Hi Bill,
Was very surprised to hear from you, and also delighted. That you are
still working on Freecell solving.
I think some of your concerns are not really relevant to my particular
interests, although I'm not sure about Shlomi's.
I'll go through your message and reply in sequence:
<<First, what does it mean to incorporate Fish's solver into FcPro? Give the
player a choice of which solver to use? Merge the best features of each into
a third solver? FCS may employ different moves than FcPro. Whose will you
use?>>
For the initial implementation, I'd just add it as an entirely separate
solver, and add another item to the selection menu for "Solver2". I'd have
no thoughts of designing a third solver based on both. As to "different
moves", see below.
What I most want a second solver for is to deal with the false
impossible problem. By far the most frequent usage that is made of the
FcPro solver is by players who want to know whether the game is solvable
from a given intermediate position. I've been getting occasional reports
from players who've identified false impossibles from such intermediate
positions. Meaning that the solver just isn't trustworthy. If it says
"Impossible", that only means "Probably impossible". Once I work out the
basic implementation of a second solver, I'd put in an automatic feature so
that any request for a solution "result only" will, if the result is
"Impossible", ask for a "second opinion" from the alternate solver. The
assumption here is that the probabillity of two solvers giving a false
impossible on the same hand will be infinitisemally low.
Re sequence moves and automatic moves. I see these questions as
relevant only if one is searching for the shortest possible solution. I'm
not really very interested in that pursuit, myself. I'd certainly agree
that if one is after such an objective, these questions are relevant, and
some rule needs to be adopted. But all I really care about for my own
purposes is a reliable and accurate "go no-go" declaration. I'm interested
in speed, also, but only to the extent of not very often requiring the user
to wait very long for his answer when testing a given position. The mass
solving of ranges of hands is something that is useful for general research
purposes, but I have yet to hear from any of our users who has actually made
use of that feature. So I'm really focussing on what the "marketplace"
seems to want, right now.
<<At the present time a competition/comparison could work if we restricted
ourselves to single-card moves only, and no auto-moves.>>
Even though I've said I couldn't care less, I think there's something to
be said for adopting the rules followed by M/S Freecell, even though we know
they are deficient. The reason for this is so that any solution is then
expressed in a language that a player can use if he wants to recreate the
solution by playing it out on M/S Freecell. It is for this reason that the
solutions we output from our solver are expressed under those rules. Even
though our solutions frequently are extremely long and sometimes include a
ridiculous number of superfluous moves.
Best regards, ------------------Adrian
Received on Sat Nov 17 2001 - 16:08:40 IST