right. the point is that it is a bit more subtle than the two cases that you've described. to answer simply, with the right kind of go knowledge, the 10,000 game machine seems, in practice, to be much stronger.
the reason is pretty straightforward: a "universal" solver that could work for any game without knowledge beyond how to make legal moves won't be as strong, for a fixed time limit game, as something that can use game knowledge. the extra game knowledge can avoid, on average, exploring some silly moves. determining what is silly enough to be worth checking for and what isn't is hard, of course. this can be formalized. s. 2010/10/31 Петр Смолов <[email protected]>: > I would like to know which program will be stronger: the one with a million > pure random games or the one with 10 000 games, but these games will not be > pure random... > > >> are you asking which converges faster asymptotically? >> there is always either a move that creates a winning situation, or no >> winning move. for every board position. so trivially the second. >> practically, however, i think that a mixture of randomness and smart >> go knowledge is what is used in the best programs. >> s. >> 2010/10/31 Петр Смолов <[email protected]>: >> > Hello all! >> > >> > How do you think what is better: maximim number of simulations (with pure >> random games) or much less simulations (but the games are not played >> randomly, >> in other words, the program tries to implement a certain move ordering in >> each >> game, and of course it takes much more time) ? >> > _______________________________________________ >> > Computer-go mailing list >> > [email protected] >> > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go >> > > > _______________________________________________ > Computer-go mailing list > [email protected] > http://dvandva.org/cgi-bin/mailman/listinfo/computer-go _______________________________________________ Computer-go mailing list [email protected] http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
