mindspore.ops.ScatterNd

class mindspore.ops.ScatterNd[源代码]

根据指定的索引将更新值散布到新Tensor上。

更多参考详见 mindspore.ops.scatter_nd()。

输入：

indices (Tensor) - 指定新Tensor中散布的索引，数据类型为int32或int64。索引的秩须至少为2，并且 indices_shape[-1] <= len(shape) 。
updates (Tensor) - 指定更新Tensor，shape为 indices_shape[:-1] + shape[indices_shape[-1]:] 。
shape (tuple[int]) - 指定输出Tensor的shape，数据类型与索引相同。 shape 为 \((x_1, x_2, ..., x_R)\) 且 shape 的长度大于或等于2。换句话说， shape 至少是 \((x_1, x_2)\) ，且 shape 中的任何元素的值都必须大于等于1。也就是说， \(x_1\) >= 1， \(x_2\) >= 1。

输出：

Tensor，更新后的Tensor，数据类型与输入 update 相同，shape与输入 shape 相同。

支持平台：

Ascend GPU CPU

样例：

>>> op = ops.ScatterNd()
>>> indices = Tensor(np.array([[0], [2]]), mindspore.int32)
>>> updates = Tensor(np.array([[[1, 1, 1, 1], [2, 2, 2, 2],
...                             [3, 3, 3, 3], [4, 4, 4, 4]],
...                            [[1, 1, 1, 1], [2, 2, 2, 2],
...                             [3, 3, 3, 3], [4, 4, 4, 4]]]), mindspore.float32)
>>> shape = (4, 4, 4)
>>> output = op(indices, updates, shape)
>>> print(output)
[[[1. 1. 1. 1.]
  [2. 2. 2. 2.]
  [3. 3. 3. 3.]
  [4. 4. 4. 4.]]
 [[0. 0. 0. 0.]
  [0. 0. 0. 0.]
  [0. 0. 0. 0.]
  [0. 0. 0. 0.]]
 [[1. 1. 1. 1.]
  [2. 2. 2. 2.]
  [3. 3. 3. 3.]
  [4. 4. 4. 4.]]
 [[0. 0. 0. 0.]
  [0. 0. 0. 0.]
  [0. 0. 0. 0.]
  [0. 0. 0. 0.]]]
>>> indices = Tensor(np.array([[0, 1], [1, 1]]), mindspore.int32)
>>> updates = Tensor(np.array([3.2, 1.1]), mindspore.float32)
>>> shape = (3, 3)
>>> output = op(indices, updates, shape)
>>> # In order to facilitate understanding, explain the operator pseudo-operation process step by step:
>>> # Step 1: Generate an empty Tensor of the specified shape according to the shape
>>> # [
>>> #     [0. 0. 0.]
>>> #     [0. 0. 0.]
>>> #     [0. 0. 0.]
>>> # ]
>>> # Step 2: Modify the data at the specified location according to the indicators
>>> # 0th row of indices is [0, 1], 0th row of updates is 3.2.
>>> # means that the empty tensor in the 0th row and 1st col set to 3.2
>>> # [
>>> #     [0. 3.2. 0.]
>>> #     [0. 0.   0.]
>>> #     [0. 0.   0.]
>>> # ]
>>> # 1th row of indices is [1, 1], 1th row of updates is 1.1.
>>> # means that the empty tensor in the 1th row and 1st col set to 1.1
>>> # [
>>> #     [0. 3.2. 0.]
>>> #     [0. 1.1  0.]
>>> #     [0. 0.   0.]
>>> # ]
>>> # The final result is as follows:
>>> print(output)
[[0. 3.2 0.]
 [0. 1.1 0.]
 [0. 0.  0.]]