The New Sudoku Players' Forum

[creint]

Difficulty should be harder than a normal sudoku at the same rating. Needing ~25 layers of forcing nets.

Adding hidden constraints makes is easier.

Code: Select all

.2...............9......7..................8.1............5.......4..........6...
AAAAABBBBACCDAAAEBCCDDFFFEBCDDFFGGEBCDFFHGEEBCDFHHGEIBCDFHGGEIICDHHGIEEIHHHGGIIII ED=10.9/8.2/2.6 becomes: < SE 10

And without adding hidden constraints, but has higher SE rating:

Code: Select all

........91.............6......4....................7......5...........8...3...... 111112222133411152334466652344667752346687552346887592346877599348879559888779999 ED=11.1/1.2/1.2 but should be < SE 10

This one seems somewhat the right rating:

Code: Select all

..3..............9....5..........7...2.......1...........4...................6... 111112222133411152334466652344667752346687552346887592346877599348879559888779999 ED=11.2/8.0/2.6

Do you have a newer version of Jigsaw Explainer 1.2.1? It took very long to rate. Why does SukakuExplainer not support Jigsaw?

Possible performance improvements:
If you only want hint rating and not easiest methods first, you could group by rating and apply that as a single step. If rating is already high you could also apply everything that is below that rating. Now its only taking the easiest chain first.

[Mathimagics]

.
Review of 8C/9C search status

For each of Bob Harris's original layouts, the table shows :

• NTV = # of transversals

• NSG = # of solution grids

• 9CP = # of 9-clue puzzles found

• 8CP = # of 8-clue puzzles found

Note that the puzzle counts reflect only a specific subset of puzzles tested, so there could well be more. No vicinity search has been done, I simply tested the 9! possible 9-clue puzzles that can be formed by a row/column transversal, with distinct givens.

Code: Select all

   JL       NTV          NSG        9CP      8CP
   ---------------------------------------------
   BB1     5120   1710356560          2        0
   BB2     3760  46770152640         25        1

   BB3     2706       112072       3113      672
   BB4     2417       493179        974       25


Intriguingly, both BB1 and BB2 have very high solution grid counts, but still yield 9-clue puzzles, and even one 8-clue puzzle.

BB3 and BB4 have low grid counts, and are rich in 9C/8C puzzles.

I have found 12 (symmetric) TJL's in my catalog that look to offer the most likely source of 9C/8C puzzles. These JL's all have grid counts less than 500,000 like BB3 and BB4. These 12 I have labelled TJL-A through TVL-L.

Some of those have very low grid counts (less than 50,000). The lowest count is just 3146 (for the JL shown below) which surely must be a record low?

The search is underway for these 12 JL's, and the early signs are that many 9C/8C puzzles are to be found in these JL's.

Code: Select all

AAAAAABBBCCCCCAAABDDDECCCCBFDEEGGGBBFDDEEEGGBFFDDDEEGBFHHHHEGGGFIIIHHHHHFFFIIIIII # TJL-B (nsg = 3146)


TJL-B-visual: Show

Code: Select all

+---+---+---+---+---+---+---+---+---+
|                       |           |
+---+---+---+---+---+   +---+---+   +
|                   |           |   |
+---+---+---+---+   +---+---+---+   +
|           |   |               |   |
+---+   +---+   +---+---+---+---+   +
|   |   |       |           |       |
+   +   +---+   +---+---+   +---+   +
|   |       |           |       |   |
+   +---+   +---+---+   +---+   +   +
|       |           |       |   |   |
+   +---+---+---+---+   +---+   +---+
|   |               |   |           |
+   +---+---+---+   +---+---+---+---+
|   |           |                   |
+   +---+---+   +---+---+---+---+---+
|           |                       |
+---+---+---+---+---+---+---+---+---+


[Mathimagics]

.
The 12 targeted TJL's proved to be a goldmine for 8-clue puzzles, with well over 15,000 found:

Code: Select all

   JL       NTV          NSG        9CP      8CP
  ----------------------------------------------
  TJL-A    2479       249592       1637       22
  TJL-B    2916         3146       2127     2921
  TJL-C    2917        11048      14133     5348
  TJL-D    2190        25318       8828     2009
  TJL-E    2137        42750      11994     1610     
  TJL-F    2688        66118       8152     2045
  TJL-G    2088        74006       7337      517     
  TJL-H    3691       148433        922      234
  TJL-I    3816       453333       1432      110
  TJL-J    3928       123289       1075      182
  TJL-K    4324       239545       3853      309
  TJL-L    4729       224145       1338      590


