Hi Danny,
On Wed, 07 Nov 2012 17:49:15 -0000
"dannyjones183" <dannyjones183_at_yahoo.com> wrote:
> An improvement is an improvement, but I feel that solving a deal must
> now take second place to providing a useful solution. A solution of 175
> moves would not be useful to me. Here's the results from my (fastest)
> solver. It's roughly 12 moves more than those typically returned by my
> (extremely slow) iterative search solver.
OK, there is a grain of truth in what you say. On the git master branch, I have
been investigating several ways to make the solutions of "-l mfi" shorter. At
first I collected statistics from one scan:
./freecell-solver-fc-pro-range-solve 1 32000 1 \
--total-iterations-limit 100000 \
--method random-dfs -to "[01]=rand()[23456789]=rand()" \
-dto "13,[0123456]=asw(1)" \
-sp r:tf -opt -opt-to 0123456789ABCDE
This solved the offending scan with 175 moves quickly and yielded a short
solution and improved on other scans.
Then I noticed that it can always yield an improvement (instead of sometimes
making things worse), if I measured the moves based on the FC-Pro count (with
Horne/Raymond play) instead of the FCS moves. So I decided to implement a
feature called "--flares-choice fcpro" that will choose the flare based on the
shortest FC-Pro solution. Some commits later (and one hard to investigate, but
trivial to actually fix, bug), it was working and now I measured the "-l mfi"
scan results with this "--flares choice fcpro":
[QUOTE]
perl /home/shlomif/progs/freecell/git/fc-solve/fc-solve/source/scripts/compare-two-fc-pro-range-solver-dumps-solution-lengths.pl
--from /home/shlomif/Arcs/fc-solve/micro-finance-improved.fc-pro-dump.txt
--to /home/shlomif/Arcs/fc-solve/micro-finance-improved--flares-choice-fcpro.fc-pro-dump.txt
--iters=1000000 --histogram
1999 1
1869 2
1663 3
1531 4
1278 5
1161 6
1043 7
906 8
751 9
649 10
493 11
390 12
331 13
286 14
205 15
176 16
118 17
97 18
69 19
61 20
42 21
32 22
21 23
17 24
13 25
5 26
2 27
2 30
[/QUOTE]
( First column is the count of states with the delta, and second column is the
delta itself. ).
In total, the FC-Pro solutions of 15,210 states have been reduced.
Then I measured the new scan against all that:
[QUOTE]
shlomif[fcs]:$trunk/fc-solve/B$
perl ../source/scripts/compare-two-fc-pro-range-solver-dumps-solution-lengths.pl
--from
~/Arcs/fc-solve/micro-finance-improved--flares-choice-fcpro.fc-pro-dump.txt
--to ~/Arcs/fc-solve/new_scan.dump.txt --iters=10000 --histogram
133 1
94 2
90 3
77 4
62 5
54 6
30 7
26 8
25 9
9 10
6 11
9 12
9 13
8 14
6 15
3 16
3 18
2 20
2 23
1 94
[/QUOTE]
I was interested in the 94 increment, but there are other good improvements
there.
The stdev of "-l mfi --flares-choice fcpro is now":
<Q>
n is 31999, min is 23, max is 175
mode is 55 _at_ 1670, median is 50, mean is 54.0131, stdev is 8.05612
</Q>
So there are no improvements to the min, max and median, but the mode is two
steps lower and (_at_ 1670 instead of _at_ 1476) and the mean is almost 3 steps lower
and the stdev is more than 1 step lower.
I am going continue to investigate new scans for reducing the solution lengths
and will report further here.
Regards,
Shlomi Fish
--
-----------------------------------------------------------------
Shlomi Fish http://www.shlomifish.org/
What Makes Software Apps High Quality - http://shlom.in/sw-quality
We don’t know his cellphone number, and even if we did, we would tell you that
we didn’t know it.
Please reply to list if it's a mailing list post - http://shlom.in/reply .
Received on Sun Nov 11 2012 - 02:03:31 IST