Re: [gnugo-devel] experimental connection test

[Arend Bayer]

On Thu, 29 Aug 2002, Evan Berggren Daniel wrote:

> I ran matcher_check --stable with gnugo (cvs) with experimental
> connections on, and normal as the stable version.  All 5 games and the
> output are at http://evand.rh.ncsu.edu/~evand/gnugo/expconn.tar
>
> I haven't yet looked at them, but hopefully this will give some help for
> refining that code.  Would it be useful to run current cvs compared to
> 3.3.6 and/or 3.2 and look at results?  If so, I'll go ahead and start
> that.

I think the main work is not so much colllecting moves where the two
versions differ (note that this can also be done with
"gnugo --experimental-connections -l x.sgf --replay white" where x.sgf
is a game played by the "stable" gnugo), but to generate good, useful
regression tests from them.

I've already considered writing a script that shows the position (e.g.
using gnugo and showboard), the two moves found, asks the user for a
list of correct moves (or lets him decide to skip that position) and then
automatically generates regression test file (or a patch adding the new
tests + new .sgf-files in their right places in CVS...)  This would
make generating these tests much more convenient.

I haven't much looked at --experimental-connections yet, but
--experimental-semeai would definitely benefit from having more
semeai-specific test cases available.


Arend