TJL-C provided the most (5348), a small sample is shown here:

Code: Select all

AAAAABBBBCCCDAAAEBCFCDDAEEBFFCCDGEBBHFFCDGEEBHHFCDGGEEHFFIDDGEGHFIIIDGGGHHHHIIIII
1.........2.........3..........5.......4..........6..........8................7..
..........2.........3...........6......4.............9....5..........7.........8.
1.........2.........3.......................9.......8.......7.....4.........5....
1.........2.........3...........6...........9...............7.....4.........5....
1...................3.............8....4..........6.......5..........7..........9
1.........2.........3.............8....4..........6..................7..........9
1.........2.........3.............8.........9....5.......4...................6...
..........2.........3..............9...4..........6.......5...........8.......7..
1...................3..............9...4..........6..........8.......7......5....
1.........2.........3..............9...4............8......6................5....


The visual layout:

TJL-C: Show

Code: Select all

+---+---+---+---+---+---+---+---+---+
|                   |               |
+---+---+---+---+   +---+---+---+   +
|           |   |           |   |   |
+   +---+   +   +---+   +---+   +   +
|   |   |   |       |   |       |   |
+---+   +   +---+   +---+   +---+   +
|       |       |   |   |   |       |
+---+   +---+   +   +   +   +---+   +
|   |       |   |   |   |       |   |
+   +---+   +   +   +   +---+   +---+
|       |   |   |   |       |       |
+   +---+   +---+   +---+   +   +---+
|   |       |   |       |   |   |   |
+   +   +---+   +---+   +   +---+   +
|   |   |           |   |           |
+   +---+---+---+   +---+---+---+---+
|               |                   |
+---+---+---+---+---+---+---+---+---+


[Mathimagics]

.
Thanks again to all who gave ratings for some of the 8-clue puzzles.

I now need to test many hundreds, maybe thousands, of puzzles.

Are you all using the JE 1.2.1 rater?

Can we arrange to deliver me a copy?

[creint]

I now need to test many hundreds, maybe thousands, of puzzles.

For what do you want to test?

[Mathimagics]

For various sets of 8-clue puzzles that I can find, I want to rank them by expected difficulty.

For example, urhegyi tested a bunch (above) and they were all high SE rated.

I think I can find sets of 8-clue puzzles that might prove to be easier ones.

The SE rating gives me a good idea of probable relative difficulty.

I don't need any info on techniques required, just a fairly reliable predictor of relative difficulty.

[1to9only]

From: http://forum.enjoysudoku.com/double-diagonal-x-sudoku-t39246.html#p307421

Edit: I've stopped updating SudokuJigsawExplainer, and it is no longer available for download. The last version was on 7th April,

Hopefully, a kind person who previously downloaded the 2021.4.7 version has sent you a copy.

Else I can describe a SukakuExplainer kludge to rate a jigsaw as a custom sudoku latin square with 9 extra regions!

The hardest (ED=12+) jigsaw sudokus can take a very very long time to rate by JE/SE. A faster solution would be to hack into PGExplainer.

[creint]

Edit: I've stopped updating SudokuJigsawExplainer, and it is no longer available for download. The last version was on 7th April,

Can I get that version too?

14-30 seconds, single thread, faster than JigsawExplainer.

Code: Select all

.2...............9......7..................8.1............5.......4..........6...
AAAAABBBBACCDAAAEBCCDDFFFEBCDDFFGGEBCDFFHGEEBCDFHHGEIBCDFHGGEIICDHHGIEEIHHHGGIIII

[Mathimagics]

Hopefully, a kind person who previously downloaded the 2021.4.7 version has sent you a copy.

Such a person, should they be willing to help, would send me a PM to obtain my email address ...

I do have two DFS solvers (one based on Mladen's FSSS2, and another a DLX solver, both functionally equivalent), and I can apply search depth limiting to crudely filter the puzzle sets ... both of these have serious deficiencies, however, when it comes to low-clue puzzles.

I like the idea of hacking into PGExplainer, and so I will petition the author about that!

[1to9only]

https://wormhole.app/nb514#RcS-5VHryeP9P1aaY3RFDA

SudokuJigsawExplainer.jar
Size: 1,037,021 bytes
MD5: b796a503d0e8c567a8ec03f471a1949a

It is only available for the next 24 hrs.

