Loading ...
Sorry, an error occurred while loading the content.

1378Performance of a new Freecell Solver preset for yielding short solutions

Expand Messages
  • Shlomi Fish
    Jan 12, 2014
    • 0 Attachment
      Hi all,

      Recently I've been playing with some scans to yield short solutions to
      Freecell deals I played in PySolFC. Some days ago I decided to put the
      qualified-seed-improved and a modified flares-based meta-scan of that sort with
      a new scan I found:

      -nf --flare-name let_roll --method a-star -asw "1" -to 01234567 -opt \
      -opt-to 0123456789ABCDE

      The older scan is called qualified-seed-improved (or qsi for short) and the new
      one was temporarily nicknamed "qualified-foo". The results of running these
      scans (without the timings and the solution) can be found here:


      (short URL : http://is.gd/IwLgRv )

      They were run with -fif 5 and --flares-choice fc_solve:

      Here is a comparison of their summaries:

      == qsi-flares-choice-fc_solve.fc-pro.summary ==

      Num FCS Moves
      Min: 64
      Max: 206
      Average: 92.5560486265196
      StdDev: 8.23845588364974
      Median: 92

      Num FCPro Moves
      Min: 23
      Max: 175
      Average: 56.8767773992937
      StdDev: 9.23174141106441
      Median: 57

      == qualified-foo-flares-choice-fc_solve.summary ==

      Num FCS Moves
      Min: 64
      Max: 120
      Average: 90.5281727553986
      StdDev: 7.20646458597401
      Median: 90

      Num FCPro Moves
      Min: 24
      Max: 89
      Average: 55.8650895340479
      StdDev: 8.63341557417936
      Median: 56


      As one can see, for both the FCS ( = fc-solve) and the FC-Pro moves, the
      maximum, average, standard deviation, and median were reduced. The minimum
      number of moves was unaffected by "FCS Moves" and grew slightly for the "FCPro"
      moves (which is not unheard of, due to the flares-choice).

      I also noticed that using "-sp r:tf" (set pruning - to-foundations)
      badly affected the solution length, at least with the new scan I tried.

      In any case, look at the following commits for my work about it:

      commit 6710bd5392c7029706647dc83ba272df2dccd6a6
      Author: Shlomi Fish <shlomif@...>
      Date: Mon Jan 6 20:02:35 2014 +0200

      Prepare summaries.

      commit aa9fe8c75d0745f4458a70b02d8494f04bd33436
      Author: Shlomi Fish <shlomif@...>
      Date: Mon Jan 6 19:55:34 2014 +0200

      The data files from running the two scans.

      qsi-flares is the old, qualified-foo is the new one.

      commit e8dce5c7d325783b82e9b3abb4e580ebb37ab8ea
      Author: Shlomi Fish <shlomif@...>
      Date: Mon Jan 6 19:43:59 2014 +0200

      Add the qualified-foo (temp name) preset.

      This was span off from qualified-seed.sh and should be used with the
      -fif 5 and --flares-choice fc_solve . Testing it revealed that it
      yielded improvements.

      commit bee68db81a31f62bc13d614fec519bdbb6d81791
      Author: Shlomi Fish <shlomif@...>
      Date: Mon Jan 6 19:40:23 2014 +0200

      Add GNU parallel scripts to run the scans.

      Of particular interest is gnu-par-wrapper.bash which provides a script
      to run a sequence of command line arguments with a prefix and an index.

      Also of particular interest is
      gnu-parallelize-measure-flares-scan__4__GENERIC.bash which abstracts the
      common parts of the scans and is used by them.


      Best regards,

      Shlomi Fish

      Shlomi Fish http://www.shlomifish.org/
      Understand what Open Source is - http://shlom.in/oss-fs

      Officer: Captain, there’s a new star system a few thousands of lightyears
      away, and its sun has an irregular shape.

      Please reply to list if it's a mailing list post - http://shlom.in/reply .