Computer Chess Club Archives


Search

Terms

Messages

Subject: Re: Automatic Eval Tuning

Author: Vincent Diepeveen

Date: 09:59:21 07/02/01

Go up one level in this thread


On June 29, 2001 at 12:23:16, Steve Maughan wrote:

Hello othello is a children game compared to chess of course.
It's like when i'm in the draughtsworld and people there still
haven't figured out to cache EGTB probes in a special cache
and they still haven't figured out there to do more probes
in hashtable.

Same is true for Othello.

Add to that that Othello is nearly solved.

If you would be searching a bit faster & deeper, then
you don't even need tuning for othello.

it's like saying that a bicycle steering wheel is exactly
what a formula 1 car needs to perform better as he does now,
because it works fine for you on your own bicycle.

How many strong programs are automatic tuned?

ZERO

I remember knightcap. Its automatic tuning was so pathetic
that in the end it was giving away pieces to give its opponent
a few checks. Crafty in those days had a pretty weak king safety,
and search depths dominated in those days. So aggressive play from
a 7 ply engine against another 7 ply engine worked quite well back
then.

If it would play crafty now you would see a 0% score and knightcap
would seem a lot less well.

In short it's always easy to conclude that something works if
all you need is to get closer to a 50% score.

If you want to get a score > 50% against someone (for example
an equal rated program) then you will run into trouble!

>Vincent,
>
>>The bigger your evaluation is, the more problematic tuning it automatic
>>is.
>
>True - as the domain space increases it is more difficult to find the optimal.
>
>>Also automatic tuners don't have any chess knowledge, so they
>>don't see the difference between tuning passed pawns negative if you happen
>>to have a testset where a passer is bad now and then.
>
>You simply need a large test set.  This is the same problem as solving a set of
>equations.  In that case you have 'x' unknowns you need 'x' equations.  In the
>case of an evaluation function you will need much more due to interaction.
>
>>Another problem for automatic tuners is that you tune for testposition set X,
>>but that in reality it has to work well also for testset Y where it has
>>not been tuned for.
>
>Again you need a large test set.
>
>>Evaluations hand tuned take into account testset Y, not only testset X.
>
>Not necessarily.
>
>>Anyway, when your number of parameters gets quite a big number then
>>automatic tuning doesn't work anyway anymore.
>
>It is certainly more difficult.
>
>>Of course it might beat random chosen parameters, but it'll never beat
>>hand chosen parameters (unless a fool choses them).
>
>This is not the case for some other games.  Othello being an example.  ALL
>decent Othello programs (e.g. mine www.maughan.clara.net) are automatically
>tuned.  Othello is a purely positional game and lends itself to self tuning.
>Chess is MUCH more complex but I predict that in the next ten years we will see
>some significant advances in automatic evaluation tuning.
>
>Regards,
>
>Steve Maughan



This page took 0.06 seconds to execute

Last modified: Thu, 07 Jul 11 08:48:38 -0700

Current Computer Chess Club Forums at Talkchess. This site by Sean Mintz.