A blazing-fast, optimal tic-tac-toe agent written in C++.
Some of the numerous optimizations made:
- Negamax search with alpha-beta pruning.
- Representation of positions as bitboards and use of bitwise operators for fast state checking.
- Optimized move-ordering for early cut-offs.
- Transposition table storing the results of previously searched nodes.
- Leaf evaluation heuristics (leaf nodes are scored based on the number of moves played and the game result).
This implementation is a proof of concept: tic-tac-toe has a relatively small search tree when compared to games like chess or Go and can be solved without such optimizations.
However, it demonstrates the application of high-level search algorithms that can be utilized on games with significantly higher complexity.
See the performance benchmarking results (each the average of 10,000 trials):
| Initial Board State | Parameters | Mean Time |
|---|---|---|
| Empty | Transposition table reset between trials | 309 μs |
| Empty | Transposition table preserved between trials | 41 μs |
You can experiment with the agent for yourself using
Benchmark.cpp.