Author: Ross Boyd
Date: 15:13:19 03/18/05
Go up one level in this thread
On March 18, 2005 at 15:01:18, Dan Honeycutt wrote: >Hi Ross: > >I know you have been writing about search instability, are you also implementing >measures with Trace to make the search more stable? Trace's search with fine >#70 looks a lot more stable than Bruja's. When Trace finds Kb1 she sticks with >it. With Bruja (output below), Kb1 fails high with a +3 score at iteration 18. >It then promptly fails low. Bruja rattles around until iteration 26 before it >locks onto Kb1 with a steadily increasing score. > Hi Dan, Well, I'm really supposed to be working on TRACE 2. But you know how it is... :-) Bruce Moreland says that if you eliminate search stability you will also eliminate a lot of powerful search/pruning methods. So, I try to reduce it if I can, but its something to be lived with.... Mostly, I avoid using alpha or beta to control extensions. With futility pruning (which depends on alpha) I keep track of the highest positional score and make sure its fairly impossible for the static evaluation to achieve alpha before pruning captures. In T2 I'll keep track of the type of nodes I'm visiting and therefore make PVS work a little more in harmony with its philosophy. In T133, it simply uses the TSCP follow_pv variable, ie, if it finds a PV move in the move list it searches it first and sets follow_pv = true. Unfortunately, my move sorting makes hashmoves get searched BEFORE PV moves regardless, which stuffs the whole thing up. No wonder my PV's are so darn choppy!! Its time to do some more experiments with that. :-) I don't know why Bruja needs ply 26. Make sure you're not nullmoving in this position, (of course). Another thing to try is to apply depth extensions before deciding whether to do a hash cut or not. (Not sure you'd get many extensions in Fine 70 anyway.) OR, it maybe that you search PV moves before hashmoves which is correct but it may delay things a little, I don't know. You should be using Bruja 1.9! :-) Cheers! Ross
This page took 0.01 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.