Apparently, I was a bit mistaken in the last estimate. I had to iron a few
bugs, both from Freecell Solver and from the script that generated the
solving preset. (and the latter still gives me a lower than run-time
approximation, which should be impossible, so there's still some work to
do)
In any case, I was able to find a preset that solves the Microsoft 32,000
in slighly less than 19,000,000 iterations. Here it is, for your
reference:
freecell-solver-range-parallel-solve 1 32000 1 \
--method soft-dfs -to 0123456789 -step 500 --st-name 1 -nst \
--method soft-dfs -to 0123467 -step 500 --st-name 2 -nst \
--method random-dfs -seed 2 -to 0[01][23456789] -step 500 --st-name 3 -nst \
--method random-dfs -seed 1 -to 0[0123456789] -step 500 --st-name 4 -nst \
--method random-dfs -seed 3 -to 0[01][23467] -step 500 --st-name 5 -nst \
--method random-dfs -seed 4 -to 0[0123467] -step 500 --st-name 9 -nst \
--method random-dfs -to [01][23456789] -seed 8 -step 500 --st-name 10 -nst \
--method random-dfs -to [01][23456789] -seed 268 -step 500 --st-name 12 \
--prelude "350_at_2,350_at_5,350_at_9,350_at_12,350_at_2,350_at_10,350_at_3,350_at_9,350_at_5,350_at_4,350_at_2,350_at_5,350_at_10,350_at_3,350_at_2,350_at_4,350_at_5,350_at_2,700_at_5,350_at_12,1050_at_9,350_at_10,350_at_2,350_at_10,1050_at_5,350_at_10,350_at_12,700_at_2,700_at_1,700_at_4,700_at_12,1400_at_2,700_at_9,350_at_10,700_at_3,700_at_4,700_at_2,5250_at_5,1050_at_10,1750_at_3,1400_at_1,1400_at_10,5600_at_1,4900_at_12,23450_at_2"
Takes 18,923,588 iterations and using Compact States runs for 35 minutes
on a Pentium III machine (making it about 15 boards per second).
You can find the code and the data that generated this preset here:
http://cvs.berlios.de/cgi-bin/cvsweb.cgi/fc-solve/presets/soft-threads/meta-moves/auto-gen/?cvsroot=fc-solve
If you have some very good iron, and can time some of the atomic scans
(run "perl time-scan.pl [ FCS args])") and send me the new scans.txt and
the files that were added to the data directory, I would appreciate it.
The script analyze.pl can be used to loosely determine which boards are
the bottlenecks of the meta-scan, and should be found an atomic scan that
solves them well.
Regards,
Shlomi Fish
----------------------------------------------------------------------
Shlomi Fish shlomif_at_vipe.technion.ac.il
Home Page:
http://t2.technion.ac.il/~shlomif/
Home E-mail: shlomif_at_iglu.org.il
He who re-invents the wheel, understands much better how a wheel works.
Received on Mon Jun 17 2002 - 20:35:58 IDT