Author: Volker Böhm
Date: 13:54:00 08/04/04
Go up one level in this thread
On August 04, 2004 at 02:46:23, Uri Blass wrote: >On August 04, 2004 at 02:11:44, Volker Böhm wrote: > >>Hi, >> >>for every pruning you make I suggest to implement a pruning test. >>In Test mode do the following: >> >>1. Remember if you are pruning >>2. Don´t do the pruning, but search normally >>3. Returning from the search find out if the pruing would have been correct. >>4. If not: Count for statistic, build a maximal error value, print out fen of >>the current position >>5. Look at the fens, find out if you have to implement additional rules to >>exclude wrong prunings. >> >>For extendet futility you have to make sure, that you are nearly not making any >>error, because the error would possible not be fixed at the next search depth. >> >>Greetings Volker >> >>P.S. I do not use futility so far, because it depends much on the engine. Thus >>if you´ll plan to improve your engine much futility can IMHO make improvements >>harder to test. >> >>Greetings Volker > >I do not understand why do you think it will make improvements harder to test. > >Uri Hi Uri, there are many settings that interacts with eval. 1. Lazy eval bounds 2. Quiescense pruning 3. extensions 4. ... the more of techniques are implemented, the higher is the complexity of the program and the interferences with changes in eval. Thus a good change to eval could interact with pruning rules and get a bad result in games. Then you have to find a new setting for the pruning rule, perhaps another margin or additional rules when not to prune. Perhaps you will even throw away your eval change because you think it works worse ... Greetings Volker
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.