Re: [Bug-gnubg] Patches for go-to-next-marked

[Joern Thyssen]

On Thu, Aug 08, 2002 at 12:27:21PM +0200, Jim Segrave wrote

> Which leads to a side point or two:
> 
> First, the analyis settings allows you to set chequer and cube
> analysis in one window, whereas you have two separate ones for
> evaluation. Personally I think evaluation should have a single window
> as well. 

Yes, it has been noted by others.

> 
> Second - the preset evaluation/analysis settings go from 0 ply in
> expert mode to 2 ply in world-class and world-class++. Is there a
> reason not to have a 1 ply setting? I've noticed huge changes in the
> equities between 0 and 2 ply, which is hardly surprising. But the 2
> ply presets are noticeably slower in processing. I have used a 1 ply,
> huge search space which seems to be a good tradeoff in first-cut
> accuracy and performance on a slower - P6 350MHz machine. I'm wondering
> if there's a reason for the break from 0 ply to 2 ply in the presets?

No particular reason, other than 1-ply is not much better than 0-ply.
As far as I recall Joseph did some experiments on chequerplay strength
between 0-ply, 1-ply and 2-ply.

I've plans to perform the following experiment:

Play a large number of non-Jacoby money games with

* 0-ply vs. 0-ply (with a number of different noise settings)
* 1-ply vs. 1-ply (with a few different search spaces)
* 2-ply vs. 2-ply (with a few different search spaces)
* 3-ply vs. 3-ply (with a few different search spaces)

and analyse these games with either 4-ply or max(2,n+1)-ply.

This should give the average error rate per move for a number of
different settings. This should give the foundation for creating the
predefined settings in gnubg.

The current settings are "guestimates".

> 
> I've incorporated your buttons - they look fine to me 

It's very simple to change them if someone draws some better ones.

Jørn

-- 
Joern Thyssen, PhD
Vendsysselgade 3, 3., DK-9000 Aalborg, Denmark
+45 9813 2791 (private) / +45 2077 2689 (mobile) / +45 9633 7036 (work)