mcts
A simple vanilla Monte-Carlo Tree Search implementation in python
TODO :
-
The MCTS logic is now using neural network as its simulation (evaluation or rollout function) backbone engine, will consider to use Deep Q-learning method later
-
Investigate the PUCT formula more in-depth especially Hoeffding’s Inequality and its weakness. Study the performance quality of exploration/exploitation solved by PUCT which is mathematically represented by Regret analysis
-
Review the BAI-MCTS paper : Monte-Carlo Tree Search by Best Arm Identification , and Adversarial Bandit Environments
Credit: Thanks to kind folks (@crem , @Naphthalin , @oscardssmith) from Leela-Zero community , and @Lattimore from DeepMind