import tensorflow as tf
import numpy as np
import datetime
import os.path
np.random.seed(4)
tf.random.set_seed(4)
from util import csv_to_dataset, history_points
# dataset
#File that will be used
#csv_path = "ETHUSDT-1d-data.csv"
csv_path = "ETHBTC-1h-data.csv"
#ohlcv_histories, technical_indicators, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset('ETHUSDT-1h-data.csv')
ohlcv_histories_train, ohlcv_histories_test, unscaled_y_test, y_train, y_test, y_normaliser, tech_ind_train, tech_ind_test = csv_to_dataset(
csv_path)
# model architecture
bs = 2048
e = 5000
# define two sets of inputs
lstm_input = tf.keras.layers.Input(shape=(history_points, 5),
name='lstm_input')
dense_input = tf.keras.layers.Input(shape=(tech_ind_train.shape[1], ),
name='tech_input')
# the first branch operates on the first input
x = tf.keras.layers.LSTM(history_points, name='lstm_0')(lstm_input)
x = tf.keras.layers.Dropout(0.2, name='lstm_dropout_0')(x)
lstm_branch = tf.keras.models.Model(inputs=lstm_input, outputs=x)
def main():
"""TODO:
"""
params = Parameters()
model = tf.keras.models.load_model(get_best_model_path(params), compile=True)
ohlcv_histories, technical_indicators, next_day_open_values, unscaled_y, y_normaliser = \
csv_to_dataset(params.CSV_DATA_PATH, params.num_history_points)
n = int(ohlcv_histories.shape[0] * params.train_split)
ohlcv_test = ohlcv_histories[n:]
tech_ind_test = technical_indicators[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
y_test_predicted = model.predict([ohlcv_test, tech_ind_test])
y_test_predicted = y_normaliser.inverse_transform(y_test_predicted)
buys = []
sells = []
# thresh = 0.7
thresh = 0.1
start = 0
end = -1
x = -1
holding_stocks = False
ratio_treshold = 0.005
def compute_earnings(buys_, sells_):
# purchase amount
purchase_amt = 10
stock = 0
balance = 0
while len(buys_) > 0 and len(sells_) > 0:
if buys_[0][0] < sells_[0][0]:
# time to buy $10 worth of stock
balance -= purchase_amt
stock += purchase_amt / buys_[0][1]
buys_.pop(0)
else:
# time to sell all of our stock
balance += stock * sells_[0][1]
stock = 0
sells_.pop(0)
print(f"earnings: ${balance}")
for ohlcv, ind in zip(ohlcv_test[start: end], tech_ind_test[start: end]):
normalised_price_today = ohlcv[-1][0]
normalised_price_today = np.array([[normalised_price_today]])
price_today = y_normaliser.inverse_transform(normalised_price_today)
predicted_price_tomorrow = np.squeeze(y_normaliser.inverse_transform(model.predict([[ohlcv], [ind]])))
###### HIS CODE ########
# delta = predicted_price_tomorrow - price_today
# price_today = price_today[0][0]
#
# ratio = predicted_price_tomorrow / price_today
# ratio_threshold = 0.02
#
# if ratio > 1 + ratio_threshold:
# buys.append((x, price_today))
# elif ratio < 1 - ratio_threshold:
# sells.append((x, price_today))
#
# if delta > thresh and ratio > 1 + ratio_threshold:
# buys.append((x, price_today))
# elif delta < -thresh and ratio < 1 - ratio_threshold:
# sells.append((x, price_today))
##### START STEFAN ######
abs_price_today = price_today[0][0]
if holding_stocks:
price = buys[-1][1]
else:
price = abs_price_today
# Absolute diff of price
ratio = predicted_price_tomorrow / price
if abs_price_today / price >= 1.3 and holding_stocks:
sells.append((x, abs_price_today))
holding_stocks = False
elif ratio >= ratio_treshold and not holding_stocks:
buys.append((x, abs_price_today))
holding_stocks = True
print("BUY: price today", abs_price_today, " predicted price tomorrow",
predicted_price_tomorrow)
elif ratio < 1 - ratio_treshold and holding_stocks:
sells.append((x, abs_price_today))
holding_stocks = False
###################################################
x += 1
print(f"buys: {len(buys)}")
print(f"sells: {len(sells)}")
