Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Fritz7 vs Crafty 18.12

Author: Roger D Davis

Date: 09:49:01 12/18/01

Go up one level in this thread



>
>Let's put it with some tact:
>
>1) if a program needs a special book to play correctly and cannot stand to use
>the same book as its opponent, then it SUCKS.
>
>2) if a program is not able to perform well with ponder=off when its opponent is
>also ponder=off, it SUCKS.
>
>I do not think 1 and 2 apply to Crafty.
>
>No matter what Bob tells, I have yet to see any proof that Crafty is handicapped
>by ponder=off. As far as I remember, results have shown that Crafty does not
>perform worse in ponder=on than in ponder=off matches.
>
>And this apply to most if not all chess engines.
>
>I also do not see any reason to believe that Crafty is more handicapped than
>Fritz by a book that has not been designed specially for it.
>
>I could even say that a commercial program, which is supposed to be helped by a
>hand tuned book, should be the most handicapped of the two.
>
>
>I find Pavel's experiment interesting and I think it tells a lot about the
>respective strength of Crafty and Fritz. I'm pretty sure additional experiments
>will confirm this result, independantly of the time controls and book, and
>ponder setting used.
>
>Those who reject the result do it for very strange reasons. Actually I think
>they would reject the result of any experiment. In this world you need to be
>able to draw conclusions (including margin of error in your conclusion) from an
>unperfect experiment setup, using your own experience and understanding of the
>experiment field.
>
>
>
>    Christophe

On the other hand, science itself is defined by its attempts to maintain what is
known as "controlled conditions." Results of experiments are considered to be
"ecologically valid" only to the extent that the conditions of the experiment
resemble "real life."

In real life, human players specialize in particular openings. They suck at
other openings, either because these do fit the players temperament or style. SO
they don't play them. Simply put, there is no moral imperative that a chess
program play all books that can be put to it equally well. Chess programs will
differ in their innerds, and it is these differences, presumably, that give the
engine a particular style. If an author develops a book that plays the strengths
of the engine while minimizing its weaknesses, then that make for smart
tournament play.

You may have different goals, Christophe, simply because your objective is to
optimize the interaction of some set of chess principles across the universe of
chess positions. But that's you...to say that a program that cannot stand to use
the same book as its opponent sucks...in my opinion, that is extreme.

In my opinion, your other arguments should also be evaluated carefully. You
stated, "I also do not see any reason to believe that Crafty is more handicapped
than Fritz by a book that has not been designed specially for it." Yet, it is
also the case that if we were to systematically match various books with various
engines and play a very, very large number of games, we would find that the
win-loss ratio would change. So we can almost certainly conclude that some books
will favor certain engines, while other books will handicap certain engines.
Take any particular book and throw it at two engines in a match, and one may
excel while the other suffers. Accordingly, you say, "I also do not see any
reason to believe that Crafty is more handicapped than Fritz by a book that has
not been designed specially for it." But I say, "I don't see any reason to
believe that Crafty is EQUALLY HANDICAPPED OR AUGMENTED than Fritz by a
particular book." The result is a loss of scientific control AND a loss of
ecological validity, since the engines are intended to play with their books.

And then there is the issue of pondering. Do all engines ponder with the same
efficiency as they play? If engines vary in terms of their pondering efficiency,
then of course pondering matters. Engines which benefit more from longer time
controls are more likely to benefit from pondering. Accordingly, if you turn
pondering off, then you are handicapping the engine.

You say that you are pretty sure these results will be confirmed, yet you fail
to say exactly what results it is to which you are referring. If you mean simply
that Fritz is stronger than Crafty, then this will probably be confirmed. I do
not believe that this is what you meant, however, since it is obviously too
small a point for this forum. If you believe that the degree to which Fritz is
stronger than Crafty will generalize across all time controls, all ponder
settings, and all books, I find that suspicious, simply because I am asked to
believe that there are no situations in which the interaction of book, ponder
setting, and time control affect one engine more than another. I find that
implausible.

Roger




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.