Author: Bruce Moreland
Date: 08:47:25 07/29/01
Go up one level in this thread
On July 29, 2001 at 09:45:19, Ferdinand S. Mosca wrote: >Hi everybody, > >How much strength by elo points is gained by adding evaluation functions to a >program, i.e the current publicly released version of TSCP? > >What are the factors that influence the high number of search depth of a >program? Branching factor, modified by move ordering, pruning, extensions, et. al. >Does search depth really matter? Yes, it matters a whole bunch against computers. I think it matters a little less against humans, because rather than using it to win material you use it to try to avoid positional blunders and attacks until the human makes a five-ply mistake. >What is the average search depth of Ferret in middle game? I have no idea. It hasn't actually played a game at slower than G/30 in a few years. Since Ferret isn't public, nobody else can look, and since I don't look, nobody knows. It doesn't go particularly fast compared with Fritz and Junior, and it doesn't go particularly deep, either. >Does it help to reduce branching factor by considering moves that influence king >safety first, or moves that influence center control first? I used to use as a secondary term the piece-square delta, which tended to try centralizing moves first. I found that this didn't have any effect, so I took it out. >In a queen and pawn ending, which is more important search depth or more >evaluation functions? Interesting question. My bet is evaluation, because passed pawns are so important, and there are enough legal moves for each side that branching factor is high enough that you can't really see anything. bruce > >Regards, >Dinan
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.