Author: Tony Werten
Date: 07:42:20 08/12/04
Go up one level in this thread
On August 12, 2004 at 10:18:45, Stuart Cracraft wrote: >Here is the result of the 30 second per position test on 300 positions >using nothing, see>=0 to search a capture, see having to bring the score >within alpha and a margin (futility/delta), and with both. I'm not sure how it is usually done, but I don't think it is safe to prune the see score against alpha but rather the captured piece value. Else it makes your qsearch even more blind. Tony > >Nothing > >$ ./qaNbatch >HERALD ga nosee30300.log 30 300 >**** 9.98/35.29 85% 255/300 7183.08 1793586432 5978621/24/249696 0/0/104784432/0 >/0/0 > >see>=0 to search a capture > >HERALD ga -DSEEQUIESCE seequiesce30300.log 30 300 >**** 10.14/31.88 85% 257/300 7144.92 1599515136 5331717/24/223868 0/0/52093584/0 >/0/0 > >see bringing score within alpha and a margin (futility/delta) > >HERALD ga -DDELTAPRUNE seedelta30300.log 30 300 >**** 9.98/35.27 85% 255/300 7183.00 1783053568 5943512/24/248232 0/0/104211232/0 >/0/0 > >delta > >HERALD ga -DSEEQUIESCE -DDELTAPRUNE seeboth.log 30 300 >**** 10.14/31.87 85% 257/300 7144.30 1600962560 5336542/24/224089 0/0/52147264/0 >/0/0 >$ > >So -DSEEQUIESCE and -DSEEQUIESCE -DDELTAPRUNE are the best with >identical ply-depth scores of 10.14 ply and very nearly equal >maximum capture-search depth of 31.88 and 31.87 plies. Other >measurements are very similar as well. > >Neither nor both is much of an improvement over not having both, >at least in this short test, however the search tree is smaller >and the depth is 1/5 of a pawn deeper so those are good signs. >The total nodes searched dropped by 10% but so did the nodes >per second. > >Probably a better test than fixed time is fixed depth. On those, >I've seen the SEEQUIESCE (see>=0 for capture search) look at >50% fewer nodes in 50% less time as predicted by others. Not >so sure about SEEDELTA. > >They don't seem as additive as I had hoped. However my SEEDELTA >is implemented with a fixed MARGIN value of 1/5 of a pawn. I'll >try it with the varying maximum positional score for the side on >move and see if that improves things. I doubt it will make much >of a difference. > >Stuart
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.