def test_balanced_random_too_many_points(classes, N_POINTS, N_CLASSES):
with pytest.raises(AssertionError) as exc:
balanced_random_indices(_dummy_indices_method,
classes,
N_POINTS + N_CLASSES,
seed=1)
assert "n_points greater than nb of points available" == str(exc.value)
def test_balanced_random_unbalance(classes, N_POINTS):
with pytest.raises(AssertionError) as exc:
balanced_random_indices(_dummy_indices_method,
classes,
N_POINTS + 1,
seed=1)
assert "n_points is not a multiple of number of classes" == str(exc.value)
def split(datasets, data_size, seed, ratio, index):
return balanced_random_indices(
method=subsample_random_indices,
classes=datasets.classes,
n_points=data_size,
seed=seed,
ratio=ratio)
def test_balanced_random_shuffle(classes, N_POINTS):
indices1 = balanced_random_indices(_dummy_indices_method,
classes,
N_POINTS,
seed=1)
indices2 = balanced_random_indices(_dummy_indices_method,
classes,
N_POINTS,
seed=2)
assert set(indices1['train']) == set(indices2['train'])
assert set(indices1['valid']) == set(indices2['valid'])
assert set(indices1['test']) == set(indices2['test'])
assert indices1['train'][0] != indices1['train'][1]
assert indices1['valid'][0] != indices1['valid'][1]
assert indices1['test'][0] != indices1['test'][1]
def test_balanced_random_size(classes, N_TRAIN, N_VALID, N_TEST, N_POINTS):
indices = balanced_random_indices(_dummy_indices_method,
classes,
N_POINTS,
seed=1)
assert len(indices['train']) == N_TRAIN
assert len(indices['valid']) == N_VALID
assert len(indices['test']) == N_TEST
def test_balanced_random_separation(classes, N_POINTS):
indices = balanced_random_indices(_dummy_indices_method,
classes,
N_POINTS,
seed=1)
assert set(indices['train']) & set(indices['valid']) == set()
assert set(indices['train']) & set(indices['test']) == set()
assert set(indices['valid']) & set(indices['test']) == set()
def test_balanced_random_deterministic(classes, N_POINTS):
indices = balanced_random_indices(_dummy_indices_method,
classes,
N_POINTS,
seed=1)
new_indices = balanced_random_indices(_dummy_indices_method,
classes,
N_POINTS,
seed=1)
for key in new_indices.keys():
assert all(new_indices[key] == indices[key])
new_indices = balanced_random_indices(_dummy_indices_method,
classes,
N_POINTS,
seed=2)
for key in new_indices.keys():
assert any(new_indices[key] != indices[key])
def test_balanced_random_balance(classes, N_TRAIN, N_VALID, N_TEST, N_POINTS,
N_CLASSES):
indices = balanced_random_indices(_dummy_indices_method,
classes,
N_POINTS,
seed=1)
for class_indices in classes:
assert len(set(indices['train'])
& set(class_indices)) == N_TRAIN // N_CLASSES
assert len(set(indices['valid'])
& set(class_indices)) == N_VALID // N_CLASSES
assert len(set(indices['test'])
& set(class_indices)) == N_TEST // N_CLASSES
def split(datasets, data_size, seed, ratio, index, balanced=True):
if balanced:
info('Using balanced bootstrap')
return balanced_random_indices(method=bootstrap_random_indices,
classes=datasets.classes,
n_points=data_size,
seed=seed,
split_ratio=ratio)
else:
info('Using unbalanced bootstrap')
n_points = len(datasets)
n_test = int(numpy.ceil(n_points * ratio))
n_valid = n_test
n_train = n_points - n_test - n_valid
rng = numpy.random.RandomState(int(seed))
return bootstrap_random_indices(rng, range(n_points), n_train, n_valid,
n_test)
def split(datasets, data_size, seed, ratio, index):
return balanced_random_indices(method=constrained_bootstrap_random_indices,
classes=datasets.classes,
n_points=data_size,
seed=seed)