Can alpha-zero learn Bejeweled?

SSSxCCC / AlphaZero-In-Unity

Several board game AI, which are trained based on AlphaZero, used in a Unity game.

MIT License

39 stars 15 forks source link

Can alpha-zero learn Bejeweled? #2

Closed tensorgpu closed 5 years ago

tensorgpu commented 5 years ago

For Bejeweled, after the same colour in a row disappeared, the jewel from row above will move down and if the new row match then they will dissappear too. One action will cause other several moves.

Assuming no new jewel will be generated, just find the best move to solve all the existing jewels.

SSSxCCC commented 5 years ago

AlphaZero is designed for two player games, its MCTS algorithm considers not only the good moves for itself but also the good moves for it's opponent. However, Bejeweled is a one player game, AlphaZero can't learn it. If you want to find the best move to play Bejeweled, a tree heuristic search algorithm should be able to solve the problem.

tensorgpu commented 5 years ago

Thank you for your suggestion, I will try tree heuristic search.