def train_model() -> None:
target_scaler = create_target_scaler()
preprocess_pipeline = create_preprocess_pipeline()
input_data = InfluxdbQuery(
dict(fields='open,high,low,close,volume',
result_limit=100000)).query()
input_data = input_data.sort_values(by='time',
ascending=True).set_index('time')
input_data_features = extract_features(input_data,
lookahead=MLTools.lookahead)
input_data_features = input_data_features.pipe(
lambda df: df[~pd.isna(df['target'])])
dataframe_x, dataframe_y = split_dataframe_on_columns(
input_data_features, ['target'])
X_train, X_val, y_train, y_val = train_test_split(dataframe_x.values,
dataframe_y.values,
test_size=0.03,
random_state=42,
shuffle=False)
X_train = preprocess_pipeline.fit_transform(X_train)
X_val = preprocess_pipeline.transform(X_val)
y_train = target_scaler.fit_transform(y_train)
y_val = target_scaler.transform(y_val)
X_train, y_train = transform_rnn_sequences(X_train,
y_train,
lookback=MLTools.lookback,
lookahead=MLTools.lookahead)
X_val, y_val = transform_rnn_sequences(X_val,
y_val,
lookback=MLTools.lookback,
lookahead=MLTools.lookahead)
joblib.dump(preprocess_pipeline, MLTools.preprocess_pipeline_path)
joblib.dump(target_scaler, MLTools.target_scaler_path)
rnn_model = build_model(input_shape=X_train.shape[1:],
output_neurons=y_train.shape[1])
rnn_model.fit(X_train,
y_train,
validation_data=(X_val, y_val),
epochs=30,
batch_size=32,
callbacks=[
EarlyStopping(restore_best_weights=True,
patience=2,
monitor='val_loss')
])
save_model(rnn_model, MLTools.rnn_model_path)
def extract(self) -> ndarray:
if self.preprocess_pipeline is None:
return
input_data = InfluxdbQuery(
dict(fields='open,high,low,close,volume',
result_limit=MLTools.lookback * 2)).query()
input_data = input_data.sort_values(by='time',
ascending=True).set_index('time')
input_data_features = extract_features(input_data)
dataframe_x, dataframe_y = split_dataframe_on_columns(
input_data_features, ['target'])
X = self.preprocess_pipeline.transform(dataframe_x.values)
X_rnn, _ = transform_rnn_sequences(X,
dataframe_y.values,
lookback=MLTools.lookback)
self.last_time_point = input_data.index.values[-1]
self.last_close_value = input_data["close"].values[-1]
return X_rnn[-1:]
def timeseries():
result_df = InfluxdbQuery(request.args).query()
return result_df.to_json(orient='split', index=False)
def test_query(self, mock_create):
mock_create.return_value.query = lambda x: {'ohlcv': self.input_data}
influxapi = InfluxdbQuery()
res = influxapi.query()
self.assertEqual(res.columns.tolist(), ["time", "a", "b"])
self.assertEqual(res.shape[0], 2)
def test_query_empty_res(self, mock_create):
mock_create.return_value.query = lambda x: {}
influxapi = InfluxdbQuery()
res = influxapi.query()
self.assertEqual(res.shape, (0, 0))
def test_transform_dataframe(self):
influxapi = InfluxdbQuery()
output_data = influxapi._transform_dataframe(self.input_data)
self.assertEqual(set(output_data.columns.tolist()), set(['time', 'a', 'b']))
self.assertEqual(len(self.input_data), len(output_data))
def test_init(self):
influxapi = InfluxdbQuery()
self.assertEqual(influxapi.request_parameters, {})
self.assertEqual(influxapi.measurement, 'ohlcv')
self.assertEqual(influxapi.fields, '*')
self.assertEqual(influxapi.asset, 'btc')