The New Sudoku Players' Forum

[mith]

As mentioned elsewhere, I have finally got some searches back up and running, now focused only on T&E depth. I may start a new thread for tracking these puzzles.

Anyway, while the generating is only based on whether they have depth 3, I am running SE on the expanded forms, and after finding a bunch of 11.7s I finally got a couple 11.8s. Here are their minimal forms:

Code: Select all

........1.......23....45.....1.54....4671.8..75.8.61...67......18.......5.4.78..6  ED=11.8/1.2/1.2
........1.......23....45.....1.54....467.18..75.86.1...67......18.......5.4.87..6  ED=11.8/1.2/1.2

The total count for expanded forms not in T&E(2) is now 1808 + any from hendrik's latest batch which I haven't incorporated yet. The count of minimals is 4513 (more than double what I had in the google doc).

I am currently running three scripts:

1. For each minimal puzzle, generate all puzzles in the following neighborhoods: {-2,+2}, {-1,+2}, {-2,+1}. Then check T&E depth, and if >2 expand and check against the expanded puzzle database.
2. For each expanded puzzle, generate all "transforms" obtained by swapping groups of disconnected givens and check against the expanded puzzle database.
3. For each expanded puzzle, generate all minimals, check against the minimal puzzle database; also expand and check against the expanded puzzle database.

The first script is easily the slowest of the three (it has gotten through 939 of the 4513 minimals so far), and I will be splitting it into multiple threads at some point. A fourth script will be adding digits to expanded puzzles and checking depth, eventually yielding all the "maximal" expanded forms for every puzzle in the database.

The lowest rated expanded form in the database so far is:

Code: Select all

........1.....234.....56.7.....8.......6.928..6821..9..59..1..86.29.8..581.5.....  ED=9.2/4.5/2.6

(Without uniqueness techniques, it jumps to 11.6.)

[denis_berthier]

The total count for expanded forms not in T&E(2) is now 1808 + any from hendrik's latest batch which I haven't incorporated yet. The count of minimals is 4513 (more than double what I had in the google doc).

When you integrate Hendrik's puzzles into the expanded database, could you have a temporary index corresponding to their place in Hendrik's original sequence, as you did with my numbering of some version of yours?

2. For each expanded puzzle, generate all "transforms" obtained by swapping groups of disconnected givens and check against the expanded puzzle database

This is probably where much "hidden" redundancy is introduced. "Hidden", I mean by quasi isomorphisms. AFAIK, this technique of expansion wasn't used in the previous database, and that would explain why we didn't have the redundancy problem. I don't mean you shouldn't do it; it's great for exploring a puzzle neighbourhood; just try to understand the why.

The lowest rated expanded form in the database so far is:

Code: Select all

........1.....234.....56.7.....8.......6.928..6821..9..59..1..86.29.8..581.5.....  ED=9.2/4.5/2.6

(Without uniqueness techniques, it jumps to 11.6.)

I'd count it as 11.6. Uniqueness makes a total mess when it comes to establishing any relationship with T&E-depth.
I'm curious to see what's the "real" min SER without uniqueness.

[mith]

Me too; at some point I'll run SE minus uniqueness on the lowest rated puzzles until I find one that is still low. [edit]Doing that now, for all grids <=10.3. The low with uniqueness is now down to 9.0, with both rating 11.7 without uniqueness.[/edit]

(And yes, I can add some sort of temporary index; may be a bit before I post an update, it's still generating at a pretty good pace right now, up to 3415 expanded forms.)

[mith]

Not too surprising, but none of the <=10.3 with uniqueness grids rated below 11.6 without.

[denis_berthier]

Me too; at some point I'll run SE minus uniqueness on the lowest rated puzzles until I find one that is still low. [edit]Doing that now, for all grids <=10.3. The low with uniqueness is now down to 9.0, with both rating 11.7 without uniqueness.[/edit]

BTW, how do you run SER without uniqueness?
The command I use for standard SER is:
java -cp SudokuExplainer.jar diuf.sudoku.test.serate --format=%r ....8...5..4..52...9...2..1.8....5.7.3..6.81.7.1..8.9.6..4...5...28..4......3....
and similarly for files of puzzles:
java -cp SudokuExplainer.jar diuf.sudoku.test.serate --format=%r --input=input.txt --output=output.txt

Is there any option for disabling rules or are you running a different version of SE?

it's still generating at a pretty good pace right now, up to 3415 expanded forms.)

