/100\n",
            " - 2s - loss: 0.0036 - acc: 0.0035 - val_loss: 0.0054 - val_acc: 0.0040\n",
            "Epoch 45/100\n",
            " - 2s - loss: 0.0034 - acc: 0.0035 - val_loss: 0.0052 - val_acc: 0.0040\n",
            "Epoch 46/100\n",
            " - 2s - loss: 0.0038 - acc: 0.0035 - val_loss: 0.0056 - val_acc: 0.0040\n",
            "Epoch 47/100\n",
            " - 2s - loss: 0.0036 - acc: 0.0035 - val_loss: 0.0058 - val_acc: 0.0040\n",
            "Epoch 48/100\n",
            " - 2s - loss: 0.0105 - acc: 0.0035 - val_loss: 0.0065 - val_acc: 0.0040\n",
            "Epoch 49/100\n",
            " - 2s - loss: 0.0064 - acc: 0.0035 - val_loss: 0.0092 - val_acc: 0.0040\n",
            "Epoch 50/100\n",
            " - 2s - loss: 0.0059 - acc: 0.0035 - val_loss: 0.0095 - val_acc: 0.0040\n",
            "Epoch 51/100\n",
            " - 2s - loss: 0.0051 - acc: 0.0035 - val_loss: 0.0083 - val_acc: 0.0040\n",
            "Epoch 52/100\n",
            " - 2s - loss: 0.0047 - acc: 0.0035 - val_loss: 0.0061 - val_acc: 0.0040\n",
            "Epoch 53/100\n",
            " - 2s - loss: 0.0043 - acc: 0.0035 - val_loss: 0.0057 - val_acc: 0.0040\n",
            "Epoch 54/100\n",
            " - 2s - loss: 0.0042 - acc: 0.0035 - val_loss: 0.0065 - val_acc: 0.0040\n",
            "Epoch 55/100\n",
            " - 2s - loss: 0.0038 - acc: 0.0035 - val_loss: 0.0062 - val_acc: 0.0040\n",
            "Epoch 56/100\n",
            " - 2s - loss: 0.0037 - acc: 0.0035 - val_loss: 0.0064 - val_acc: 0.0040\n",
            "Epoch 57/100\n",
            " - 2s - loss: 0.0037 - acc: 0.0035 - val_loss: 0.0053 - val_acc: 0.0040\n",
            "Epoch 58/100\n",
            " - 2s - loss: 0.0037 - acc: 0.0035 - val_loss: 0.0051 - val_acc: 0.0040\n",
            "Epoch 59/100\n",
            " - 2s - loss: 0.0035 - acc: 0.0035 - val_loss: 0.0054 - val_acc: 0.0040\n",
            "Epoch 60/100\n",
            " - 2s - loss: 0.0033 - acc: 0.0035 - val_loss: 0.0049 - val_acc: 0.0040\n",
            "Epoch 61/100\n",
            " - 2s - loss: 0.0035 - acc: 0.0035 - val_loss: 0.0049 - val_acc: 0.0040\n",
            "Epoch 62/100\n",
            " - 2s - loss: 0.0035 - acc: 0.0035 - val_loss: 0.0059 - val_acc: 0.0040\n",
            "Epoch 63/100\n",
            " - 2s - loss: 0.0034 - acc: 0.0035 - val_loss: 0.0047 - val_acc: 0.0040\n",
            "Epoch 64/100\n",
            " - 2s - loss: 0.0033 - acc: 0.0035 - val_loss: 0.0047 - val_acc: 0.0040\n",
            "Epoch 65/100\n",
            " - 2s - loss: 0.0032 - acc: 0.0035 - val_loss: 0.0060 - val_acc: 0.0040\n",
            "Epoch 66/100\n",
            " - 2s - loss: 0.0036 - acc: 0.0035 - val_loss: 0.0048 - val_acc: 0.0040\n",
            "Epoch 67/100\n",
            " - 2s - loss: 0.0036 - acc: 0.0035 - val_loss: 0.0048 - val_acc: 0.0040\n",
            "Epoch 68/100\n",
            " - 2s - loss: 0.0036 - acc: 0.0035 - val_loss: 0.0058 - val_acc: 0.0040\n",
            "Epoch 69/100\n",
            " - 2s - loss: 0.0035 - acc: 0.0035 - val_loss: 0.0055 - val_acc: 0.0040\n",
            "Epoch 70/100\n",
            " - 2s - loss: 0.0033 - acc: 0.0035 - val_loss: 0.0048 - val_acc: 0.0040\n",
            "Epoch 71/100\n",
            " - 2s - loss: 0.0034 - acc: 0.0035 - val_loss: 0.0049 - val_acc: 0.0040\n",
            "Epoch 72/100\n",
