Log-Softmax
EasyMachine LearningMathArray
Description
Given an array of logits, return the log-softmax: each element minus the log-sum-exp of the array. Use the max-subtraction trick for numerical stability. Round each result to 4 decimal places.
Examples
Input:
[1,2,3]Output:
[-2.4076,-1.4076,-0.4076]Explanation:
Each logit has the log-sum-exp of all logits subtracted from it, giving the logarithm of the softmax probabilities.
Input:
[0,0]Output:
[-0.6931,-0.6931]Explanation:
Each logit has the log-sum-exp of all logits subtracted from it, giving the logarithm of the softmax probabilities.
Input:
[5]Output:
[0]Explanation:
Each logit has the log-sum-exp of all logits subtracted from it, giving the logarithm of the softmax probabilities.
Constraints
- •
1 ≤ length ≤ 10³