# we create new lists so we don't modify the original
compute_earnings([b for b in buys], [s for s in sells])
import matplotlib.pyplot as plt
plt.gcf().set_size_inches(22, 15, forward=True)
real = plt.plot(unscaled_y_test[start:end], label='real')
pred = plt.plot(y_test_predicted[start:end], label='predicted')
if len(buys) > 0:
plt.scatter(list(list(zip(*buys))[0]), list(list(zip(*buys))[1]), c='#00ff00', s=50)
if len(sells) > 0:
plt.scatter(list(list(zip(*sells))[0]), list(list(zip(*sells))[1]), c='#ff0000', s=50)
# TRAIN TEST
# ohlcv_train = ohlcv_histories[:n]
# tech_ind_train = technical_indicators[:n]
# y_train = next_day_open_values[:n]
# real = plt.plot(unscaled_y[start:end], label='real')
# pred = plt.plot(y_predicted[start:end], label='predicted')
plt.legend(['Real', 'Predicted', 'Buy', 'Sell'])
plt.show()
import keras
from keras.models import Model
from keras.layers import Dense, Dropout, LSTM, Input, Activation
from keras import optimizers
import numpy as np
np.random.seed(4)
import tensorflow as tf
tf.random.set_seed(4)
from util import csv_to_dataset, history_points
import matplotlib.pyplot as plt
if __name__ == '__main__':
data_histories, _, next_day, unscaled_y, y_normaliser = csv_to_dataset(
'MSFT_daily.csv')
test_split = 0.9
n = int(data_histories.shape[0] * test_split)
data_train = data_histories[:n]
y_train = next_day[:n]
data_test = data_histories[n:]
y_test = next_day[n:]
unscaled_y_test = unscaled_y[n:]
print(data_train.shape)
print(data_test.shape)
# model code
lstm_input = Input(shape=(history_points, 5), name='input')
x = LSTM(50, name='lstm_layer')(lstm_input)
def main():
"""
:return:
"""
params = Parameters()
if not params.SAVE_MODEL_PATH.exists():
params.SAVE_MODEL_PATH.mkdir()
print("Write params in JSON")
json_string = Parameters().to_json()
params_path = params.SAVE_MODEL_PATH / "params.json"
with open(params_path.as_posix(), "w") as f:
f.write(json_string)
ohlcv_histories, technical_indicators, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
params.CSV_DATA_PATH, params.num_history_points)
print("----", ohlcv_histories.shape, technical_indicators.shape,
next_day_open_values.shape)
n = int(ohlcv_histories.shape[0] * params.train_split)
ohlcv_train = ohlcv_histories[:n]
tech_ind_train = technical_indicators[:n]
y_true_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
tech_ind_test = technical_indicators[n:]
# y_true_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
print("train_val split", ohlcv_train.shape)
print("test split", ohlcv_test.shape)
callbacks: List[tf.keras.callbacks.Callback] = [
tf.keras.callbacks.TensorBoard(params.SAVE_MODEL_PATH.as_posix()),
tf.keras.callbacks.ModelCheckpoint(
get_best_model_path(params),
monitor='val_loss',
verbose=1,
save_best_only=True,
)
]
# model = tech_net(technical_indicators.shape[1:], params)
#
# adam = optimizers.Adam(lr=params.LR)
# model.compile(optimizer=adam, loss='mse')
#
# model.fit(x=[ohlcv_train, tech_ind_train], y=y_true_train, batch_size=params.BATCH_SIZE, epochs=params.EPOCHS,
# shuffle=True, validation_split=params.val_split_out_of_train,
# callbacks=callbacks)
evaluate(params, ohlcv_test, tech_ind_test, unscaled_y_test, y_normaliser)
from keras.models import Model
from keras.layers import Dense, Dropout, LSTM, Input, Activation
from keras import optimizers
from util import csv_to_dataset, history_points
import numpy as np
import tensorflow
# np.random.seed(4)
# from tensorflow import set_random_seed
# set_random_seed(4)
# np.random.seed(4)
# set_random_seed(4)
tensorflow.random.set_seed(4)
# dataset testing and trainig
ohlcv_histories, _, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
'MSFT_intraday.csv')
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
