def test_different_seeds_for_subset_of_trials(self):
n_trials = 25
cols = ["col_" + str(i) for i in range(4)]
dfs = [
pd.DataFrame(np.random.rand(10, len(cols)), columns=cols)
for _ in range(n_trials)
]
self.write_dataframes_to_file(dfs, self.hdf_path)
first_dataset = dataset.read_dataset(self.hdf_path,
shuffle=True,
n_trials=4,
seed=90)
second_dataset = dataset.read_dataset(self.hdf_path,
shuffle=True,
n_trials=4,
seed=0)
different_order = False
for df1, df2 in zip(first_dataset, second_dataset):
if not df1.equals(df2):
different_order = True
break
self.assertTrue(different_order)
def test_cols_that_do_not_exist(self):
n_trials = 5
dfs = [pd.DataFrame(np.random.rand(10, 4)) for _ in range(n_trials)]
self.write_dataframes_to_file(dfs, self.hdf_path)
inexisting_cols = ["inexisting_col_a", "inexisting_col_b"]
with self.assertRaises(KeyError):
dataset.read_dataset(self.hdf_path,
cols=inexisting_cols,
add_class_columns=False)
def test_n_trials_is_none(self):
n_trials = 5
dfs = [pd.DataFrame(np.random.rand(10, 4)) for _ in range(n_trials)]
self.write_dataframes_to_file(dfs, self.hdf_path)
read_dataset = dataset.read_dataset(self.hdf_path)
self.assertEqual(n_trials, len(read_dataset))
def test_n_trials_larger_than_available(self):
n_trials = 5
dfs = [pd.DataFrame(np.random.rand(10, 4)) for _ in range(n_trials)]
self.write_dataframes_to_file(dfs, self.hdf_path)
read_dataset = dataset.read_dataset(self.hdf_path,
n_trials=n_trials * 5)
self.assertEqual(n_trials, len(read_dataset))
def test_datasets_sizes(self):
gps.create_passive_datasets_for_training(**self.args)
for path, dataset_trials in zip(
[self.train_hdf_path, self.val_hdf_path, self.test_hdf_path], [
self.args["n_simulations_train"],
self.args["n_simulations_val"], self.args["n_simulations_test"]
]):
dataset = read_dataset(path)
self.assertEqual(len(dataset), dataset_trials)
def test_reading_all_trials_with_different_seeds_but_without_shuffle(self):
n_trials = 25
cols = ["col_" + str(i) for i in range(4)]
dfs = [
pd.DataFrame(np.random.rand(10, len(cols)), columns=cols)
for _ in range(n_trials)
]
self.write_dataframes_to_file(dfs, self.hdf_path)
# Specifying shuffle=False (or not specifying at all) should make the method ignore
# the seed argument
first_dataset = dataset.read_dataset(self.hdf_path,
shuffle=False,
seed=90)
second_dataset = dataset.read_dataset(self.hdf_path,
shuffle=False,
seed=0)
for df1, df2 in zip(first_dataset, second_dataset):
pd.testing.assert_frame_equal(df1, df2)
def test_reproducibility_for_subset_of_trials(self):
n_trials = 25
cols = ["col_" + str(i) for i in range(4)]
dfs = [
pd.DataFrame(np.random.rand(10, len(cols)), columns=cols)
for _ in range(n_trials)
]
self.write_dataframes_to_file(dfs, self.hdf_path)
first_dataset = dataset.read_dataset(self.hdf_path,
shuffle=True,
n_trials=4,
seed=90)
second_dataset = dataset.read_dataset(self.hdf_path,
shuffle=True,
n_trials=4,
seed=90)
for df1, df2 in zip(first_dataset, second_dataset):
pd.testing.assert_frame_equal(df1, df2)
def test_cols_is_none(self):
n_trials = 5
cols = ["col_" + str(i) for i in range(4)]
dfs = [
pd.DataFrame(np.random.rand(10, 4), columns=cols)
for _ in range(n_trials)
]
self.write_dataframes_to_file(dfs, self.hdf_path)
read_dataset = dataset.read_dataset(self.hdf_path,
cols=None,
add_class_columns=False)
self.assertTrue(self.are_columns_in_all_trials(read_dataset, cols))
def test_add_class_columns(self):
n_trials = 5
cols = ["col_" + str(i) for i in range(4)]
cols_to_write = cols + list(MASS_CLASS_COLS) + list(FORCE_CLASS_COLS)
dfs = [
pd.DataFrame(np.random.rand(10, len(cols_to_write)),
columns=cols_to_write) for _ in range(n_trials)
]
self.write_dataframes_to_file(dfs, self.hdf_path)
read_dataset = dataset.read_dataset(self.hdf_path,
cols=cols,
add_class_columns=True)
self.assertTrue(
self.are_columns_in_all_trials(read_dataset, cols_to_write))
def test_duplicated_cols_are_only_read_once(self):
n_trials = 5
cols = ["col_" + str(i) for i in range(4)]
dfs = [
pd.DataFrame(np.random.rand(10, 4), columns=cols)
for _ in range(n_trials)
]
self.write_dataframes_to_file(dfs, self.hdf_path)
read_dataset = dataset.read_dataset(self.hdf_path,
cols=cols + cols,
add_class_columns=False)
self.assertTrue(self.are_columns_in_all_trials(read_dataset, cols))
self.assertTrue(
reduce(lambda x, y: x and y,
[len(df.columns) == 4 for df in read_dataset]))
def test_empty_dataset(self):
args_copy = self.args.copy()
args_copy["n_simulations_train"] = 0
gps.create_passive_datasets_for_training(**args_copy)
for path, dataset_trials in zip(
[self.train_hdf_path, self.val_hdf_path, self.test_hdf_path], [
args_copy["n_simulations_train"],
args_copy["n_simulations_val"], args_copy["n_simulations_test"]
]):
if dataset_trials > 0:
dataset = read_dataset(path)
self.assertEqual(len(dataset), dataset_trials)
else:
self.assertFalse(os.path.exists(path))
def test_do_not_add_class_columns_when_cols_unspecified(self):
n_trials = 5
cols = ["col_" + str(i) for i in range(4)]
dfs = [
pd.DataFrame(np.random.rand(10, len(cols)), columns=cols)
for _ in range(n_trials)
]
self.write_dataframes_to_file(dfs, self.hdf_path)
try:
read_dataset = dataset.read_dataset(self.hdf_path,
cols=None,
add_class_columns=True)
except KeyError:
self.fail("read_dataset failed unexpectedly")
self.assertTrue(self.are_columns_in_all_trials(read_dataset, cols))
def test_datasets_sizes_with_previously_existing_dataset(self):
# Create initial datasets with default key_prefix
gps.create_passive_datasets_for_training(**self.args)
# Create new datasets with a different key_prefix and number of trials
# If the previous dataset isn't deleted: either there will be more trials (added to the
# same path) or an incorrect number of trials.
args_copy = self.args.copy()
args_copy["trial_hdf_key_prefix"] = "newtrials_"
args_copy["n_simulations_train"] = 1
args_copy["n_simulations_val"] = 1
args_copy["n_simulations_test"] = 1
gps.create_passive_datasets_for_training(**args_copy)
for path, dataset_trials in zip(
[self.train_hdf_path, self.val_hdf_path, self.test_hdf_path], [
args_copy["n_simulations_train"],
args_copy["n_simulations_val"], args_copy["n_simulations_test"]
]):
dataset = read_dataset(path)
self.assertEqual(len(dataset), dataset_trials)
