A3C: Policy Entropy
Beginner
The problem, from first principles
The entropy of the action distribution measures how random the policy is. Adding an entropy bonus to the objective keeps the policy exploring instead of collapsing onto one action too early.
From theory to code
Implement entropy(p), the Shannon entropy of a discrete distribution, with the convention that zero probabilities contribute zero.
Constraints
- Use the natural logarithm.
Hints
Hint 1
Compute -p log p element-wise, treating zero entries as zero, then sum.
Keywords
research-papersreinforcement-learningactor-criticasynchronous
Python
Run your code to see results
No results yet
Run your code to check it against the examples, or submit to run every test.