Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Question for Vincent Diepeveen (other programmers might answer).

Author: Vincent Diepeveen

Date: 08:34:42 12/10/99

Go up one level in this thread


On December 09, 1999 at 14:50:58, José de Jesús García Ruvalcaba wrote:

>	There are some passed pawns which are a dangerous threat to the opponent,
>because they can promote and are therefore very valuable.
>	There are some passed pawns which are safely blocked or which can not advance
>without being taken. They are not as valuable.
>	How does Diep tell a good passed pawn from an inoffensive passed pawn? This is
>related to a somewhat old post of you, which must be buried behind a few days of
>posts.
>	I ask because I have seen many programs overvaluing passed pawns that are not
>dangerous, prefering them over healthy pawn majority which eventually would
>produce a stronger passed pawn, and I wonder if Diep is better at making these
>decisions.
>José.

I'm sure in all programs in the world a lot can be improved upon
passed pawns, especially when concerning endgame.

DIEP has quite a bit of knowledge concerning passed pawns,
a lot more factors than you describe above are taken into account.

I'm not gonna describe here algorithmicly nor chesstechnical
what is in DIEP. in general all terms a 2000 player or worse player
than that has ever heart from are in DIEP.

I don't need to mention that still this doesn't mean that the
positional evaluation is like a 2000 player.

I'm 2254 FIDE myselve, but getting up. All my TPRs are 2400+, but
i play too little games.

Like Jaap v/d Herik already told once a big problem is the relative weights
of terms to each other. a human is still superior in that, which is very
good otherwise we could create a human easily in software which could
take all decisions on all terrains.

Yet i'm amazed how well the knowledge in general works. Just the summation
of bonusses and penalties and some meta knowledge as described here
have a big impact on the positional style.

Where most chessprogrammers are only busy fulltime busy tuning the few
weights they have in DIEP i'm mainly busy with making new terms or expanding
current ones and fixing old ones. Tuning i have done a lot less.

I have a very simple vision on tuning. The less terms you have the more
important the exact tuning of it is. With little terms tuning means
that simply in x positions your program will play bad. improving tuning
means in the commercial chessworld usual tuning the program to the openings
it plays and to the style those opening represent.

Nimzo is able to just give away a pawn for some vague development or
pressure, yet that usual doesn't matter as the aggressive and huge
book of it gives it positions where it can affort to do that.

Same we see with fritz6 now. Where fritz used to be a major pawngrabber
in the past, fritz6 now is a lot more aggressive in this area.
This has very obviously to do with the new book of it, which
is the nimzo book!

>P. S. I have access to a dual UltraSparc machine, and I migth be able to run
>some test of reasonable length (say two-three hours) for you. If you are
>interested, just let me know.

Though diep works on the sparc i have done so far no effort to get it
to work SMP on the sparc. Note that sparc is not very fast processor for
chess. it is not a very fast processor in general.
small L1 cache (8 kilobyte if i read well!!)
and performance though not that bad as in the past at a sparc of 300Mhz 3d
for diep is comparable with a PII233.

Some overnight analyzes at positions with a vaste amount of hashtables,
say 2 gigabyte or something, that's of course very interesting though.

diep@xs4all.nl for further contact info.

Vincent







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.