Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: move ordering and node count

Author: martin fierz

Date: 15:08:10 03/29/04

Go up one level in this thread


On March 29, 2004 at 15:46:01, Dieter Buerssner wrote:

>On March 29, 2004 at 15:22:50, martin fierz wrote:
>
>>On March 29, 2004 at 14:30:17, Dieter Buerssner wrote:
>>
>>>This has not so much to do with your question, but I doubt the last part of your
>>>sentence. I believe, it will be impossible to become as bad as the minimax tree,
>>>even when by purporse ordering the moves "perfectly wrong". You will still have
>>>plenty of situations, where many different moves give a cutoff. In a previous
>>>experiment, I got 50% beta cutoffs for the first move, when randomizing the move
>>>order. Note that this is far away, from a minimax tree (where you would need
>>>100% beta cutoffs in the last tried moves - there are in general many more move,
>>>you try before).
>
>>hmm, everybody writes this that making A/B MO as bad as possible you return to
>>minimax. somewhere below gerd just made the same point as you did here. and if
>>two experienced programmers like you say so, i am of course afraid to contradict
>>you :-)
>>but i have to contradict you all the same. in a perfectly misordered tree you
>>will *never* fail high. which also means that the case that you and gerd were
>>thinking of never happens.
>
>martin, I might have not thought well enough about it. But some things, that
>might happen: 2 moves have exactly identical minimax score somewhere inside the
>tree (I think this will happen quite often like piece1x... opponetx... piece2x
>and piece2x... opponetx... piece1x). I think, you cannot avoid to get cutoffs.
>But, I have to agree, I thought of it too naively first.
>
>Cheers,
>Dieter

hi dieter,

i see your point though: if you have an evaluation function which always returns
0, then you get cutoffs everywhere. in this sense, i think that a tree is only
maximally disordered if you have a >, not a >= in the values of siblings.

cheers
  martin



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.