Forwarding.
Begin forwarded message:
Date: Fri, 27 Mar 2015 15:37:14 -0400
From: Tom Holroyd <tomh_at_kurage.nimh.nih.gov>
To: Shlomi Fish <shlomif_at_shlomifish.org>
Cc: Freecell Solving Discussions <fc-solve-discuss_at_yahoogroups.com>
Subject: Re: Patsolve appears to solve the same initial Freecell layout
differently based on the order of the columns
On Sat, 21 Feb 2015 15:34:45 +0200
Shlomi Fish <shlomif_at_shlomifish.org> wrote:
> One question to Tom Holroyd (= the original author of patsolve) is
> whether he used the collection of Microsoft Freecell boards as they
> are when providing input for the algorithm that yielded the Xparams
> and Yparams.
Hi. Sorry for the lateness of my reply. Yeah, patsolve doesn't
change the order of the stacks. The dataset used to train the
parameters was static---I don't remember all the details, but the
boards would have been the same, permutations of the stacks were not
considered, as far as I can recall. I'm not even sure how one would
implement that, though on modern hardware I suppose you could have
different threads solving differently permuted boards, but there are
too many permutations. You could also throw a random element in there,
just permute them differently each time and retrain. No idea how that
would work. It can do "depth-first" instead of "breadth-first", and
picking a random child-node would be easy to try. I haven't looked at
the code in years, maybe it does that.
I've been more busy with "real" work these days ...
--
Dr. Tom
--
"There are not more than five musical notes,
yet the combinations of these five give rise to
more melodies than can ever be heard." -- Sun Tzu
--
-----------------------------------------------------------------
Shlomi Fish http://www.shlomifish.org/
http://www.shlomifish.org/humour/bits/facts/Summer-Glau/
Larry Wall dreams in Perl.
Please reply to list if it's a mailing list post - http://shlom.in/reply .
Received on Fri Mar 27 2015 - 23:07:56 IDT