LSTM.py

from urllib import error as urllib_err
# coding: utf-8

# # Stockex LSTM predictive model

# In[1]:
import datetime
import math
import time
from StockInfoProvider import StockInfoProvider
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from keras.layers.core import Dense, Activation, Dropout
from keras.layers.recurrent import LSTM
from keras.models import Sequential
from sklearn import linear_model
from sklearn import preprocessing
from sklearn.preprocessing import MinMaxScaler
import random
import os.path

# # Fetch historical stock data using google finance (uk)
# We've chosen Google Finanace because of the ability to handle adjustment close prices
class StockModel:
    def __init__(self):
        self.sip = StockInfoProvider()
        self.stockDataDict = {}

    # In[13]:
    def fetch_stocks_data(self, symbol, start_date, end_date):
        ''' Daily quotes from Google. Date format='yyyy-mm-dd' '''
        symbol = symbol.upper()
        start = datetime.date(int(start_date[0:4]), int(start_date[5:7]), int(start_date[8:10]))
        end = datetime.date(int(end_date[0:4]), int(end_date[5:7]), int(end_date[8:10]))
        url_string = "https://finance.google.co.uk/bctzjpnsun/historical?q=NASDAQ:{0}".format(symbol)
        url_string += "&startdate={0}&enddate={1}&num={0}&ei=KKltWZHCBNWPuQS9147YBw&output=csv".format(
            start.strftime('%b%d,%Y'), end.strftime('%b%d,%Y'), 4000)
        print(symbol)
        col_names = ['Date', 'Open', 'High', 'Low', 'Close', 'Volume']
        stocks = pd.read_csv(url_string, header=0, names=col_names)
        df = pd.DataFrame(stocks)
        return df


    # In[16]:
    def remove_data(self, data): #TODO: TURN INTO remove_data_with_sentiment
        """
        Remove columns from the data
        :param data: stock data containing the following columns:  ['Date','Open','High','Low','Close','Volume']
        :param sentiment_df: dataframe containing the sentiment for each day ['Date','Positive', 'Neutral', 'Negative', 'Compound']
        :return: a DataFrame with columns as  ['index','Open','Close','Volume']
        """
        # Define columns of data to keep from historical stock data
        item = []
        open = []
        close = []
        volume = []
        # Loop through the stock data objects backwards and store factors we want to keep
        i_counter = 0
        for i in range(len(data) - 1, -1, -1):
            item.append(i_counter)
            open.append(data['Open'][i])
            close.append(data['Close'][i])
            volume.append(data['Volume'][i])
            i_counter += 1

        # Create a data frame for stock data
        stocks = pd.DataFrame()


        # Add factors to data frame
        stocks['Item'] = item
        stocks['Open'] = open
        stocks['Close'] = pd.to_numeric(close)
        stocks['Volume'] = pd.to_numeric(volume)
        # return new formatted data
        return stocks

    def remove_data_with_sentiment(self, data, sentiment_df): #TODO: TURN INTO remove_data_with_sentiment
        """
        Remove columns from the data
        :param data: a record of all the stock prices with columns as  ['Date','Open','High','Low','Close','Volume']
        :param sentiment_df: dataframe containing the sentiment for each day ['Date','Positive', 'Neutral', 'Negative', 'Compound']
        :return: a DataFrame with columns as  ['index','Open','Close','Volume']
        """
        # Define columns of data to keep from historical stock data
        item = []
        open = []
        close = []
        volume = []
        # TODO: should we include pos, neut, neg + compound OR just compound?
        positive = []
        neutral = []
        negative = []
        compound = []
        # Loop through the stock data objects backwards and store factors we want to keep
        i_counter = 0
        for i in range(len(data) - 1, -1, -1):
            item.append(i_counter)
            open.append(data['Open'][i])
            close.append(data['Close'][i])
            volume.append(data['Volume'][i])
            # positive.append(sentiment_df['Positive'][i])
            # neutral.append(sentiment_df['Neutral'][i])
            # negative.append(sentiment_df['Negative'][i])
            # compound.append(sentiment_df['Compound'][i])
            i_counter += 1

        # Create a data frame for stock data
        stocks = pd.DataFrame()


        # Add factors to data frame
        stocks['Item'] = item
        stocks['Open'] = open
        stocks['Close'] = pd.to_numeric(close)
        stocks['Volume'] = pd.to_numeric(volume)
        # SENTIMENTS

