def _create_dummy_controller(self, simulation_end, simulation_start, temp_dir, calendar_name,
oracle_config,
scheduling_config):
data_source_config = {
"filename": TEST_HDF5FILE_NAME,
"exchange": calendar_name,
"start": datetime.datetime(1999, 1, 11, tzinfo=pytz.utc),
"end": datetime.datetime(1999, 1, 11, tzinfo=pytz.utc)
}
datasource = DummyDataSource(data_source_config)
oracle = ConstantOracle(
calendar_name=calendar_name,
oracle_configuration=oracle_config,
scheduling_configuration=scheduling_config
)
scheduler = Scheduler(
simulation_start,
simulation_end,
calendar_name,
oracle.prediction_frequency,
oracle.training_frequency
)
controller_configuration = {
'start_date': simulation_start.strftime('%Y-%m-%d'),
'end_date': simulation_end.strftime('%Y-%m-%d')
}
oracle_performance = OraclePerformance(
temp_dir.name, 'test'
)
controller = Controller(
configuration=controller_configuration,
oracle=oracle,
scheduler=scheduler,
datasource=datasource,
performance=oracle_performance
)
return controller
def test_controller_with_xarray(self):
calendar_name = "NYSE"
data_source_config = {
"exchange": calendar_name,
"data_timezone": "America/New_York",
"filename": "tests/resources/19990101_19990301_3_stocks.nc",
"start": datetime.datetime(2006, 12, 31, tzinfo=pytz.utc),
"end": datetime.datetime(2011, 12, 31, tzinfo=pytz.utc)
}
datasource = XArrayDataSource(data_source_config)
scheduling_configuration = {
"prediction_frequency":
{
"frequency_type": "DAILY",
"days_offset": 0,
"minutes_offset": 15
},
"training_frequency":
{
"frequency_type": "WEEKLY",
"days_offset": 0,
"minutes_offset": 15
},
}
oracle_config = {
"prediction_horizon": {
"unit": "days",
"value": 10
},
"prediction_delta": {
'unit': 'days',
'value': 10
},
"training_delta": {
'unit': 'days',
'value': 20
},
"model": {
"constant_variance": 0.1,
"past_horizon": datetime.timedelta(days=7),
"target_feature": "close"
}
}
oracle = ConstantOracle(
calendar_name=calendar_name,
scheduling_configuration=scheduling_configuration,
oracle_configuration=oracle_config
)
# these dates need to be within [start, end] of the data source
simulation_start = datetime.datetime(1999, 1, 10, tzinfo=pytz.utc)
simulation_end = datetime.datetime(1999, 2, 10, tzinfo=pytz.utc)
scheduler = Scheduler(simulation_start,
simulation_end,
calendar_name,
oracle.prediction_frequency,
oracle.training_frequency
)
controller_configuration = {
'start_date': simulation_start.strftime('%Y-%m-%d'),
'end_date': simulation_end.strftime('%Y-%m-%d')
}
temp_dir = TemporaryDirectory()
oracle_performance = OraclePerformance(
temp_dir.name, 'test'
)
controller = Controller(
configuration=controller_configuration,
oracle=oracle,
scheduler=scheduler,
datasource=datasource,
performance=oracle_performance
)
controller.run()
# Check if files have been writter
assert len(glob.glob(temp_dir.name + "/*hdf5")) == 3
def test_run(self):
oracle_configuration = {
'prediction_delta': {
'unit': 'days',
'value': 10
},
'training_delta': {
'unit': 'days',
'value': 20
},
'prediction_horizon': {
'unit': 'days',
'value': 1
},
'data_transformation': {
'feature_config_list': [
{
'name': 'close',
'transformation': {
'name': 'log-return'
},
'normalization': 'standard',
'is_target': True,
'local': False,
'length': 5
},
],
'features_ndays': 10,
'features_resample_minutes': 15,
'fill_limit': 5,
},
"model": {
'train_path': OUTPUT_DIR,
'covariance_method': 'NERCOME',
'covariance_ndays': 9,
'model_save_path': OUTPUT_DIR,
'tensorboard_log_path': OUTPUT_DIR,
'd_type': 'float32',
'tf_type': 32,
'random_seed': 0,
# Training specific
'predict_single_shares': True,
'n_epochs': 1,
'n_retrain_epochs': 1,
'learning_rate': 2e-3,
'batch_size': 100,
'cost_type': 'bayes',
'n_train_passes': 30,
'n_eval_passes': 100,
'resume_training': False,
'classify_per_series': False,
'normalise_per_series': False,
# Topology
'n_series': 3,
'n_assets': 3,
'n_features_per_series': 271,
'n_forecasts': 1,
'n_classification_bins': 12,
'layer_heights': [270, 270],
'layer_widths': [3, 3],
'activation_functions': ['relu', 'relu'],
# Initial conditions
'INITIAL_ALPHA': 0.2,
'INITIAL_WEIGHT_UNCERTAINTY': 0.4,
'INITIAL_BIAS_UNCERTAINTY': 0.4,
'INITIAL_WEIGHT_DISPLACEMENT': 0.1,
'INITIAL_BIAS_DISPLACEMENT': 0.4,
'USE_PERFECT_NOISE': True,
# Priors
'double_gaussian_weights_prior': False,
'wide_prior_std': 1.2,
'narrow_prior_std': 0.05,
'spike_slab_weighting': 0.5,
},
"universe": {
"method": "liquidity",
"n_assets": 3,
"ndays_window": 5,
"update_frequency": 'weekly',
"avg_function": 'median',
"dropna": False
},
}
scheduling_configuration = {
"prediction_frequency": {"frequency_type": "WEEKLY", "days_offset": 0, "minutes_offset": 75},
"training_frequency": {"frequency_type": "WEEKLY", "days_offset": 0, "minutes_offset": 60}
}
oracle = MetaCrocubotOracle(
calendar_name="NYSE",
oracle_configuration=oracle_configuration,
scheduling_configuration=scheduling_configuration
)
simulation_start = datetime.datetime(2009, 1, 5, tzinfo=pytz.utc)
simulation_end = datetime.datetime(2009, 1, 31, tzinfo=pytz.utc)
calendar_name = 'NYSE'
scheduler = Scheduler(
simulation_start,
simulation_end,
calendar_name,
oracle.prediction_frequency,
oracle.training_frequency,
)
oracle_performance = OraclePerformance(OUTPUT_DIR, 'oracle')
datasource = SyntheticDataSource({
"start_date": datetime.datetime(2006, 12, 31),
"end_date": datetime.datetime(2011, 12, 31),
"n_sin_series": 3
})
controller = Controller(
configuration={
'start_date': '2009-01-05',
'end_date': '2009-01-31'
},
oracle=oracle,
scheduler=scheduler,
datasource=datasource,
performance=oracle_performance
)
controller.run()
expected_files = ["oracle_correlation_coefficient.pdf",
"oracle_cumulative_returns.pdf",
"oracle_data_table.csv",
"oracle_oracle_results_actuals.hdf5",
"oracle_oracle_results_covariance_matrix.hdf5",
"oracle_oracle_results_mean_vector.hdf5",
"oracle_performance_table.csv",
"time-series-plot.pdf"
]
for filename in expected_files:
self.assertTrue(os.path.isfile(os.path.join(OUTPUT_DIR, filename)))
"training_frequency": {"frequency_type": "MONTHLY", "days_offset": -1, "minutes_offset": 0}
}
oracle = MetaCrocubotOracle(
calendar_name=CALENDAR_NAME,
oracle_configuration=oracle_configuration,
scheduling_configuration=scheduling_configuration
)
simulation_start = datetime.datetime(2007, 11, 30, tzinfo=pytz.utc)
simulation_end = datetime.datetime(2008, 11, 30, tzinfo=pytz.utc)
scheduler = Scheduler(
simulation_start,
simulation_end,
CALENDAR_NAME,
oracle.prediction_frequency,
oracle.training_frequency,
)
oracle_performance = OraclePerformance(OUTPUT_DIR, 'oracle')
datasource = DataSource({
'data_file': os.path.join(RESOURCES_DIR, 'test_stock_data.hdf5')
})
controller = Controller(
configuration={
'start_date': simulation_start.strftime('%Y-%m-%d'),
'end_date': simulation_end.strftime('%Y-%m-%d')
},
def test_run(self):
oracle_configuration = {
'prediction_delta': {
'unit': 'days',
'value': 3
},
'training_delta': {
'unit': 'days',
'value': 12
},
'prediction_horizon': {
'unit': 'days',
'value': 1
},
'data_transformation': {
'feature_config_list': [
{
'name': 'Returns',
'transformation': {
'name': 'value'
},
'normalization': None,
'is_target': True,
'local': False,
'length': 5
},
],
'features_ndays':
5,
'features_resample_minutes':
15,
'fill_limit':
5,
},
"model": {
'train_path': OUTPUT_DIR,
'covariance_method': 'NERCOME',
'covariance_ndays': 9,
'model_save_path': OUTPUT_DIR,
'tensorboard_log_path': OUTPUT_DIR,
'd_type': 'float32',
'tf_type': 32,
'random_seed': 0,
# Training specific
'predict_single_shares': True,
'n_epochs': 1,
'n_retrain_epochs': 1,
'learning_rate': 2e-3,
'batch_size': 100,
'cost_type': 'bayes',
'n_train_passes': 30,
'n_eval_passes': 100,
'resume_training': False,
'classify_per_series': False,
'normalise_per_series': False,
# Topology
'n_series': 324,
'n_assets': 324,
'n_correlated_series': 1,
'n_features_per_series': 271,
'n_forecasts': 1,
'n_classification_bins': 12,
'layer_heights': [270, 270],
'layer_widths': [3, 3],
'activation_functions': ['relu', 'relu'],
# Initial conditions
'INITIAL_ALPHA': 0.2,
'INITIAL_WEIGHT_UNCERTAINTY': 0.4,
'INITIAL_BIAS_UNCERTAINTY': 0.4,
'INITIAL_WEIGHT_DISPLACEMENT': 0.1,
'INITIAL_BIAS_DISPLACEMENT': 0.4,
'USE_PERFECT_NOISE': True,
# Priors
'double_gaussian_weights_prior': False,
'wide_prior_std': 1.2,
'narrow_prior_std': 0.05,
'spike_slab_weighting': 0.5,
}
}
scheduling_configuration = {
"prediction_frequency": {
"frequency_type": "MONTHLY",
"days_offset": -1,
"minutes_offset": 0
},
"training_frequency": {
"frequency_type": "MONTHLY",
"days_offset": -1,
"minutes_offset": 0
}
}
oracle = MetaCrocubotOracle(
calendar_name=CALENDAR_NAME,
oracle_configuration=oracle_configuration,
scheduling_configuration=scheduling_configuration)
simulation_start = datetime.datetime(2007, 12, 31, tzinfo=pytz.utc)
simulation_end = datetime.datetime(2008, 12, 31, tzinfo=pytz.utc)
scheduler = Scheduler(
simulation_start,
simulation_end,
CALENDAR_NAME,
oracle.prediction_frequency,
oracle.training_frequency,
)
oracle_performance = OraclePerformance(OUTPUT_DIR, 'oracle')
datasource = DataSource(
{'data_file': os.path.join(RESOURCES_DIR, 'test_stock_data.hdf5')})
controller = Controller(configuration={
'start_date':
simulation_start.strftime('%Y-%m-%d'),
'end_date':
simulation_end.strftime('%Y-%m-%d')
},
oracle=oracle,
scheduler=scheduler,
datasource=datasource,
performance=oracle_performance)
controller.run()
expected_files = [
"oracle_correlation_coefficient.pdf",
"oracle_cumulative_returns.pdf", "oracle_data_table.csv",
"oracle_oracle_results_actuals.hdf5",
"oracle_oracle_results_covariance_matrix.hdf5",
"oracle_oracle_results_mean_vector.hdf5",
"oracle_performance_table.csv", "time-series-plot.pdf"
]
for filename in expected_files:
self.assertTrue(os.path.isfile(os.path.join(OUTPUT_DIR, filename)))