This project is only for Zhejiang University AI Course.
This project works well on mo.zju.edu.cn and actually takes first place over many advanced algorithms on a computer with restricted computing power.
- Minimax search with limited depth
- Believe the final result will be reversed even if our model is at disadvantage in the middle of a game
- Differential test with a slightly stronger RL model to fix some flaws in heuristic function until our model defeat it
64 in total, winning from 40 - 24 to 64 - 0. Detailed pictures are in doc.
First place. The only one winning all matches.