Re: Scoring Go games (fwd)

[Jan Ramon <Jan.Ramon@xxxxxxxxxxxxxxxxx>]

On Thu, 17 Jun 1999, Patricia Hughes and David Elsdon wrote:

> Hi there,
> 
> I am developing a strong Go-playing program using AI/KBS techniques. I
> expect the project to take several years. The development of the program
> 
> is progressing very well at the moment - it plays novice level go on a
> 9x9 board and absolutely slaughters my random go player. It would have
> been very embarrassing if this were not the case !! :-)
> 
> I am writing my program in LPA Prolog, so if anybody wants some prolog
> code to use as a starting point for developing Go playing software just
> let me know and I'll send you mine. And if anybody has some wizzy bits
> of prolog code for playing go I would be very interested.
> 

I'm interested in your prolog code.  

We're doing some research on go in relation to Inductive Logic
Programming (mainly machine learning). 

I also have some sicstus prolog + C routines i've written some year ago.
I'll make my routines publicly available as soon as they are cleaned up a
bit (i.e. made readable + documented).


> I have a small problem. There is a research project being run by Dr
> Andras Lorincz at Eotvos Lorand University, one of the Budapest
> Universities, where they would like to use my software as part of a Go
> learning system. The learning is based on Time Difference learning and
> Function Approximation. Interfacing my software to theirs should be no
> problem.
> 
> There must be something wrong with that last sentence: it contains
> "interfacing", "software" and "no problem".
> 
> Now to the problem. Scoring the game is a real pain. My program is not
> very good at scoring yet and accurate scoring is important to the
> learning software. So does anybody know where I can find a scoring
> program which I can interface with our learning system. It would be
> useful to have source code and C or C++ would be fine. If anybody has
> some prolog code that scores the game I would be "very" interested.
> 
difficult, very difficult.

in fact it depends on what you want:

do you want a program that decides which groups are dead, which alive,
which seki, etc.?  Scoring needs search/evaluation in itself 
(e.g. Professional games are sometimes difficult to
score for beginners because semeai are not played out)
One of the most difficult problems with learning go is that the
evaluation function is very difficult.

if you only have to score games with dame filled and prisonners removed
... well, then you use the scoring algorithm of any go server.


> I came to Budapest last October and I will be living here for several
> years, so if any one fancies a few days, or weeks, in Budapest feel free
> to contact me. I can even offer accomodation (a sofa bed in the lounge)
> for a short period.
> 

I would be interested in knowing in which domain you're looking for your
techniques.  You talk about AI/KBS techniques. what is KBS?

Jan


                                          \\\|/// 
                                        \\  ^ ^  // 
                                         (  @ @  ) 
+--------------------------------------oOOo-(_)-oOOo----+
|  Jan Ramon                                            |
|  Department Of Computerscience K.U.Leuven             |
|  Celestijnenlaan 200 A 03.46                          |
|  3001 Heverlee                                        |
|  Belgium                                              |
|  janr@xxxxxxxxxxxxxxxxx; jan.ramon@xxxxxxxxxxxxxxxxx  |
|  +32-16-32.75.67 (A03.46)                             |
|  +32-16-32.75.52 (Secretary deptcw)                   |
+-----------------------------------------------Oooo----+ 
                                        oooO   (   ) 
                                        (   )   ) / 
                                         \ (   (_/ 
                                          \_)