def verify_train_NN(self):
# Load parameters used during model training and if trend different than current param re-create model and train and save param
with open(os.getcwd()+self.ML_dataset_parameters_path, 'r') as json_file:
ML_Parameters = json.load(json_file)
# Verify if the dataset has to be redone
if ML_Parameters['ML_trend_length'] != self.ML_trend_length or self.parameters['Crypto?'] != ML_Parameters['Crypto?']:
# Hist loading and dataset creation
my_hist = Dataset(Parameters)
my_hist.load()
ML_dataset = my_hist.new_format(len(my_hist.hist))
# Create the dataset
ML_dataset = my_hist.create_ML_dataset(ML_dataset)
# Save dataset parameters
with open(os.getcwd()+self.parameters['ML_path']['ML_dataset_parameters_path'], 'w') as json_file:
json.dump(self.parameters, json_file)
else:
ML_dataset = pd.read_csv(os.getcwd()+Parameters['ML_path']['ML_dataset_path'])
if not(os.path.exists(os.getcwd()+self.NN_model_path+str(self.ML_trend_length))):
# Train the dataset
self.train_NN(ML_dataset)
def verify_train_LSTM(self):
# Load parameters used during model training and if trend different than current param re-create model and train and save param
with open(os.getcwd()+self.ML_dataset_parameters_path, 'r') as json_file:
ML_Parameters = json.load(json_file)
full_hist = Dataset(Parameters)
full_hist.load()
if full_hist.date_name != '1m':
full_hist.hist[full_hist.date_name] = full_hist.hist[full_hist.date_name].astype('datetime64[ns]')
else:
full_hist.hist[full_hist.date_name] = full_hist.hist[full_hist.date_name].dt.floor('min')
dataset = full_hist.new_format(len(full_hist.hist))
companies_to_train = []
for company in self.companies_list:
if not(os.path.exists(os.getcwd()+self.LSTM_model_path+'_'+str(company))):
companies_to_train.append(company)
self.train_LSTM(dataset, companies_to_train)
# --------------------
with open(os.getcwd()+'\\parameters\\Parameters.json', 'r') as json_file:
Parameters = json.load(json_file)
# Overwrite mesh
Parameters['Mesh'] = '1m'
# --------------------
# MAIN
# --------------------
# ------- LOAD -------
# Loading history csv
full_hist = Dataset(Parameters)
full_hist.load()
# Setting back the timestamp format
if full_hist.date_name != '1m':
full_hist.hist[full_hist.date_name] = full_hist.hist[full_hist.date_name].astype('datetime64[ns]')
else:
full_hist.hist[full_hist.date_name] = full_hist.hist[full_hist.date_name].dt.floor('min')
# Create the porfolio
Portfolio = Portfolio_class(Parameters)
Portfolio.create()
# ------ UPDATE ------
# Fetching Data to complete history
if Parameters['Update_Values'] == True:
print('##### Crypto #####')
else:
sys.stdout = open(
os.getcwd() + '\\models\\logs\\NASDAQ\\Model_Eval_' +
str(Parameters['trend_length']) + '-' +
str(Parameters['ML_trend_length']) + '-' +
str(Parameters['study_length']) + '.txt', 'w')
print('##### NASDAQ #####')
# --------------------
# MAIN
# --------------------
# ------- LOAD -------
# Loading history csv
full_hist = Dataset(Parameters)
full_hist.load()
# Setting back the timestamp format
if full_hist.date_name != '1m':
full_hist.hist[full_hist.date_name] = full_hist.hist[
full_hist.date_name].astype('datetime64[ns]')
else:
full_hist.hist[full_hist.date_name] = full_hist.hist[
full_hist.date_name].dt.floor('min')
# Load the porfolio
Portfolio = Portfolio_class(Parameters).reset()
# ------ UPDATE ------
# Update Companies list if needed
else:
BS_dict[company] = "Hold"
# Specific rules
if add_specific_rules:
# Sell after a high rise
if prediction_dict[company] > sell_after_high_rise_ratio * last_values.loc[company]:
BS_dict[company] = "Sell"
# Sell after a high loss
if prediction_dict[company] < last_values.loc[company] / sell_after_high_loss_ratio:
BS_dict[company] = "Sell"
return(BS_dict, prediction_dict, next_variation_dict)
if __name__ == "__main__":
study_length = 10
# Loading history csv
full_hist = pd.read_csv(os.getcwd() + '\\resources\\full_NASDAQ_history.csv', usecols=['Close', 'Company', 'Date', 'Dividends', 'High', 'Low', 'Open', 'Stock Splits', 'Volume'])
# Setting back the timestamp format
full_hist['Date'] = full_hist['Date'].astype('datetime64[ns]')
dataset = Dataset(full_hist).new_format(study_length)
BS_dict, Trend_dict = BuySell(dataset, Parameters).trend()
print(BS_dict, Trend_dict)
from functions.Plot_Class import Plot
# --------------------
# PARAMETERS
# --------------------
with open(os.getcwd() + '\\parameters\\Parameters.json', 'r') as json_file:
Parameters = json.load(json_file)
# --------------------
# MAIN
# --------------------
# ------- LOAD -------
# Loading history csv
full_hist = Dataset(Parameters)
full_hist.load()
# Setting back the timestamp format
if full_hist.date_name != '1m':
full_hist.hist[full_hist.date_name] = full_hist.hist[
full_hist.date_name].astype('datetime64[ns]')
else:
full_hist.hist[full_hist.date_name] = full_hist.hist[
full_hist.date_name].dt.floor('min')
# Load the porfolio
Portfolio = Portfolio_class(Parameters).reset()
# ------ UPDATE ------
# Update Companies list if needed