Author: Heiner Marxen
Date: 15:33:24 10/17/99
Go up one level in this thread
On October 17, 1999 at 16:14:36, Jeremiah Penery wrote: >On October 17, 1999 at 14:51:21, Heiner Marxen wrote: > > ><snip move ordering> > >>Summary: a seemingly stupid move first will not hurt much, IF it spans >>a comparatively small tree. > >True, but won't bad move ordering still hurt a lot more because you're doing >this at every point in the tree (for all the subtrees)? In general bad move ordering will hurt. But my reasoning was a general one: it applies to all points in the tree: IF small tree THEN not so bad. It still will waste some time, sometimes. Of course, I do not plea for bad move ordering :-) I just wanted to point out, that doing check moves first (leonid) need not be THAT bad, as you might think, first. Consider to be at a node, where beta is so low, that not only the "best" move, but the majority of the moves will produce a value above beta, and thus a fail high. Assume we suspect this with a high hit rate. Now, which move should we search first? Shouldn't we pick from those, that do cut off, that one, which does it with the smallest search? Yes, we should. Normally the "best" move is good in producing small searches, because within its tree it does cut off more often due to its comparatively high intermediate values. But another move, which also cuts off, might have a much smaller tree intrinsically, so that this search may span also a very small tree, although not doing as many internal cut offs. It greatly depends on the exact situation, and what we know or (often correctly) suspect about the situation at hand. If you think, the above does not apply to normal chess playing programs, you may be right. I'm a bit theoretically, here. And in the context of a chess problem solver (which does an alpha/beta with just two possible values) I have learnt, that "best" is not always fastest. Heiner
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.