Hello all, first, I thank you all for the very interesting comments you made. I answer quickly the various comments in that email to avoid spamming the list :-).
In fact I was no aware of all the results available in 7x7. So I think you are right, 7x7 is no more interesting. I though it could be interesting to see programs beating humans in Go, even if it is on a small board, but it is already the case. The disapointing thing is that the opening was made by hand, and as the opening is so important in 7x7, it is more the human knowledge which makes programs so strong and not really the level of the program. The other disapointing thing is that it seems that the meta-UCT doesn't manage to find this good opening in 7x7, even with 50 CPUs working for two days now. The meta-UCT still find that black is loosing with komi 7.5... :-(. But perhaps I am doing something wrong. It would be great to see a program find this "perfect opening" alone. It even could be a benchmark for the improvements: how many moves the program could find in 1000000 simulations? Again thank you all for your comments, and sorry for trying to put interest in an uninteresting problem :-). Sylvain _______________________________________________ computer-go mailing list [email protected] http://www.computer-go.org/mailman/listinfo/computer-go/