I'll tax you for giving me too much analysis work

[denis_berthier]

Not too surprising, but none of the <=10.3 with uniqueness grids rated below 11.6 without.

That's also what I was expecting and probably true above 10.3 too.
It also means that in the previous database, there may be hidden jewels with SER 11.9 without uniqueness.

[mith]

I'm using: https://github.com/SudokuMonster/SukakuExplainer (1.17.8 I believe, which is not the latest)

Code: Select all

--techs=1111000011010111011011011011000010100010111111 (1.2.1 mode)
--techs=1111000011010111011011001011000000100010111111 (without uniqueness)

I don't know if the original has this parameter, but at least in this fork you can get a list of techniques with "--techs=" (it will tell you there is an error because you didn't give characters, and then a list of the techniques by index).

[1to9only]

See this post: http://forum.enjoysudoku.com/pgexplainer-a-minimal-sudokuexplainer-in-56-712-bytes-t39049-15.html#p310358
PGExplainer with the -M option will do the same, and much faster!!
-M was implemented for sudokus with multiple solutions, with uniqueness (UR/UL and BUG) techniques disabled.

java.exe -jar PGExplainer.jar --input=92.45.26.txt --output=output.txt
........1.....234.....56.7.....8.......6.928..6821..9..59..1..86.29.8..581.5..... ED=9.2/4.5/2.6

java.exe -jar PGExplainer.jar -M --input=92.45.26.txt --output=output.txt
........1.....234.....56.7.....8.......6.928..6821..9..59..1..86.29.8..581.5..... ED=11.6/8.4/2.6

[mith]

Now have some 37c expanded forms not in T&E(2); here was the first found:

Code: Select all

.......12..3.1456.16..524.3...2.513.3.1.78...6..391....3.....5141.5..62.5..1..3.4  ED=11.7/11.7/3.4

[edit]Make that 38c.

Code: Select all

.......12.13.245.6.5261.34.....3.6.11....6.353...5142..7816....23....16.691.4....  ED=11.7/11.7/3.4

I just started the "adder" script (the fourth script mentioned on the previous page), maybe it will go even higher.

[mith]

Down to 6.7 for lowest SE-with-uniqueness for a depth 3 puzzle:

Code: Select all

........1....23.45..51.4.6....2.54.626.34....54..16.23.51...632672.31...8....2...  ED=6.7/4.5/2.6

11.6 without

[mith]

The pace really picked up with the fourth script; over 11k expanded puzzles now, with over 45k minimals.

The first script has a lot of catching up to do, it has only made it through about 1600 of those minimals. Really need to split it.

[mith]

Alright, reached the lower limit for this:

Code: Select all

........1..1..2.34.34.156.....2.6....7.....4..89....5....5.32..3..16.4.55...24..3  ED=4.5/2.0/2.0

11.7 without uniqueness.

[mith]

And I'm now over 23k expanded forms and 100k minimals. This is growing incredibly fast.

[denis_berthier]

.
Hi with + 1-to-9-only
Thanks for the information. I think it will not be long before I use it.

1-to-9-only, does PGP-Explainer always produce the same result as standard SE?

[denis_berthier]

And I'm now over 23k expanded forms and 100k minimals. This is growing incredibly fast.

Great how fast [the auto-correct wrote fat] that grows.

This also implies that we would now really need a way of extracting the subset of "minimals" from the whole database of these expanded forms. Maybe in the short term, just a 3rd .xls table (keeping the same IDs and all the same information).

Perhaps we should also choose a name for this kind of "minimals":
- "expanded minimals" would be ambiguous (that's more accurate for what you now have in the "expanded" database
- what do you think of something like min-expands (shortcut for "minimal expanded forms")?

IMO, in the future, these min-expands (or whatever name we give them) will play the same role as the minimals (in the usual sense) now do.
Expanded forms correspond to a broader view of equivalence (expanded equivalence ?) and play the same role in avoiding redundancy.
Introducing their minimals "officially" by naming them is also recognising the role of the new form of proximity search allowed by considering the expanded forms.