Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: MVV/LVA verses MVV/MVA

Author: Tony Werten

Date: 03:39:31 08/04/04

Go up one level in this thread


On August 04, 2004 at 06:04:52, martin fierz wrote:

>On August 04, 2004 at 05:44:57, Tony Werten wrote:
>
>>On August 03, 2004 at 22:07:58, Dann Corbit wrote:
>>
>>>In a thesis paper on hardware move generation, the author found better success
>>>with MVV/MVA than MVV/LVA for normal search (as opposed to quiescent).
>>>http://www.macs.ece.mcgill.ca/~mboul/ICGApaper.pdf has this:
>>
>>How serious do you want to take a paper that uses a chessprogram that plays best
>>if killer move, transpositiontables, check extensions and positional evaluation
>>are disabled ?
>>
>>Tony
>
>to be fair, that project was about writing an FPGA-based move generator.
>obviously, they only had a toy program to play with, because they concentrated
>on that hardware part. so i guess if you want to take anything in that paper
>seriously, it would be that move-generator part, and you should forget about the
>rest.

That was what I meant, sorry if it wasn't clear. My exact point was that you
can't take any claims about node reductions serious.

I wouldn't want to be negative about the FPGA move generator. Specially since I
don't understand that stuff myself. Read about it, tried some things and
concluded I really don't get it.

Tony

>
>cheers
>  martin
>
>>
>>>
>>>"The arbiters are also capable of dynamically reversing priorities, thereby
>>>permitting two different move ordering schemes: most-valuable-victim /
>>>least-valuable-aggressor (MVV/LVA) and most-valuable-victim /
>>>most-valuableaggressor
>>>(MVV/MVA). This is labeled MVV/XVA. It was observed that MVV/MVA is the better
>>>of the two move ordering methods during full-width tree searching (13% fewer
>>>nodes, 10 opening-game test positions used).  However, in quiescence search,
>>>MVV/LVA is the preferred ordering (9% fewer nodes, same test positions). It
>>>seems logical that during capture search, it is better to capture with the
>>>least-valued pieces first. In full-width searching, the stronger pieces
>>>typically cause the most damage and/or board control, explaining the somewhat
>>>unorthodox MVV/MVA move ordering."
>>>
>>>Has anyone else tried this reversal for search/qsearch?



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.