print(ohlcv_train.shape)
print(ohlcv_test.shape)
import keras
import tensorflow as tf
from keras.models import Model
from keras.layers import Dense, Dropout, LSTM, Input, Activation, concatenate
from keras import optimizers
import numpy as np
np.random.seed(16)
import tensorflow as tf
tf.random.set_seed(16)
from util import csv_to_dataset, history_points
import matplotlib.pyplot as plt
# dataset
if __name__ == '__main__':
data_histories, moving_av, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
'MSFT_daily.csv')
test_split = 0.9
n = int(data_histories.shape[0] * test_split)
data_train = data_histories[:n]
moving_av_train = moving_av[:n]
y_train = next_day_open_values[:n]
data_test = data_histories[n:]
moving_av_test = moving_av[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
#print(data_train.shape)
#print(data_test.shape)
# model code
lstm_input = Input(shape=(history_points, 5), name='lstm_input')
def get_predict(model_path='technical_model.h5',
symbol='GOOGL',
update='daily'):
stock = './data/' + symbol.upper() + "_" + update.lower() + ".csv"
print(stock)
print(model_path)
model = load_model(model_path)
ohlcv_histories, technical_indicators, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
stock)
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
tech_ind_train = technical_indicators[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
tech_ind_test = technical_indicators[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
y_test_predicted = model.predict([ohlcv_test, tech_ind_test])
y_test_predicted = y_normaliser.inverse_transform(y_test_predicted)
buys = []
sells = []
thresh = 0.1
start = 0
end = -1
x = -1
for ohlcv, ind in zip(ohlcv_test[start:end], tech_ind_test[start:end]):
normalised_price_today = ohlcv[-1][0]
normalised_price_today = np.array([[normalised_price_today]])
price_today = y_normaliser.inverse_transform(normalised_price_today)
predicted_price_tomorrow = np.squeeze(
y_normaliser.inverse_transform(model.predict([[ohlcv], [ind]])))
delta = predicted_price_tomorrow - price_today
if delta > thresh:
buys.append((x, price_today[0][0]))
elif delta < -thresh:
sells.append((x, price_today[0][0]))
x += 1
print(f"buys: {len(buys)}")
print(f"sells: {len(sells)}")
def compute_earnings(buys_, sells_):
purchase_amt = 10
stock = 0
balance = 0
while len(buys_) > 0 and len(sells_) > 0:
if buys_[0][0] < sells_[0][0]:
# time to buy $10 worth of stock
balance -= purchase_amt
stock += purchase_amt / buys_[0][1]
buys_.pop(0)
else:
# time to sell all of our stock
balance += stock * sells_[0][1]
stock = 0
sells_.pop(0)
print(f"earnings: ${balance}")
# we create new lists so we dont modify the original
compute_earnings([b for b in buys], [s for s in sells])
import matplotlib.pyplot as plt
plt.gcf().set_size_inches(22, 15, forward=True)
real = plt.plot(unscaled_y_test[start:end], label='real')
pred = plt.plot(y_test_predicted[start:end], label='predicted')
if len(buys) > 0:
plt.scatter(list(list(zip(*buys))[0]),
list(list(zip(*buys))[1]),
c='#00ff00',
s=50)
if len(sells) > 0:
plt.scatter(list(list(zip(*sells))[0]),
list(list(zip(*sells))[1]),
c='#ff0000',
s=50)
# real = plt.plot(unscaled_y[start:end], label='real')
# pred = plt.plot(y_predicted[start:end], label='predicted')
plt.legend(['Real', 'Predicted', 'Buy', 'Sell'])
plt.show()
import tensorflow as tf
from keras.models import Model
from keras.layers import Dense, Dropout, LSTM, Input, Activation, concatenate
from keras import optimizers
import numpy as np
np.random.seed(4)
from tensorflow import set_random_seed
set_random_seed(4)
from util import csv_to_dataset
history_points = 50
# dataset
ohlcv_histories, technical_indicators, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
