def provide_few_shot_dataset_splits(
self, dataset_splits: DatasetSplits
) -> FewShotDatasetSplits:
valid = create_few_shot_datasets(
dataset_splits.valid, self.config.dataset.num_samples
)
test = create_few_shot_datasets(
dataset_splits.test, self.config.dataset.num_samples
)
return FewShotDatasetSplits(dataset_splits.train, valid, test)
def test_shouldNotHaveDupplicatedSamples(self):
dataset = create_random_dataset(NUM_ITEMS, NUM_CLASSES, SHAPE)
train_dataset, test_dataset = create_few_shot_datasets(dataset, 5)
samples = unique_samples([train_dataset, test_dataset])
self.assertEqual(len(samples), len(train_dataset) + len(test_dataset))
def test_givenOneNumSample_trainDatasetShouldHaveOneSamplePerClass(self):
dataset = create_random_dataset(NUM_ITEMS, NUM_CLASSES, SHAPE)
train_dataset, test_dataset = create_few_shot_datasets(dataset, 1)
classes_train = unique_classes(train_dataset)
self.assertEqual(len(train_dataset), NUM_CLASSES)
self.assertEqual(len(classes_train), NUM_CLASSES)
def test_given5NumSamples_trainDatasetShouldHave5SamplesPerClass(self):
num_samples = 5
# We add more items to make sure all classes are included
dataset = create_random_dataset(NUM_ITEMS * num_samples, NUM_CLASSES,
SHAPE)
train_dataset, test_dataset = create_few_shot_datasets(
dataset, num_samples)
classes_train = unique_classes(train_dataset)
self.assertEqual(len(train_dataset), num_samples * NUM_CLASSES)
self.assertEqual(len(classes_train), NUM_CLASSES)
def setUp(self):
samples = create_random_dataset(NUM_ITEMS, NUM_CLASSES,
SHAPE).samples # type: ignore
self.dataset = create_few_shot_datasets(ListDataset(samples), N_WAY)
self.dataloader_factory = DataLoaderFactory(BATCH_SIZE, SHUFFLE,
PIN_MEMORY)
def test_shouldKeepAllSamples(self):
dataset = create_random_dataset(NUM_ITEMS, NUM_CLASSES, SHAPE)
train_dataset, test_dataset = create_few_shot_datasets(dataset, 5)
self.assertEqual(NUM_ITEMS, len(train_dataset) + len(test_dataset))