mindspore.ops.one_hot

View Source On Gitee
mindspore.ops.one_hot(indices, depth, on_value=1, off_value=0, axis=- 1)[source]

Generate a new tensor, where the positions specified by indices are assigned on_value, and all other positions are assigned off_value.

Note

If the input indices has rank N, the output will have rank N+1. The new axis is created at dimension axis. On Ascend, if on_value is int64 dtype, indices must be int64 dtype, and the value for on_value and off_value can only be 1 and 0.

Parameters

  • indices (Tensor) – The input tensor of indices.

  • depth (int) – The depth of the one-hot.

  • on_value (Union[Tensor, int, float], optional) – The value used to fill indexed positions. Default 1 .

  • off_value (Union[Tensor, int, float], optional) – The value used to fill non-indexed positions. Default 0 .

  • axis (int, optional) – Specify the axis for computation. Default -1 .

Returns

Tensor

Supported Platforms:

Ascend GPU CPU

Examples

>>> import mindspore
>>> indices = mindspore.tensor([0, 1, 2, 4])
>>> mindspore.ops.one_hot(indices, depth=5)
Tensor(shape=[4, 5], dtype=Int64, value=
[[1, 0, 0, 0, 0],
 [0, 1, 0, 0, 0],
 [0, 0, 1, 0, 0],
 [0, 0, 0, 0, 1]])
>>>
>>> mindspore.ops.one_hot(indices, depth=3)
Tensor(shape=[4, 3], dtype=Int64, value=
[[1, 0, 0],
 [0, 1, 0],
 [0, 0, 1],
 [0, 0, 0]])
>>> # If shape of indices is (N, C), and axis=-1, the returned shape will be (N, C, depth).
>>> indices = mindspore.tensor([[0, 2], [1, -1]])
>>> mindspore.ops.one_hot(indices, depth=3, on_value=10, off_value=4, axis=-1)
Tensor(shape=[2, 2, 3], dtype=Int64, value=
[[[10,  4,  4],
  [ 4,  4, 10]],
 [[ 4, 10,  4],
  [ 4,  4,  4]]])
>>> # If axis=0, the returned shape will be (depth, N, C).
>>> mindspore.ops.one_hot(indices, depth=3, on_value=10, off_value=4, axis=0)
Tensor(shape=[3, 2, 2], dtype=Int64, value=
[[[10,  4],
  [ 4,  4]],
 [[ 4,  4],
  [10,  4]],
 [[ 4, 10],
  [ 4,  4]]])