Author: Stan Arts
Date: 05:32:13 06/08/04
Go up one level in this thread
On June 08, 2004 at 04:35:24, Jeff GAZET wrote: >Good morning, > >i notice that my engine always searches x3 or x4 nodes more than others. >If you disable razoring or any method like this in your engine, does it searches >x3 or x4 nodes more ? >Where should i work on to dicrease this number, move ordering ? > >Regards. Hi Jeff Wondering the same with Neurosis, but in my case my node-numbers are becomming better all the time, by finding something small, and then some days/weeks later something small again, and all these improvements narrow down the tree.. so I figure in a few years I'll have a narrow tree. :) There are a lot of things that determine the size of your searchtree. And here "depth" doesn't mean anything because if, for instance, I add checks to my qsearch, I'm losing a depth but finding lots of combinations at the same speed or faster (and others less fast, because I need perhaps 3x 4x as many nodes to get to a depth.) And you can think of lots of other things that easely make up for a factor 3x 4x. (But not nescesarily strength) That makes comparing node- counts impossible. It's probably mostly extentions that can make for a factor 2 easely, and indeed at depths higher then 3-4-5 pruning will become important. For instance Neurosis uses null-move, but not recursive, and verified. (..My goal is to have a sort of strong program that doesn't miss anything upto the depth it's searching, maybe because of zugzwang or because it prunes something else away, etc. Neurosis is for fun and interest and so estetics can be important. :) ) Used to be adaptive reduction of 1 or 2, but now, I've experimented with allowing a reduction of 3 in some cases and with a lot of safeguards, that can possibly set in at depths higher then 6. Well, by allowing 3 sometimes instead of 2 at higher depths, my nodes needed for depths higher then 7-8 can be nearly half of before. Then if I'd be less worried about pruning and prune a lot more it'd be much better again. So I guess things like that make all the difference too. Also, look at strong-program's nodes-at-lower-depths when they have empty hash, because in games when they reach search depths >10 it usually seems that they need really small amount of nodes till depth 8 and so, but that's not so accurate. Then, their node-counts at lower depths seem "normal" to me. Hmm with the exception of very agressive pruners. But then you can feed in a lot of positions where they are entirely blind or need huge depths to see something simple, so then depth and nodes doesn't mean much again. Goodluck with WJchess. It's of similair strength as Neurosis and I think WJChess plays agressive, and makes it fun to test with. Greetings Stan
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.