I can't really tell you the true specifics, so if you're doing this
interested in the AI components, all I can really say is to read the
code.  I don't know how to read the code, as of now.

What I can do is clarify what exactly is going on in layman's terms. 
Quackle is different than Maven in that it is a much simpler and more
eloquent program: it has much fewer eccentricities so it is much
easier to follow what is going on.  It uses storage files to record
data so that it can get them out of tables, such as win% and
superleaves, rather than calculating them all out at the time.  

(This reminds me that I should talk to Jason or olaugh about possibly
changing the win% algorithm to something more realistic, incorporating
probability distributions and standard deviations, but that is for
another time)

The simulation is Monte Carlo, just like Maven.  It works faster than
Maven because its evaluation heuristic is simpler, involving just
points, leave, opponents points, and opponents leave, all things that
are simple to calculate.  

As for estimation, *sigh*.  Estimation is done very crudely, in I
believe all computer programs, unless you're looking for something
more complicated.  I don't think Quackle does any real estimating, but
the program involves estimation as it makes the calculations more
simple.  Basically, if you study game theory, half of that field is
all about estimation.  I'd be happy to point you to plenty of sources
but that has nothing to do with computing and Quackle, but only the
theory of what the actual correct play is.  Computers cannot thing,
and as such they cannot do any estimation that is not in the code. 
Why would it bother when it can compute at record speeds?

If you mean how numbers are rounded, that's very simple computing and
is just done with a rounding off function.  I'm not sure what you're
talking about.  

As for move generation, umm, maybe someone else can help with that. 
It just takes the top scoring N plays according to points+leave and
posts them.  How it finds those plays I have very little idea, but I'm
sure that someone with computer experience can tell you.

There is at this moment no Quackle FAQ.  If I ever get around to it I
might write it, or I'll get to something like that, but I'd really
suggest someone else do it first.  It would only take roughly a day to
write.  I do think it would be helpful, especially since there are so
many people who do not understand how Quackle or Maven works in even
the most fundamental terms.

Kenji

--- In [email protected], "hirsch.horst" <hirsch.ho...@...> wrote:
>
> Hey there,
> 
> as I'm doing a presentation on Scrabble AI, I wanted to know if there
> is a Quackle documentation which describes the steps (Move generation,
> evaluation, simulation, estimation) in a more detailed way. 
> 
> I'm interested in the differences to Maven and the improvements.
> 
> Greetings
> Michael
>


Reply via email to