'data_daily/AAPL_daily.csv', history_points, 0, False)
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
tech_ind_train = technical_indicators[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
tech_ind_test = technical_indicators[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
print(ohlcv_train.shape)
def train_model(options):
info = get_info_from_options(options)
file_name = info['file_name']
moving_average = info['moving_average']
epochs = info['epochs']
save_name = info['model_name']
model_base_name = info['model_base_name']
print(f'{model_base_name}')
if os.path.isfile(save_name) == True:
print(f'{model_base_name} trained')
return
try:
ohlcv_histories, technical_indicators, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
file_name)
except ():
print('data parse false')
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
tech_ind_train = technical_indicators[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
tech_ind_test = technical_indicators[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
# model architecture
# define two sets of inputs
lstm_input = Input(shape=(history_points, 5), name='lstm_input')
dense_input = Input(shape=(technical_indicators.shape[1], ),
name='tech_input')
# the first branch operates on the first input
x = LSTM(moving_average, name='lstm_0')(lstm_input)
x = Dropout(0.2, name='lstm_dropout_0')(x)
lstm_branch = Model(inputs=lstm_input, outputs=x)
# the second branch opreates on the second input
y = Dense(20, name='tech_dense_0')(dense_input)
y = Activation("relu", name='tech_relu_0')(y)
y = Dropout(0.2, name='tech_dropout_0')(y)
technical_indicators_branch = Model(inputs=dense_input, outputs=y)
# combine the output of the two branches
combined = concatenate(
[lstm_branch.output, technical_indicators_branch.output],
name='concatenate')
z = Dense(64, activation="sigmoid", name='dense_pooling')(combined)
z = Dense(1, activation="linear", name='dense_out')(z)
# our model will accept the inputs of the two branches and
# then output a single value
model = Model(
inputs=[lstm_branch.input, technical_indicators_branch.input],
outputs=z)
adam = optimizers.Adam(lr=0.0005)
model.compile(optimizer=adam, loss='mse')
model.fit(x=[ohlcv_train, tech_ind_train],
y=y_train,
batch_size=64,
epochs=epochs,
shuffle=True,
validation_split=0.1)
# evaluation
y_test_predicted = model.predict([ohlcv_test, tech_ind_test])
# print('y_test_predicted.shape', y_test_predicted.shape)
y_test_predicted = y_normaliser.inverse_transform(y_test_predicted)
y_predicted = model.predict([ohlcv_histories, technical_indicators])
y_predicted = y_normaliser.inverse_transform(y_predicted)
assert unscaled_y_test.shape == y_test_predicted.shape
real_mse = np.mean(np.square(unscaled_y_test - y_test_predicted))
scaled_mse = real_mse / (np.max(unscaled_y_test) -
np.min(unscaled_y_test)) * 100
# print(scaled_mse)
plt.gcf().set_size_inches(22, 15, forward=True)
start = 0
end = -1
real = plt.plot(unscaled_y_test[start:end], label='real')
pred = plt.plot(y_test_predicted[start:end], label='predicted')
plt.legend(['Real', 'Predicted'])
# plt.show()
model.save(save_name)
import numpy as np
from tensorflow import keras
from util import csv_to_dataset, history_points
from binance.client import Client
from binance.enums import *
model = keras.models.load_model('./save_model')
ohlcv_histories, technical_indicators, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset('BNBUSDT_1d_2021-01-01 00-00-00_2021-04-15 00-00-00.csv')
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
tech_ind_train = technical_indicators[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
tech_ind_test = technical_indicators[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
