MONTE CARLO TETRIS BOT
A tetris battle bot that plays using Monte Carlo Tree Search. (with some modifications)
Made at the Agency at Georgia Tech (the AI Club), for theaigames.com competition.
Consistently placed at the top 150 bots.
WHAT I DID
- Implemented MCTS, using the tetris game board as states
- Optimized search to not need to expand states before they're visited (40x performance boost).
- Optimized state evaluation function so it only takes O(n) (where n is number of tiles in the board).
- Came up with the idea of 'relaxed' evaluation functions, as failsafe alternatives to vanilla MCTS.