        # sentiment_df.drop('Date',axis=1)
        # print(str(stocks))
        # print(str(sentiment_df))
        stocks = stocks.add(sentiment_df,fill_value=0,axis=1) #sentiment_df[['Positive']]

        # END
        # stocks['Neutral'] = sentiment_df[['Neutral']]
        # stocks['Negative'] = sentiment_df[['Negative']]
        # stocks['Compound'] = sentiment_df[['Compound']]
        # return new formatted data
        return stocks


    # In[19]:
    def price(self, x):
        """
        format the coords message box
        :param x: data to be formatted
        :return: formatted data
        """
        return '$%1.2f' % x


    # # In[20]:
     #def plot_basic(self, stocks, title='Google Trading', y_label='Price USD', x_label='Trading Days'):
    #"""
    #    Plots basic pyplot
    #     :param stocks: DataFrame having all the necessary data
    #     :param title:  Title of the plot
    #     :param y_label: yLabel of the plot
    #     :param x_label: xLabel of the plot
    #     :return: prints a Pyplot againts items and their closing value
    #     """
    #   fig, ax = plt.subplots()
    #    ax.plot(stocks['Item'], stocks['Close'], '#0A7388')
    #
    #     ax.format_ydata = price
    #     ax.set_title(title)
    #
    #     # Add labels
    #     plt.ylabel(y_label)
    #     plt.xlabel(x_label)
    #
    #     plt.show()


    # In[21]:
    def plot_prediction(self, actual, prediction, title='Google Trading vs Prediction', y_label='Price USD',
                        x_label='Trading Days'):
        """
        Plots train, test and prediction
        :param actual: DataFrame containing actual data
        :param prediction: DataFrame containing predicted values
        :param title:  Title of the plot
        :param y_label: yLabel of the plot
        :param x_label: xLabel of the plot
        :return: prints a Pyplot againts items and their closing value
        """
        fig = plt.figure()
        ax = fig.add_subplot(111)

        # Add labels
        plt.ylabel(y_label)
        plt.xlabel(x_label)

        # Plot actual and predicted close values

        plt.plot(actual, '#00FF00', label='Adjusted Close')
        plt.plot(prediction, '#0000FF', label='Predicted Close')

        # Set title
        ax.set_title(title)
        ax.legend(loc='upper left')

        plt.show()


    # In[22]:
    def plot_lstm_prediction(self, actual, prediction, title='Google Trading vs Prediction', y_label='Price USD',
                             x_label='Trading Days'):
        """
        Plots train, test and prediction
        :param actual: DataFrame containing actual data
        :param prediction: DataFrame containing predicted values
        :param title:  Title of the plot
        :param y_label: yLabel of the plot
        :param x_label: xLabel of the plot
        :return: prints a Pyplot againts items and their closing value
        """
        fig = plt.figure()
        ax = fig.add_subplot(111)

        # Add labels
        plt.ylabel(y_label)
        plt.xlabel(x_label)

        # Plot actual and predicted close values
        #print("ACTUALLLL: "+str(actual))
        # print("PREDICTTTT: " + str(prediction))
        plt.plot(actual, '#00FF00', label='Adjusted Close')
        plt.plot(prediction, '#0000FF', label='Predicted Close')

        # Set title
        ax.set_title(title)
        ax.legend(loc='upper left')

        plt.show()
        # print("EYYY: "+str(len(prediction) - len(actual)))#actual.tail(1)[list(actual)[0]])
        # list(prediction)[0]
        # print("PRE-SCORE: "+)

    #plot_basic(stocks)
    #In[24]:
    def get_normalised_data(self, data):
        """
        Normalises the data values using MinMaxScaler from sklearn
        :param data: a DataFrame with columns as  ['index','Open','Close','Volume']
        :return: a DataFrame with normalised value for all the columns except index
        """

        scaler = preprocessing.MinMaxScaler()
        # Initialize a scaler, then apply it to the features
        scaler = MinMaxScaler()
        numerical = ['Open', 'Close', 'Volume']

        data[numerical] = scaler.fit_transform(data[numerical])

        return data

    def get_normalised_data_with_sentiment(self, data):
        """
        Normalises the data values using MinMaxScaler from sklearn
        :param data: a DataFrame with columns as  ['index','Open','Close','Volume']
        :return: a DataFrame with normalised value for all the columns except index
        """

