How to find Attribute for good moves in nineman morris

Question

Suppose the program has no clue about rules.
How can it learn good combinations?
Thinking about Reinforcement algorithm.

Answer 1

Oma Falk
You could make it learn by punishment instead of rewards!!!! This means if the Neural Network loses you say that to every move you make a other move (maybe opposite move) as desired output

Moreover nineman morris seems to be a game with only a “small” anount of possibilites, so you can in fact decide to every single situation who can win and who not, so you can decide between good and bad moves!!!

In addition to that if we are talking about “Mühle” you could split the game in start phase middel phase and end phase and generate an Network for each phase. And you should limit the number of moves in a game otherwise you could get in an endless game.

Answer 2

I have no clue about nineman Morris but I guess it depends on the feedback and what moves contributed to wins, provided a neural network had enough neurons to 'deduce' that?

Answer 3

Oma Falk I wanted to do something similar to, so I already though about it

Answer 4

Alexander Thiem yeah...
I will give it a try.