mindspore.mint.logsumexp

mindspore.mint.logsumexp(input, dim, keepdim=False)[source]

Computes the logarithm of the sum of exponentiations of all elements along the specified dim dimension of the input (with numerical stabilization), and retains the dimension based on the keepdim parameter.

\[logsumexp(input) = \log(\sum(e^{input-input_{max}})) + input_{max}\]

Warning

This is an experimental API that is subject to change or deletion.

Parameters

  • input (Tensor) – Input Tensor.

  • dim (Union[int, tuple(int), list(int)], optional) – The dimension to be reduced (the value should be within [0, len(input.shape) - 1]), when the dim is (), all dimensions are reduced.

  • keepdim (bool, optional) – Whether the output tensor retains the dimension dim, default: False.

Returns

Tensor, the dtype changes according to the input.dtype, and the shape changes according to the values of dim and keepdim.

  • If input.dtype is in [float16, float32, bfloat16], the output dtype is the same as the input.dtype.

  • If input.dtype is an integer or boolean type, the output dtype is float32.

  • If dim is (), and keepdim is False, the output is a 0-D tensor representing the logarithm of the sum of exponentiations of all elements in the input tensor.

  • If dim is 1, and keepdim is False, the shape of output is \((input.shape[0], input.shape[2], ..., input.shape[n])\).

  • If dim is (1, 2), and keepdim is False, the shape of output is \((input.shape[0], input.shape[3], ..., input.shape[n])\).

Raises

  • TypeError – If input is not a Tensor.

  • TypeError – If dtype of input is not one of: bool, int8, int16, int32, int64, uint8, float16, float32, bfloat16.

  • TypeError – If dim is not an int or tuple(int) or list(list).

  • TypeError – If keepdim is not a bool.

  • ValueError – If the value of any elements of dim is not in the range [0, len(input.shape) - 1].

  • RuntimeError – If any element of dim is repeated.

Supported Platforms:

Ascend

Examples

>>> import numpy as np
>>> from mindspore import Tensor, mint
>>> x = Tensor(np.random.randn(3, 4, 5, 6).astype(np.float32))
>>> output = mint.logsumexp(x, 1, keepdim=True)
>>> print(output.shape)
(3, 1, 5, 6)