            " - 2s - loss: 0.0034 - acc: 0.0035 - val_loss: 0.0060 - val_acc: 0.0040\n",
            "Epoch 73/100\n",
            " - 2s - loss: 0.0036 - acc: 0.0035 - val_loss: 0.0056 - val_acc: 0.0040\n",
            "Epoch 74/100\n",
            " - 2s - loss: 0.0034 - acc: 0.0035 - val_loss: 0.0053 - val_acc: 0.0040\n",
            "Epoch 75/100\n",
            " - 2s - loss: 0.0039 - acc: 0.0035 - val_loss: 0.0050 - val_acc: 0.0040\n",
            "Epoch 76/100\n",
            " - 2s - loss: 0.0034 - acc: 0.0035 - val_loss: 0.0051 - val_acc: 0.0040\n",
            "Epoch 77/100\n",
            " - 2s - loss: 0.0031 - acc: 0.0035 - val_loss: 0.0054 - val_acc: 0.0040\n",
            "Epoch 78/100\n",
            " - 2s - loss: 0.0033 - acc: 0.0035 - val_loss: 0.0049 - val_acc: 0.0040\n",
            "Epoch 79/100\n",
            " - 2s - loss: 0.0034 - acc: 0.0035 - val_loss: 0.0052 - val_acc: 0.0040\n",
            "Epoch 80/100\n",
            " - 2s - loss: 0.0035 - acc: 0.0035 - val_loss: 0.0070 - val_acc: 0.0040\n",
            "Epoch 81/100\n",
            " - 2s - loss: 0.0039 - acc: 0.0035 - val_loss: 0.0053 - val_acc: 0.0040\n",
            "Epoch 82/100\n",
            " - 2s - loss: 0.0032 - acc: 0.0035 - val_loss: 0.0049 - val_acc: 0.0040\n",
            "Epoch 83/100\n",
            " - 2s - loss: 0.0031 - acc: 0.0035 - val_loss: 0.0047 - val_acc: 0.0040\n",
            "Epoch 84/100\n",
            " - 2s - loss: 0.0032 - acc: 0.0035 - val_loss: 0.0050 - val_acc: 0.0040\n",
            "Epoch 85/100\n",
            " - 2s - loss: 0.0031 - acc: 0.0035 - val_loss: 0.0058 - val_acc: 0.0040\n",
            "Epoch 86/100\n",
            " - 2s - loss: 0.0031 - acc: 0.0035 - val_loss: 0.0054 - val_acc: 0.0040\n",
            "Epoch 87/100\n",
            " - 2s - loss: 0.0032 - acc: 0.0035 - val_loss: 0.0049 - val_acc: 0.0040\n",
            "Epoch 88/100\n",
            " - 2s - loss: 0.0031 - acc: 0.0035 - val_loss: 0.0047 - val_acc: 0.0040\n",
            "Epoch 89/100\n",
            " - 2s - loss: 0.0031 - acc: 0.0035 - val_loss: 0.0045 - val_acc: 0.0040\n",
            "Epoch 90/100\n",
            " - 2s - loss: 0.0032 - acc: 0.0035 - val_loss: 0.0056 - val_acc: 0.0040\n",
            "Epoch 91/100\n",
            " - 2s - loss: 0.0031 - acc: 0.0035 - val_loss: 0.0047 - val_acc: 0.0040\n",
            "Epoch 92/100\n",
            " - 2s - loss: 0.0031 - acc: 0.0035 - val_loss: 0.0046 - val_acc: 0.0040\n",
            "Epoch 93/100\n",
            " - 2s - loss: 0.0030 - acc: 0.0035 - val_loss: 0.0052 - val_acc: 0.0040\n",
            "Epoch 94/100\n",
            " - 2s - loss: 0.0033 - acc: 0.0035 - val_loss: 0.0044 - val_acc: 0.0040\n",
            "Epoch 95/100\n",
            " - 2s - loss: 0.0030 - acc: 0.0035 - val_loss: 0.0044 - val_acc: 0.0040\n",
            "Epoch 96/100\n",
            " - 2s - loss: 0.0031 - acc: 0.0035 - val_loss: 0.0048 - val_acc: 0.0040\n",
            "Epoch 97/100\n",
            " - 2s - loss: 0.0030 - acc: 0.0035 - val_loss: 0.0043 - val_acc: 0.0040\n",
            "Epoch 98/100\n",
            " - 2s - loss: 0.0030 - acc: 0.0035 - val_loss: 0.0054 - val_acc: 0.0040\n",
            "Epoch 99/100\n",
            " - 2s - loss: 0.0030 - acc: 0.0035 - val_loss: 0.0046 - val_acc: 0.0040\n",
            "Epoch 100/100\n",
