Thanks for you input Don! You are absolutely right. I tried to bias the search in the wrong location. I fixed this now and hacked the GenerateLegalMoves moves function, which is used by the tree search.
This seems to fix the problem. Now i have to run tests for evaluation. Regards, David 2013/6/24 Don Dailey <dailey....@gmail.com> > It seems to me that you want to change the TREE portion of the playouts to > only consider moves your predictor sanctions. Am I misunderstanding > this? The tree portion of the search is still going to try every move at > least a few times (depending on the level set) and the only way to > guarantee that it never choose a move to play is if you somehow invalid > these. You could use a hack which is to give them very low bogus scores > or make it appear to the program that these moves do not even exist - but > the normal flow of the program will cause them to still be considered > regardless of anything you do with playouts. > > Don > > > On Mon, Jun 24, 2013 at 7:58 AM, David Briemann <dbriem...@gmail.com>wrote: > >> Hi Aja, >> >> What the picture shows is just the result of the current board situation >> fed into the predictor, namely the first 30 moves the predictor recommends. >> >> The only thing I changed in how Fuego works is how the random playouts >> (GOUCT_RANDOM) are done. They are not just random anymore but semi-random >> biased by the predictor. >> >> So Fuego does still try to generate moves with heuristics and only if all >> heuristics fail it uses random playouts(which I biased). >> >> >> And by "play different moves" I mean the moves Fuego actually plays after >> finishing calculations. What I want is to forbid Fuego to even consider >> certain moves(the bad ones, according to the predictor). If you look at my >> screenshot.. I want Fuego to only consider the moves marked with numbers. >> >> I'm beginning to think that I didn't understand the tree search part >> correctly. You say the tree search generates moves too. I thought moves >> were only generated in playouts and the tree search part was to follow >> already played lines until it reaches a position which has not been played >> out. Probably that's the location were I have too look into. >> >> David >> >> >> 2013/6/24 Aja Huang <ajahu...@gmail.com> >> >>> Hi David, >>> >>> 2013/6/24 David Briemann <dbriem...@gmail.com> >>>> >>>> To give you an impression, this is what it looks like in fuego for a >>>> well known opening position: http://www.abload.de/img/board7brdj.png >>>> >>>> So what is puzzling me right now is this: Even if I limit the possible >>>> playout moves to the best Y predictions, fuego will play different moves. >>>> For example in the linked picture it could play K11, which is not in the >>>> predictor move list. This happens too if I disable all heuristics and just >>>> do the biased playouts. >>>> >>> >>> By "play different moves" do you mean the moves generated by the tree >>> search or playout policy? From the linked picture it looks like the moves >>> proposed by priors and being searched, since Fuego doesn't consider global >>> patterns playouts in the current version. >>> >>> Aja >>> >>> _______________________________________________ >>> Computer-go mailing list >>> Computer-go@dvandva.org >>> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >>> >> >> >> _______________________________________________ >> Computer-go mailing list >> Computer-go@dvandva.org >> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >> > > > _______________________________________________ > Computer-go mailing list > Computer-go@dvandva.org > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >
_______________________________________________ Computer-go mailing list Computer-go@dvandva.org http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
