Re: [gnugo-devel] assignment with gnu-go

[Evan Berggren Daniel]

On Tue, 12 Nov 2002 address@hidden wrote:

> Hello,
>
> I have an assignment in my A.I. course where I have the opportunity to go work
> on whatever I want. I was wondering if there is any little projects relating 
> to
> AI that I could help out with in Gnu-go. I'm familiar with go at little more
> than beginner level.

I started working when I was about 20k, and I'd like to think that my
first contribution, the matcher_check script, was at least moderately
useful.  I think the regression suites provide a good set of problems to
work on, and since there are solutions and sometimes comments, it
shouldn't be too hard to see why the correct answer is better than gnugo's
even at a beginner level.  Stepping through the problem having gnugo
generate moves and report status of groups can often help you understand
what is going on.

>
> I have read the Tasklist and am working through the overview and code as a
> start.
>
> couple things that caught my eye were
>
> ->support for ko in owl.c and optics.c
>
> ->global alpha-beta tree
>
> Are these things that I could help with? or any other suggestions?
> How long does it normally take people to get familiar with the code and start
> working in previous experience?

I found it took a couple days to get a basic familiarity, but even now
there is lot left for me to learn.  The tactics code is probably the
simplest to learn, but in some ways one of the less interesting pieces of
the engine.

As for something that might have more of an AI (or at least game theory)
feel to it, there are papers on how to mathematically solve endgames.
There is also a book available, I believe it is called Mathematical
Endgame Analysis in Go.  Making gnugo play a perfect endgame would be a
worthy goal, and would probably not interfere with anything else going on.
Also, endgame could be tackled as a subset of a global alpha beta tree.

Hope this helps

Evan Daniel