        scaler = preprocessing.MinMaxScaler()
        # Initialize a scaler, then apply it to the features
        scaler = MinMaxScaler()
        numerical = ['Open', 'Close', 'Volume']#,'Positive','Neutral','Negative','Compound'] #SENTIMENT
        data[numerical] = scaler.fit_transform(data[numerical]) #TODO: UNCOMMENT!!!!!!!!!!!!!!!!KBJFEVDWFVWEFVGIWEVWEFVGR

        return data

    # In[29]:
    def scale_range(self, x, input_range, target_range):
        """

        Rescale a numpy array from input to target range
        :param x: data to scale
        :param input_range: optional input range for data: default 0.0:1.0
        :param target_range: optional target range for data: default 0.0:1.0
        :return: rescaled array, incoming range [min,max]
        """

        range = [np.amin(x), np.amax(x)]
        x_std = (x - input_range[0]) / (1.0 * (input_range[1] - input_range[0]))
        x_scaled = x_std * (1.0 * (target_range[1] - target_range[0])) + target_range[0]
        return x_scaled, range


    # In[30]:
    def train_test_split_linear_regression(self, stocks):
        """
            Split the data set into training and testing feature for Linear Regression Model
            :param stocks: whole data set containing ['Open','Close','Volume'] features
            :return: X_train : training sets of feature
            :return: X_test : test sets of feature
            :return: y_train: training sets of label
            :return: y_test: test sets of label
            :return: label_range: scaled range of label used in predicting price,
        """
        # Create numpy arrays for features and targets
        feature = []
        label = []
        # print("STOCKEROOS: "+ str(stocks))
        # Convert dataframe columns to numpy arrays for scikit learn
        for index, row in stocks.iterrows():
            # print([np.array(row['Item'])])
            feature.append([(row['Item'])])
            label.append([(row['Close'])])

        # Regularize the feature and target arrays and store min/max of input data for rescaling later
        feature_bounds = [min(feature), max(feature)]
        feature_bounds = [feature_bounds[0][0], feature_bounds[1][0]]
        label_bounds = [min(label), max(label)]
        label_bounds = [label_bounds[0][0], label_bounds[1][0]]

        feature_scaled, feature_range = self.scale_range(np.array(feature), input_range=feature_bounds, target_range=[-1.0, 1.0])
        label_scaled, label_range = self.scale_range(np.array(label), input_range=label_bounds, target_range=[-1.0, 1.0])

        # Define Test/Train Split 80/20
        split = .315
        split = int(math.floor(len(stocks['Item']) * split))

        # Set up training and test sets
        X_train = feature_scaled[:-split]
        X_test = feature_scaled[-split:]

        y_train = label_scaled[:-split]
        y_test = label_scaled[-split:]

        return X_train, X_test, y_train, y_test, label_range


    # In[31]:
    # TODO: HOW TO DETERMINE TEST_DATA_SIZE AND UNROLL_LENGTH, PREDICTION_TIME?
    def train_test_split_lstm(self, stocks, prediction_time=1, test_data_size=160, unroll_length=40):# TODO: 80,20
        """
            Split the data set into training and testing feature for Long Short Term Memory Model
            :param stocks: whole data set containing ['Open','Close','Volume'] features
            :param prediction_time: no of days
            :param test_data_size: size of test data to be used
            :param unroll_length: how long a window should be used for train test split
            :return: X_train : training sets of feature
            :return: X_test : test sets of feature
            :return: y_train: training sets of label
            :return: y_test: test sets of label
        """
        # training data
        test_data_cut = test_data_size + unroll_length + 1

        x_train = stocks[0:-prediction_time - test_data_cut].as_matrix()
        y_train = stocks[prediction_time:-test_data_cut]['Close'].as_matrix()

        # test data
        x_test = stocks[0 - test_data_cut:-prediction_time].as_matrix()
        # print("prediction_time:-test_data_cut  {}  :  {}  ".format(prediction_time,-test_data_cut))
        y_test = stocks[prediction_time - test_data_cut:]['Close'].as_matrix()
        # print("stocks[0 - test_data_cut:-prediction_time]: " + str(stocks[0 - test_data_cut:-prediction_time]))
        # print("Y)TEST: "+str(y_test))
        return x_train, x_test, y_train, y_test


