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

[David G Doshay <ddoshay@xxxxxxx>]

On 14, Feb 2005, at 8:37 AM, Don Dailey wrote:

> I think you are getting hung up on the analogy I used and not getting
> the deeper point.  I think most of the group understood it was about 
> the
> relationship between computing power and GO playing strength.
>
> You see, the computer is purely hypothetical.  It is not something you
> are supposed to try to build.  Think of it like you might think of a
> "plane" in geometry, not a real physical object, just a mathematical
> concept.
>
> When I was in high school the teacher made the mistake of comparing a
> "plane" to a "very large flat sheet of paper" and several of the
> children couldn't get past the physical properties of a piece of
> paper!
Indeed, this is where Don and I have a very different
perspective, one we both have come to terms with.

To me, a computer is something I buy and then put to work.
To me, being told that an easy algorithm scales as N! and
N is larger than 1 digit is the same as saying "I cannot think
of an algorithm that scales," because to me N! algorithms
are the definition of "does not scale." To a theory guy this is
not the case because the scaling does have a specific form.
I guess that once it has a name it is not as scary.

And just like the above example with the paper and the
plane, this discussion went off into the weeds (and it looks
like it is still going further out there) because an analogy
was given that made perfect sense to one person and
meant something entirely different to others. Some of us
are interested in theory and others in practice.

To be concrete: increasing search depth takes time and
results in SlugGo playing better. This was never in dispute.

My nature is to find the point at which my software works
and then discuss the advantages and disadvantages to
the approach. I find it interesting that folks who glibly
reference times in excess of the lifetime of the universe
still get up in arms about the difference between 1 hour
and 3, but I accept it.

And just to be concrete again, SlugGo will rarely take 3
hours per game, and also will rarely take 1 hour per game
at what we consider normal settings against Many Faces
when giving Many Faces between 4 and 6 stones. Most
of the games take about 2 hours.

And one more component of my curiosity about this
insistence that computer Go tournaments only give 1
hour to each program is that when we are playing our
automated games against version 11 of Many Faces,
we have Many Faces playing on some old PC that was
a leftover department spare, and Many Faces usually
plays so fast that we do not see SlugGo's move and then
Many Faces' move played sequentially ... all we see is
the new MF move and then we have to search for the
last SlugGo move. On very rare occasion MF will take
2 or 3 seconds to play. I do not see that Many Faces wants
an hour.

Cheers,
David


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