Author: Richard Pijl
Date: 00:44:44 09/02/03
Go up one level in this thread
On September 01, 2003 at 18:53:48, Geoff wrote: >Hi > >I have been improving my chess program so far by concentrating on speeding up >the search. I have got this reasonably fast now in terms of nodes per second, >but the program is still searching far too many nodes to get down to a >particular depth. > >I noticed that quite a few other programs have a separate search root function, >with a specific root move ordering function, I haven't bothered doing this so >far as I couldn't really see how it would give that good an improvement ? Is it >really worth doing? > >For info, my program currently uses fairly standard PVS search routine, alpha >beta with aspiration window, history array ordering, null move R=2, TT table. > >Any other techniques I should be looking at next to get the nodes search down >another notch ? Thanks > > Geoff There are a number of things special for the rootnode: - There is only one, so extra work done here is almost for free - It has large subtrees, so only slightly improved ordering can give big improvements - Moves are generated only once, so it is easy to keep statistics on the moves to help you improve ordering, e.g. number of nodes in the subtrees, scores returned by the search (especially when you use fail soft) etc. 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.