mindspore.ops.logsumexp

mindspore.ops.logsumexp(input, dim, keepdim=False)

Calculate the log of summed exponentials of the tensor along a specified dimension.

$$logsumexp(input)=\log (\sum (e^{input-inpu{t}_{max}}))+inpu{t}_{max}$$

Parameters

• input (Tensor) – The input tensor.

• dim (Union[int, tuple(int), list(int)]) – Specify the dimension for computation. If () , compute all elements in the input .

• keepdim (bool, optional) – Whether the output tensor has dim retained. Default False .

Returns

Tensor

Supported Platforms:

Ascend GPU CPU

Examples

>>> import mindspore
>>> input = mindspore.tensor([[9., 3, 4, 5],
...                           [5, 2, 7, 4],
...                           [8, 1, 3, 6]])
>>> # case 1: By default, compute the log of summed exponential of all elements.
>>> output = mindspore.ops.logsumexp(input, ())
>>> print(output)
9.475807
>>>
>>> # case 2: Compute the log of summed exponential along dim 0.
>>> output = mindspore.ops.logsumexp(input, 0)
>>> print(output)
[9.326562  3.4076054 7.065884  6.4076056]
>>>
>>> # case 3: If keepdim=True, the output shape will be same of that of the input.
>>> output = mindspore.ops.logsumexp(input, 1, True)
>>> print(output)
[[9.02716 ]
 [7.175515]
 [8.133643]]