DeepStock/stockpredictor/train/PricePredictor.py

import os
import json
import sqlite3
import numpy as np
from keras.models import Sequential
from keras.layers import Dense
from keras.callbacks import EarlyStopping, ModelCheckpoint
from keras.layers import LSTM

class PricePredictor:

    trainingRate = 0.7
    TYPE = 0 # OPEN = 0,  CLOSE = 1, HIGH = 2, LOW = 3, VOLUME = 4
    MAX_VALUE = 500000
    MAX_VOLUME_VALUE = 300000000

    def getNormalPrice(self, value):
        value = value / self.MAX_VALUE
        if value>1:
            value = 1
        return value

    def getNormalVolume(self, value):
        value = value / self.MAX_VOLUME_VALUE
        if value > 1:
            value = 1
        return value

    def read (self, inFileName):
        conn = sqlite3.connect(inFileName)
        cursor = conn.cursor()

        stocks = []
        rowid = 1
        cursor.execute('SELECT * FROM stock WHERE rowid=?', (rowid,))
        result = cursor.fetchone()
        while result != None:
            prices = json.loads(result[2])
            stock = []
            for price in prices:
                stock.append([self.getNormalPrice(price['open']), self.getNormalPrice(price['close']), self.getNormalPrice(price['high']), self.getNormalPrice(price['low']), self.getNormalVolume(price['volume'])])
            stocks.append(stock)

            rowid += 1
            cursor.execute('SELECT * FROM stock WHERE rowid=?', (rowid,))
            result = cursor.fetchone()

        cursor.close()
        conn.close()

        return stocks

    def make_dataset(self, stocks, window_size=5):
        feature = []
        label = []
        test_feature = []
        test_label = []

        for i in range(len(stocks)):
            stock = stocks[i]
            for j in range(len(stock) - window_size):
                feature.append(stock[j:j+window_size])
                label.append([stock[j+window_size][self.TYPE]])

            test_feature.append(stock[len(stock)-window_size-1:len(stock)-1])
            test_label.append(stock[len(stock)-1][self.TYPE])

        index = int(len(feature) * self.trainingRate)
        train_feature = feature[:index]
        train_label = label[:index]
        valid_feature = feature[index:]
        valid_label = label[index:]

        test_feature = feature[len(feature) - 1]
        test_label = label[len(feature) - 1]

        return np.array(train_feature), np.array(train_label), np.array(valid_feature), np.array(valid_label), np.array(test_feature), np.array(test_label)

    def train(self, inFileName, model_path):
        stocks = self.read(inFileName)
        train_feature, train_label, valid_feature, valid_label, test_feature, test_label = self.make_dataset(stocks)
        print (train_feature.shape)
        print (train_label.shape)

        model = Sequential()
        model.add(LSTM(32,
                       input_shape=(train_feature.shape[1], train_feature.shape[2]),
                       activation='relu',
                       return_sequences=False))

        model.add(Dense(1))

        model.compile(loss='mean_squared_error', optimizer='adam')
        early_stop = EarlyStopping(monitor='val_loss', patience=5)

        filename = ''
        if self.TYPE==0 :
            filename = os.path.join(model_path, 'open_checkpoint.h5')
        elif self.TYPE==1 :
            filename = os.path.join(model_path, 'close_checkpoint.h5')
        elif self.TYPE==2 :
            filename = os.path.join(model_path, 'high_checkpoint.h5')
        elif self.TYPE==3 :
            filename = os.path.join(model_path, 'low_checkpoint.h5')

        checkpoint = ModelCheckpoint(filename, monitor='val_loss', verbose=1, save_best_only=True, mode='auto')

        history = model.fit(train_feature, train_label,
                            epochs=1000,
                            batch_size=100000,
                            validation_data=(valid_feature, valid_label),
                            callbacks=[early_stop, checkpoint])

        test_feature = np.array([test_feature])
        predition = model.predict(test_feature)
        print(predition[0], test_label)
        print(predition[0] * self.MAX_VALUE, test_label * self.MAX_VALUE)
        return

    def predict(self, inFileName, model_path):
        stocks = self.read(inFileName)
        train_feature, train_label, valid_feature, valid_label, test_feature, test_label = self.make_dataset(stocks)

        filename = ''
        if self.TYPE==0 :
            filename = os.path.join(model_path, 'open_checkpoint.h5')
        elif self.TYPE==1 :
            filename = os.path.join(model_path, 'close_checkpoint.h5')
        elif self.TYPE==2 :
            filename = os.path.join(model_path, 'high_checkpoint.h5')
        elif self.TYPE==3 :
            filename = os.path.join(model_path, 'low_checkpoint.h5')

        model = Sequential()
        model.add(LSTM(32,
                       input_shape=(train_feature.shape[1], train_feature.shape[2]),
                       activation='relu',
                       return_sequences=False))
        model.add(Dense(1))
        model.compile(loss='mean_squared_error', optimizer='adam')
        model.load_weights(filename)

        test_feature = np.array([test_feature])
        predition = model.predict(test_feature)
        print(predition[0], test_label)
        print(predition[0] * self.MAX_VALUE, test_label * self.MAX_VALUE)

        return

if __name__ == "__main__":
    PROJECT_HOME = "../.."
    inFileName = PROJECT_HOME + '/resources/stock.db'
    model_path = PROJECT_HOME + '/resources/model'

    pricePredictor = PricePredictor()
    pricePredictor.train(inFileName, model_path)
    #pricePredictor.predict(inFileName, model_path)