Author: Dan Newman
Date: 15:12:19 03/30/00
Go up one level in this thread
On March 30, 2000 at 11:04:01, Robert Hyatt wrote: >On March 30, 2000 at 07:57:12, Bruce Cleaver wrote: > >>All - >> >>Has anyone experimented with dynamically adjusting R for nullmove reduction? >>The idea is that some positions deserve R = 1, others R = 2, maybe some even >>deserve R = 3, if your stomach is strong enough. The value of R could be based >>either on the distance from Beta, or on other factors (as used in fractional ply >>extensions). >> >>Perhaps this could save search time or boost confidence that you aren't missing >>deep tactics with a static R value. >> >> If it has been tried before , and found wanting, well then obviously some >>imposter hijacked my email ID and posted this message.... :^) >> >> >>Bruce > > >It has been done (and is still being done.) I have been doing it for well over >a year. Ernst wrote a paper describing how he implemented the same idea (we >apparently tried the same idea independently and liked the results although I >am not sure we are doing it in identical ways). > >I have been using R=3 near the root and R=2 near the leaves for a long while. >I have had the idea on my 'todo list" for trying continuous values rather than >discrete values, since I already do fractional plies. IE R=3 near the root >tapering to R=2 near the leaves, but maybe R=2.5 in the middle, roughly. I do fractional null-move reductions in Shrike (ever since I read about the variable reductions here). I'm not sure how well it works though. I don't think it improved my testsuite results much... What I do is fill out an array with R values to be indexed by the depth (which, since it's fractional, I shift down a bit to allow the R array to be smaller). Currently I start out at three, shift to two and a half, then two, and finally one ply. I may try something finer grained, but this idea creates a really huge parameter space to explore... -Dan.
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.