HexAIpy

Play Hex by yourself or against AI!

Installation

Install dependencies easily using:

pip install -r requirements.txt

How to run?

You can run the main program where you can play Hex against another (local) player or an AI using the following command:

python hexai.py

Optionally you can add the following flags:

Option	Action	Choices	Default
-p1	Choose Player 1 (Blue) type	{human, alphabeta, mcts}	human
-p2	Choose Player 2 (Red) type	{human, alphabeta, mcts}	mcts
-s, --size	Specify board size	int	4
-t, --use_tt	Use transposition table	bool	False
-b, --begin	Which player starts the game (1/2)	int	1

You can place Hex pieces by typing the coordinate when prompted. The coordinate should be contain a letter and a number to indicate the column and row (e.g., b3).

Experiments

The accompanying report contains various experiments that were performed on the different AI variants. To perform the experiments yourself, use the hexexperiments.py file.

The following experiments have been performed:

Alpha-Beta Comparison

Compares AB with a random evaluation function against AB's with the Dijkstra evaluation function:

python hexexperiments.py comp -o base -m

Iterative Deepening + Tranposition Table Comparison

Compares the IDTT agent against base AB agents:

python hexexperiments.py comp -o idtt -m

Monte-Carlo Tree Search Optimisation

Compares different MCTS agents to find the optimal parameter settings

python hexexperiments.py mcts

This experiment takes about 17 minutes to complete on a 12-core AMD Ryzen 5900x processor. The processing time can be decreased by changing the values in the n_options and cp_options arrays or by lowering the number of matches played by using the -n flag.

Monte-Carlo Tree Search Comparison

Compares the best MCTS agent against the IDTT agent

python hexexperiments.py comp -o mcts -m

(More statistics)

The exp_competition_old.py file contains a singlethreaded implementation of the experiments returning some more statistics. You can run it using:

python hexexperiments.py comp_old

Warning: These experiments can take a long time as only a single thread is used!

Other Experiment Flags

To play around more with the experiments, you can change the following flags:

Option	Action	Choices	Default
<test>	Experiment type	{comp, comp_multi, mcts}	-
-o, --option	What agents are tested when using comp / comp_multi	{base, idtt, mcts}	base
-s, --size	Specify board size	int	5
-n, --no_matches	Manually specify number of matches	int	None
-m, --start_moves	Use predefined start moves	bool	False

williamcorsel/hexai