RE: [Bug-gnubg] gnubg defends poorly against outer primes

[Misja Alma]

I have a few questions about the training of gnubg. Perhaps one of you can
help me out?
In reply to my mail about gnubg's poor handling of outer primes, Øystein
wrote that it was extremely hard to train those positions. Why is that?
I have read the interesting article 'the gnubg training program' on the
internet, and I know some about neural nets myself, but I still don't
understand? If the problem is in the evaluating of benchmarks because
gnubg's rollouts are biased, why not let them be rolled out by Snowie? This
seems to handle them pretty well.

My other question is about the fact that gnubg uses three different neural
nets; the crashed net, the race net and the contact net. I can see how gnubg
knows when to use the race net, but how exactly does it know when to use the
crashed net and when the contact net? I could imagine there are some
borderline positions there.

Ciao,
Misja

-----Oorspronkelijk bericht-----
Van: Øystein Johansen [mailto:address@hidden
Verzonden: Sunday, November 23, 2003 9:43 PM
Aan: Misja Alma
CC: address@hidden
Onderwerp: Re: [Bug-gnubg] gnubg defends poorly against outer primes


Misja Alma wrote:
> Hi,

Hi!

Thanks for your observations. We're aware of the bad evaluations of
outside primes, but these kind of positions are extreamly hard to train.
I don't think JellyFish does thes kinds of positions any better, and
maybe not even Snowie. This is still a weak point of most bots. Thanks
anyway.

<quote>
..... I read in your excellent manual that ......
</quote>

Excellent manual? Where? Which manual are you talking about? ;-)

> PS: importing the rollout trials into Snowie was quite a hassle,
 > because the JellyFish .gam format does not support games which
 > do not start from the initial position. So I had to create a
> fake start of the game where finally my position would come up,
 > and this start I had to paste at the start of every saved
 > rollout trial.

Pasting in a fake start could be a simple solution with some scripting,
however I really believe JellyFish .gam format do support positions in
the game line, but I'm not sure if Snowie can import .gam files with
positions in it. (At least I don't think GNU backgammon can handle
positions in .gam files.)

Maybe someone can work out the format based on the last game of this match:
http://www.nbgf.no/matches/hans_karsten_nordic1997.mat

> As far as I can see the only solution would be to implement an
 > export to Snowie format? I read in an earlier post that Snowie
 > .swm format is just a zipped Paradox database. So I tried to unzip
 > such a file and open it in MS Access, but this gave an error message.
>     So does anybody know which kind of a Paradox database this is?
 > Then I could perhaps make an attempt to write an export/ import
 > function in Delphi..

It would be excellent if someone can work this out. I'm not sure if
anyone have tried this seriously, but I think it will be a hard and a
time consuming task.

Hei og hopp!
-Øystein