def test_backward_indirectly(
fix_seed,
test_input,
test_weight,
test_bias,
test_mode,
expected_weight_grad,
expected_input_grad,
):
binarized_conv2d(test_input, test_weight, test_bias, 1, 0, 1, 1,
test_mode).backward()
assert torch.allclose(
input=test_input.grad,
other=expected_input_grad,
rtol=1e-04,
atol=1e-04,
equal_nan=True,
)
assert torch.allclose(
input=test_weight.grad,
other=expected_weight_grad,
rtol=1e-04,
atol=1e-04,
equal_nan=True,
)
def test_forward(fix_seed, test_input, test_weight, test_bias, test_mode,
expected):
assert torch.allclose(
input=binarized_conv2d(test_input, test_weight, test_bias, 1, 0, 1, 1,
test_mode),
other=expected,
rtol=1e-04,
atol=1e-04,
equal_nan=True,
)
def forward(self, input: torch.Tensor) -> torch.Tensor:
self.clipping()
return binarized_conv2d(
input,
self.weight,
self.bias,
self.stride,
self.padding,
self.dilation,
self.groups,
self.mode,
)
def test_forward(fix_seed, test_input, test_weight, test_bias, test_mode, expected):
answer = binarized_conv2d(test_input, test_weight, test_bias, 1, 0, 1, 1, test_mode)
logger.debug(f"answer: {answer}")
logger.debug(f"test mode: {test_mode}")
logger.debug(f"expected: {expected}")
assert torch.allclose(
input=answer,
other=expected,
rtol=1e-04,
atol=1e-04,
equal_nan=True,
)
def test_supported_mode(fix_seed, test_input, test_weight, test_bias,
test_mode):
with pytest.raises(RuntimeError):
binarized_conv2d(test_input, test_weight, test_bias, 1, 0, 1, 1,
test_mode)