def test_classification_error():
batch_size = 13
output_dim = 32
rng = np.random.RandomState(42)
targets = rng.randint(output_dim, size=(batch_size, 1))
dataset = DummyDataset()
dataset.symb_targets = targets
loss = ClassificationError(DummyModel(), dataset)
model_output = rng.rand(batch_size, output_dim).astype(np.float32)
# Test the shape of the output of _compute_losses.
losses = loss._compute_losses(model_output).eval()
assert_equal(losses.shape, (batch_size, ))
def test_L2_distance():
batch_size = 13
input_dim = 32
rng = np.random.RandomState(42)
targets = (rng.rand(batch_size, input_dim) > 0.5).astype(np.float32)
dataset = DummyDataset()
dataset.symb_targets = targets
loss = L2Distance(DummyModel(), dataset)
model_output = rng.rand(batch_size, input_dim).astype(np.float32)
# Test the shape of the output of _compute_losses.
losses = loss._compute_losses(model_output).eval()
assert_equal(losses.shape, (batch_size, ))