[computer-go] ZG1bot

[Łukasz Lew]

On 9/12/06, Don Dailey <drd at mit.edu> wrote:
> Łukasz,
>
> If you did it exactly according to my spec,  it should achieve a rating
> very close to GenericMC_100K.   But since you are using the All as first
> heuristic it could be stronger.   Right now, it is showing about 80
> points higher.
>
> GenericMC_VAR  may be close to the upper bound of how high you can get -
> it starts out with several hundred thousand simulations and gradually
> decreases as the game proceeds.   It DOES NOT use "all as first".
> Many of my experiments show that most of the effort should be spent on
> the earlier moves - thus GenericMC_VAR.
>
> I don't have an easy way to test where the ceiling is - I only suspect
> it's between 1500 and 1600.    All as first gets you closer if you are
> not doing a LOT of simulations, but probably limits you in the long run,
> perhaps at higher levels than we can achieve in practical play.    So
> your experiment will help us understand that.

What I wanted to say is that I have a lot of examples of SO bad moves.
Big semeai, both groups in atari. All as first do NOT capture
It tries to save a black group by playing where white should play to capture,
and then white captures by playing elsewhere (if it is not All as first :) ).

Understanding why it plays so is also valuable.

BTW
If You want I can put 200k or 400k, should be no problem as
ZG1bot has 45k pps on 2.2GHz

Best Regards,
Łukasz

>
> - Don
>
>
> On Tue, 2006-09-12 at 00:30 +0200, Łukasz Lew wrote:
> > ZG1bot-100k is online on CGOS
> > It is raw monte carlo, standard local eye definition, 100k playouts,
> > all as first heuristics,
> > played move with a highest probability of win.
> >
> > BTW all as first heuristics leads to a terrible mistakes :)
> >
> > Best,
> > Łukasz Lew
> > _______________________________________________
> > computer-go mailing list
> > computer-go at computer-go.org
> > http://www.computer-go.org/mailman/listinfo/computer-go/
>
> _______________________________________________
> computer-go mailing list
> computer-go at computer-go.org
> http://www.computer-go.org/mailman/listinfo/computer-go/
>