Re: [computer-go] Computer Go tournament at EGF

[David G Doshay <ddoshay@xxxxxxx>]

On 12, Feb 2005, at 10:03 AM, Chris Fant wrote:
> This is silly.
I agree. I find it very silly for people to say that it is easy to write
a strong Go program.

On 12, Feb 2005, at 10:05 AM, Evan Daniel wrote:

> Be reasonable.
Perhaps this is coming off as an unreasonable flame-war, but
that is not my intent. On one hand folks are treating the arbitrary
choice of 1 hour per program as if it is carved in stone, and on
the other hand they are referencing a theory that has no
possible use in an implementation to "prove" how easy it is to
do something that has not been done.

> Full search (full breadth, full depth) WILL play a
> perfect game.  It just won't make its first move before the end of the
> universe, let alone in our lifetimes.
Well, then the only difference between that "Go program" and one
with pseudo code:

	opponent_move goes_to devnull
	wait

is that the above pseudo-code won't have an out of memory crash
when running on any computer ever built. Yes, I know ... silly. That
is exactly the problem with theoretical algorithms that effectively
never finish.

I accept fully that all computation takes place with finite resources
and only has value when it completes in reasonable time. The
discussion should focus upon what is reasonable time. I only
started this discussion to point out that while one hour per program
is often used, and has a specific convenience factor, it is arbitrary
and that it might be interesting to try something longer.

Considering that SlugGo is the first time in years that a new
program has leap-frogged another program at or near the top of
the pack, it might be of interest to some folks to see it run. If I am
mistaken and nobody wants to know about a another strong
program until it plays in under an hour then it seems to me that
computer Go will not progress at a very fast pace. But that is just
my opinion.

> So, your challenge has an implied time limit, which I will take to be
> "within the attention span of this mailing list."
Which surely must be reaching its limit.

> Therefore, your challenge is whether a 1d program can be written that
> will play within a few days (weeks?) per move, at a maximum.
I left that to him. As I specifically stated in an earlier email, I have
a long history of far more patience than most when it comes to
waiting for a computation.

I would be thrilled to see a 1 dan program. One move per day or
week would not matter to me. I accept that your milage may vary
and have no problem recognizing the inconvenience of one that
takes a month per move(17 years per game if both players play
that slow). I have played Go by US Mail and that is slow too ...
and in my case nowhere near the 1 dan level.

> All that said, my program is slow too, and I think that slow programs
> can be interesting, so I'm in favor of there being a longer time-limit
> tournament.
I would love it for this discussion to return to that issue, so I won't
respond again to the side issue of theoretical constructs that have
no meaningful use, or of how easy it might be to write a program
that depends upon such theories.

Cheers, (and I do mean it cheerfully!)
David


_______________________________________________
computer-go mailing list
computer-go@xxxxxxxxxxxxxxxxx
http://www.computer-go.org/mailman/listinfo/computer-go/