Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Root move ordering - an experiment

Author: Dan Honeycutt

Date: 07:33:14 09/30/04

Go up one level in this thread


On September 29, 2004 at 17:59:58, Ed Schröder wrote:

>On September 29, 2004 at 10:43:12, Dan Honeycutt wrote:
>
>>On September 28, 2004 at 18:44:13, Ed Schröder wrote:
>>
>>>
>>>Interesting idea. Are you sure it's the best way of doing things? Any speed
>>>increase percentage to offer?
>>>
>>>My way of doing: after iteration one I do a full sort, after iteration 2 and up
>>>I increment the root scores with a decreasing value first (move-1 + 256, move-2
>>>+ 248, move-3 +240 etc.) before I sort. This is meant to keep the order of
>>>iteration one as good as possible and in the case certain moves had serious
>>>score increases these moves automatically will make it to the top.
>>>
>>>As last item I make sure that previous best-moves are stored as second, third,
>>>fourth in the list and finally that fail-high-errors (fail-high's that after a
>>>research did not produce a best move) are stored as second.
>>>
>>>I never found a better system. I will try to mix your idea with system and see
>>>what happens.
>>>
>>>My best,
>>>
>>>Ed
>
>>Hi Ed:
>>How do get scores for the other root moves?  In my case, for a typical iteration
>>(no fail hi or lo), the first move sets alpha.  All the rest of the moves come
>>back with that same value.
>
>Exactly.
>
>That's why I said to modify the scores before you sort after iteration one, see
>above.
>
>So at iteration one you just sort, every next iteration add a decreasing value
>first and then sort. It will keep the moves in place.
>
>My best,
>
>Ed

I think I see.  The only actual new scores (iteration 2 and up) are when a move
brings back a value greater than alpha.

Thanks.
Dan H.




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.