[Computer-go] CGOS standard engine proposal

Michael Williams michaelwilliams75 at gmail.com
Mon Oct 18 04:53:31 PDT 2010

I think it's worth doing.  Only programs that allow you to specify
options that completely remove the use of the clock should be
considered.  I think most programs allow this if you can figure out
the correct options.  Also, the options should be set such that only
very old hardware would run out of time.  I would look first at gnugo,
pachi and fuego.  If you could create both a linux archive and a
windows archive, that would be great.  Please continue to solicit
input from the group on the details.

On Mon, Oct 18, 2010 at 6:27 AM, Robert Finking
<robertfinkng555 at o2.co.uk> wrote:
> I've been running a few instances of GnuGo at various levels on 9x9 over the
> weekend. I've just tried a couple of other engines this morning as the Elo
> difference between level 1 and level 10 GnuGo is tiny. The monte-carlo
> variations are stronger but see below.
> My original plan was to zip up a package of standard players so that anybody
> could run them. However there are a couple of issues:
> - The monte-carlo variations appear to adjust the number of playouts
> depending on time left. I discovered this by doing more and more playouts
> and never finding a limit that went over time. This means that the
> "standard" will be different depending on the hardware. For reference I
> using a 2.67Ghz i7 (i.e. 4 cores).
> - One of the players (Go169( doesn't eat up captured stones at the end
> - Apart from GnuGo, couldn't see how to specify komi, superko etc.
> I have only been looking at 9x9 so far. Is it worth continuing with this
> idea, or do we think there is no point in packaging and publishing?
> Opinions appreciated.
> Cheers
> Don Dailey wrote:
>> On Fri, Oct 15, 2010 at 6:42 PM, Robert Finking <robertfinkng555 at o2.co.uk
>> <mailto:robertfinkng555 at o2.co.uk>> wrote:
>>    Can we add
>>      7) We publish the passwords for the standard players?
>> That's up to you.    If you want to share passwords for standard players
>> you can.    I have access to all passwords however if any of them are lost.
>> Don
>>    I've just connected for the first time this evening with GnuGo...
>>    but since I don't know the standard login name and password for
>>    this configuration I can't log my player in with the standard name.
>>    Given the ease of misconfiguring a standard engine (how do I know
>>    I've used the right command line?), I am wondering whether we need
>>    a tarball containing all the standard players with a cgos config
>>    file containing them all (including user name and password). That
>>    would make it very easy A. for people to get going on CGOS and B.
>>    to avoid configuration mistakes (and hence getting non-standard
>>    settings for "standard" players).
>>    On a related note, where can I get Fatman from (or is it not public)?
>>    Cheers
>>    Raffles
>>    Brian Sheppard wrote:
>>        It makes sense to have more than one standard program.
>>        Preferably covering a
>>        wide range of skills.
>>        There is no need to fix the ratings of standard programs,
>>        provided that they
>>        play a large number of games.
>>        The problem is to find the resources to host a wide range of
>>        standard
>>        players.
>>        I experienced a period on CGOS where 3 programs were stronger
>>        than Pebbles
>>        by 300 rating points, and the standard 9x9 programs were
>>        weaker by 300
>>        rating points, and no opponents were in Pebbles range. I
>>        resolved that by
>>        running two copies of Pebbles, and by creating a fixed version
>>        of Fuego
>>        running on a spare computer. Now there was always a good
>>        matchup to be
>>        found.
>>        But that solution only works when Pebbles is playing.
>>        I feel like a collaborative solution should work. Here is a
>>        proposal:
>>           1) We adopt a naming convention for standard players.
>>           2) We create a set of standard players, scaled over a range
>>        from far
>>        below the anchor to far above.
>>           3) CGOS blocks creating names that follow the convention
>>        except by the
>>        server admin.
>>           4) We publish the commands that execute the standard players.
>>           5) Anyone can create an instance of a standard version
>>        (provided that no
>>        other user is currently running an instance).
>>           6) Now you can run instances of standards that scale to the
>>        rating range
>>        you care about.
>>        For example, suppose Fuego will be a standard. Our naming
>>        convention could
>>        be Std-Fuego-0.4-2000, to indicate that the player is a
>>        standard player,
>>        running Fuego, version 0.4, fixed to 2000 trials per turn.
>>        It makes sense to have a diversity of programs, so that we
>>        don't get in the
>>        rut of tuning to a specific program's weaknesses. Open source
>>        engines like
>>        Fuego, Pachi, and Orego would make excellent choices.
>>        We would need to archive binary versions that the standards
>>        depend upon.
>>        Thoughts?
>>        Brian
>>        _______________________________________________
>>        Computer-go mailing list
>>        Computer-go at dvandva.org <mailto:Computer-go at dvandva.org>
>>        http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>>    _______________________________________________
>>    Computer-go mailing list
>>    Computer-go at dvandva.org <mailto:Computer-go at dvandva.org>
>>    http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
>> ------------------------------------------------------------------------
>> _______________________________________________
>> Computer-go mailing list
>> Computer-go at dvandva.org
>> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go
> _______________________________________________
> Computer-go mailing list
> Computer-go at dvandva.org
> http://dvandva.org/cgi-bin/mailman/listinfo/computer-go

More information about the Computer-go mailing list