Re: [Bug-gnubg] questions about saving rollout games

[Jim Segrave]

On Thu 20 Nov 2003 (19:30 +0100), Misja Alma wrote:
> Hi All,
> 
> Thanks for your tips about how to save rollout trials to .sgf files. I tried
> this with a position where I didn't trust gnubg's rollout results. It
> worked, I got a whole lot of .sgf files.
> Now my plan was to import those into Snowie (3.2), so that I could check if
> gnubg was playing correctly. Snowie does not accept .sgf files, so I export
> every file from gnubg to a .txt file. But when I import this into Snowie, it
> says that the file does not contain a position, game or match. Is there any
> other way to do this?

I don't actually have Snowie, so I can't tell you how you can import a
partial file

> My second problem is that there is there are many files to import into
> Snowie. If I would have to export those to .txt format (or whatever else),
> this would be a lot of work. Now I heard that gnubg can run from the command
> line, and that it even works with Python commands. So I could perhaps build
> a script that let gnubg convert all files in a directory! That would be
> great. Could you please help me to find out which commands gnubg accepts? I
> can't find them anywhere..

Joern or Joseph may be able to help there - I haven't played with the
Python interface at all.

> Finally I imported one of the rollout trials into gnubg itself to check how
> it had been playing. But it seems that the file contains wrong information,
> the moves don't make sense or are even illegal. Are you sure that the 'set
> rollout log yes' command works correctly? I included one of the .sgf files
> in the e-mail. The position which was rolled out was ObsdAAhi2zYQEA
> (matchscore 0-0 to 7). gnubg evaluates this correctly as double-take, but a
> rollout says double pass.

Arrgh. My routines for writing the files out only worked correctly
when player 0 was on roll, not player 1. If you look, you'll see that
the players checkers are reversed (player 0's checkers are where
player 1's should be (and are the wong colour) and vice-versa.

I can run a simple script over the files to fix
this (attached is the same .sgf file, but with the board set
correctly). If you want to zip up the files and mail me the result (at
address@hidden), I can run a script over them and send you back a
zip file with things the right way around. 

I have the fix commited to CVS and in a day or two later the gnubg download
sites will have the fixed version.

-- 
Jim Segrave           address@hidden