Author: Bruce Moreland
Date: 11:38:47 06/14/98
Go up one level in this thread
On June 14, 1998 at 10:39:16, Thorsten Czub wrote: >When 5K would have no chance against 300K, this is what some people say, >and what I cannot reproduce, because cstal HAS a high chance to win >against Fritz, why e.g. was Gandalf able to kill Fritz ? Gandalf was a >slow program in the old days (before Jakarta) and was easily able to >kill Fritz. I have sent the games to chessBase. NPS isn't very meaningful, a low NPS value doesn't make one program a sort of underdog against another one, you have your processor, it is the same processor the other guy uses, and if one guy wants to traverse nodes, and another guy wants to do a lot of expensive stuff in an eval function, I don't see that this has any independent meaning. These programs are all doing search, and although it is impossible to know for sure, I doubt that any of them are "knowledge" programs in the sense that they would satisfy the AI people. By the way, exactly how is CST getting good depth with low node count? It has to be doing pruning in the search. Is it doing this in a smart, domain-dependent fashion, or is its search some sort of Fritz-with-rabies, meaning null move with a huge R? If you draw a dotted line around any sub-tree with 3 plies of draft under it, and call this whole mess a node, and don't recognize any smaller unit as a node, your NPS would drop tremendously, and you'd have a pretty sophisticated "eval function", it would detect numerous tactical threats "in the eval", as it were. There is some difference between this and an eval function where someone goes to a lot of trouble trying to "manually" detect tactical threats, but it isn't clear how much difference there is. I find the whole topic to be speculative and insubstantial, without specifics of how the eval functions in question actually work, and how the search functions in question actually work. My understanding of what other people are actually doing is so limited that, if others have the same problem I do, I don't see what use it is forming opinions in order to actually argue. The whole thing boils down to an argument about whether you want "the taste of a new generation" or something that "adds life" -- an argument between different varieties of marketing puff. bruce
This page took 0.01 seconds to execute
Last modified: Thu, 15 Apr 21 08:11:13 -0700
Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.