I no longer have the source code for the 2021.4.7 version, meaning no support is available!
Edit: To clarify: I keep tweaking the code to align it with my other Explainers, but I haven't used the program much lately.

Usage: http://forum.enjoysudoku.com/hardest-sudoku-jigsaws-js-ed-11-x-t38196.html#p299179

Edit: The 2021.4.7 supports 1-9 jigsaw layouts, there is some (untested) code in places to translate A-I to 1-9, but I mostly use 1-9 jigsaw layouts.

[Mathimagics]

Many thanks, we got there in the end.

It is indeed curious that the file as jpeg in a zip wrapper works, but the same trick can't be used for an EXE.

So there is still a need for my binary/text encoder tool, which I see has been downloaded over 70 times ... it's my most successful software app ever

[1to9only]

File types you can't include as attachments
To protect your account, Gmail doesn't allow you to attach certain types of files. Gmail often updates the types of files not allowed to keep up with harmful software that is constantly changing.
File types blocked by Gmail are:
.ade, .adp, .apk, .appx, .appxbundle, .bat, .cab, .chm, .cmd, .com, .cpl, .dll, .dmg, .ex, .ex_, .exe, .hta, .ins, .isp, .iso, .jar, .js, .jse, .lib, .lnk, .mde, .msc, .msi, .msix, .msixbundle, .msp, .mst, .nsh, .pif, .ps1, .scr, .sct, .shb, .sys, .vb, .vbe, .vbs, .vxd, .wsc, .wsf, .wsh

I think gmail inspects the zip contents and only checks it doesn't contain any blocked extensions. So renaming an exe, zipping and sending may work.

I suspect sending an exe masquarading as a jpeg as an attachment (without zipping) might still fail if gmail then inspects the file contents (checks headers) to verify file type. I've not checked that this is true or not!

[Mathimagics]

So renaming an exe, zipping and sending may work.

.

No, that's just what I was trying to do.

I think of the exclusion list as simply a convenient way to tell gmail that you are a file that should be blocked!

A file called "fred" definitely has to be inspected. Headers at least - windows exe's are notoriously easy to identify, so no matter what size they are, they are impossible to smuggle through security. So exe's are very hard to smuggle, but jar's seem to be easier.

[Mathimagics]

I no longer have the source code for the 2021.4.7 version, meaning no support is available!

Edit: To clarify: I keep tweaking the code to align it with my other Explainers, but I haven't used the program much lately.

Hmmmm ... then how do you tweak the code if you no longer have it?

A mystery, perhaps, but anyway, thank you very much for supplying a working JE. And thanks also to creint, who has kindly given me a version of his solver with a command-line rating option. (I will call these rating tools SJE and CJE respectively.)

So I can now process a set of 8-clue puzzles for a given JL by:

• using CJE to rate them by crein]'s scale of (0 to 6). See below for rating list.

• selecting any cases with rating < 5, then getting the SE rating using SJE

There is a strong correlation between the puzzles which I want to find (ie not TOO difficult), and how long SJE takes to rate them. Anything with an SE rating of 11+ takes a very long time (5 to 15 minutes). This java app is multi-threaded, tso these puzzles use all of my CPU resources (4-core x 2-threads) for most of that time.

By contrast, these 11+ puzzles take at most 5 seconds for CJE, and since CJE is single-threaded, only 16% CPU usage.

Cheers
MM

CJE ratings: Show

Code: Select all

0 = hidden singles
1 = naked singles
2 = locked singles
2 = sets
3 = fish
4 = easy chains
5 = hard chains
6 = chains on chains (~SE > 10)
? = unsolved


[1to9only]

I no longer have the source code for the 2021.4.7 version, meaning no support is available!

Edit: To clarify: I keep tweaking the code to align it with my other Explainers, but I haven't used the program much lately.

Hmmmm ... then how do you tweak the code if you no longer have it?

A mystery, perhaps, but anyway, thank you very much for supplying a working JE. And thanks also to creint, who has kindly given me a version of his solver with a command-line rating option. (I will call these rating tools SJE and CJE respectively.)

I have the code in 1 folder which is not version controlled, so I no longer have the source code to build the specific version 2021.4.7 that is out there, therefore no support/fixes are possible!

I still hack the code now and then, there is no log of changes, it builds cleanly, it runs and brings up the GUI, I do minimal testing. That's its current state...