Author: Tim Foden
Date: 15:46:48 11/28/00
Go up one level in this thread
On November 28, 2000 at 16:32:15, David Rasmussen wrote: >I know we can't probably agree on a way to define nps that works for all >programs, but couldn't we at least state some of the most common ways that >people do it that are wrong? >Maybe we should expand this a bit, and make this thread a "common mistakes in >chessprogramming" thread. To know other peoples mistakes would be very valuable >when writing a program yourself. I'm talking about any kind of error here, from >the positively wrong (like not correcting for ply when storing a matescore in >the hashtable which was an error I made when I had just added hashtables) Me too. Even then I got it wrong with the ply correction (wrong sign)! > to the >not-so-smart things that people tend to do, like having one-ply extensions on >everything with no bound so your search explodes. Hmmm. In GreenLight I currently have extensions: Check +1.0 Recapture +1.0 Pawn push +1.0 1 reply to check +0.5 What do people think the best values should be? (At the moment in GreenLight I am limited by the fact that my fractional depths are only accurate to 0.5, but in the future I play to fix this) >Coming back to nps, the biggest mistake I made was very simple, but I >incremented my nodes counter before I went into queiscence and then incremented >it again in the beginning of quiescence, with a ego-boosting nps as a result... >:) When I removed that error, my nps dropped to a lowly 120 knps on a PII450 >instead of 300 knps :( > >Discuss amongst yourselves now.
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.