def train_and_test_by_ticker(test_epochs, test_display_step, buy_threshold,
sell_threshold, use_random_data):
# GET DATA
data_df = dml.get_all_ml_data()
tickers = list({t for t in data_df['ticker']})
training_df = data_df[data_df.date < test_data_date].copy()
test_df = data_df[data_df.date >= test_data_date].copy()
del data_df
prediction_files = []
print("BULK TRAINING for {} tickers.".format(len(tickers)))
for ticker in tickers:
try:
print(" ----- Begin Training for {} ----- ".format(ticker))
# DATA
training_data_class = td.TrainingDataTicker(
training_df, feature_series_count, feature_count, label_count,
ticker)
testing_data_class = td.TrainingDataTicker(test_df,
feature_series_count,
feature_count,
label_count, ticker)
# TRAIN
ticker_path = _model_path + ticker + '/'
if os.path.exists(ticker_path):
print("Found ticker directory, skipping {}".format(ticker))
# add existing file to predictions list
file_list = [
ticker_path + a_file for a_file in os.listdir(ticker_path)
]
if len(file_list) > 0:
latest_file = max(file_list, key=os.path.getmtime)
if latest_file.endswith('.csv'):
print("adding prediction file: {}".format(latest_file))
prediction_files.append(latest_file)
else:
print("couldn't find csv as latest file in {}".format(
ticker_path))
else:
os.makedirs(ticker_path)
saved_model_file = train_rnn(training_data_class, ticker_path,
use_random_data)
saved_model_file = saved_model_file + '.meta'
prediction_files.append(saved_model_file + '.csv')
# TEST
test_rnn(testing_data_class, test_epochs, test_display_step,
buy_threshold, sell_threshold, [saved_model_file])
except ValueError as ve:
print(ve)
# GENERATE PREDICTION AGGREGATE FILE
merge_csv_files(prediction_files, prediction_file)
def get_data_and_test_rnn(test_epochs,
test_display_step,
buy_threshold,
sell_threshold,
specific_file=None):
# GET DATA
data_df = dml.get_all_ml_data()
test_df = data_df[data_df.date >= test_data_date].copy()
del data_df
# TEST
testing_data_class = td.TrainingData(test_df, feature_series_count,
feature_count, label_count)
test_rnn(testing_data_class, test_epochs, test_display_step, buy_threshold,
sell_threshold, specific_file)
def get_data_and_test_rnn_by_ticker(test_epochs, test_display_step,
buy_threshold, sell_threshold,
specific_file):
# Get ticker
parse_filename = specific_file.replace('\\', "/")
print(parse_filename)
ticker = "WIKI/" + parse_filename.split("/")[-2]
print("Using ticker - {}".format(ticker))
# GET DATA
data_df = dml.get_all_ml_data()
test_df = data_df[data_df.date >= test_data_date].copy()
del data_df
# TEST
testing_data_class = td.TrainingDataTicker(test_df, feature_series_count,
feature_count, label_count,
ticker)
test_rnn(testing_data_class, test_epochs, test_display_step, buy_threshold,
sell_threshold, [specific_file])
def get_data_train_and_test_rnn(test_epochs, test_display_step, buy_threshold,
sell_threshold, use_random_data):
# GET DATA
data_df = dml.get_all_ml_data()
training_df = data_df[data_df.date < test_data_date].copy()
test_df = data_df[data_df.date >= test_data_date].copy()
del data_df
# TRAIN
training_data_class = td.TrainingData(training_df, feature_series_count,
feature_count, label_count)
# TODO: switch rnn to use batch data, testing below
# fff, lll, ddd = training_data_class.get_batch(3)
train_rnn(training_data_class, _model_path, use_random_data)
# TEST
testing_data_class = td.TrainingData(test_df, feature_series_count,
feature_count, label_count)
test_rnn(testing_data_class, test_epochs, test_display_step, buy_threshold,
sell_threshold)
merge_csv_files(_get_meta_prediction_files(), prediction_file)