AI that can win over a human at gomoku under 0.5 second per move with minimax algorithm

You need the golang version 1.16.6 or less and to install go-sdl2 depending your OS : install go-sdl2.

After this install you can execute

go mod init gomoku
go mod tidy
go build

You must align 5 stones to win.

You can remove a pair of your opponent’s stones from the board by flanking them with your own stones.

This rule adds a win condition : If you manage to capture ten of your opponent’s stones, you win the game.

A player that manages to align five stones only wins if the opponent can not break this alignment by capturing a pair, or if he has already lost four pairs and the opponent can capture one more, therefore winning by capture.

It is forbidden to play a move that introduces two free-three alignments, which would guarantee a win beny alignment.

We used minimax algorithm with alphabeta prunning. To achieve a move under 0.5 second, we limited the exploration at 5 depth in default settings.

Alpha-beta pruning reduces the number of nodes that are evaluated by the minimax algorithm in its search tree.

We keep track of two values: Alpha and Beta.

• Alpha = the minimum score that the player is assured of.
• Beta = the maximum score that the opponent is assured of. Initially, alpha is negative infinity and beta is positive infinity.

It stops evaluating a move when at least one possibility has been found that proves the move to be worse than a previously examined move.

During the search, we don't evaluate all the Goban but only the neighbors to limited the tree search.

The interface is very simply, and developed with sdl2.

In the menu, you can set all the rules, and disable/enable the captures, and no double-three rule.

You can also play against a bot, or an another player. We have developed an helper who can suggest a move, but warning, it's not the best move ! This option can be disable.

Feel free to use the "thinking time" option. The algorithm will have more time to process and search the best move in the tree search.