Re: [computer-go] Re: Move Prediction and Strength in Monte-Carlo Go
荒木伸夫 wrote: I have considered this, and I think that this may be caused by wrong training model. In my master thesis, I mentioned that the relationship between top 1 accuracy of move prediction and the strength of Monte-Carlo is not simple (I increased the number of matches to 600, and similar tendency appeared). Therefore, it might be wrong to use only one human move (top 1 move) as a positive example (such training will highten top 1 accuracy). We may need to use another training model... Unfortunately, I don't believe a usable training model exists, besides playing plenty of games with the full MC tree search to figure out which weights produce the best playing strength. A big problem is the sample distribution. Whatever patterns we use, they are general rules with exceptions. That is to say it is always possible to make up a weird (or not so weird) position where patterns fail. And when a MC program is using patterns, it is naturally attracted towards positions that are evaluated wrongly. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Move Prediction and Strength in Monte-Carlo Go
Rémi Coulom wrote: 荒木伸夫 wrote: I have considered this, and I think that this may be caused by wrong training model. In my master thesis, I mentioned that the relationship between top 1 accuracy of move prediction and the strength of Monte-Carlo is not simple (I increased the number of matches to 600, and similar tendency appeared). Therefore, it might be wrong to use only one human move (top 1 move) as a positive example (such training will highten top 1 accuracy). We may need to use another training model... Unfortunately, I don't believe a usable training model exists, besides playing plenty of games with the full MC tree search to figure out which weights produce the best playing strength. A big problem is the sample distribution. Whatever patterns we use, they are general rules with exceptions. That is to say it is always possible to make up a weird (or not so weird) position where patterns fail. And when a MC program is using patterns, it is naturally attracted towards positions that are evaluated wrongly. This all (combined with the results of the study) makes me think mogo and probably the other UCT programs should be searching a little wider at long time controls.At normal levels they are probably very well balanced. - Don Rémi ___ 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/
Re: [computer-go] Re: Move Prediction and Strength in Monte-Carlo Go
Hi, I would like to confirm your experiments: I have noticed already that adding shapes of radius 4 improves prediction a lot, but does not improve playing strength (from progressive widening). Also, even worse than that, for a given set of features, the pattern urgencies computed by MM are not optimal. That is to say, it is possible to manually tweak urgencies and get a stronger program. So, as Gian-Carlo puts it, optimizing a Go program is still black magic. There is no way to avoid playing games to measure playing strength. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
Re: [computer-go] Re: Move Prediction and Strength in Monte-Carlo Go Program
荒木伸夫 wrote: Hello, Coulom. I'm Nobuo Araki. Thank you for reading my thesis. However, this thesis is first version, not final version. Therefore, there are too few experiments. And Mr. Hideki Kato sent me many warnings about this thesis, for example English is too bad. You may be confused while reading my English...sorry. Anyway, thanks again. ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/ Hi, Sorry for announcing too early. Your English is maybe a bit exotic, but not too difficult to understand. I appreciate your effort to write in English. I was very frustrated in Hakone with all those papers in Japanese that looked so interesting. Also, I believe it is not such a bad idea to release preliminary versions to the Go-programming community before producing a final version. I did it with my previous computer-go papers, and the feedback I got here helped me to improve the final version a lot. Rémi ___ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/