y_test_predicted = model.predict([ohlcv_test, tech_ind_test])
y_test_predicted = y_normaliser.inverse_transform(y_test_predicted)
buys = []
sells = []
thresh = 0.1
def predict_model(options):
info = get_info_from_options(options)
file_name = info['file_name']
modal_name = info['model_name']
start = info['start']
end = info['end']
model_base_name = info['model_base_name']
predict_delay_session_list = info['predict_delay_session_list']
check_result_already_exist_value = check_result_already_exist(options)
if check_result_already_exist_value == True:
print(f'{model_base_name} predicted')
return False
model = load_model(modal_name)
ohlcv_histories, technical_indicators, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
file_name)
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
tech_ind_train = technical_indicators[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
tech_ind_test = technical_indicators[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
y_test_predicted = model.predict([ohlcv_test, tech_ind_test])
y_test_predicted = y_normaliser.inverse_transform(y_test_predicted)
buys = []
sells = []
thresh = 0.1
# start = 0
# end = -1
x = 0
todayPriceList = []
predictedPriceList = []
length = len(ohlcv_test[start:end])
for ohlcv, ind in zip(ohlcv_test[start:end], tech_ind_test[start:end]):
normalised_price_today = ohlcv[-1][0]
normalised_price_today = np.array([[normalised_price_today]])
price_today = y_normaliser.inverse_transform(normalised_price_today)
ohlcv_predict = np.array(ohlcv, ndmin=3) # fix dimension
ind_predict = np.array(ind, ndmin=2) # fix dimension
predicted_price_tomorrow = np.squeeze(
y_normaliser.inverse_transform(
model.predict([ohlcv_predict, ind_predict])))
todayPriceList.append(price_today[0][0])
predictedPriceList.append(float(predicted_price_tomorrow))
delta = predicted_price_tomorrow - price_today
if delta > thresh:
buys.append((x, price_today[0][0]))
elif delta < -thresh:
sells.append((x, price_today[0][0]))
x += 1
# print(x, '/', length)
# print(f"buys: {len(buys)}")
# print(f"sells: {len(sells)}")
# def compute_earnings(buys_, sells_):
# purchase_amt = 10
# stock = 0
# balance = 0
# while len(buys_) > 0 and len(sells_) > 0:
# if buys_[0][0] < sells_[0][0]:
# # time to buy $10 worth of stock
# balance -= purchase_amt
# stock += purchase_amt / buys_[0][1]
# buys_.pop(0)
# else:
# # time to sell all of our stock
# balance += stock * sells_[0][1]
# stock = 0
# sells_.pop(0)
# print(f"earnings: ${balance}")
# we create new lists so we dont modify the original
# compute_earnings([b for b in buys], [s for s in sells])
# plt.gcf().set_size_inches(15, 10, forward=True)
# real = plt.plot(todayPriceList, label='real')
# pred = plt.plot(predictedPriceList, label='predicted')
# if len(buys) > 0:
# plt.scatter(list(list(zip(*buys))[0]),
# list(list(zip(*buys))[1]), c='#00ff00', s=30)
# if len(sells) > 0:
# plt.scatter(list(list(zip(*sells))[0]),
# list(list(zip(*sells))[1]), c='#ff0000', s=30)
# caculate buy sell with difference predicted price
buyWithDifference = []
sellWithDifference = []
buy_sell_difference = []
rootPredictedPriceList = predictedPriceList[0]
for index, predictedValue in enumerate(predictedPriceList):
if index > 0:
if predictedValue > predictedPriceList[index - 1]:
buyWithDifference.append(
(index - 1, predictedPriceList[index - 1]))
buy_sell_difference.append('buy')
else:
sellWithDifference.append(
(index - 1, predictedPriceList[index - 1]))
buy_sell_difference.append('sell')