    # In[32]:
    def unroll(self, data, sequence_length=24):
        """
        use different windows for testing and training to stop from leak of information in the data
        :param data: data set to be used for unrolling
        :param sequence_length: window length
        :return: data sets with different window.
        """
        result = []
        for index in range(len(data) - sequence_length):
            result.append(data[index: index + sequence_length])
        return np.asarray(result)


    # In[34]:
    def build_model(self, X, y):
        """
        build a linear regression model using sklearn.linear_model
        :param X: Feature dataset
        :param y: label dataset
        :return: a linear regression model
        """
        linear_mod = linear_model.LinearRegression()  # defining the linear regression model
        X = np.reshape(X, (X.shape[0], 1))
        y = np.reshape(y, (y.shape[0], 1))
        linear_mod.fit(X, y)  # fitting the data points in the model

        return linear_mod


    # In[35]:
    def build_model(self, X, y):
        """
        build a linear regression model using sklearn.linear_model
        :param X: Feature dataset
        :param y: label dataset
        :return: a linear regression model
        """
        linear_mod = linear_model.LinearRegression()  # defining the linear regression model
        X = np.reshape(X, (X.shape[0], 1))
        y = np.reshape(y, (y.shape[0], 1))
        linear_mod.fit(X, y)  # fitting the data points in the model
        return linear_mod


    # In[36]:
    def predict_prices(self, model, x, label_range):
        """
        Predict the label for given test sets
        :param model: a linear regression model
        :param x: testing features
        :param label_range: normalised range of label data
        :return: predicted labels for given features
        """
        x = np.reshape(x, (x.shape[0], 1))
        predicted_price = model.predict(x)
        predictions_rescaled, re_range = self.scale_range(predicted_price, input_range=[-1.0, 1.0], target_range=label_range)

        return predictions_rescaled.flatten()


    # In[44]:
    def build_basic_model(self, input_dim, output_dim, return_sequences):
        """
        Builds a basic lstm model
        :param input_dim: input dimension of the model
        :param output_dim: output dimension of the model
        :param return_sequences: return sequence of the model
        :return: a basic lstm model with 3 layers.
        """
        model = Sequential()
        model.add(LSTM(
            input_shape=(None, input_dim),
            units=output_dim,
            return_sequences=return_sequences))

        model.add(LSTM(
            100,
            return_sequences=False))

        model.add(Dense(
            units=1))
        model.add(Activation('linear'))
        model.add(Dropout(0.2))
        return model


    # In[45]:
    def build_improved_model(self, input_dim, output_dim, return_sequences):
        """
        Builds an improved Long Short term memory model using keras.layers.recurrent.lstm
        :param input_dim: input dimension of model
        :param output_dim: ouput dimension of model
        :param return_sequences: return sequence for the model
        :return: a 3 layered LSTM model
        """
        model = Sequential()
        model.add(LSTM(
            input_shape=(None, input_dim),
            units=output_dim,
            return_sequences=return_sequences))

        model.add(Dropout(0.5))

        model.add(LSTM(
            128,
            return_sequences=False))

        model.add(Dropout(0.5))

        model.add(Dense(
            units=1))
        model.add(Activation('linear'))

        return model


    def start(self, symbol, history_start_date, history_end_date, predict_start_date, predict_end_date):
        #symbol="GOOG" #TODO:COOMENT!!!!f!!!!!!!!!!!!!!!!!!!!!!!!
        #TODO: train for every stock combined
        if not os.path.exists('./csv/'+symbol+'.csv'):
            data = self.fetch_stocks_data(symbol, history_start_date, history_end_date)
            data.to_csv('./csv/'+symbol+'.csv', index=False) #TODO: UNCOMMENT


