Bill, I meant to comment on this point, and neglected to:
[AE]
> > ------------------------------------------------
> > I'd say there are three distinct possible goals for a Freecell Solver
> > design:
> >
> > (1) To find as many solutions as possible, regardless of size, to as
many
> > boards as possible in the least time.
> >
> > (2) To find as many solutions as short as possible, to as many boards
as
> > possible in the least time.
> >
> > (3) To deliver a completely trustworthy verdict as to the solvabililty
of
> as many boards as possible in the least time.
[BR]
> I don't think you've defined these goals carefully. To be specific, there
is
> no useful difference between your (1) and (3).
I think you may not grasp the essence of (1). This was an attempt to
define what I analyzed to be one of Shlomi's goals, and I believe Shlomi
concurred that it was an accurate description. The goal being to generate
the largest possible volume of freecell solutions in the shortest possible
time, ignoring all deals that take too much time too solve, and not being
concerned whether such deals are "Impossible" or "Intractable", and also
with no concern for the number of moves in the solution. I. e., no attempt
whatsoever to deliver a trustworthy verdict. (3) emphasizes the delivery of
a trustworthy verdict.
<<<< Let me suggest these:
> 1. Given a range of deal numbers (32 bits for now) and the number of
> freecells to be used, print the numbers: (of the unsolvable deals for 2-4
> freecells, of the solvable deals for 0-1 freecells). Same as your (3).
> 2. Given a legal freecell position, give a solution or pronounce
"unsolvable."
> 3. Find a shortest solution: (from a given legal position or from a deal
> number and number of freecells).>>
1. and 2. here, do not seem to me to define "goals" but rather describe
two modes of operation for a solver which has the same goal -- to provide a
trustworthy verdict -- and presumably in the shortest possible time. 1. is
test mode to evaluate speed, and 2. is a mode which allows input of an
arbitrary position as opposed to operating on a suite of random positions
generated by a dealing algorithm. Also, these criteria do not mention the
speed factor. 3. would need, I'd think some sort of speed criteria, such as
"Find the shortest solution you can within one minute of operation, using no
more than x megs. of virtual memory, on a standardized machine of
such-and-such design.
For the purpose of designing a "competition", it would be valid to
define the conditions as performance over a given small sample of standard
positions for some of the goals, but to evaluate a solver's lack of false
impossibles is more difficult, requiring testing over a very large sample,
including not only original positions but also intermediate positions
reached through human play.
Best regards, ---------------Adrian
Received on Tue Dec 11 2001 - 04:33:45 IST