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, ))