The New Sudoku Players' Forum

[denis_berthier]

.
Here is another Pandiag(13x13):

Code: Select all

. 3 C B 9 . . . . 8 . 1 A
7 6 4 . 8 . 1 A . . . . .
D 1 . . . . . . . . . 2 .
. . 9 . . . 2 8 . . . . .
4 2 . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . 8 . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. . . . . . . . . . . . .
. A . . . . . . . . . 8 .
. . . . . 2 . . . A . . .
. . . . A . . . . 9 . . .

.3CB9....8.1A764.8.1A.....D1.........2...9...28.....42.............................8...............................................A.........8......2...A.......A....9...

The general method of construction was as for the first, but I think there remains more givens that could be deleted (second exercise, apart from solving it).

BTW, is there a public version of SE for rating such puzzles?

[1to9only]

ED=8.3/1.5/1.5

BTW, is there a public version of SE for rating such puzzles?

Currently: No.

This is really down to having little time for packaging proper releases of the many versions of SE I've created, and writing up how to use them!! There are a few issues I'm aware of in most of the SEs, but I don't fix these until I use/need the features. Also the SE licence require program releases to be accompanied by source code, and I dont want the hassle of maintaining umpteenth versions of SE!! The PD13 SE is also problematic, as serate for PDs > ED=10 takes hours, and PDs > ED=11 takes days!!

If anyone wants just the PD5/7/11/13 jar programs (!without source!), post here.

[denis_berthier]

ED=8.3/1.5/1.5

Thanks for the SE rating. It confirms what I thought, i.e. that it was slightly harder than the 1st (in spite of having more givens and being also in W4).

BTW, is there a public version of SE for rating such puzzles?

Currently: No.
This is really down to having little time for packaging proper releases of the many versions of SE I've created, and writing up how to use them!! There are a few issues I'm aware of in most of the SEs, but I don't fix these until I use/need the features. Also the SE licence require program releases to be accompanied by source code, and I dont want the hassle of maintaining umpteenth versions of SE!! The PD13 SE is also problematic, as serate for PDs > ED=10 takes hours, and PDs > ED=11 takes days!!
If anyone wants just the PD5/7/11/13 jar programs (!without source!), post here.

I understand the problems of distributing software with various applications and user manual. I still have to find time for writing how to use the Pandiag extension of LatinRules before I can publish it.
On MacOS, one can't use jar files if they are not signed by a registered developer or compiled on the local computer; so, they'd be of no use to me. Otherwise, I would have used them for ratings. But this is not essential for creating puzzles in T&E(1).

As for the time SER takes for hard puzzles, there's an easy way out of it, depending on your goals: it should be easy to put an upper bound defining which puzzles you consider as interesting and declare that anything above gets the same rating 20 or 10,000,000 or whatever.

[creint]

Does it really matters how detailed the rating is?
Basic chains < 7.0
Advanced chains < 8.8
Dynamic chains < 10
>= 10 is TE 2 or more
If it fully solves it is in TE 2.

Above puzzle solves within 20 seconds.

[denis_berthier]

Does it really matters how detailed the rating is?

It depends on one's goals.
There are many universal classification or rating systems (T&E-depth, W, B, gW, ...) but there's no universally useful one. A good rule of thumb can be: solvable by a human solver => in T&E(1). But even this rough classification is not always true; and many puzzles in T&E(1) are much too hard for a real human solver.

Basic chains < 7.0
Advanced chains < 8.8
Dynamic chains < 10

All the bounds in SER are arbitrary (and logically inconsistent, as the number of inferences can't be defined in logical terms). I don't see the point of adding more arbitrariness.

>= 10 is TE 2 or more

In Sudoku, T&E(1) ends much before SER 10. Generally around 9.1 or 9.2 but I've seen an exceptional 9.8. There's no fixed correspondence.
Moreover, checking if a puzzle is in T&E(1) can be done very fast - much faster than SE could do it approximately.

If it fully solves it is in TE 2.

Fully solves by what? A puzzle in T&E(4) fully solves by T&E(4).
If you mean by SE, it's not even sure, as there's no correspondence between the rules in SE and T&E levels.

Above puzzle solves within 20 seconds.

Meaningless. Solves by what? What resolution path does it produce (if any)?

[creint]

If it fully solves it is in TE 2.

Above puzzle solves within 20 seconds.

Solving with my variant solver.
You could define an upper bound of difficulty.
It is for me unclear what the definition of T&E 2 is.

Can you design a good rating system?
It is probably best if computer rating is separate from user rating.
Should chain length affect rating?
It should include bottleneck size, but how to determine bottleneck?
Layer count should affect rating. How to handle back doors?
How to rate larger structures with more than one link?

A singles only puzzle can be rated like something: each possible states take average and minimal(available singles / all pencil mark count).
A puzzle with 0.3/0.2 is easier than 0.2/0.1.
For user rating you could factor in the difficulty on how hard the user estimates it, not all constraints are easy to see.

[denis_berthier]

If it fully solves it is in TE 2.

Fully solves by what? A puzzle in T&E(4) fully solves by T&E(4).
If you mean by SE, it's not even sure, as there's no correspondence between the rules in SE and T&E levels.

Solving with my variant solver.

Above puzzle solves within 20 seconds.

Meaningless. Solves by what? What resolution path does it produce (if any)?

Those are non-answers, as nobody knows how your "variant solver" works. Variant of what?
This puzzle could be solved in a fraction of a second by a properly coded T&E procedure.

Can you design a good rating system?

Yes, I can and I have. Several. And you perfectly know it.

[coloin]

Can you design a good rating system?

Dennis ... if I can intervve here .....
I think creint was been philisophical and his [rhetorical] question was not really a direct question at you
I interpret it as ... " can one actually design a rating system that is perfect ? "
and his suuggestions are pertinent to our understanding of SE at > 11 and with varients with "harder" puzzzles.

[creint]

With variant I mean free design with all-different constraints/other type constraints.
Larger puzzles, jigsaw, windoku, sudoku-x.
More constraints means less chance on finding the right structures.
A rating on bottlenecks tells much more than one outlier that determines the full rating. I would rate a puzzle that requires more effort higher.
Rating is always solver-based because no solver has every tactic.

[denis_berthier]

A rating on bottlenecks tells much more than one outlier that determines the full rating. I would rate a puzzle that requires more effort higher.

I'd be glad to hear any concrete idea about how to do this.

Rating is always solver-based because no solver has every tactic.

This is essentially false.
There is a universal backbone for pure logic ratings and it is invariant under isomorphisms (a minimum property any serious rating should satisfy). This backbone is provided by whips.
Any additional rules (Subsets, exotic patterns, g-whips, braids, forcing-chains, ...) can only modify it marginally, as I've proven here: http://forum.enjoysudoku.com/pattern-based-constraint-satisfaction-2nd-edition-t32567-11.html

This also explains why the Sudoku Explainer Rating, in spite of having some of these additional rules, is strongly correlated to the W rating, as shown in [PBCS].
Notice that "universal" means this rating applies to any finite binary Constraint Satisfaction Problem (and much beyond, by the addition of relevant CSP-Variables).

I think creint was been philisophical and his [rhetorical] question was not really a direct question at you
I interpret it as ... " can one actually design a rating system that is perfect ?

Perfect, no - in particular because it has to be completed by something else beyond T&E(1). Almost perfect, yes.