"Machinelearning Stratego"

wessel
wessel New Altair Community Member
edited November 5 in Altair RapidMiner
Dear All,

I'm a big fan of the stratego* game and I would like to create a smart computer player.
There is already a stratego-bot that is pretty good: http://www.imersatz.com/probe/index.html

What machine-learning techniques would be well suited for learning to play stratego?

I have a big data-set of stratego games played by the world's best players.
My plan is to convert learning stratego into a regression problem.

The regression algorithm will assign a "score" to each board position.
By computing all possible moves given a board position, and playing the move with the best associated score, we convert board scores into moves.
So now the performance of the regression algorithm can be measured as the number of correct moves played (where a move is correct if it is played by a strong human player in the data-set).

Best regards,

Wessel



*Stratego is a mixture between poker and chess: http://www.imersatz.com/probe/index.html
Tagged: