Author: Dann Corbit
Date: 15:32:19 10/16/03
Go up one level in this thread
On October 16, 2003 at 17:04:01, Charles Roberson wrote: > > I've run numerous tests with several versions of NoonianChess vs 8 sparring > partners at G/10 and G/90. Below I show tables summarizing the results. > ><pre> > NoonianChess version Time Control Score Total Games Percent won > 3.4.20030118 G/90 8 28 28.57% > 3.4.20030118 G/10 17 32 53.12% > > 3.6e2 G/90 14 42 33.33% > 3.6g G/90 7 42 16.67% > > 3.6e2 G/10 19.5 32 60.94% > 3.6g G/10 11.5 32 35.94% > > 3.5 G/90 11.5 28 41.07% > 3.5 G/10 13 32 40.63% > ></pre> > > The machines for G/90 were AMD 1.3Ghz while the machines for G/10 were > AMD 1.6 GHz. The 8 sparring partners were consistent across all runs. Also, > they were 8 different winboard engines that I picked up off the net instead > of different versions of NoonianChess. > > While version 3.5 did perform consistently. The other versions did not. > Also, we see that 3.6e2 and 3.6g both performed worse at G/90 than G/10. > > I think this shows that some programs (atleast 1) doesn't perform similarly >at disparate time controls. Why is this so? Maybe those that claim there is no >difference across time controls could say. > > Personally, I think the problem is in my use of extensions. > > So, what say you Christophe?? Here is my opinion: At fast time controls, microoptimization is important. Little tweaky things like unfolding loops and what-not. However, at long time controls, the fundamental algorithm driving the program will dominate. Something I would investigate for those programs is to see what the branching factor is for each one of them with a couple dozen positions. Experiment from 5 to 11 ply or so to get a good measure. I suspect that the one that held its own has a branching factor similar to that of its opponents, and the ones that fell off had a worse branching factor.
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.