Author: Michel Langeveld
Date: 03:45:31 10/16/03
Go up one level in this thread
On October 15, 2003 at 15:09:32, Michel Langeveld wrote: >On October 15, 2003 at 13:21:20, Gerd Isenberg wrote: > >>On October 15, 2003 at 12:37:43, Michel Langeveld wrote: >> >>>[D]8/k7/3p4/p2P1p2/P2P1P2/8/8/K7 w - - >>> >>>This position really bothers me ... and bothers nullmover... >>>nullmover takes a really long time to see that the pawn can be really won. >>> >>>At ply 23 we see the keymove Kb1, but it needs 26 ply after 6 minutes we see >>>that Kb1 wins a true pawn. >>> >>>The hashhit percentage is only 3%. >>>The FH is only 30% >>> >>>Do I have to be worried ... or is this still all normal? >> >>Hi Michel, >> >>no, absolutely not. There is something wrong with your hash-table. >>Hash-hits in IsiChess are about 70-80%, exact hits ~25%. >>Depth 30 with >3.00 score after 0.5 seconds on Athlon XP 2.6+. > >Thanks for the hashnumbers ... I calculated my hashhits wrong. >I casted to an int ... and this made it go wrong because the hashhit >counter is bigger as 32 bit. I cast it now to a double and I hit around 50% >I will calculate the exact hits also. > >>I'll hope you don't fix this severe bug until saturday ;-) > >Yes I hope to fix it also :-) The problem is complicated ... and all things in Nullmover seem to be correlated with not finding this move: hash, hashreplace algoritm, evaluation, moveordering, lazy evaluation. In order words the error is not one thing alone but a sum-up of things that works in the most of the positions but sub-optimal in others. And in this particulair position it ends up in horror. I found out that if I use: - an always replace algoritm - take away my lazy evaluation I find Kb1 on ply 22 I get 33 ply almost instantly I get 40 ply in a minute with a nice score of +2.38 score on Kb1 and Kxf5 in the PV. The problem with always replace is that on WAC I score only 210 out of 300 with 2 seconds a position and else 265 / 300 on 2 seconds...... still puzzling
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.