Author: Uri Blass
Date: 08:01:37 11/27/00
I already posted this idea in a Rebel Century thread in this club but I corrected steps 2 and 3 in the experiment. Many top programs do preprocessing and the interesting question is the size of the preproccesing. It is a question that only the programmer can find out because we do not know the static evaluation of the programs. The test should be done by the following way. 1)Take a random node that you evaluate(you can use database of the program's games for getting these nodes). 2)Calculate the exact static evaluation of a random node in the tree that the program generates when the node in 1 is the root position. 3)Calculate the static evaluation of the same node assuming the root position is 2 plies earlier. 4)Calculate the difference between the evaluations. You get a number. Continue with this experience and get a lot of numbers. I will be interested to know the distribution of these numbers for all the top programs. You can do the same experiment with 1 ply if you want to check also preprocessing that is result of asymmetry in the evaluation. I think that some information about these numbers average,variance and the maximal number can help people to decide which programs do more preprocessing in the root. We need also to translate the evaluation because 1 pawn advantage has not the same meaning for all programs but this difference is smalland usually not more than 20-30%. Uri
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.