# plt.scatter(list(list(zip(*buyWithDifference))[0]),
# list(list(zip(*buyWithDifference))[1]), c='#00ff00', s=10)
# plt.scatter(list(list(zip(*sellWithDifference))[0]),
# list(list(zip(*sellWithDifference))[1]), c='#ff0000', s=10)
# plt.legend(['Real', 'Predicted', 'Buy', 'Sell',
# 'buyWithDifference', 'sellWithDifference'])
# caculate accuracy
# print('buyWithDifference', len(buyWithDifference))
# print('sellWithDifference', len(sellWithDifference))
# print('todayPriceList', len(todayPriceList))
def caculate_accuracy_buy_sell_with_delay_session(deal_type,
delay_session):
total_caculate_deal = 0
true_deal = 0
false_deal = 0
if (deal_type == 'buy'):
list_data = buyWithDifference
else:
list_data = sellWithDifference
# remove last item because list predict from 0 to length - 1
length_real_data = len(todayPriceList) - 1
for index, deal_predicted_price in list_data:
# print(index, deal_predicted_price)
index_has_delay_session = index + delay_session
if index_has_delay_session <= length_real_data:
total_caculate_deal += 1
deal_real_price = todayPriceList[index]
delay_session_price = todayPriceList[index_has_delay_session]
# print('dealRealPrice', deal_real_price,
# 'delay_sessionPrice', delay_session_price)
if (deal_type == 'buy'):
if (delay_session_price > deal_real_price):
true_deal += 1
else:
false_deal += 1
else:
if (delay_session_price < deal_real_price):
true_deal += 1
else:
false_deal += 1
if total_caculate_deal > 0:
print(deal_type, ': total deal: ', len(list_data),
', total caculate deal: ', total_caculate_deal, ', true: ',
true_deal, ', false: ', false_deal, ' accuracy: ',
round(true_deal / total_caculate_deal * 100, 1), '%')
return [
deal_type,
len(list_data), total_caculate_deal, true_deal, false_deal,
round(true_deal / total_caculate_deal * 100, 1)
]
# return [deal_type, len(list_data), total_caculate_deal, true_deal, false_deal,'no_signed_deal']
else:
print(deal_type, ': total deal: ', len(list_data),
', total caculate deal: ', total_caculate_deal, ', true: ',
true_deal, ', false: ', false_deal)
return [
deal_type,
len(list_data), total_caculate_deal, true_deal, false_deal,
'no_signed_deal'
]
def caculate_accuracy_predicted_with_delay_session(delay_session):
print('delay_session: ', delay_session)
buy_result = caculate_accuracy_buy_sell_with_delay_session(
'buy', delay_session)
# sell_result = caculate_accuracy_buy_sell_with_delay_session('sell', delay_session)
# print('total: total deal: ', buy_result[1]+sell_result[1], ', total caculate deal: ',
# buy_result[2] + sell_result[2], ', true: ', buy_result[3] +
# sell_result[3], ', false: ', buy_result[4] +
# sell_result[4], ' accuracy: ',
# ((buy_result[3] +
# sell_result[3])/(buy_result[2] + sell_result[2])*100), '%')
return buy_result
buy_result_list = {}
for delay_session in predict_delay_session_list:
buy_result = caculate_accuracy_predicted_with_delay_session(
delay_session)
buy_result_list[delay_session] = buy_result
return {
'options': options,
'buy_result_list': buy_result_list,
'predicted_price_list': predictedPriceList,
'buy_sell_difference': buy_sell_difference
}
np.random.seed(4)
import tensorflow
tensorflow.random.set_seed(4)
from util import csv_to_dataset, history_points
import argparse
parser = argparse.ArgumentParser(description='File to process')
parser.add_argument('dataFile', help='Data source file path')
parser.add_argument('saveToFile', help='Where to save the built model')
args = parser.parse_args()
# dataset
ohlcv_histories, _, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(args.dataFile)
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
print(ohlcv_train.shape)
print(ohlcv_test.shape)
def model_training(filepath, model, history_points=50, offset=0):
ohlcv_histories, _, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(filepath, history_points,
offset=offset)
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
