As a player somewhere near 1600, it seems to me some words come to mind
if the boards and the rack carry the right 'suggestion' in some way.
I hope you can
imagine what I mean by that - a stupid example might be if FOX plays
for 72 and the
other four tiles on my rack are EIST, where possibly I could miss the
bingo if the
board didn't lead me to notice the beginning of the word ...

So trying to get Quackle to play like a 1600 human realistically will
be very tricky.

Trying to build a perfect rating system, I've thought it could be
extremely helpful
to have a machine who could play human-like at different strengths.
But so far the large dimensionality
of the problem has me pretty baffled.

(The crux of a technical problem in ratings is: If player A generally
beats B at rate Pab,
and B beats C at Pbc, what rate will A beat C at?  Generally we answer by saying
the logistic curve.)

One hope I have is that Quackle random errors can be generated so they
have the same
effect as a Human's random error from combination of human error sources

I can say that, but the layers of complexity are too deep for me to
comprehend what it
means after I've said it :-)

Human errors might be classified somehow, and Quackle could be programmed
to try to mimic them.  Possible categories:
  Don't know a word at all
  Know a word, but only x % of the time (x is between 0 and 100).
   (This would take a lot of research to develop)
  Not very good at using hot spots yet
  Too afraid of setting up bingo hooks
  Too afraid of setting up hot spots on triples (both red & blue)
  Too fond of using 3 letter words like COW
  Noplays the opponent when more than 3 bingos are made
  Fails to notice the number of tiles in the bag

I think the list is both incomplete, yet too long now.







On Wed, Nov 5, 2008 at 2:09 PM,  <[EMAIL PROTECTED]> wrote:
> We can only guess the playing vocabulary of a top human player, so who
> knows.  We know that Quackle has a perfect vocabulary-hakim withstanding.
>
> So could we guess how much of advantage having a perfect vocabulary would be
> by simulating Quackle perfect against Quackle not perfect (QNP)?  Would it
> be possible to load a second dictionary file or have two instances of
> Quackle play each other?
>
> A simple scenario would have QNP have a limited vocabulary of all 3, 4, 5's
> and say 80% of the 6, 7, 8, 9's.  It would never challenge QP...
>
> Anybody up to this?
>
> Ron
>
>
>
> 

Reply via email to