import os import json import joblib import argparse import numpy as np import pandas as pd from sklearn.preprocessing import StandardScaler, MinMaxScaler from tensorflow.keras.models import Sequential from tensorflow.keras.layers import GRU, LSTM, Dense, Dropout from tensorflow.keras.callbacks import EarlyStopping, ModelCheckpoint from warnings import filterwarnings filterwarnings('ignore') class DataProcessor: def __init__(self, datasets_path): self.datasets_path = datasets_path self.datasets = self._get_datasets() def _get_datasets(self): return sorted([ item for item in os.listdir(self.datasets_path) if os.path.isfile(os.path.join(self.datasets_path, item)) and item.endswith('.csv') ]) @staticmethod def create_sequences(df, sequence_length): labels, sequences = [], [] for i in range(len(df) - sequence_length): seq = df.iloc[i:i + sequence_length].values label = df.iloc[i + sequence_length].values[0] sequences.append(seq) labels.append(label) return np.array(sequences), np.array(labels) @staticmethod def preprocess_data(dataframe): for col in dataframe.columns: if dataframe[col].isnull().any(): if dataframe[col].dtype == 'object': dataframe[col].fillna(dataframe[col].mode()[0], inplace = True) else: dataframe[col].fillna(dataframe[col].mean(), inplace = True) return dataframe @staticmethod def scale_data(dataframe, scaler_cls): scaler = scaler_cls() dataframe['Close'] = scaler.fit_transform(dataframe[['Close']]) return scaler, dataframe class ModelBuilder: """ GRU (Gated Recurrent Units) Model """ @staticmethod def gru_model(input_shape): model = Sequential([ GRU(50, return_sequences = True, input_shape = input_shape), Dropout(0.2), GRU(50, return_sequences = True), Dropout(0.2), GRU(50, return_sequences = True), Dropout(0.2), GRU(50, return_sequences = False), Dropout(0.2), Dense(units = 1) ]) model.compile(optimizer = 'nadam', loss = 'mean_squared_error') return model """ LSTM (Long Short-Term Memory) Model """ @staticmethod def lstm_model(input_shape): model = Sequential([ LSTM(50, return_sequences = True, input_shape = input_shape), Dropout(0.2), LSTM(50, return_sequences = True), Dropout(0.2), LSTM(50, return_sequences = True), Dropout(0.2), LSTM(50, return_sequences = False), Dropout(0.2), Dense(units = 1) ]) model.compile(optimizer = 'nadam', loss = 'mean_squared_error') return model """ LSTM (Long Short-Term Memory) and GRU (Gated Recurrent Units) Model """ @staticmethod def lstm_gru_model(input_shape): model = Sequential([ LSTM(50, return_sequences = True, input_shape = input_shape), Dropout(0.2), GRU(50, return_sequences = True), Dropout(0.2), LSTM(50, return_sequences = True), Dropout(0.2), GRU(50, return_sequences = False), Dropout(0.2), Dense(units = 1) ]) model.compile(optimizer = 'nadam', loss = 'mean_squared_error') return model class Trainer: def __init__(self, model, model_file, sequence_length, epochs, batch_size): self.model = model self.model_file = model_file self.sequence_length = sequence_length self.epochs = epochs self.batch_size = batch_size def train(self, X_train, y_train, X_test, y_test): early_stopping = EarlyStopping(monitor = 'val_loss', patience = 5, mode = 'min') model_checkpoint = ModelCheckpoint( filepath = self.model_file, save_best_only = True, monitor = 'val_loss', mode = 'min' ) history = self.model.fit( X_train, y_train, epochs = self.epochs, batch_size = self.batch_size, validation_data = (X_test, y_test), callbacks = [early_stopping, model_checkpoint] ) return history class PostProcessor: @staticmethod def inverse_transform(scaler, data): return scaler.inverse_transform(data) @staticmethod def save_json(filename, data): with open(filename, 'w') as f: json.dump(data, f) def main(algorithm: str, sequence_length: int, epochs: int, batch_size: int): datasets_path = './datasets' models_path = './models' posttrained = './posttrained' pickle_file = './pickles' data_processor = DataProcessor(datasets_path) for dataset in data_processor.datasets: print(f"[TRAINING] {dataset.replace('.csv', '')} ") dataframe = pd.read_csv(os.path.join(datasets_path, dataset), index_col='Date')[['Close']] model_file = os.path.join(models_path, f"{dataset.replace('.csv', '')}.keras") # dataframe = data_processor.preprocess_data(dataframe) dataframe.dropna(inplace = True) standard_scaler, dataframe = data_processor.scale_data(dataframe, StandardScaler) minmax_scaler, dataframe = data_processor.scale_data(dataframe, MinMaxScaler) sequences, labels = data_processor.create_sequences(dataframe, sequence_length) input_shape = (sequences.shape[1], sequences.shape[2]) if algorithm == "GRU": model = ModelBuilder.gru_model(input_shape) elif algorithm == "LSTM": model = ModelBuilder.lstm_model(input_shape) elif algorithm == "LSTM_GRU": model = ModelBuilder.lstm_gru_model(input_shape) else: model = ModelBuilder.lstm_model(input_shape) train_size = int(len(sequences) * 0.8) X_train, X_test = sequences[:train_size], sequences[train_size:] y_train, y_test = labels[:train_size], labels[train_size:] trainer = Trainer(model, model_file, sequence_length, epochs, batch_size) trainer.train(X_train, y_train, X_test, y_test) dataframe_json = {'Date': dataframe.index.tolist(), 'Close': dataframe['Close'].tolist()} PostProcessor.save_json( os.path.join(posttrained, f'{dataset.replace(".csv", "")}-posttrained.json'), dataframe_json ) joblib.dump(minmax_scaler, os.path.join(pickle_file, f'{dataset.replace(".csv", "")}_minmax_scaler.pickle')) joblib.dump(standard_scaler, os.path.join(pickle_file, f'{dataset.replace(".csv", "")}_standard_scaler.pickle')) model.load_weights(model_file) model.save(model_file) print("\n\n") if __name__ == "__main__": parser = argparse.ArgumentParser(description = "Tebakaja Model Trainer") parser.add_argument('-a', '--algorithm', type = str, required = True, help = 'select the algorithm to be trained (LSTM, GRU, LSTM_GRU)') parser.add_argument('-e', '--epochs', type = int, required = True, help = 'epochs') parser.add_argument('-b', '--batchs', type = int, required = True, help = 'batch length') parser.add_argument('-s', '--sequences', type = int, required = True, help = 'sequences length') args = parser.parse_args() main( epochs = args.epochs, batch_size = args.batchs, algorithm = args.algorithm, sequence_length = args.sequences )