Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: How Rebel plays at SSDF the bare facts, just statistics and thoughts

Author: Enrique Irazoqui

Date: 05:56:51 06/17/98

Go up one level in this thread


On June 17, 1998 at 01:29:46, Ed Schröder wrote:

>>Posted by Enrique Irazoqui on June 16, 1998 at 20:22:19:
>
>>In Reply to: Re: How Rebel plays at SSDF the bare facts, just statistics and
>>thoughts posted by Ed Schröder on June 16, 1998 at 19:16:46:
>
>>On June 16, 1998 at 19:16:46, Ed Schröder wrote:
>
>>>>Posted by Enrique Irazoqui on June 16, 1998 at 18:02:56:
>
>>>>Not true. The first version of Fritz’s autoplayer didn’t allow some of the
>>>>opponents to save their games.
>
>>>>As a consequence, Rebel, and only Rebel, could not learn.
>
>>>How do you know?
>
>>>Did you ask other chess programmers too then?
>
>>What for? No other program that played against Fritz 5 in the SSDF matches went
>>twice for the same losing line, except Comet32 in 2 games. All the other double
>>games in there are the Fritz5-Rebel9 ones.
>
>>Knowing what kind of aggressive learner Fritz 5 has and knowing that
>>Rebel 9 did not behave like this in other matches, this losing double
>>games indicate that Rebel's learner was the only one hampered by
>>Fritz's autoplayer.
>
>And who is to blame for that?

First of all, it happened. Rebel 9 lost to Fritz 5 repeatedly because Rebel's
learner did not work well under non auto232 conditions. Reasons: R9 did not save
its games; games were played in w-b-w sequence. You complained about this
before.

As for who is to blame, I'll try to go a step further than my prior "go figure".
I think, and always said, that Fritz's autoplayer should be made public, even if
only to put an end to all this repetitive discussion, and not because it does
not adhere to auto232 conventions. As far as I can tell, an autoplayer has to do
its job: let you plug the serial cable and autoplay machine-machine. Anything
else is in my opinion arguable and artificial.

On the other hand, Rebel's learner is faulty. No one says a game must be saved
before learning takes place. No one says a player must play all games as white
first and as black later. White-black-white is the common way and as such
acceptable. A learner that requires saving games and playing them in the auto232
sequence is a learner good for the SSDF and mostly useless against human
opponents. When I play a program, most of the time I won't save my games and I
will play with any color I want. Then Rebel 9 won't learn better than in its
match against Fritz 5. This is a design flaw, independently of whether the
autoplayer Fritz 5 is released to the public or not, whether or not it complies
to auto232 standards. That's why the learners of other programs worked just fine
against Fritz 5 in the SSDF matches, while Rebel's did not. I have the same
pattern in the matches I autoplayed with Fritz 5: the only double games were
played (and lost) by Rebel 9.

Another way to put it: I am sure that the learner in Rebel 10 will correct these
shortcomings. Any bets? :)

>Rebel9 learner doing well against *ALL* other SSDF participants but not
>against a hidden secret non public autoplayer

Not because is "hidden", "secret" and "non public". See above. And one more
point: making this autoplayer public would allow you to adapt Rebel 10, which
would be only fair, but it wouldn't change a thing for Rebel 9. Making it public
wouldn't have any impact on the SSDF list, the first one with Fritz 5 on top or
the next one coming this week.

> that doesn't stick to
>the general excepted rules of AUTO232 which instead of that created
>his own rules and what's worse nobody can check.

We checked it, you and I together, and nothing was fishy with it. Yes, it does
not play games in the same order as auto232. So what? So Rebel couldn't handle
it, unlike all other learners.

>As a result I can't judge what went wrong in these games, next it
>prevents me to correct it because I don't know the reason.

You complained before about Rebel 9 not being able to save its games when
autoplaying Fritz 5 and about the w-b-w sequence, because then Rebel's learner
wouldn't work properly. This is why R9 lost so many doubles to Fritz.

>Nice way of competing :)

No, it is not... I agree about releasing the autoplaying version of Fritz 5, so
everybody can compete on equal grounds. I also believe that a learner demanding
auto232-SSDF conditions, as opposed to normal human playing, is severely
handicapped and of not much value to human players.

Enrique

>- Ed -



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.