On 7/3/07, chrilly <[EMAIL PROTECTED]> wrote:
>> > >> > We just presented our paper describing MoGo's improvements at ICML, >> > and we thought we would pass on some of the feedback and corrections >> > we have received. >> > (http://www.machinelearning.org/proceedings/icml2007/papers/387.pdf) >> > > > They are probably referring to this paper: > http://www.cs.ualberta.ca/~mmueller/ps/silver-ijcai2007.pdf > No, I am referring to the icml2007 paper.
What I said is that ICML paper mentions RLGO evaluation function which is described in this paper: http://www.cs.ualberta.ca/~mmueller/ps/silver-ijcai2007.pdf
> > It's because Go is not only game in the world and certainly not only > reinforcement learning problem. They are using a widely accepted > terminology. > But a very inappropriate one. I have read Suttons book and all the things I know (e.g. TD-Gammon) are completly obfuscated. Its maybe suitable to present generel concepts, but it is extremly complicated to formulate an algorithm in this framework. But the main point is: I think game programmers should be more proud of their work and should present their results in the language of game programming. We are the ones which make progress, not these paper tigers.
I agree that game programmers make most of the progress in game domain.
When I wrote the Null-Move article, some referees pointed out, that the article is not scentific enough. The NullMove was already known and even published before. But only after this "non-scientific" article it became a "must have" in chess programming. The pseudo-code had a bug and I see till today this bug in open-source chess programms. > > Can You share the source? > Yes. See attached Archive. The "interface" and the "UCT" part are rather primitive. The move-generator is better/faster than the ones I have seen in other public code (but can be certainly improved). The evaluation was published by G.Tesauro, but the implementation is more efficient. It is - according to G.Tesauro - a baseline for an evaluation. Every usefull evaluation function should clearly beat this one. The purpose of the code was to study the effect of Monte-Carlo sampling. I was deeply impressed how much better the Rollout/Monte-Carlo version plays.
Thanks, I asked for the source, mostly because I expect to learn something from you. Best regards, Lukasz
Chrilly _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
