Hello, I answer here on the discussion because I think it is more relevant to what I wanted to say. Sorry not to answer to the lasts posts.
The computer go guys don't think of performance as a function of time,
only as a kind of absolute, it plays good or it doesn't.
I think it simply comes from the fact that the majority of strong Go programs do not scale well (do not improve a lot) with the thinking time. This is a fact, so the experiments done by Hiroshi are totally relevant, even if the programs did not take the same amont of time. With more time I bet the results would have been very close. Comparing to chess, I think also that we experiment 19x19 computer Go in blitz games. I used to play chess, and 20 minutes games was called semi-blitz. 19x19 Go has a lot more moves to do, so for me 30 minutes games are blitz. I set up MoGo playing in 30 minutes for 19x19 in KGS. I think people would simply not play if the games were longer. So GnuGo and commercial programs have been optimized to play strong in very few time. For them, long thinking time is simply not relevant. It is not the same goal. The scale of MoGo in 19x19 with the thinking time is very different (against gnugo 3.7.10 level 8): - with 35000 simulations/move (~30seconds/move in a P4 3.0Ghz) the winning rate is 39% - with 70000 it becomes 59% My preliminary experiments before the slow tournaments showed very high winning rates with high thinking time, and at this time the 70000 simulations winning rate was closer to 30%. So it depends on what your goals are? Playing strong very fast (gnugo is incredibely good at that, mogo will play random)? Playing strong with a big computer power (UCT approaches scale better)? For commercial programs and also GnuGo, the first goal is clearly better, so comparison with more thinking time would not be fair. For computer science perspective, the second goal seems more appealing, so comparison with less thinking time would not be fair. Sylvain
_______________________________________________ computer-go mailing list computer-go@computer-go.org http://www.computer-go.org/mailman/listinfo/computer-go/
