def test_data_model_types_forecasting_pipeline_fit():
train_data, test_data = get_ts_data(forecast_length=10)
pipeline = get_multiscale_pipeline()
pipeline.fit(train_data)
pipeline.save('data_model_forecasting')
expected_len_nodes = len(pipeline.nodes)
actual_len_nodes = len(PipelineTemplate(pipeline).operation_templates)
assert actual_len_nodes == expected_len_nodes
def configure_experiment():
""" Generates a time series of 100 elements. The prediction is performed
for five elements ahead
"""
# Default number of validation blocks
validation_blocks = 3
forecast_len = 5
time_series, _ = get_ts_data(n_steps=105, forecast_length=forecast_len)
log = default_log(__name__)
return log, forecast_len, validation_blocks, time_series
def test_ts_pipeline_with_stats_model():
""" Tests PipelineTuner for time series forecasting task with AR model """
train_data, test_data = get_ts_data(n_steps=200, forecast_length=5)
ar_pipeline = Pipeline(PrimaryNode('ar'))
# Tune AR model
tuner_ar = PipelineTuner(pipeline=ar_pipeline,
task=train_data.task,
iterations=3)
tuned_ar_pipeline = tuner_ar.tune_pipeline(input_data=train_data,
loss_function=mse)
is_tuning_finished = True
assert is_tuning_finished
def get_dataset(task_type: str):
if task_type == 'regression':
data = get_synthetic_regression_data()
train_data, test_data = train_test_data_setup(data)
threshold = np.std(test_data.target) * 0.05
elif task_type == 'classification':
data = get_iris_data()
train_data, test_data = train_test_data_setup(data, shuffle_flag=True)
threshold = 0.95
elif task_type == 'clustering':
data = get_synthetic_input_data(n_samples=1000)
train_data, test_data = train_test_data_setup(data)
threshold = 0.5
elif task_type == 'ts_forecasting':
train_data, test_data = get_ts_data(forecast_length=5)
threshold = np.std(test_data.target)
else:
raise ValueError('Incorrect type of machine learning task')
return train_data, test_data, threshold
def test_ts_models_fit_correct():
train_data, test_data = get_ts_data(forecast_length=5)
logger = default_log('default_test_logger')
with OperationTypesRepository() as repo:
model_names, _ = repo.suitable_operation(
task_type=TaskTypesEnum.ts_forecasting, tags=['time_series'])
for model_name in model_names:
logger.info(f"Test time series model: {model_name}.")
model = Model(operation_type=model_name)
_, train_predicted = model.fit(data=deepcopy(train_data))
test_pred = model.predict(fitted_operation=_,
data=test_data,
is_fit_pipeline_stage=False)
mae_value_test = mean_absolute_error(y_true=test_data.target,
y_pred=test_pred.predict[0])
mae_threshold = np.var(test_data.target) * 2
assert mae_value_test < mae_threshold
def test_ts_forecasting_decomposition():
""" The function checks whether, after the decompose operation, the pipeline
actually models the original target (not decomposed) for the time series
forecasting task
"""
# Generate synthetic data for time series forecasting
train_data, _ = get_ts_data(forecast_length=5)
# Distort original values
train_data.features = train_data.features + 150
train_data.target = train_data.target + 150
_, pipeline_decompose_finish, pipeline = get_refinement_pipeline(lagged=10)
pipeline.fit(train_data)
pipeline_decompose_finish.fit(train_data)
full_output = pipeline.predict(train_data)
decompose_output = pipeline_decompose_finish.predict(train_data)
full_level = np.mean(full_output.predict)
decompose_level = np.mean(decompose_output.predict)
assert full_level > (decompose_level + 100)