def test_dataset_with_transform(self):
for r in self.rs:
for k in self.ks:
for d in self.ds:
for n in self.ns:
data = random.RandomData(k=k, r=r, d=d)
X, _, Y = data.generate_samples(n)
ds = dataset.RandomDataset(data)
self.assertTrue(
len(ds) == len(data.data_points_labels) == n)
indexes = np.random.choice(np.arange(len(ds)),
size=max(10,
len(ds) // 10))
for index in indexes:
x, y = ds[index]
self.assertTrue(x.dtype == torch.float32)
self.assertTrue(y.dtype == torch.float32)
self.assertSequenceEqual(x.shape, (d, ))
self.assertTrue(y.detach().item() in [-1., 1.])
np.testing.assert_allclose(
x.detach().cpu().numpy(),
data.data_points[index, :],
rtol=1e-6,
atol=1e-6)
self.assertTrue(y.detach().item() ==
data.data_points_labels[index])
def test_dataloader_without_shuffle(self):
for r in self.rs:
for k in self.ks:
for d in self.ds:
for n in self.ns:
data = random.RandomData(k=k, r=r, d=d)
x, _, y = data.generate_samples(n)
ds = dataset.RandomDataset(data)
dataloader = DataLoader(ds,
shuffle=False,
batch_size=BATCH_SIZE)
i = 0
x_shapes = []
y_shapes = []
for index, (x, y) in enumerate(dataloader):
self.assertTrue(index == i)
x_shape = x.shape
y_shape = y.shape
self.assertTrue(len(x_shape) == 2)
self.assertTrue(len(y_shape) == 2)
self.assertTrue(x_shape[1] == d)
self.assertTrue(y_shape[1] == 1)
self.assertTrue(x.dtype == torch.float32)
self.assertTrue(y.dtype == torch.float32)
self.assertTrue((y == 1).sum() +
(y == -1).sum() == len(y))
np.testing.assert_allclose(
x.detach().cpu().numpy(),
data.data_points[index *
BATCH_SIZE:(index + 1) *
BATCH_SIZE, :],
rtol=1e-6,
atol=1e-6)
self.assertTrue(
(y.detach().cpu().numpy()[:, 0] == data.
data_points_labels[index *
BATCH_SIZE:(index + 1) *
BATCH_SIZE]).all())
x_shapes.append(x_shape)
y_shapes.append(y_shape)
i += 1
self.assertTrue(i == len(x_shapes) == len(y_shapes))
self.assertTrue(
len(x_shapes) == math.ceil(
len(dataloader.dataset) / BATCH_SIZE))
self.assertTrue(
all([
x_shapes[j][0] == y_shapes[j][0]
for j in range(len(x_shapes))
]))
self.assertTrue(x_shapes[-1][0] <= BATCH_SIZE)
self.assertTrue(
all([
y_shapes[j] == (BATCH_SIZE, 1)
for j in range(len(x_shapes) - 1)
]))
def test_train_val_test_split(self):
r, k, d, n = 1., 4, 20, 10000
n_train = 6000
n_val = 2000
n_test = 2000
shuffled_indexes = np.arange(n)
np.random.shuffle(shuffled_indexes)
train_indexes = shuffled_indexes[:n_train]
val_indexes = shuffled_indexes[n_train:n_train + n_val]
test_indexes = shuffled_indexes[n_train + n_val:]
data = random.RandomData(k=k, r=r, d=d)
data.generate_samples(n)
ds = dataset.RandomDataset(data)
training_dataset = Subset(ds, train_indexes)
val_dataset = Subset(ds, val_indexes)
test_dataset = Subset(ds, test_indexes)
self.assertTrue(len(training_dataset) == n_train)
self.assertTrue(len(val_dataset) == n_val)
self.assertTrue(len(test_dataset) == n_test)
train_dataloader = DataLoader(training_dataset,
shuffle=True,
batch_size=BATCH_SIZE)
val_dataloader = DataLoader(val_dataset,
shuffle=True,
batch_size=BATCH_SIZE)
test_dataloader = DataLoader(test_dataset,
shuffle=True,
batch_size=BATCH_SIZE)
i = 0
for index, (x, y) in enumerate(train_dataloader):
i += len(x)
self.assertTrue(i == n_train)
i = 0
for index, (x, y) in enumerate(val_dataloader):
i += len(x)
self.assertTrue(i == n_val)
i = 0
for index, (x, y) in enumerate(test_dataloader):
i += len(x)
self.assertTrue(i == n_test)
def setUp(self) -> None:
config_file = os.path.join('../../pytorch/configs', 'wide_two_layer_net.yaml')
with open(config_file, 'r') as stream:
try:
config_dict = yaml.safe_load(stream)
except yaml.YAMLError as e:
raise Exception("Exception while reading yaml file {} : {}".format(config_file, e))
self.config = ModelConfig(config_dict=config_dict)
self.two_layer_net = TwoLayerNet(self.config, train_hidden=True)
r, k, n = 1., 4, 5000
d = config_dict['architecture']['input_size']
data = random.RandomData(k=k, r=r, d=d, n=n)
data.generate_samples()
self.ds = dataset.RandomDataset(data)
self.data_loader = DataLoader(self.ds, shuffle=True, batch_size=BATCH_SIZE)
def test_dataset_without_transform(self):
for r in self.rs:
for k in self.ks:
for d in self.ds:
for n in self.ns:
data = random.RandomData(k=k, r=r, d=d)
x, _, y = data.generate_samples(n)
ds = dataset.RandomDataset(data, transform=None)
self.assertTrue(
len(ds) == len(data.data_points_labels) == n)
indexes = np.random.choice(np.arange(len(ds)),
size=max(10,
len(ds) // 10))
for index in indexes:
x, y = ds[index]
self.assertSequenceEqual(x.shape, (d, ))
self.assertTrue(y in [-1, 1])
np.testing.assert_allclose(
x,
data.data_points[index, :],
rtol=1e-9,
atol=1e-9)
self.assertTrue(
y == data.data_points_labels[index])
def _generate_data(k, r, d, n):
data = random.RandomData(k=k, r=r, d=d, n=n)
data.generate_samples()
return dataset.RandomDataset(data)
def _generate_data(k, r, d, n) -> torch.utils.data.Dataset:
data = random.RandomData(k=k, r=r, d=d, n=n)
data.generate_samples()
return dataset.RandomDataset(data)