megengine.functional.nn.logsoftmax

logsoftmax(inp, axis)[源代码]

对一个n维的输入张量做 \(\log(\text{softmax}(x))\) 函数 \(\text{logsoftmax}(x)\) 的公式可以简化为:

\[\text{logsoftmax}(x_{i}) = \log(\frac{\exp(x_i) }{ \sum_j \exp(x_j)} )\]

为了提高数值稳定性,实现根据以下的变换:

\[\text{logsoftmax}(x) = \log (\frac{\exp (x)}{\sum_{i}(\exp (x_{i}))}) = x - \log (\sum_{i}(\exp (x_{i}))) = x - \text{logsumexp}(x)\]

实际案例

import numpy as np
from megengine import tensor
import megengine.functional as F

x = tensor(np.arange(-5, 5, dtype=np.float32)).reshape(2,5)
y = F.logsoftmax(x, axis=1)
print(y.numpy().round(decimals=4))

输出:

[[-4.4519 -3.4519 -2.4519 -1.4519 -0.4519]
 [-4.4519 -3.4519 -2.4519 -1.4519 -0.4519]]
返回类型

Tensor