Author: Mogens Larsen
Date: 12:16:23 12/19/00
Go up one level in this thread
On December 19, 2000 at 13:39:53, Christophe Theron wrote: >As far as I know, the speed (in term of NPS) of a program is not an indicator of >its favorite time control. For example, Hiarcs is a very slow program, but it >performs extremely well at very fast time controls. It is the opposite for >CSTal. I agree. Using NPS alone won't get you anywhere in general. >The level of selectivity is not a good indicator either. The King and Hiarcs are >very selective (they have to, as they are also very slow in NPS), and as I >pointed out, Hiarcs is very good at blitz. I don't know if the King shows any >preference for a given time control. No argument from here. >Tiger is a very selective program, and it does not favor a time control over >another one. What kind of "selectivity" do you use? >I would add that the presumed level of "knowledge" of a program does not >indicate either the time control it will excel in. Once again, Hiarcs is >supposed to have a lot of knowledge, and once again it performs the best at >blitz. Which is not to say that it is bad a long time controls. I may talk nonsense in the next paragraphs or so, but what else is new :o). Just a few (probably erroneous) thoughts. In my opinion, the timecontrol at which a computer program excels at, is based on a combination of speed, selectivity and knowledge, which are intertwined in an unique way from program to program. Often the unknown factor is knowledge. To me, as a non programmer, the obvious distinction between performance of various programs would be the search/knowledge ratio in the eval, which then presumably influences selectivity. This could mean that some programs patches short term eval problems in time, while others may not improve very much by using more time. I don't know anything substantial on preprocessing, extensions or other methods to guess what eval "problems" they patch. My impression, which is probably wrong, is that extensions are good tactically, which would complement a complex eval well. A guess would be that complex eval and extensions benefit from time. My second impression is that programs using preprocessing are quite stable at most timecontrols. The difference is probably the amount, but my knowledge is insignificant. Even if everything I say is absolute nonsense, there's ample opportunity for variations for arguing the time issue IMO. I think it's possible to estimate the strength of a program at blitz and standard from very little information, but not just NPS and depth. Mogens.
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.