        # Calculate Mean, Std , Min, Max for current dataset
        # In[14]:
        data = pd.read_csv('./csv/'+symbol+'.csv') #TODO: UNCOMMENT
        # data = pd.read_csv('./csv/' + 'google' + '.csv')
        # print(data.head())
        # print(data.tail())
        #
        # print("\n")
        # print("Open   --- mean :", np.mean(data['Open']), "  \t Std: ", np.std(data['Open']), "  \t Max: ",
        #       np.max(data['Open']), "  \t Min: ", np.min(data['Open']))
        # print("High   --- mean :", np.mean(data['High']), "  \t Std: ", np.std(data['High']), "  \t Max: ",
        #       np.max(data['High']), "  \t Min: ", np.min(data['High']))
        # print("Low    --- mean :", np.mean(data['Low']), "  \t Std: ", np.std(data['Low']), "  \t Max: ", np.max(data['Low']),
        #       "  \t Min: ", np.min(data['Low']))
        # print("Close  --- mean :", np.mean(data['Close']), "  \t Std: ", np.std(data['Close']), "  \t Max: ",
        #       np.max(data['Close']), "  \t Min: ", np.min(data['Close']))
        # print("Volume --- mean :", np.mean(data['Volume']), "  \t Std: ", np.std(data['Volume']), "  \t Max: ",
        #       np.max(data['Volume']), "  \t Min: ", np.min(data['Volume']))

        # # Preprocessing # #

        # In[15]:
        #TODO: CHANGE pddf TO SENTIMENT DATAFRAME WE RECEIVED FROM SENTIMENT ANALYSIS
        # pddf = pd.DataFrame(2 * np.random.random_sample(size=(len(data), 1)) -1 , columns=['Compound'])
        pddf = pd.DataFrame([1] *len(data), columns=['Compound'])
        # pddf['Date']= 1
        # pddf['Positive']= 1
        # pddf['Neutral']= 1
        # pddf['Negative']= 1
        #pddf['Compound'] = 1.0
        stocks = self.remove_data_with_sentiment(data, pddf)

        # Print the dataframe head and tail
        # print(stocks.head())
        # print("---")
        # print(stocks.tail())

        # Remove least prominent features - Date, Low and High value
        # In[17]:
        stocks = self.remove_data_with_sentiment(data, pddf)


        # # Plotting and Visualization

        # In[18]:

        plt.rcParams['figure.figsize'] = (18, 12)
        # Raw plotting

        # In[23]:
        # Normalize the data
        # In[25]:
        stocks= self.get_normalised_data_with_sentiment(stocks)
        # print(stocks.head())
        #
        # print("\n")
        # print("Open   --- mean :", np.mean(stocks['Open']), "  \t Std: ", np.std(stocks['Open']), "  \t Max: ",
        #       np.max(stocks['Open']), "  \t Min: ", np.min(stocks['Open']))
        # print("Close  --- mean :", np.mean(stocks['Close']), "  \t Std: ", np.std(stocks['Close']), "  \t Max: ",
        #       np.max(stocks['Close']), "  \t Min: ", np.min(stocks['Close']))
        # print("Volume --- mean :", np.mean(stocks['Volume']), "  \t Std: ", np.std(stocks['Volume']), "  \t Max: ",
        #       np.max(stocks['Volume']), "  \t Min: ", np.min(stocks['Volume']))

        # In[26]:
        #plot_basic(stocks)

        # In[27]:
        stocks.to_csv('./csv/'+symbol+'_preprocessed.csv', index=False)# TODO: UNCOMMENT
        # stocks.to_csv('./csv/' + 'google' + '_preprocessed.csv', index=False)
        # # Stock Data Manipulation

        # In[28]:

        # # linear Regression Benchmark Model

        # In[33]:

        # Load the preprocessed data

        # In[37]:

        # stocks = pd.read_csv('./csv/'+symbol+'_preprocessed.csv')
        # display(stocks.head())
        #
        # # Split data into train and test pairs
        #
        # # In[38]:
        # X_train, X_test, y_train, y_test, label_range = self.train_test_split_linear_regression(stocks)

        # print("x_train", X_train.shape)
        # print("y_train", y_train.shape)
        # print("x_test", X_test.shape)
        # print("y_test", y_test.shape)

        # Train a Linear regressor model on training set and get prediction

        # In[39]:
        # model = self.build_model(X_train, y_train)
        # model.add(Dropout(0.5))
        # Get prediction on test set

        # In[40]:
        # predictions = self.predict_prices(model, X_test, label_range)

        # Plot the predicted values against actual

        # In[41]:
        # self.plot_prediction(y_test, predictions)

        # measure accuracy of the prediction

        # In[42]:
        # trainScore = mean_squared_error(X_train, y_train)
        # print('Train Score: %.4f MSE (%.4f RMSE)' % (trainScore, math.sqrt(trainScore)))
        #
        # testScore = mean_squared_error(predictions, y_test)
        # print('Test Score: %.8f MSE (%.8f RMSE)' % (testScore, math.sqrt(testScore)))

