818492efd4238bd8fedcff105bd46044a714f762,pymanopt/autodiff/backends/_pytorch.py,_PyTorchBackend,compute_gradient,#_PyTorchBackend#,71

Before Change

    @Backend._assert_backend_available
    def compute_gradient(self, function, arguments):
        flattened_arguments = flatten_arguments(arguments)

        if len(flattened_arguments) == 1:
            def unary_gradient(argument):
                torch_argument = self._from_numpy(argument)
                torch_argument.requires_grad_()
                function(torch_argument).backward()
                return self._sanitize_gradient(torch_argument)
            return unary_gradient

        def nary_gradient(arguments):
            torch_arguments = []
            for argument in flatten_arguments(arguments):
                torch_argument = self._from_numpy(argument)

After Change

                torch_arguments.append(torch_argument)
            function(*torch_arguments).backward()
            return self._sanitize_gradients(torch_arguments)
        if len(arguments) == 1:
            return unpack_singleton_iterable_return_value(gradient)
        return gradient

    @Backend._assert_backend_available
    def compute_hessian_vector_product(self, function, arguments):
        def hessian_vector_product(*args):

Instances