khuangaf

ground_true_df = pd.DataFrame()
ground_true_df['times'] = ground_true_times
ground_true_df['value'] = ground_true

prediction_df = pd.DataFrame()
prediction_df['times'] = validation_output_times
prediction_df['value'] = predicted_inverted

prediction_df = prediction_df.loc[(prediction_df["times"].dt.year == 2017 )&(prediction_df["times"].dt.month > 7 ),: ]
ground_true_df = ground_true_df.loc[(ground_true_df["times"].dt.year == 2017 )&(ground_true_df["times"].dt.month > 7 ),:]

khuangaf

plt.figure(figsize=(20,10))
plt.plot(ground_true_df.times,ground_true_df.value, label = 'Actual')
plt.plot(prediction_df.times,prediction_df.value,'ro', label='Predicted')
plt.legend(loc='upper left')
plt.show()

khuangaf

def fit_lstm(reg):
    global training_datas, training_labels, batch_size, epochs,step_size,nb_features, units
    model = Sequential()
    model.add(CuDNNLSTM(units=units, bias_regularizer=reg, input_shape=(step_size,nb_features),return_sequences=False))
    model.add(Activation('tanh'))
    model.add(Dropout(0.2))
    model.add(Dense(output_size))
    model.add(LeakyReLU())
    model.compile(loss='mse', optimizer='adam')
    model.fit(training_datas, training_labels, batch_size=batch_size, epochs = epochs, verbose=0)

khuangaf

def experiment(validation_datas,validation_labels,original_datas,ground_true,ground_true_times,validation_original_outputs, validation_output_times, nb_repeat, reg):
    error_scores = list()
    #get only the close data
    ground_true = ground_true[:,:,0].reshape(-1)
    ground_true_times = ground_true_times.reshape(-1)
    ground_true_times = pd.to_datetime(ground_true_times, unit='s')
    validation_output_times = pd.to_datetime(validation_output_times.reshape(-1), unit='s')
    for i in range(nb_repeat):
        model = fit_lstm(reg)
        predicted = model.predict(validation_datas)

khuangaf

results.describe().boxplot()
plt.show()

khuangaf

class DataSrc(object):
    """Acts as data provider for each new episode."""

    def __init__(self, df, steps=252, scale=True, scale_extra_cols=True, augment=0.00, window_length=50):
        """
        DataSrc.
        df - csv for data frame index of timestamps
             and multi-index columns levels=[['LTCBTC'],...],['open','low','high','close',...]]
             an example is included as an hdf file in this repository
        steps - total steps in episode

khuangaf

class DataSrc(object):
    
    def _step(self):
        # get history matrix from dataframe
        data_window = self.data[:, self.step:self.step +
                                self.window_length].copy()

        # (eq.1) prices
        y1 = data_window[:, -1, 0] / data_window[:, -2, 0]
        y1 = np.concatenate([[1.0], y1])  # add cash price

khuangaf

class PortfolioSim(object):
  def _step(self, w1, y1):
    """
    Step.
    w1 - new action of portfolio weights - e.g. [0.1,0.9, 0.0]
    y1 - price relative vector also called return
        e.g. [1.0, 0.9, 1.1]
    Numbered equations are from https://arxiv.org/abs/1706.10059
    """
    w0 = self.w0

khuangaf

class PortfolioEnv(gym.Env):
  def _step(self, action):
    """
    Step the env.
    Actions should be portfolio [w0...]
    - Where wn is a portfolio weight between 0 and 1. The first (w0) is cash_bias
    - cn is the portfolio conversion weights see PortioSim._step for description
    """
    logger.debug('action: %s', action)

khuangaf

import gym.spaces
import gym.wrappers
import numpy as np


def concat_states(state):
    history = state["history"]
    weights = state["weights"]
    weight_insert_shape = (history.shape[0], 1, history.shape[2])
    if len(weights) - 1 == history.shape[0]: