def test_train_simple_model(model_name, module_workspace, workspace,
tiny_feats_csv):
thisconfig = copy.deepcopy(config)
thisconfig.data['training']['iteration_write_frequency'] = 2
thisconfig.data['training']['max_iterations'] = 10
thisconfig.data['training']['batch_size'] = 12
thisconfig.data['training']['max_files_per_class'] = 1
thisconfig.data['paths']['model_dir'] = workspace
thisconfig.data['paths']['feature_dir'] = module_workspace
# The features get loaded by tiny_feats_csv anyway
thisconfig.data['features']['cqt']['skip_existing'] = True
experiment_name = "testexperiment"
hold_out = "rwc"
driver = hcnn.driver.Driver(thisconfig,
model_name=model_name,
experiment_name=experiment_name,
dataset=tiny_feats_csv, load_features=True)
driver.setup_partitions(hold_out)
result = driver.train_model()
assert result is True
# Expected files this should generate
new_config = os.path.join(workspace, experiment_name, "config.yaml")
train_loss_fp = os.path.join(workspace, experiment_name, hold_out,
"training_loss.pkl")
assert os.path.exists(new_config)
assert os.path.exists(train_loss_fp)
def train(config,
experiment_name,
test_set,
model_name):
"""Run training loop.
Parameters
----------
config : str
Full path
experiment_name : str
Name of the experiment. Files are saved in a folder of this name.
test_set : str
String in ["rwc", "uiowa", "philharmonia"] specifying which
dataset to use as the test set.
model_name : str
Name of the model to use for training.
"""
print(utils.colored("Training experiment: {}".format(experiment_name)))
logger.info("Training model '{}' with test_set '{}'"
.format(model_name, test_set))
driver = hcnn.driver.Driver(config, test_set,
model_name=model_name,
experiment_name=experiment_name,
load_features=True)
return driver.train_model()
def test_find_best_model(config_with_workspace, model_name, workspace):
experiment_name = "testexperiment"
hold_out = "rwc"
driver = hcnn.driver.Driver(config_with_workspace, model_name=model_name,
experiment_name=experiment_name,
load_features=True)
driver.setup_partitions(hold_out)
result = driver.train_model()
assert result is True
# Create a vastly reduced validation dataframe so it'll take less long.
validation_size = 3
driver.valid_set.df = driver.valid_set.df.sample(n=validation_size,
replace=True)
assert len(driver.valid_set.df) == validation_size
driver.test_set.df = driver.test_set.df.sample(n=validation_size,
replace=True)
results_df = driver.find_best_model()
# check that the results_df is ordered by iteration.
assert all(results_df["model_iteration"] ==
sorted(results_df["model_iteration"]))
# Get the best param
param_iter = driver.select_best_iteration(results_df)
assert param_iter is not None
# load it again to test the reloading thing.
# Just making sure this runs through
results_df2 = driver.find_best_model()
assert all(results_df == results_df2)
# Shrink the dataset so this doesn't take forever.
driver.dataset.df = driver.dataset.df.sample(n=10, replace=True)
predictions_df = driver.predict(param_iter)
assert not predictions_df.empty
predictions_df_path = os.path.join(
workspace, experiment_name, hold_out,
"model_{}_predictions.pkl".format(param_iter))
assert os.path.exists(predictions_df_path)