        # ## Long-Sort Term Memory Model
        #
        # LSTM  train and test phases

        # In[43]:

        # In[46]:

        ## Data setup: ##
        stocks = pd.read_csv('./csv/'+symbol+'_preprocessed.csv') #TODO: UNCOMMENT
        # stocks = pd.read_csv('./csv/'+'google'+'_preprocessed.csv')
        stocks_data = stocks.drop(['Item'], axis=1)

        #display(stocks_data.head()) #TODO: UNCOMMENT

        # Split train and test data sets and Unroll train and test data for lstm model
        # TODO: WHY DOES MODEL PREDICT ACCORDING TO HISTORY AND IGNORES SENTIMENT? MAYBE BECAUSE OF SHAPE?
        # In[47]:
        z = self.train_test_split_lstm(stocks_data)
        X_train, X_test, y_train, y_test = z

        # print("PRINT MODEL:" + str(X_train))

        unroll_length = 50
        # X_train = unroll(X_train, unroll_length)
        # X_test = unroll(X_test, unroll_length)
        y_train = y_train[-X_train.shape[0]:]
        y_test = y_test[-X_test.shape[0]:]

        # y_train = np.reshape(y_train, (y_train.shape[0], 1, y_train.shape[1]))
        # y_test = np.reshape(y_test, (y_test.shape[0], 1, y_test.shape[1]))

        X_train = np.reshape(X_train, (X_train.shape[0], 1, X_train.shape[1]))# TODO: problem is probably here, original values were ['Open', 'Close', 'Volume'], but now we have ['Open', 'Close', 'Volume','Positive','Neutral','Negative','Compound']
        X_test = np.reshape(X_test, (X_test.shape[0], 1, X_test.shape[1]))# TODO: problem is probably here, original values were ['Open', 'Close', 'Volume'], but now we have ['Open', 'Close', 'Volume','Positive','Neutral','Negative','Compound']
        #
        # y_train = np.reshape(y_train, (-X_train.shape[0], 1, y_train.shape[1]))
        # y_test = np.reshape(y_test, (-X_test.shape[0], 1, y_test.shape[1]))

        ## TODO PRINTS ##
        # print("X_train: "+str(X_train[0]))
        # print("x_train", X_train.shape)
        # print("y_train", y_train.shape)
        # print("x_test", X_test.shape)
        # print("y_test", y_test.shape)

        #  Build a basic Long-Short Term Memory mode

        # In[48]:
        # build basic lstm model
        # model = self.build_basic_model(input_dim=X_train.shape[-1], output_dim=unroll_length, return_sequences=True)
        #
        # # Compile the model
        # start = time.time()
        # model.compile(loss='mean_squared_error', optimizer='adam')
        # print('compilation time : ', time.time() - start)

        # # Train the model
        # # In[49]:
        # model.fit(
        #     X_train,
        #     y_train,
        #     batch_size=1,
        #     epochs=1,
        #     validation_split=0.05) # TODO: WHAT DOES THIS DO
        #
        # # Predict


        # In[149]:
        '''data = pd.read_csv('goog.csv')
        print(data.head())
        print(data.tail())
        
        stocks = remove_data(data)
        stocks = get_normalised_data(stocks)
        stocks = stocks.drop(['Item'], axis = 1)
        
        #Print the dataframe head and tail
        print(stocks.head())
        
        X = stocks[:].as_matrix()
        Y = stocks[:]['Close'].as_matrix()
        X = sd.unroll(X,1)
        Y = Y[-X.shape[0]:]
        
        print(X.shape)
        print(Y.shape)
        
        # Generate predictions 
        predictions = model.predict(X)
        
        #get the test score
        testScore = model.evaluate(X, Y, verbose=0)
        print('Test Score: %.4f MSE (%.4f RMSE)' % (testScore, math.sqrt(testScore)))'''

        ## Basic model Plotting ##
        # predictions = model.predict(X_test)
        # print('X_test: '+str(X_test))
        # # Plot results
        #
        # # In[150]:
        # print("PRINTGIN BLUE")
        # self.plot_lstm_prediction(predictions, y_test)

