def test_forward_pass_with_invalid_padding_value(self):
# given
w = np.random.rand(5, 5, 3, 16)
b = np.random.rand(16)
activation = np.random.rand(16, 11, 11, 3)
padding = 'lorem ipsum'
# when
layer = SuperFastConvLayer2D(w=w, b=b, padding=padding)
with pytest.raises(InvalidPaddingModeError):
_ = layer.forward_pass(activation, training=True)
def test_backward_pass_only_size_valid_padding(self):
# given
activation = np.random.rand(64, 11, 11, 3)
w = np.random.rand(5, 5, 3, 16)
b = np.random.rand(16)
layer = SuperFastConvLayer2D(w=w, b=b, padding='valid')
# when
forward_result = layer.forward_pass(activation, training=True)
backward_result = layer.backward_pass(forward_result)
# then
assert backward_result.shape == activation.shape
def test_forward_pass_with_valid_padding(self):
# given
w = np.random.rand(5, 5, 3, 16)
b = np.random.rand(16)
activation = np.random.rand(16, 11, 11, 3)
padding = 'valid'
# when
layer = SuperFastConvLayer2D(w=w, b=b, padding=padding)
result = layer.forward_pass(activation, training=True)
assert result.shape == (16, 7, 7, 16)
expected_val = np.sum(
w[:, :, :, 0] * activation[0, 0:5, 0:5, :]) + b[0]
assert abs(expected_val - result[0, 0, 0, 0]) < 1e-8