>On 17-11-17 02:15, Hideki Kato wrote: >> Stephan K: ><CAO6hqCq-82X=Pk=cbgfz-vnv7gdgox_kurxpleejufyu984...@mail.gmail.com>: >>> 2017-11-16 17:37 UTC+01:00, Gian-Carlo Pascutto <[email protected]>: >>>> Third, evaluating with a different rotation effectively forms an >>>> ensemble that improves the estimate. >>> >>> Could you expand on that? I understand rotating the board has an >>> impact for a neural network, but how does that change anything for a >>> tree search? Or is it because the monte carlo tree search relies on >>> the policy network? >> >> The author of AQ told me that majority-voting based on the >> orientation (4 or 8) makes AQ stronger. > >How do you majority vote with 4 (or 8...) real numbers?
I haven't heard the detail but I guess the voting is based on not the value but the move; if a move is selected by more than 4 (case of 8) node computers then the move will be played, for example. The source code of AQ on Github might help. Best, Hideki -- Hideki Kato <mailto:[email protected]> _______________________________________________ Computer-go mailing list [email protected] http://computer-go.org/mailman/listinfo/computer-go
