mindspore.ops.scatter_div

mindspore.ops.scatter_div(input_x, indices, updates)[源代码]

根据指定更新值和输入索引通过除法操作更新输入数据的值。 该操作在更新完成后输出 input_x ,这样方便使用更新后的值。

对于 indices.shape 的每个 \(i, ..., j\)

\[\text{input_x}[\text{indices}[i, ..., j], :] \mathrel{/}= \text{updates}[i, ..., j, :]\]

输入的 input_xupdates 遵循隐式类型转换规则,以确保数据类型一致。如果数据类型不同,则低精度数据类型将转换为高精度的数据类型。当 updates 不支持转成 input_x 需要的数据类型时,则会抛出RuntimeError异常。

参数:
  • input_x (Parameter) - scatter_div的输入,数据类型为Parameter。

  • indices (Tensor) - 指定相除操作的索引,数据类型必须为mindspore.int32或者mindspore.int64。

  • updates (Tensor) - 指定与 input_x 相除的Tensor,数据类型与 input_x 相同,shape为 indices.shape + input_x.shape[1:]

返回:

Tensor,更新后的 input_x ,shape和类型与 input_x 相同。

异常:
  • TypeError - indices 不是int32或者int64。

  • ValueError - updates 的shape不等于 indices.shape + input_x.shape[1:]

  • RuntimeError - 当 input_xupdates 类型不一致,需要进行类型转换时,如果 updates 不支持转成参数 input_x 需要的数据类型,就会报错。

  • RuntimeError - 在Ascend平台上,输入的 input_xindicesupdates 的数据维度大于八维。

支持平台:

Ascend GPU CPU

样例:

>>> input_x = Parameter(Tensor(np.array([[6.0, 6.0, 6.0], [2.0, 2.0, 2.0]]), mindspore.float32), name="x")
>>> indices = Tensor(np.array([0, 1]), mindspore.int32)
>>> updates = Tensor(np.array([[2.0, 2.0, 2.0], [2.0, 2.0, 2.0]]), mindspore.float32)
>>> output = ops.scatter_div(input_x, indices, updates)
>>> print(output)
[[3. 3. 3.]
 [1. 1. 1.]]
>>> # for input_x will be updated after the operation is completed. input_x need to be re-initialized.
>>> input_x = Parameter(Tensor(np.array([[105.0, 105.0, 105.0],
...                                      [315.0, 315.0, 315.0]]), mindspore.float32), name="x")
>>> # for indices = [[0, 1], [1, 1]]
>>> # step 1: [0, 1]
>>> # input_x[0] = [105.0, 105.0, 105.0] / [1.0, 1.0, 1.0] = [105.0, 105.0, 105.0]
>>> # input_x[1] = [315.0, 315.0, 315.0] / [3.0, 3.0, 3.0] = [105.0, 105.0, 105.0]
>>> # step 2: [1, 1]
>>> # input_x[1] = [105.0, 105.0, 105.0] / [5.0, 5.0, 5.0] = [21.0, 21.0, 21.0]
>>> # input_x[1] = [21.0, 21.0, 21.0] / [7.0, 7.0, 7.0] = [3.0, 3.0, 3.0]
>>> indices = Tensor(np.array([[0, 1], [1, 1]]), mindspore.int32)
>>> updates = Tensor(np.array([[[1.0, 1.0, 1.0], [3.0, 3.0, 3.0]],
...                            [[5.0, 5.0, 5.0], [7.0, 7.0, 7.0]]]), mindspore.float32)
>>> output = ops.scatter_div(input_x, indices, updates)
>>> print(output)
[[105. 105. 105.]
 [  3.   3.   3.]]
>>> # for input_x will be updated after the operation is completed. input_x need to be re-initialized.
>>> input_x = Parameter(Tensor(np.array([[105.0, 105.0, 105.0],
...                                      [315.0, 315.0, 315.0]]), mindspore.float32), name="x")
>>> # for indices = [[1, 0], [1, 1]]
>>> # step 1: [1, 0]
>>> # input_x[0] = [105.0, 105.0, 105.0] / [3.0, 3.0, 3.0] = [35.0, 35.0, 35.0]
>>> # input_x[1] = [315.0, 315.0, 315.0] / [1.0, 1.0, 1.0] = [315.0, 315.0, 315.0]
>>> # step 2: [1, 1]
>>> # input_x[1] = [315.0, 315.0, 315.0] / [5.0, 5.0, 5.0] = [63.0 63.0 63.0]
>>> # input_x[1] = [63.0 63.0 63.0] / [7.0, 7.0, 7.0] = [9.0, 9.0, 9.0]
>>> indices = Tensor(np.array([[1, 0], [1, 1]]), mindspore.int32)
>>> updates = Tensor(np.array([[[1.0, 1.0, 1.0], [3.0, 3.0, 3.0]],
...                            [[5.0, 5.0, 5.0], [7.0, 7.0, 7.0]]]), mindspore.float32)
>>> output = ops.scatter_div(input_x, indices, updates)
>>> print(output)
[[35. 35. 35.]
 [ 9.  9.  9.]]