# model architecture
model = model(history_points)
adam = optimizers.Adam(lr=0.0005)
model.compile(optimizer=adam, loss='mse')
trained_model = model.fit(x=ohlcv_train, y=y_train, batch_size=32, epochs=100, shuffle=True, validation_split=0.1)
# evaluation
y_test_predicted = model.predict(ohlcv_test)
y_test_predicted = y_normaliser.inverse_transform(y_test_predicted)
y_predicted = model.predict(ohlcv_histories)
y_predicted = y_normaliser.inverse_transform(y_predicted)
assert unscaled_y_test.shape == y_test_predicted.shape
real_mse = np.mean(np.square(unscaled_y_test - y_test_predicted))
scaled_mse = real_mse / (np.max(unscaled_y_test) - np.min(unscaled_y_test)) * 100
print(scaled_mse)
import matplotlib.pyplot as plt
plt.gcf().set_size_inches(22, 15, forward=True)
start = 0
end = -1
real = plt.plot(unscaled_y_test[start:end], label='real')
pred = plt.plot(y_test_predicted[start:end], label='predicted')
# real = plt.plot(unscaled_y[start:end], label='real')
# pred = plt.plot(y_predicted[start:end], label='predicted')
plt.title(filepath)
plt.legend(['Real', 'Predicted'])
plt.show()
from datetime import datetime
model.save(f'basic_model.h5')
history = trained_model.history
plt.title(filepath)
plt.plot(history['loss'], label="Loss")
plt.plot(history['val_loss'], label="Val Loss")
plt.plot([0, len(history['val_loss'])], [history['val_loss'][-1], history['val_loss'][-1]], ls='-')
plt.show()
return mse(unscaled_y_test, y_test_predicted)
from keras.layers import Dense, Dropout, LSTM, Input, Activation
from keras import optimizers
import numpy as np
np.random.seed(4)
from tensorflow import set_random_seed
set_random_seed(4)
from util import csv_to_dataset
from params import num_history_points
# dataset
CSV_DATA_PATH: str = "/media/dorel/DATA/work/stock-trading-ml/data/msft_daily.csv"
ohlcv_histories, _, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
CSV_DATA_PATH, )
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
print(ohlcv_train.shape)
print(ohlcv_test.shape)
import keras
from keras.models import Model
from keras.layers import Dense, Dropout, LSTM, Input, Activation
from keras import optimizers
import numpy as np
np.random.seed(4)
import tensorflow as tf
tf.random.set_seed(4)
from util import csv_to_dataset, history_points
from binance.client import Client
from binance.enums import *
# dataset
ohlcv_histories, _, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
csv_path='BNBUSDT_1d_2021-01-01 00-00-00_2021-04-15 00-00-00.csv')
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
print(ohlcv_train.shape)
print(ohlcv_test.shape)
import numpy as np
from tensorflow import keras
from util import csv_to_dataset, history_points
from binance.client import Client
from binance.enums import *
model = keras.models.load_model('./save_model')
ohlcv_histories, technical_indicators, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(symbol="BNBUSDT", interval=Client.KLINE_INTERVAL_1MINUTE, start_time=1618419600000)
test_split = 0.4
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
tech_ind_train = technical_indicators[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
tech_ind_test = technical_indicators[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
y_test_predicted = model.predict([ohlcv_test, tech_ind_test])
y_test_predicted = y_normaliser.inverse_transform(y_test_predicted)
start = 0
end = -1
import matplotlib.pyplot as plt
import keras
from keras.models import Model
from keras.layers import Dense, Dropout, LSTM, Input, Activation
from keras import optimizers
import numpy as np
np.random.seed(4)
import tensorflow
tensorflow.random.set_seed(4)
from util import csv_to_dataset, history_points
