Hi all!
TL;DR: I recently improved the performance of the fc-solve's MS 32K benchmark,
from about 5.52s down to 4.15s, after following Bill Raymond's excellent advice
in this post:
https://groups.yahoo.com/neo/groups/fc-solve-discuss/conversations/messages/584
(short URL:
https://is.gd/Lsmnll )
and implementing a move-card-from-freecells-to-empty-stacks-and-immediately-put
-a-child-card-above-it move.
----------
I recall reading that post in the past, so I can refer to the speed stats of
Bill Raymond's solver there, but I did not register the rest of the advice in
the message until a few days ago. Then I tried implementing a new move (titled
'j') for this and after some coding and debugging had a working version.
Trying to do:
./freecell-solver-range-parallel-solve 1 32000 100 -mi 100000 -to 0123456j89 |
tee new.dump
vs.:
./freecell-solver-range-parallel-solve 1 32000 100 -mi 100000 -to 0123456789 |
tee old.dump
Indicated that in the new version there were much fewer intractable deals (but
also some new ones), and the total iterations count decreased by a factor of 4
, while the speed increased by a factor of 4.
So with this in mind, I created the obf-mod4.sh preset file which was based on
obf-mod3.sh (which I wrote about in the previous post) only replacing all '7'
moves with 'j'. This time the speedup was less dramatic but still
significant: from 5.52seconds down to 4.20seconds (and with no introduced
intractables).
Some further tweaking reduced it to about 4.15s.
I think there is more room for improvement there in Raymond's advice that
"I thrive on these little details. My solver is fast (~10000000 deals per hour
_at_733mHz) not because of any sophisticated algorithm but because I've
experimented endlessly with every tiny variation and the best order to apply
them. " and it would better be investigated further.
Regards,
Shlomi Fish
--
-----------------------------------------------------------------
Shlomi Fish
If you wish to live wisely, ignore sayings — including this one.
— Unknown Source (via fortune-mod).
Received on Tue Sep 12 2017 - 21:42:33 IDT