Hi Danny,
On Sat, 01 Sep 2012 17:53:47 -0000
"dannyjones183" <dannyjones183_at_yahoo.com> wrote:
>
>
> An impressive number of 8x2 deals examined -- considering the
> execution time for my solver on the first 32,000 deals.
>
See below where I explain how they were all solved.
> My solver found 6,619 unsolvables in the first 32,000 deals, but 13
> of them were Intractable.
Yes, all the deals that were found to be intractable by your solver in the first
32,000 deals turned out to be impossible after exhaustive scans.
> Of couse, I'm running on an Intel P-4
> computer with 1.25GB of RAM.
Yes, it seems underpowered compared to what I now have (an x86-64
Core i3 computer with 8 GB of RAM). And, naturally, I deployed my
solvers on three x86-64 HPC (= High Performance Computing) machines
with 64 GB, 128 GB and then 512 GB of RAM.
> Of the solveable deals, the fewest
> number of moves used was 23, the most number of moves used was 84,
> and the median number of moves used was 52.
>
OK.
Here is roughly my solving strategy:
1. At first I covered the deals using a fast meta-moves and meta-scan preset, which may
yield some false negatives (= deals that are reported as unsolvable while actually can be
solved). This yielded many deals that were shown to solve.
2. Then I passed the remaining unsolvable deals locally using some atomic move presets -
either those of fc-solve or later the dbm_fc_solver. These guarantee an accurate verdict,
but sometimes report the deal as intractable. If that's not the case, then I have an accurate
verdict for it.
3. The remaining deals were iterated using a script that ran a Random-DFS scan with a certain
limit with incremented numeric seeds on the deals one after the other. Once a deal was solved,
we mark it as solved and record its solved seed, and only check the intractable/unsolvable
deals with the remaining seeds.
This allows us to find solutions for some of the remaining deals.
4. The still intractable deals were deployed using the dbm_fc_solver or the depth_dbm_fc_solver
on the HPC machines, which either yielded the final verdict, or in the case of the remaining
intractable deals, did not yield a verdict and the deal remained intractable.
Hope it helps. I hope to make some of the scripts used in this range solving
framework available in the fc-solve repository (without revealing the
identities of some of the people who gave me access to the HPC
machines and chosen to remain anonymous), soon.
Regards,
Shlomi Fish
> --- In fc-solve-discuss_at_yahoogroups.com, Shlomi Fish <shlomif_at_...>
> wrote:
> >
> > I ran my solvers on the first 400,000 Windows Freecell deals
> > with only two available freecells to see how many of them can be
> > solved.
> >
> > Here is the report:
> >
> > [REPORT]
> >
> > 1 - 32,000:
> > -----------
> >
> > Solved: 25,381 ; Impossible: 6,619 ; Intractable: 0 .
> >
>
>
>
--
-----------------------------------------------------------------
Shlomi Fish http://www.shlomifish.org/
Escape from GNU Autohell - http://www.shlomifish.org/open-source/anti/autohell/
Doing linear scans over an associative array is like trying to club someone to
death with a loaded Uzi. — Larry Wall
Please reply to list if it's a mailing list post - http://shlom.in/reply .
Received on Sat Sep 01 2012 - 12:42:25 IDT