Author: Tony Werten
Date: 02:06:48 07/14/01
Go up one level in this thread
On July 13, 2001 at 15:27:17, Mike S. wrote: >On July 13, 2001 at 03:26:09, Tony Werten wrote: > >>I've been working on this now for a couple of years. Haven't had much succes >>yet. Found a way to reduce the computertime a bit and am using 4 computers now >>to try to build a decent book. It looks ok. Not good, not bad. > >Sounds interesting. Can you give an overwiew of your results, i.e. the amount of >moves or lines produced, details of the method etc.? The basic idea is this: Extend the line with the highest score. Now suppose d4 scores .25 and e4 scores 0.30 This would mean e4 gets extended every time until it gets below 0.25 That's not what you want. So I not only give back the minimax score, but also the depth of the pv. When I start to extend, I take the line with the highest score-(delta*pv_depth) That means "bad" lines get searched after a while. If you take a big delta you get a wide book, take a small delta and you get a deep book. This method doesn't work all by itself. ie Gambits will not be searched because the overcoming of a pawn loss will take to deep search. For this I take an ECO file (all the standard openings) and force the search to use the first couple of moves and then start extendeding by itself. > >I have the rough impression, that many programs like (after 1.e4) 1...d5 very >much (too much probably), is this true? Depends. My program gives a big penalty for taking out the queen if there are still small pieces (bishops,knights) on the back rank. If you don't then this will happen. > >I'd also be interested if longer theory line have been (re-)produced in your >experiment, and which ones. It's running now on 4 machines and I can add pc's without a problem. Everyone examines a positions for 12 minutes, but I used a trick so it only costs (on average) 1 minute. I'll use the book at WMCC. cheers, Tony > >Thanks, >M.Scheidl
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.