Author: Andrew Williams
Date: 13:32:37 03/28/00
Go up one level in this thread
On March 28, 2000 at 15:34:02, Peter McKenzie wrote: >[D]3qkbr1/1b1n1p1p/4pp2/rB6/P2p1B2/5N2/1P3PPP/R2Q1RK1 b - - > >The above position is number 30 from the GMG2 test suite (positions from a >Larsen book), and causes my program's search to explode. > >Node counts for Lambchop 9.50, without recapture extension: >6ply: 82K, 7ply: 193K, 8ply 1262K > >Node counts for Lambchop 9.50, normal version with recapture extension: >6ply: 186K, 7ply: 2997K, this is a branching factor of about 15!! > >Does anyone else have a similar problem with this position? > >I'm not exactly sure why the search explosion happens because the recapture >extension fires at only about 0.7% of all nodes. >I restrict it to the following conditions: >- don't allow 2 recapture extensions in a row >- previous move must have been a swap (eg. BxN, RxR etc) >- current move must be a capture by a lesser piece, or piece >being captured must be unguarded. >- material balance after the current move must be between -2 pawns and >+2 pawns. > >Note: the recapture extension extends a whole ply. >Perhaps my recapture conditions aren't very good, or maybe I just have a bug... >input welcome! > >I also hacked together a version of LambChop using partial ply. > >Node counts for Lambchop 9.50, using 0.75ply recapture extension: >6ply: 68K, 7ply: 159K, 8ply 1328K > >So maybe partial play (crafty does similar I believe) is a good solution. More >testing required... Mine extends a full ply, but only if the recapture is of the same value and on the same square. I had forgotten putting in this last condition, so I tried it without it as well. Here are my results: ONLY EXTEND IF RECAPTURING ON SAME SQUARE (normal postmodernist): 6ply=144k, 7ply=509k, 8ply=1147k EXTEND EVEN IF NOT RECAPTURING ON SAME SQUARE: 6ply=179k, 7ply=612k, 8ply=1425k NO RECAPTURE EXTENSION AT ALL: 6ply=88k, 7ply=313k, 8ply=729k Hope this is helpful Andrew
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.