explorationbias.hashnode.devBatch MCTS, Part 2The previous post described a batch-inference MCTS implementation for self-play. Batches consisted of one state from each of N concurrent episodes. Unfortunately that approach doesn’t work for playing a competitive episode since in that case there’s ...Dec 1, 2024·2 min read
explorationbias.hashnode.devBatch MCTS, Part 1It’s not straightforward to get good performance out of AlphaZero’s Monte Carlo Tree Search. Search wants to perform inference for one game state at a time — the new leaf state visited at the end of each game simulation — and the results of that infe...Dec 1, 2024·2 min read
explorationbias.hashnode.devAlphaZero modificationsI had to make a couple of changes to AlphaZero to apply it to games with randomness and more than two players. More than two players Handling more than two players is the easy one. AlphaZero uses values in the range 0-1 to represent state values for ...Nov 29, 2024·3 min read
explorationbias.hashnode.devIntroductionHi. I plan to use this blog to document my personal project to build a platform for playtesting board games against AI opponents, starting with the game Kingdomino. A primary motivation for this project is to learn a bunch of new-to-me tools like Typ...Nov 28, 2024·3 min read