mindspore.get_grad

mindspore.get_grad(gradients, identifier)[source]

When return_ids of mindspore.grad() or mindspore.grad() is set to True , use return value of mindspore.grad, or the second return value of mindspore.grad as gradients. Then find the specific gradient from gradients according to identifier .

As for gradient, two typical cases are included:

identifier is the position of the specific tensor to get gradient.
identifier is a parameter of a network.

Parameters

gradients (Union[tuple[int, Tensor], tuple[tuple, tuple]]) – The return value of mindspore.grad() when return_ids is set to True.
identifier (Union[int, Parameter]) – The position number of a tensor, or a parameter that is used in mindspore.grad().

Returns

The gradient of the tensor on the position or in the parameter that specified by the identifier.

Raises

RuntimeError – If gradient is not found.
TypeError – If type of Args does not belong to required ones.

Supported Platforms:: Ascend GPU CPU

Examples

>>> import mindspore
>>> from mindspore import Tensor, nn
>>> from mindspore import grad, get_grad
>>>
>>>  # Cell object to be differentiated
>>> class Net(nn.Cell):
...     def construct(self, x, y, z):
...         return x * y * z
>>> x = Tensor([1, 2], mindspore.float32)
>>> y = Tensor([-2, 3], mindspore.float32)
>>> z = Tensor([0, 3], mindspore.float32)
>>> net = Net()
>>> out_grad = grad(net, grad_position=(1, 2), return_ids=True)(x, y, z)
>>> output = get_grad(out_grad, 1)
>>> print(output)
[0. 6.]