Author: scott farrell
Date: 18:39:10 09/12/03
Go up one level in this thread
On September 12, 2003 at 15:34:41, Dann Corbit wrote: >On September 12, 2003 at 15:23:02, Omid David Tabibi wrote: > >>On September 12, 2003 at 13:46:42, Dann Corbit wrote: >> >>>This has the source code, and Visual Studio.NET 2003 solution file: >>>ftp://cap.connx.com/chess-engines/new-approach/CRAFTY_Mpc.ZIP >> >>Have you tried running it on some tactical test suits? > >Yes. It clearly hurts tactical performance at fast time control. (I got the ftp to work, windows couldnt do it, but wget on linux was fine) Is it just the section headed probcut? I see no comments for mpc or multi-probe-cut. Can you give me a short English run down on what you are achieving? I read the essay you pointed to as well. As far as I can work out the paper, and your code does: 1. reduce depth searches as a proxy for deeper ones 2. uses a bound that is harder to prove/cut, but my C skills are not good enough to fathom this: bound = rint((pa->t * pa->s + beta - pa->b) / pa->a); 3. The depth reduction changes depending how deep you are in the tree depth - (depth / INCPLY - pa->d) * INCPLY 4. is it recursive (it looks like it to me). My implementation is different, so maybe its a different idea, multi-probe-cut. This is roughly what I do: 1.try null move first 2. try IID if no good move from hash 3. generate all moves, and order all moves 4. try MultiProbeCut which is: 4a. iterate first 6 moves from sorted list 4b. doMove 4c. search at beta, beta+1 (ZWS search at beta) at depth -2 4d. undoMove 4e. if I get 2 fail-hi, then fail high 4f. if multiprobe fails, re-order moves with new knowledge 5. start iterating all moves, normal searching with PVS etc. I'd been interested if you think this is similar, I think its 2 different ideas witha simlar name. Scott
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.