Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: MTD(f) and hash table size

Author: Tord Romstad

Date: 11:09:19 08/17/03

Go up one level in this thread


On August 17, 2003 at 04:27:06, Rudolf Huber wrote:

>First scan you source files for occurences of alpha and beta.

There are no alphas or betas left -- I rewrote search() and qsearch() from
scratch
when implementing MTD(f).  I only have the single parameter (gamma), like in
Aske Plaat's paper.

>If you find any look hard and try to understand what you are doing there. In
>mtd(f) there ist NO alpha and NO real beta. All the things (lazy eval?)
>you do with alpha and beta are most likely wrong. Use the last score instead.

What does the "last score" mean in the above sentence?

>Also I think it is a good idea to keep it simple. Why not disabling
>forward pruning and your "convergence accelerator" and first try to make
>your mtd(f) implementation better than negascout.

I agree entirely.  This is basically what I have done (except that I have no
negascout
implementation to compare it to).  I started with a very rudimentary full-width
search
without any extensions, reductions or pruning.  When everything appeared to
work,
I gradually added the usual extensions, recursive null move pruning, and checks
in
the qsearch.  When I was satisfied with this, I implemented a more efficient
search
driver.  It has now reached the stage where it works rather well, except that my
program is so slow (it has a heavy and time-consuming eval) that I need to
implement
some kind of forward pruning.  The techniques I use with aspiration search do
not
combine very well with MTD(f), because they use alpha and beta.  I have to look
for something better.

Tord



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.