[Top][All Lists]

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

RE: [Bug-gnubg] BUG: Rollout statistics wrong

From: Ian Shaw
Subject: RE: [Bug-gnubg] BUG: Rollout statistics wrong
Date: Mon, 8 Mar 2004 08:40:26 -0000

Jim Segrave wrote on Sunday, March 07, 2004 11:55 AM:
> On Thu 04 Mar 2004 (09:23 -0000), Ian Shaw wrote:
> > 
> > The rollout statistics in this file look completely rubbish. 
> > There are negative numbers in some entries , and the final 
> percentage reads 1.#J%.
> > 
> > The two positions rolled out are the two double/takes.
> > 
> > Windows version 1.1444 031101.
> > The rollouts were stopped and extended several times, but 
> all within the same session of gnubg.
> I loaded both of these into gnubg and don't see anything odd at
> all. Linux, gnubg 1.1504, 2 feb 2004. But nothing should have been
> changed in the sgf processing in a long time. The values in the .sgf
> file are reasonable - there are no negative numbers there (well, there
> are negative exponents on some floats - 3.972246486e-005 for example,
> but that's normal.
It isn't the cube decision stats that are the problem. It is the numbers in the 
View Statistics window that are dodgy.

Try setting your rollout trials to a low number like 1296 (since I've already 
done about 5000 trials) and select to rollout the cube decision. It should be 
immediately complete. If you then View Statistics you should see, for example, 
that -1 games ended on a 128 cube.

Secondary issue:

When resuming a rollout, the window shows 0 trials until the first new trial is 
complete. It would be nice to immediately show the start position. It doesn't 
matter much on 0-ply, but it is unnerving on 2-ply, where it looks for a while 
as if all your previous hard work has been lost and gnubg is starting again.

-- Ian

reply via email to

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