            " - 2s - loss: 0.0029 - acc: 0.0035 - val_loss: 0.0047 - val_acc: 0.0040\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "<keras.callbacks.History at 0x7fe703d0e5f8>"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 28
        }
      ]
    },
    {
      "metadata": {
        "id": "zrw8WYerXxaD",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "p = model.predict(X_test)"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "j_koxvrFYV4-",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        "newp =  p\n",
        "newy_test =  y_test"
      ],
      "execution_count": 0,
      "outputs": []
    },
    {
      "metadata": {
        "id": "vPbYzSpPYYhQ",
        "colab_type": "code",
        "outputId": "e99c6e70-7e53-48bf-e382-290303c3ed30",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 68
        }
      },
      "cell_type": "code",
      "source": [
        "def model_score(model, X_train, y_train, X_test, y_test):\n",
        "    trainScore = model.evaluate(X_train, y_train, verbose=0)\n",
        "    print('Train Score: %.5f MSE (%.2f RMSE)' % (trainScore[0], math.sqrt(trainScore[0])))\n",
        "\n",
        "    testScore = model.evaluate(X_test, y_test, verbose=0)\n",
        "    print('Test Score: %.5f MSE (%.2f RMSE)' % (testScore[0], math.sqrt(testScore[0])))\n",
        "    return trainScore[0], testScore[0]\n",
        "\n",
        "\n",
        "model_score(model, X_train, y_train, X_test, y_test)"
      ],
      "execution_count": 31,
      "outputs": [
        {
          "output_type": "stream",
          "text": [
            "Train Score: 0.00240 MSE (0.05 RMSE)\n",
            "Test Score: 0.00190 MSE (0.04 RMSE)\n"
          ],
          "name": "stdout"
        },
        {
          "output_type": "execute_result",
          "data": {
            "text/plain": [
              "(0.002403963824530572, 0.0018989782154003284)"
            ]
          },
          "metadata": {
            "tags": []
          },
          "execution_count": 31
        }
      ]
    },
    {
      "metadata": {
        "id": "tGwhdx40Ybae",
        "colab_type": "code",
        "outputId": "fa847765-6f12-4c04-f080-3bec45282b24",
        "colab": {
          "base_uri": "https://localhost:8080/",
          "height": 347
        }
      },
      "cell_type": "code",
      "source": [
        "import matplotlib.pyplot as plt2\n",
        "\n",
        "plt2.plot(newp,color='red', label='Prediction')\n",
        "plt2.plot(newy_test,color='blue', label='Actual')\n",
        "plt2.legend(loc='best')\n",
        "plt2.show()"
      ],
      "execution_count": 32,
      "outputs": [
        {
          "output_type": "display_data",
          "data": {
            "image/png": "iVBORw0KGgoAAAANSUhEUgAAAeEAAAFKCAYAAAAqkecjAAAABHNCSVQICAgIfAhkiAAAAAlwSFlz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\n",
            "text/plain": [
              "<Figure size 576x396 with 1 Axes>"
            ]
          },
          "metadata": {
            "tags": []
          }
        }
      ]
    },
    {
      "metadata": {
        "id": "skFYb3l5Yeel",
        "colab_type": "code",
        "colab": {}
      },
      "cell_type": "code",
      "source": [
        ""
      ],
      "execution_count": 0,
      "outputs": []
    }
  ]
}