John, Dallas...
I had a similar situation last March, while simming a 37-pt move
against a 16-pt move with a close valuation. The preferred move
seemed to depend on which "yellow" tiles were left on the board
("yellow" = suggested, but not played).
I still have the screen-shots and .gcg file, if there's somewhere I
could email them...
Zev
--- In [email protected], "John O'Laughlin" <[EMAIL PROTECTED]>
wrote:
>
> Dallas,
>
> I think the rack ordering thing is probably a red herring, but there
> might still be something to this. Quackle's random numbers might be
> embarrassing cyclic in a way that might cause results to repeat like
> this (but since the CP uses wall clock time, it's not so easy to
> recreate). I'll try to look into it soon.
>
> I don't really recommend using the Championship Player for things
like
> this. It's kind of a big hack that we threw together at the last
> minute for the Toronto Open, and unfortunately I haven't had time to
> improve it. It's nice that it gives you a way to play against
> something that sort of resembles a human expert player, and right
now
> it's the only way to run the endgame solver, but it's got a lot of
> problems. I don't use it to go through games, which is probably why
> its bugs haven't bothered me enough to fix them right away. I just
> generate choices, prune the list by hand and sim, pretty much like I
> used to use Maven.
>
> John
>
> On Fri, May 23, 2008 at 4:56 PM, John O'Laughlin <[EMAIL PROTECTED]>
wrote:
> >> It appears that the order of letters on the rack, at least in
this isolated
> >> case, is impacting the sim (as well as win percentages and
valuation). Does
> >> anyone know why that would be happening?
> >>
> >> FYI, I am running Quackle 0.95 in a Windows XP environment, in
case that
> >> matters.
> >
> > Dallas,
> >
> > I can't think see reason why this would happen. The sim results
should
> > be slightly different every time, and the order of tiles on the
rack
> > shouldn't have any meaning. Are the results really exactly the
same
> > for every AAGIRSS run or just similar, and could this be a case of
> > seeing a pattern in a too-small sample size? This definitely
doesn't
> > happen for me on Linux (I can try it on Mac and Windows when I get
> > home).
> >
> > John
> >
>
