Author: Ren Wu
Date: 13:39:52 12/20/01
Go up one level in this thread
On December 20, 2001 at 16:03:38, Dann Corbit wrote: > >Beowulf had serious problems with move ordering. It still has some problems, >but an increase of about 10% improved play tremendously. Thanks for share this info. 10% increase is huge, so i can understand it can play better game. But how about 5%, 3%, for example from 85% increase to 90%? >Mathematically, bad move ordering will cause a pure minimax search, if you >always had it inverted. Perfect move ordering would reduce the search to the >square root of nodes. Hence, any improvement should make it search better. Right, but what i still not sure are 1. this one only consider the first move cufoff, which can be a problem, for example, if I have a great hashtable, and so hastable move will be searched first. any improvement i made in other part of move ordering will not be refelected in here. 2. seems to me this is bias to use SEE, or any other techniques that will pick the first move in case the hashmove doesnot exist. It will not help for any other techniques, such as a. quality of your history table. b. second killer move c. killer move in 2 plies above d. capture last moved piece, etc. while these techniques will certainly help the move ordering, it will not increase the percentage much, if at all. 2. the cost, how much more calculation you want to spend here, will these extra calculations slow down your program too much? For myself, i had some cases where i increase this number by a few percent without slowdown the program, but still, the new version lost the matchs against old versions. I was thinking this is almost like the mobility, while good program trend to have higher number, maximize it will not necessary result a better program. But I don't really sure. Ren.
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.