Author: Richard Pijl
Date: 15:48:49 05/08/03
Go up one level in this thread
On May 08, 2003 at 13:55:14, Djordje Vidanovic wrote: >On May 08, 2003 at 05:49:23, Richard Pijl wrote: > >>On May 07, 2003 at 22:32:31, Will Singleton wrote: >> >>>On May 07, 2003 at 17:46:45, Richard Pijl wrote: >>> >>>>On May 07, 2003 at 04:40:43, Enrique Irazoqui wrote: >>>> >>>>>King majority wins in this game Pitschak - Foerder played in 1930. >>>>> >>>>>[D]1R6/p3k1p1/7p/2b1pP2/P1r3P1/B7/7P/7K w - - 0 1 >>>>> >>>>>1.Rc8! Kd6 2.Rxc5! Rxc5 3.h4!+- 1-0 >>>>> >>>>>Enrique >>>> >>>>The Baron finds Rc8 in 10 seconds, but it takes much longer to show convincing >>>>scores (AMD MP 2000+, single CPU version): >>>> >>> >>>Hi Richard, >>> >>>Looks like you've been working. The latest Baron I'm aware of takes over a hour >>>to find Rc8, might've been several hours for a good score. Did you add specific >>>eval code for this type of position? >>> >>>Will >> >>The last few weeks I've been working a lot on (passed) pawn evaluation and also >>added some more general endgame knowledge. The job is far from finished though >>as I suspect my recent changes may have some negative side effects in game play. >>I hope I get it stable before Leiden ... >> >>I think the 10 seconds for Rc8 might be a little luck though. >>Richard. > > >No need for modesty here. We are discussing facts :) From what I've seen of >Baron it is an excellent endgame player, relative to most WB engines... Coupled >with faster search it could boost its strength a lot, hopefully before Leiden. >Even though the hour glass is speedily running out... > >-- Djordje Thanks Djordje, However, I really meant the 'luck' part. After inspection of my passed pawn code I saw some bad errors. After correcting them, Baron needs 27 seconds to find Rc8. But at least the score is more convincing a little faster. You're right that the slow search is really becoming a pain now. The Baron used to do quite well at bullet (0.99.3 and earlier), but after I removed futility pruning and razoring, that is no longer true. In standard games (and blitz) progress is still there though. There it does better without these mechanisms. I have been experimenting with different kinds of reductions and other selective search mechanisms, but none of them was stable enough in games (although some were great in test suites). 1 week to go before Leiden ... Richard.
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.