# dataset
ohlcv_histories, _, next_day_open_values, unscaled_y, next_day_open_higher, y_normaliser = csv_to_dataset('final/TX_daily.csv')
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
print(ohlcv_train.shape)
print(ohlcv_test.shape)
# model architecture
import numpy as np
from keras.models import load_model
from util import csv_to_dataset, history_points
model = load_model('technical_model.h5')
ohlcv_histories, technical_indicators, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
'MSFT_daily.csv')
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
tech_ind_train = technical_indicators[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
tech_ind_test = technical_indicators[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
y_test_predicted = model.predict([ohlcv_test, tech_ind_test])
y_test_predicted = y_normaliser.inverse_transform(y_test_predicted)
buys = []
sells = []
thresh = 0.1
start = 0
end = -1
def train(filePath, symbol):
# dataset
ohlcv_histories, technical_indicators, next_day_open_values, technical_normalizer, y_normaliser, data_normaliser = csv_to_dataset(
filePath, 3)
# model architecture
# define two sets of inputs
lstm_input = Input(shape=(history_points, 5), name='lstm_input')
dense_input = Input(shape=(technical_indicators.shape[1], ),
name='tech_input')
# the first branch operates on the first input
x = LSTM(50, name='lstm_0')(lstm_input)
x = Dropout(0.2, name='lstm_dropout_0')(x)
lstm_branch = Model(inputs=lstm_input, outputs=x)
# the second branch opreates on the second input
y = Dense(20, name='tech_dense_0')(dense_input)
y = Activation("relu", name='tech_relu_0')(y)
y = Dropout(0.2, name='tech_dropout_0')(y)
technical_indicators_branch = Model(inputs=dense_input, outputs=y)
# combine the output of the two branches
combined = concatenate(
[lstm_branch.output, technical_indicators_branch.output],
name='concatenate')
z = Dense(64, activation="sigmoid", name='dense_pooling')(combined)
z = Dense(1, activation="linear", name='dense_out')(z)
# our model will accept the inputs of the two branches and
# then output a single value
model = Model(
inputs=[lstm_branch.input, technical_indicators_branch.input],
outputs=z)
adam = optimizers.Adam(lr=0.0005)
model.compile(optimizer=adam, loss='mse')
model.fit(x=[ohlcv_histories, technical_indicators],
y=next_day_open_values,
batch_size=32,
epochs=50,
shuffle=True,
validation_split=0.1)
return model, technical_normalizer, y_normaliser, data_normaliser
import keras
from keras.models import Model
from keras.layers import Dense, Dropout, LSTM, Input, Activation
from keras import optimizers
import numpy as np
np.random.seed(4)
import tensorflow as tf
tf.random.set_seed(4)
from util import csv_to_dataset, history_points
# dataset
ohlcv_histories, _, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
'stock_data.csv')
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
print(ohlcv_train.shape)
print(ohlcv_test.shape)
# model architecture
import keras
import tensorflow as tf
from keras.models import Model
from keras.layers import Dense, Dropout, LSTM, Input, Activation, concatenate
from keras import optimizers
import numpy as np
np.random.seed(4)
import tensorflow
tensorflow.random.set_seed(4)
import sys
from util import csv_to_dataset, history_points
# dataset
ohlcv_histories, technical_indicators, next_day_open_values, unscaled_y, y_normaliser = csv_to_dataset(
sys.argv[1] + "_" + sys.argv[2] + ".csv")
test_split = 0.9
n = int(ohlcv_histories.shape[0] * test_split)
ohlcv_train = ohlcv_histories[:n]
tech_ind_train = technical_indicators[:n]
y_train = next_day_open_values[:n]
ohlcv_test = ohlcv_histories[n:]
tech_ind_test = technical_indicators[n:]
y_test = next_day_open_values[n:]
unscaled_y_test = unscaled_y[n:]
print(ohlcv_train.shape)