        # Get Test Scores
        # trainScore = model.evaluate(X_train, y_train, verbose=0)
        # print('Train Score: %.8f MSE (%.8f RMSE)' % (trainScore, math.sqrt(trainScore)))
        #
        # testScore = model.evaluate(X_test, y_test, verbose=0)
        # print('Test Score: %.8f MSE (%.8f RMSE)' % (testScore, math.sqrt(testScore)))

        ## Advanced LSTM Model ##

        # In[152]:
        # Set up hyperparameters
        batch_size = 512
        epochs = 20

        # build improved lstm model
        model = self.build_improved_model(X_train.shape[-1], output_dim=unroll_length, return_sequences=True)

        start = time.time()
        # final_model.compile(loss='mean_squared_error', optimizer='adam')
        model.compile(loss='mean_squared_error', optimizer='adam')
        print('compilation time : ', time.time() - start)

        # Train improved LSTM model

        # In[153]:
        model.fit(X_train,
                  y_train,
                  batch_size=batch_size,
                  epochs=epochs,
                  verbose=2,
                  validation_split=0.05 # WHAT DOES THIS DO
                  )

        # Make prediction on improved LSTM model

        # In[159]:
        # Generate predictions
        predictions = model.predict(X_test, batch_size=batch_size)
        # print('ACTUAL: ' + str(stocks))
        # print('PREDICT: ' + str(predictions))
        # mx = predictions.max()
        # print("MX: "+str(mx))
        # mn = predictions.min()

        # mx = df.max()
        # mn = df.min()

        # TODO: UNCOMMENT HALP #
        # print("PREDICITONS HALP:"+str(predictions))
        # print("STOCKS HALP:" + str(stocks))
        #
        # # df = df.transform(lambda x: (x * (mx - mn)) + mn)
        # mx = predictions.max()
        # mn = predictions.min()
        # df = pd.DataFrame(predictions)
        # df = df.transform(lambda x: (x * (mx - mn)) + mn)
        # df.to_csv("AAPL_PREDICTIONS.csv")
        #
        # mx = stocks["Open"].max()
        # mn = stocks["Open"].min()
        # df = pd.DataFrame(stocks)
        # df = df.transform(lambda x: (x * (mx - mn)) + mn)
        #
        # df.to_csv("AAPL_STOCKS.csv")
        # TODO: UNCOMMENT HALP #
        # df = pd.DataFrame(predictions)
        # df.to_csv("GOOGLE_PREDICTIONS.csv")
        #
        # stocks.to_csv("GOOGLE_STOCKS.csv")
        # TODO: UNCOMMENT HALP #
        #predictions.to_csv('./csv/'+"AAPL2"+'.csv', index=False)
        # In[160]:

        #self.plot_lstm_prediction(predictions, y_test) #TODO: UNCOMMENT

        # Get test score

        trainScore = model.evaluate(X_train, y_train, verbose=0)

        ## TODO PRINTS ##
        # print('Train Score: %.8f MSE (%.8f RMSE)' % (trainScore, math.sqrt(trainScore)))
        testScore = model.evaluate(X_test, y_test, verbose=0)
        ## TODO PRINTS ##
        # print('Test Score: %.8f MSE (%.8f RMSE)' % (testScore, math.sqrt(testScore)))
        range = [np.amin(stocks_data['Close']), np.amax(stocks_data['Close'])]

        # Calculate the stock price delta in $
        true_delta = testScore * (range[1] - range[0])
        ## TODO PRINTS ##
        # print('Delta Price: %.6f - RMSE * Adjusted Close Range' % true_delta)
        # print("PREDICTIONS: "+str(predictions["Close"]))
        # print("Actual: "+str(stocks))

        # TODO: self.stockDataDict[symbol] = <calculate score for symb> #

        self.stockDataDict[symbol] = random.randint(0,100)# TODO: REMOVE PLACEHOLDER

        # def get_recommendation():
            # grab the last 3 items in the most current frame (last nested loop) and covert closing prices to a 1-D vector
        step_matrix = np.array(X_test[-1][-3:][:, 1])
        predicted_close = step_matrix.mean()
        predicted_open = X_test[-1][-1][1]
        score = ((predicted_close - predicted_open) * 1000)
        return score



sm = StockModel()
dict = {}
sip = StockInfoProvider()
for symbol in sip.getAllStocks():
    try:
        sm.stockDataDict[symbol] = sm.start(symbol, "2005-01-01", "2018-05-27")
    except (urllib_err.HTTPError, TypeError):
        continue