mindspore.Tensor.scatter

Tensor.scatter(dim, index, src) → Tensor[source]

Update the value in src to self according to the specified index.

For a 3-D tensor, the output will be:

output[index[i][j][k]][j][k] = src[i][j][k]  # if dim == 0

output[i][index[i][j][k]][k] = src[i][j][k]  # if dim == 1

output[i][j][index[i][j][k]] = src[i][j][k]  # if dim == 2

Note

The backward is supported only for the case src.shape == index.shape when src is a tensor. The rank of the input tensor self must be at least 1. Each value of index must be in the valid range [0, self.shape[dim]) for the normalized dim. Negative or out-of-range index values are illegal input and will lead to undefined behavior.

Warning

Non-backward-compatible change after version 2.9.0: overload scatter(axis, index, src) will be removed, and only scatter(dim, index, src) will be supported.

Parameters:

  • dim (int) – Which axis to scatter. Accepted range is [-r, r) where r = rank(self).

  • index (Tensor) – The index to do update operation whose data type must be int32 or int64. Same rank as self .

  • src (Tensor, float) – The data doing the update operation with self. Can be a tensor with the same data type as self or a float number to scatter.

Returns:

Tensor, has the same shape and type as self .

Raises:

  • TypeError – If index is neither int32 nor int64.

  • ValueError – If rank of any of self , index and src is less than 1.

  • ValueError – If the rank of src is not equal to the rank of self .

  • TypeError – If the data types of self and src have different dtypes.

Supported Platforms:

Ascend GPU CPU

Examples

>>> import numpy as np
>>> import mindspore as ms
>>> from mindspore import Tensor
>>> input = Tensor(np.array([[1, 2, 3, 4, 5]]), dtype=ms.float32)
>>> src = Tensor(np.array([[8, 8]]), dtype=ms.float32)
>>> index = Tensor(np.array([[2, 4]]), dtype=ms.int64)
>>> out = input.scatter(dim=1, index=index, src=src)
>>> print(out)
[[1. 2. 8. 4. 8.]]
>>> input = Tensor(np.zeros((5, 5)), dtype=ms.float32)
>>> src = Tensor(np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]), dtype=ms.float32)
>>> index = Tensor(np.array([[0, 0, 0], [2, 2, 2], [4, 4, 4]]), dtype=ms.int64)
>>> out = input.scatter(dim=0, index=index, src=src)
>>> print(out)
[[1. 2. 3. 0. 0.]
 [0. 0. 0. 0. 0.]
 [4. 5. 6. 0. 0.]
 [0. 0. 0. 0. 0.]
 [7. 8. 9. 0. 0.]]
>>> input = Tensor(np.zeros((5, 5)), dtype=ms.float32)
>>> src = Tensor(np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]), dtype=ms.float32)
>>> index = Tensor(np.array([[0, 2, 4], [0, 2, 4], [0, 2, 4]]), dtype=ms.int64)
>>> out = input.scatter(dim=1, index=index, src=src)
>>> print(out)
[[1. 0. 2. 0. 3.]
 [4. 0. 5. 0. 6.]
 [7. 0. 8. 0. 9.]
 [0. 0. 0. 0. 0.]
 [0. 0. 0. 0. 0.]]
Tensor.scatter(axis, index, src) → Tensor[source]

Update the value in src to self according to the specified index. Refer to mindspore.ops.tensor_scatter_elements() for more details.

Note

The backward is supported only for the case src.shape == index.shape. The rank of the input tensor self must be at least 1. Each value of index must be in the valid range [0, self.shape[axis]) for the normalized axis. Negative or out-of-range index values are illegal input and will lead to undefined behavior.

Parameters:

  • axis (int) – Which axis to scatter. Accepted range is [-r, r) where r = rank(self).

  • index (Tensor) – The index to do update operation whose data type must be int32 or int64. Same rank as self .

  • src (Tensor, float) – The data doing the update operation with self. Can be a tensor with the same data type as self or a float number to scatter.

Returns:

Tensor, has the same shape and type as self .

Raises:

  • TypeError – If index is neither int32 nor int64.

  • ValueError – If rank of any of self , index and src is less than 1.

  • ValueError – If the rank of src is not equal to the rank of self .

  • TypeError – If the data types of self and src have different dtypes.

Supported Platforms:

Ascend GPU CPU

Examples

>>> import numpy as np
>>> import mindspore as ms
>>> from mindspore import Tensor
>>> input = Tensor(np.array([[1, 2, 3, 4, 5]]), dtype=ms.float32)
>>> src = Tensor(np.array([[8, 8]]), dtype=ms.float32)
>>> index = Tensor(np.array([[2, 4]]), dtype=ms.int64)
>>> out = input.scatter(axis=1, index=index, src=src)
>>> print(out)
[[1. 2. 8. 4. 8.]]
>>> input = Tensor(np.zeros((5, 5)), dtype=ms.float32)
>>> src = Tensor(np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]), dtype=ms.float32)
>>> index = Tensor(np.array([[0, 0, 0], [2, 2, 2], [4, 4, 4]]), dtype=ms.int64)
>>> out = input.scatter(axis=0, index=index, src=src)
>>> print(out)
[[1. 2. 3. 0. 0.]
 [0. 0. 0. 0. 0.]
 [4. 5. 6. 0. 0.]
 [0. 0. 0. 0. 0.]
 [7. 8. 9. 0. 0.]]
>>> input = Tensor(np.zeros((5, 5)), dtype=ms.float32)
>>> src = Tensor(np.array([[1, 2, 3], [4, 5, 6], [7, 8, 9]]), dtype=ms.float32)
>>> index = Tensor(np.array([[0, 2, 4], [0, 2, 4], [0, 2, 4]]), dtype=ms.int64)
>>> out = input.scatter(axis=1, index=index, src=src)
>>> print(out)
[[1. 0. 2. 0. 3.]
 [4. 0. 5. 0. 6.]
 [7. 0. 8. 0. 9.]
 [0. 0. 0. 0. 0.]
 [0. 0. 0. 0. 0.]]