Author: Mike S.
Date: 12:16:05 06/26/03
Go up one level in this thread
On June 26, 2003 at 11:14:36, Bob Durrett wrote: >As a user of chess engines, I would like to know if the operator's procedures >can impact the engine's move choices [assuming limited time]. Definetely! From my experience as a user, it is a very common effect for example, that solving times are not reproduceable in a second run, if you don't delete the hash content in between. That is so since it has become common, that the hash content isn't deleted between the move anymore (like it was in old chess programs and engines). This is also the reason why the Fritz GUI has a button "clear hashtable" in the F3 dialogue (since Fritz 6 IIRC). But the permanent position learning - using learning files - which some engines have, can also lead to different results in the same position and within the same time than before. >This question is in context of post-mortem analysis. In analysis, it will most often be a useful effect which improves analysis results, when you step through a difficult variant *from the end to the beginning*, IOW use backwards analysis. Some automatic analysis functions also work like this. By properly keeping the hash content, the engine kind of "carries" information from later in the game (i.e. surprising results of deep sacrifices) updwards, thus being able to provide better results in a given time than by the forward analysis with pv insertion, because the engine may not yet see deep enough yet on the critical points. Some engine are especially good at this, i.e. Shredder,Hiarcs,Nimzo or Yace. Some others are not that good in keeping the important hash entries. (The forward analyis OTOH is useful, if you don't focus on learning about the *game*, but more about the *engine*, in the sense of "what would the engine have played during that game...") Regards, M.Scheidl
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.