def main(_):
config = cfg.Config()
data_builder = DataBuilder(config)
train_set, evaluate_set = data_builder.build_data(config.path_data)
model = Model(config)
model.build()
model.train(train_set, evaluate_set)
class RNNWrapper:
def __init__(self, trading_data):
self.config = trading_data.config
self.company = trading_data.company
data_set = pd.read_csv(trading_data.csv_dataset)
self.data_builder = DataBuilder(data_set, trading_data.entry_columns,
trading_data.prediction_column)
self.data_builder.build_data()
def build(self):
training_entry_set = self.data_builder.training_entry_set
input_shape = (training_entry_set.shape[1], training_entry_set.shape[2])
self.sequential = Sequential()
for i in range(input_shape[1]):
last_iteration = i == input_shape[1]-1
self.sequential.add(LSTM(units = self.config.layer_units,
recurrent_activation = self.config.activation_function,
return_sequences = not last_iteration,
input_shape = input_shape))
self.sequential.add(Dense(units = 1))
optimizer = Adam(learning_rate=self.config.learning_rate,
beta_1=self.config.beta_1,
beta_2=self.config.beta_2)
self.sequential.compile(optimizer = optimizer, loss = self.config.loss)
def fit(self):
entry_set = self.data_builder.training_entry_set
result_set = self.data_builder.training_result_set
rlrop = ReduceLROnPlateau(monitor=self.config.monitor,
factor=self.config.factor,
patience=self.config.patiente)
with tf.device(self.config.device):
self.sequential.fit(entry_set,
result_set,
epochs = self.config.epochs,
batch_size = self.config.batch_size,
callbacks=[rlrop])
def predict(self):
entry_set = self.data_builder.test_entry_set
scaler_prediction = self.sequential.predict(entry_set)
self.predicted_results = self.data_builder.inverse_transform(scaler_prediction)
def load_real_results(self):
self.real_results = self.data_builder.real_test_results
def to_image(self):
plt.clf()
plt.plot(self.real_results, color = 'red', label = 'Real Results')
plt.plot(self.predicted_results, color = 'blue', label = 'Predicted Results')
plt.title(self.company + ' Prediction')
plt.xlabel('Time')
plt.ylabel(self.company + ' Price')
plt.legend()
return plt
def save(self, path):
self.sequential.save(path)
