We have enabled 'Play Against The Computer' for Amazons. This uses the 'Fairy Stockfish' engine (github.com/fairy-stockfish/Fairy-Stockfish) as the opponent. This is the same engine that we use for the AI for Shogi, Xiangqi and Othello. However for Amazons this AI is very weak - I easily beat level 8 in my first attempt - something I shouldn't be able to do.
Having spoken with people involved with the Fairy Stockfish project, they have indicated that only a simple evaluation function has been applied to the Amazons game.
So for now, we have opted to enable 'Play Against The Computer' for Amazons as it gives another option alongside @PS-Greedy-Two-Move and @PS-Random-Mover. However if any of our users are technically minded and have experience of AIs for Amazons you may be able to help provide Fairy Stockfish with a much more advanced evaluation. If Fairy Stockfish upgrade their Amazons evaluation, PlayStrategy will be able to make use of this. Then, if we felt the evaluation were suitably strong we would be able to also turn on 'Request a Computer Analysis' for Amazons games which would be able to give feedback about blunders etc.
Amazons 'Play Against The Computer' - can you help us?
I have some experience with Amazons AI — a good evaluation function is non-trivial for Amazons. I do have some ideas of what works (and what doesn’t) for evaluation functions using MCTS.
I’ll play against the Fairy Stockfish and see what I think.
@testingqwerty Fairy-SF basically uses the same evaluation function as for chess, at least the terms that are applicable, since I only implemented the rules of the game and did not make any further adjustments as non-chess games are not really the focus of Fairy-SF. Therefore, in practice mostly only mobility bonuses will apply for evaluation in game of the amazons. Not being too much into the game, I assume mobility bonuses do make a lot of sense, but probably are still a way too simplistic evaluation.
Fairy-SF could likely also benefit from NNUE for game of the amazons, but I did not find the time yet to add the support for blocked squares to the NNUE code. So for the time being if you have suggestions for basic heuristics of game of the amazons positions that could be used in the handcrafted evaluation, let me know.
This topic has been archived and can no longer be replied to.