The New Sudoku Players' Forum

[tdillon]

I found this puzzle has nearly perfect balance in the templates survived after direct eliminations from the given pencilmarks.

That sounds like an interesting metric. Have you found that this balance is correlated with difficulty? If so it would be interesting to explore whether this can be used as a heuristic to aid in searching for difficult puzzles.

My solver finds the solution to tdillon's sukaku in 0.45s, but has never completed the proof of uniquness.

Perhaps your solver's heuristics are tuned for minlex presentation of puzzles?

[tarek]

I found this puzzle has nearly perfect balance in the templates survived after direct eliminations from the given pencilmarks.

That sounds like an interesting metric. Have you found that this balance is correlated with difficulty? If so it would be interesting to explore whether this can be used as a heuristic to aid in searching for difficult puzzles.

I think this was the latest observation/metric that dobrichev observed and with it eleven was able to expand the hardest list even further:
http://forum.enjoysudoku.com/the-hardest-sudokus-new-thread-t6539-270.html#p210708

[dobrichev]

No. This is unrelated, entirely different observation.

[mith]

Ran into another puzzle skfr fails to rate; Output 2 is:

...........1..2..3.4..5.16.........7.5.....1.1...8.54...5.1.48.4..9...3.86....... ED=0.0/0.0/0.5

(SE 1.17.7.1 rates this an 11.0.)

[champagne]

Ran into another puzzle skfr fails to rate; Output 2 is:

...........1..2..3.4..5.16.........7.5.....1.1...8.54...5.1.48.4..9...3.86....... ED=0.0/0.0/0.5

(SE 1.17.7.1 rates this an 11.0.)

I ran skfr and got 10.9/1.2/1.2 ???

[mith]

Strange. Maybe my version is old.

[champagne]

This old wersion of SKFR remains relatively slow for me rating millions of puzzles in the high levels to update the database of potential hardest.
SKFR has been written years ago, starting from a frame of code not at all oriented to produce an efficient clone of Sudoku Explainer.

As the meter for the length, Sudoku Explainer rating is somehow a reference of the possible difficulty of a sudoku that everybody understand and that nobody wants to change.

And as for the meter, small adjustments of the definition remain possible to avoid small problems linked to the original definition.

Years later, I think possible to write a new version of SKFR much better for hard puzzles. Expected improvement is big enough to push me to start a new design, the target remaining to be as close as possible to Sudoku Explainer's rating.

This will be in C++ and using intel set of instructions.
Anybody willing to contribute to this new version is welcome.

[champagne]

in line with the lowest activity in the forum, nobody came.

As I need this update of SKFR to continue to work on the n digits end pattern, I try to do this alone.
I reached cloning the area of UR/UL, trying to find a good process in this place where no bypass is possible (unpredictable effect on the final rating).

Hope to reach the end in 2/3 months

[champagne]

starting now cloning of chains,
this is the place where a significant improvement in performance is expected.
Nobody showed interest to see the draft.
bi values chains should be quickly covered

[yzfwsf]

I wonder if the new SKFR will offer support for Sukaku?

[champagne]

I never worked on suklakus, so I am not sure at all to give the right answer,
but with small adjustments, I think that the process is valid for any start PM with a unique solution.

The first step in the process is to apply the brute solver to find the solution and to setup the start PM.
Later, only unsolved cells and the current PM are used to apply serate rules.

If I am right, a sukaku is a PM that you can't get from a list of given, so you can force the start and the solution.