def test_binary_balanced_length(self):
"""Test sampler length with mode='balanced' for binary-1"""
sampler = ClassificationDataSampler(
self.binary_1_dataset, shuffle=True,
target_transform=self.binary_target_transform,
mode='balanced')
self.assertEqual(len(sampler), 4)
def test_binary_balanced_no_shuffle(self):
"""Test sampling without shuffling on mode='balanced' for binary"""
sampler = ClassificationDataSampler(
self.binary_1_dataset, shuffle=False,
target_transform=self.binary_target_transform,
mode='balanced')
indices = list(sampler)
assert_array_equal(indices, [0, 1, 3, 2])
def test_default_no_shuffle(self):
"""Test sampling without shuffling on mode='default'"""
sampler = ClassificationDataSampler(
self.binary_1_dataset, shuffle=False,
target_transform=self.binary_target_transform, mode='default')
indices = list(sampler)
true_indices = [0, 1, 2, 3, 4]
# without shuffle=True, the ordering should be the default ordering
self.assertEqual(indices, true_indices)
def test_multi_balanced_no_shuffle(self):
"""Test sampling on mode='balanced' for multiclass no shuffle"""
sampler = ClassificationDataSampler(
self.multi_dataset, shuffle=False,
target_transform=self.multi_target_transform,
mode='balanced')
indices = list(sampler)
true_indices = [2, 1, 0]
# with shuffle=False, the first occurence of each
# class should be returned
self.assertEqual(indices, true_indices)
def test_binary_1_balanced_shuffle(self):
"""Test sampling with shuffling on mode='balanced' for binary-1"""
sampler = ClassificationDataSampler(
self.binary_1_dataset, shuffle=True,
target_transform=self.binary_target_transform,
mode='balanced')
indices = list(sampler)
for i, index in enumerate(indices):
if not i % 2:
target_label = 0
else:
target_label = 1
self.assertEqual(sampler.labels[index], target_label)
def test_multi_balanced_with_shuffle(self):
"""Test sampling on mode='balanced' for multiclass with shuffle"""
sampler = ClassificationDataSampler(
self.multi_dataset, shuffle=True,
target_transform=self.multi_target_transform,
mode='balanced')
indices = list(sampler)
for i, index in enumerate(indices):
if not i % 3:
target_label = 0
elif i % 3 == 1:
target_label = 1
else:
target_label = 2
self.assertEqual(sampler.labels[index], target_label)