Author: Robert Hyatt
Date: 08:22:38 10/09/02
Go up one level in this thread
On October 08, 2002 at 01:52:16, Gian-Carlo Pascutto wrote: >On October 08, 2002 at 00:52:38, Eugene Nalimov wrote: > >>He also says that their Q-search is much more powerful than the one that is >>usually used in the software-only programs. > >How can he tell? Nobody knows what the commercials are doing. You can't _know_ everything, but you can decipher a lot by looking at the program output... > >>Hsu gave some details why they don't use null-move: >>(1) He thinks that singular extensions and null-move gave more-or-less the same >>rating difference (100-200 points IIRC). > >Bull. Their own reported findings directly contradict this! It depends. Early findings suggested SE was tremendously effective. It turned out to be true mainly in test positions. Later findings suggested that the effect was much more modest. But that was a restricted form of SE. Their most recent version was far different from anything they had done in the past, including extending when there are only two or three good moves, rather than just one, etc. So nothing they have previously posted really contradicts the above statement since the two approaches are different. > >>(2) It's very hard to implement both null move and singular extensions in the >>same program. > >Bull. I don't believe that is "bull" at all. I had a very effective SE implementation in Cray Blitz, which used a very limited null-move search. I have _never_ gotten an implementation in Crafty to work nearly as well. In fact, I haven't (so far) gotten an implementation to break even in self-play testing. The two things are diametrically opposed, one trying to extend, one trying to reduce the depth. Add the hash table and it gets messy. Add the sticky hash table they mentioned and it gets messier still... > >I'm sorry, I don't believe a word of it. > >-- >GCP
This page took 0 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.