Author: Djordje Vidanovic
Date: 20:19:18 03/11/00
Go up one level in this thread
On March 10, 2000 at 12:53:57, Bruce Moreland wrote: >On March 10, 2000 at 11:50:42, Soren Riis wrote: > >>I am a mathematician on IM level who only have >>done armature work (like testing evaluation >>functions etc) in chess programming. >> >>I would like to hear the expert opinion on the >>following questions: >> >>It is well known that programs which have the same >>strength at one time control, might have different >>strengths at different time controls. Or equivalently >>that two programs relative strength might depend on >>the speed of the hardware. >> >>Q1: What is factors contribute to this? >> >>Humans players, when compared to programs plays relatively >>better with long time controls. While a program increase >>its strength by typically 70 rating points when the speed is >>doubled, the human (for fast time controls like 5 min) rather >>seem to gain 200 rating points when the speed, so to speak, is >>doubled (i.e. when the available time is doubled). >>This suggest that with slow speed computers also gain much more >>than 70 rating points. Is this correct? >> >>Q2: Does very knowledge strong program gain more playing strength >>relative to less knowledge strong programs (on slow processors)? >> >>On fast processors knowledge is sometimes even a disadvantage. >> >>Q3: How does knowledge fare as a function of speed? >> >>The questions are a bit vague, but I would like to hear the >>expert opinions. Any thoughts? > >There's never been any consensus as to what knowledge is. People assume that >programs with low NPS are high knowledge programs, but these programs could be >doing tactical evaluation that may as well be search. > >If I have a program that evaluates at the tips, maybe it will do 200K nps. But >if I suddenly decide that my "eval function" is the last three plies of search, >it will fall to a few K nps. It didn't get any more knowledeable. > >bruce Well said. Perfect example of what you said is Hiarcs -- it is all tactics even though it does less K nps than most other top programs. Proof: great bullet player, perhaps one of the best there is. *** Djordje
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.