AlphaZero: The Visit Distribution
Intermediate
The problem, from first principles
After search, AlphaZero uses the visit counts of the root moves as the training target for the policy. A temperature controls how sharp that target is: high temperature explores, low temperature plays the most-visited move.
From theory to code
Implement visit_distribution(counts, tau), returning the normalized counts raised to 1 / tau.
Constraints
- Counts are positive.
Hints
Hint 1
Raise each count to 1 / tau, then divide by the total.
Keywords
research-papersreinforcement-learningself-playsearch
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.