def test_format_of_x(self):
dataset = SequencesDataset()
x, y = dataset[0]
self.assertIsInstance(x, torch.FloatTensor)
self.assertEqual(len(x.shape), 2)
self.assertGreater(x.shape[0], 8)
self.assertEqual(x.shape[1], 1)
def test_train_one_epoch(self):
dataset = SequencesDataset(set_size=10)
mean_loss = self.interface.train_one_epoch(dataset,
weight_decay=0,
learning_rate=0.02,
shuffle=False)
self.assertIsInstance(mean_loss, float)
def test_performance_on_samples(self):
dataset = SequencesDataset(set_size=10)
performance = self.interface.performance_on_samples(dataset)
sample = performance[0]
# raw_sample = dataset[0]
self.assertIsInstance(sample.x, torch.FloatTensor)
self.assertEqual(len(sample.x.shape), 2)
self.assertGreater(sample.x.shape[0], 8)
self.assertEqual(sample.x.shape[1], 1)
self.assertIsInstance(sample.loss, float)
self.assertIsInstance(sample.actual_y, float)
self.assertIsInstance(sample.predicted_y, float)
def test___preprocessed_batch__(self):
batch_size = 5
batch = SequencesDataset(set_size=batch_size)[:batch_size]
x, y, normalization_factors, normalization_offsets = self.interface.__preprocessed_batch__(
batch)
self.assertIsInstance(x, torch.nn.utils.rnn.PackedSequence)
x_unpacked, x_lengths = torch.nn.utils.rnn.pad_packed_sequence(
x, batch_first=True)
self.assertEqual(x_unpacked.shape[0], batch_size)
self.assertGreater(x_unpacked.shape[1], 8)
self.assertEqual(x_unpacked.shape[2], 1)
self.assertIsInstance(y, torch.FloatTensor)
# y_unpacked, y_lengths = torch.nn.utils.rnn.pad_packed_sequence(y, batch_first=True)
self.assertEqual(y.shape, (batch_size, 1))
self.assertFalse(y.requires_grad)
self.assertIsInstance(normalization_factors, torch.FloatTensor)
self.assertEqual(normalization_factors.shape, (batch_size, ))
self.assertIsInstance(normalization_offsets, torch.FloatTensor)
self.assertEqual(normalization_offsets.shape, (batch_size, ))
def test___len__(self):
dataset = SequencesDataset()
self.assertEqual(len(dataset), 50000)
def test_format_of_y(self):
dataset = SequencesDataset()
x, y = dataset[0]
self.assertIsInstance(y, torch.FloatTensor)
self.assertEqual(y.shape, (1, 1))
def test___getitem__(self):
dataset = SequencesDataset()
samples = dataset[:2]
self.assertIsInstance(samples[0], tuple)
self.assertIsInstance(samples[1], tuple)
import dataset_parties from datasets.sequences_dataset import SequencesDataset dataset = SequencesDataset() train_set, dev_set, test_set = dataset_parties.ng_style(dataset)
def test_performance_on_set(self):
dataset = SequencesDataset(set_size=10)
performance = self.interface.performance_on_set(dataset)
self.assertIsInstance(performance, float)