I retrieved and read the "paper" to which you refer. I'd read it
before and it probably contributed to my use of the term meta-move.
Too bad they went to all that hard work and their logic for ending a
search path had such a terrible flaw -- including the number of cards
remaining in their (expected) count of moves needed to complete the
path. They ended up aborting move paths where one more move might have
produced a cascade that cleared the board. Because of this, they
appear to have seldom come close to an optimal solution.
I've run 2653 "difficult" deals and never needed more than 78 atomic
moves to find a solution -- with a median solution of 50 moves. My
definition of difficult is: deals that PatSolve -E -M64 found
intractable or required 69 atomic moves or more to solve -- excluding
auto-moves to the home cells. Of course, PatSolve found its solutions
in 1/8-th the time my solver took, but the storage required was
approximately the same and my solver never encountered an intractable
state. Of the 2653 deals, PatSolve found 140 of them intractable even
when re-run with -E -M96.
In any event, I've dropped the phrase "meta-move" and am now using
"sequence of moves" and "multi-card moves", instead.
Question: Do you know why sequences of moves aren't supported by MS
FreeCell and FcPro if there isn't at least one free cell open?
It seems to me that using multiple empty columns, without any free
cells open, would still be viable -- but not to these solvers!
--- In fc-solve-discuss_at_yahoogroups.com, Shlomi Fish <shlomif_at_v...> wrote:
OK. Note that I borrowed the term meta-move from:
>
> http://kevin.atkinson.dhs.org/freecell/report/paper.pdf
>
> The report Kevin Atkinson and Shari Holstege prepared for their Freecell
> Solver project. I used to use the term "multi-move" for it, but
apparently
> "meta-move" is the standard term in game AI. (or so I hope).
>
> Regards,
>
> Shlomi Fish
Received on Tue Nov 18 2003 - 06:02:48 IST