Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Thanks for extracting this!

Author: Vincent Diepeveen

Date: 17:50:30 12/31/05

Go up one level in this thread


On December 31, 2005 at 20:20:47, Greg Simpson wrote:

>Vasik had very logical and persuave ideas.  I particularly liked the point that
>trading one third speed searching for twenty times the evauation per position
>almost has to be good if done right.

If that describes what he's doing then it seems however Vasik has taken the
other way around, the junior way. The utmost minimum of knowledge in leafs.

3 (Rybka?) or 4 ply depthleft (junior=4?),
do set a few pointers which search tables to use,
then search with shitzero knowledge last few plies.

Very old concept.

Called preprocessor.

Big disadvantage of it is that those last few plies mess up your positional
evaluation and that at 15 0 you play a similar playstrength like 150 0. Your
engine goes nullmoving a lot instead of a direct breakthrough.

Advantage is you can search at millions of nodes a second. Debugging those
tables is relative easy giving a bugfree search, and you're tactical very very
strong thanks to a huge nps. If your opponents make big mistakes you'll win
every game then of course. Nullmoving in won positions is usually possible.

I saw that in the past a lot too. This is how genius killed a lot of engines in
the 80s. My amazement would be great however if Rybka keeps killing others in
this way. The shock and awe has been delivered at a wrong time in history.

That's sad for Vasik. He wins paderborn with it, good for him. He deserves that
tournament victory.

But you would nearly wish him to win a world title knowing he shows up with a
very old concept and gets it to work in a way i'm amazed.

It's very nice to know therefore that finally someone else other than junior
team is using a by now an alternative approach.

Note this concept is especially suited for small slow processors. This concept
was used extensively for example by Eric van Rietpaap for those 1 Mhz
processors. He used last 3 plies a preprocessor. His explanation for that was:
"you need to get some reasonable search depth and a full preprocessor is too
weak for that".

In his eyes it gave too many positional weaknesses, so his second try for a
chessprogram that ran on a pro200Mhz was using a full leaf eval concept. It was
running around 200k nps.

However if your preprocessor is using some chessknowledge covering important
areas and has things very well tuned, then obviously you kick butt with such a
concept as no one is prepared to play against engines like that. If Vasik is
using this concept in Rybka the main differece between Rybka and junior is that
junior is hyper agressive. Whereas some might prefer rybka because it kicks
butt, of course from chessplayer viewpoint, junior plays positional more sound
as it plays more active thanks to the agressiveness.

Yet it has been proven in history that it is very hard to win tournaments and
titles with agressive play and there is way more champions that win by passive
play. There has been a lot of world champions chess over the past centuries, but
i could only write down 3 names that showed real hyperagressive games.

Kasparov, Tal and Topalov.

I dare to say that from tuning viewpoint, Rybka, Zappa and Fruit really are
setting a new standard here. Soon many programmers will follow that. Some 'old'
guys and some 'new' guys. Especially the free fruit 2.1 source code will give a
big boom there.

Just take over its eval, use its extensions, implement it in some other
datastructure, add a kingsafety and you have... ...

It will happen.

What will be more frustrating to Vasik is that he will discover that it is very
hard to win rating points with the Eric van Rietpaap concept. It plays great
now, but improving it in 2006 now will be a big problem.

I've seen how Junior team struggled to get endgame improved and they managed,
but it took them 7 years or so, and in middlegame it still makes positional
mistakes.

That's why most left that concept a long while ago, because the more knowledge
you put in the preprocessor, the bigger the positional ratingpoint loss will be
in the plies that come after it.

So if the concept described is the concept used in Rybka, winning world champs
2006 would mean others won't progress the coming months, which would be wishful
thinking with that fruit source code laying for free around.

Vincent



This page took 0.02 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.