Author: Michael Henderson
Date: 21:17:53 09/19/04
Go up one level in this thread
On September 19, 2004 at 23:58:28, Stuart Cracraft wrote: >I use null move only to return the null move search >value if >= beta, otherwise fall through to everything else > >But I hear that one can use the null to adjust alpha. > >When I try to adjust alpha this way: > > nullmovestuff > alpha = MAX(alpha,nullmovescore); > >my results drop precipitously. > >Perhaps my method is wrong or adjusting alpha has fallen >out of favor. > >Comments anyone? > >Stuart Ok assuming nullmovescore >= beta, cutoff. That leaves you with only fail-low scores to work with (assuming null window). Fail low score, of course, means that the true value is <= nullmove score value. So you can't use a fail low score because it means *nothing* to alpha. It's not possible to use a null move score for anything but cutoffs and threat detection. I have heard of using beta cutoff entries in the hash table. Beta entry--true value is >= beta. So alpha can be increased if the beta bound is greater than alpha. This usually gains you about 0% improvement in speed and requires some well thought out code. For example, you have to avoid corrupting the PV (easy to do) if you do this and avoid messing up your hashprobe code...with 2-tier it's very messy. But don't worry, almost 100% your windows are null and that's about as efficient as you can get. Michael
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.