In a message dated 11/19/01 6:57:18 PM Pacific Standard Time,
aettlinger_at_worldnet.att.net writes:
> I do have the feeling that Freecell can be very diabolical in how it
can
> treat solver software. I. e., for "best-first" searching, some deals are
> going to show up for which, whatever choosing rules you adopt, the "best"
> is going to turn out to be "worst", and the solver will run on and on and
> on.
Mine has that problem with just one deal so far; one of the zero-freecell
deals cannot be solved quickly with any of my current approaches. It does get
solved, but needs much more time than other games. I'll need a new techique
for that one. Not a problem - it's near the bottom of my list.
> My hunch is that in the final analysis, an approach is going to be
> needed in which, when a given set of rules results in the solver bogging
> down, you give up and try it again with a different set of rules.
That's a sure thing.
> We do something like that in FcPro in a primitive manner, trying out
> various permutations of column order searching for one that makes it work
> within a given limit of time or number of iterations.
And Fish alters the order of his best-first weights (or allows the user to do
so), while mine keeps weights constant but blocks other factors.
All of this will be obsolete when we can use the starting position as an
index into an array of solutions -- all 10^55 (Pascal) or 10**55 (Fortran) of
them.
Bill Raymond
Received on Mon Nov 19 2001 - 20:23:38 IST