[Top][All Lists]

[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]

RE: [Bug-gnubg] Is it time for Gnubg 0.15? Re-rolling the position datab

From: Ian Shaw
Subject: RE: [Bug-gnubg] Is it time for Gnubg 0.15? Re-rolling the position database.
Date: Mon, 17 Jul 2006 16:30:47 +0100

> -----Original Message-----
> From: Joseph Heled [mailto:address@hidden 
> Sent: 17 July 2006 16:05
> To: Ian Shaw
> Cc: address@hidden
> Subject: Re: [Bug-gnubg] Is it time for Gnubg 0.15? 
> Re-rolling the position database.
> Hi Ian,
> Yes, it would be great to improve the playing level of gnubg 
> once more. Here are (yet again) my thoughts and comments.
> The net was not trained from the rollout results, but by 
> using 2 ply evaluations. The best I could come up with 
> resulted from the choice which positions to include in the 
> training set. And for my particular training method, more was 
> frequently not better.
I've just been reading old archive posts, e.g.
http://lists.gnu.org/archive/html/bug-gnubg/2003-11/msg00104.html, and
realised that my description was incorrect. I guess I'm a victim of
wanting to have a simple explanation to a question, but the reality is
much more complex.

So how exactly were the rollout results used? Are they just used as a
benchmark for trying out new neural nets. As such, can we safely assume
that they are still accurate enough, three years on, to be the gold
standard? If so, I would be wasting my time re-rolling them. Should I
just drop the idea?

> My personal view is that to move past the next step we need a 
> new method to generate an evaluation net. this requires some 
> thought and research by someone other than me, as I am set in my ways.

I have some ideas, but that's another thread. It's by far the most
radical approach, but quite likely the most rewarding, both in terms of
playing strength and enjoyable design and implementation.

-- Ian

reply via email to

[Prev in Thread] Current Thread [Next in Thread]