> Hi Jonas, > > welcome to the list. > > The idea of using f(score) instead of sign(score) is interesting. Long > ago, I tried tanh(K*score) on 9x9 (that was before the 2006 Olympiad, so > it may be worth trying again), and I found that the higher K, the > stronger the program. Still, I believe that other f may be worth trying.
If you increase K the program tends more to win/loss scoring. With very high K you will reach it. So it's no surprise it works well. > By the way, Olivier and Sylvain mentioned earlier on this list that they > were using floating point in their tree data structure. So MoGo may be > using a floating point function to estimate the score of a playout, > otherwise there would be no reason to use floating point. But I may be > guessing wrong. Maybe they can tell us ? These ideas are all old, see for example "Old fashioned Computer Go vs Monte Carlo Go": http://ewh.ieee.org/cmte/cis/mtsc/ieeecis/tutorial2007/Bruno_Bouzy_2007.pdf Page 89, "which kind of outcome". This method is better than the above and similar to what Jonas seems to propose. The improvement is minor. I believe one of the MoGo people stated earlier that they use the exact mean score update rules as published in the ICML paper. Then floating point is required. I do not believe they use the above idea based on the games I see. -- GCP _______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
