English | 中文版
jiangsier-gobang is a reinforcement-learning-based AI Gomoku game. It supports multiple AI players (each using different model structures). These AI players can learn through self-play or playing against each other.
You can start from scratch using the Monte Carlo tree for training. Alternatively, you can use game records from approximately 5000 human games (in Smart Game Format) for training to obtain better initial parameters.
It supports multiple AI players. Model complexity: Ace < Fox < Baker < Casey < Darling < Ellis
Special players:
- Gill: Combines the predictions from Ace, Baker, Casey, Darling, and Fox with different weights. The weight settings are based on the results of games played among the AI players. Tests show that its playing strength is higher than that of individual model AI player.
- Monkey: Uses random moves as priors and employs the Monte Carlo tree to calculate the next move.
- Teacher: A role based on strategy.
