mindspore.ops.tensor_split

mindspore.ops.tensor_split(input, indices_or_sections, axis=0)

Splits a tensor into multiple sub-tensors along the given axis.

Parameters

• input (Tensor) – A Tensor to be divided.

• indices_or_sections (Union[int, tuple(int), list(int)]) –

  • If indices_or_sections is an integer n, input tensor will be split into n sections.

    • If $input.shape[axis]$ can be divisible by n, sub-sections will have equal size $input.shape[axis]/n$ .

    • If $input.shape[axis]$ is not divisible by n, the first $input.shape[axis]modn$ sections will have size $input.shape[axis]//n+1$ , and the rest will have size $input.shape[axis]//n$ .

  • If indices_or_sections is of type tuple(int) or list(int), the input tensor will be split at the indices in the list or tuple. For example, given parameters $indices\mathrm{\_}or\mathrm{\_}sections=[1,4]$ and $axis=0$ , the input tensor will be split into sections $input[:1]$ , $input[1:4]$ , and $input[4:]$ .

• axis (int) – The axis along which to split. Default: 0 .

Returns

A tuple of sub-tensors.

Raises

• TypeError – If argument input is not Tensor.

• TypeError – If argument axis is not int.

• ValueError – If argument axis is out of range of $[-input.ndim,input.ndim)$ .

• TypeError – If each element in 'indices_or_sections' is not integer.

• TypeError – If argument indices_or_sections is not int, tuple(int) or list(int).

Supported Platforms:

Ascend GPU CPU

Examples

>>> import numpy as np
>>> from mindspore import Tensor, ops
>>> input_x = np.arange(9).astype("float32")
>>> output = ops.tensor_split(Tensor(input_x), 3)
>>> print(output)
(Tensor(shape=[3], dtype=Float32, value= [ 0.00000000e+00,  1.00000000e+00,  2.00000000e+00]),
Tensor(shape=[3], dtype=Float32, value= [ 3.00000000e+00,  4.00000000e+00,  5.00000000e+00]),
Tensor(shape=[3], dtype=Float32, value= [ 6.00000000e+00,  7.00000000e+00,  8.00000000e+00]))