Sudoku Explainer Ratings and Chain Classes

Everything about Sudoku that doesn't fit in one of the other sections

Re: Sudoku Explainer Ratings and Chain Classes

Postby denis_berthier » Tue Dec 28, 2021 6:04 am

AnotherLife wrote:I have come to the point where anomalies arise. If a theory does not accord with the experimental results, it needs to be corrected.


Let's keep all this real.

1) There has never been any theory in Sudoku Explainer - so, nothing to be corrected in this regard. There's a bunch of techniques that have never been popular, some very general (forcing chains..), some very specific (uniqueness...), all mixed in very arbitrary ways, with arbitrary thresholds on the size of chains for the associated ratings.
If that was not clear enough: from a theoretical point of view, SE is pure bullshit. But as I wrote in several places, the interest of SE is, in the mean, it gives some idea of the difficulty of the hardest step of a puzzle. That's why SE has been and remains the common reference. And that's why the author should be given the credit he deserves.
I wrote (and everybody knows) SE rates the "hardest step". As a result, any try to apply it to deal with the number of steps is totally absurd. Like using a clock to measure a red cell count.
Also, SER is based on a fixed set of rules. Adding rules will obviously change the rating. Criticisng SER for not producing the same rating as when one adds new rules is totally absurd.

2) In all this thread, I can't see any serious "experimental result".
AnotherLife's "results" are based on manual solving hundreds of very hard puzzles in a few days. Who can seriously believe in the resulting "classifications"?
yzfwsf computations seem to be more serious, as they are made by a computer. Unfortunately, I can't see anything related to rating puzzles. Only vague classifications not involving the lengths of chains.

The only thing that seems to be suggested form all this is, ALS chains are very weak patterns and are unable to be, in and of themselves, the backbone of a rating system.

Before trying to correct a theory that doesn't even exist, some thinking about rating systems seems to be necessary.
denis_berthier
2010 Supporter
 
Posts: 3166
Joined: 19 June 2007
Location: Paris

Re: Sudoku Explainer Ratings and Chain Classes

Postby AnotherLife » Tue Dec 28, 2021 11:22 am

denis_berthier wrote:2) In all this thread, I can't see any serious "experimental result".
AnotherLife's "results" are based on manual solving hundreds of very hard puzzles in a few days. Who can seriously believe in the resulting "classifications"?

I have checked 1414 puzzles with the help of HoDoKu. Such an attitude to my work is insulting.
Bogdan
AnotherLife
 
Posts: 172
Joined: 07 January 2021
Location: Moscow, Russia

Re: Sudoku Explainer Ratings and Chain Classes

Postby denis_berthier » Tue Dec 28, 2021 1:00 pm

AnotherLife wrote:
denis_berthier wrote:2) In all this thread, I can't see any serious "experimental result".
AnotherLife's "results" are based on manual solving hundreds of very hard puzzles in a few days. Who can seriously believe in the resulting "classifications"?

I have checked 1414 puzzles with the help of HoDoKu. Such an attitude to my work is insulting.

You always said you didn't use a solver.

Moreover, instead of feeling insulted, you'd better deal with the real objections I raise.
denis_berthier
2010 Supporter
 
Posts: 3166
Joined: 19 June 2007
Location: Paris

Re: Sudoku Explainer Ratings and Chain Classes

Postby AnotherLife » Tue Dec 28, 2021 2:52 pm

Do you think that my efforts were aimed at deceiving people? Do you think that I have spent so much time just for fun? When I say 'I check the puzzles manually', I mean that I copy every puzzle into a program and look at the methods used, and I cannot run this process automatically. This is really a manual work because sometimes the hand even aches. I am not sure if I want to discuss anything with a person who distrusts my work. Distrust is always insulting.
Bogdan
AnotherLife
 
Posts: 172
Joined: 07 January 2021
Location: Moscow, Russia

Re: Sudoku Explainer Ratings and Chain Classes

Postby denis_berthier » Tue Dec 28, 2021 2:59 pm

Oh, so that's what a manual solver is for you: someone who types the puzzles manually into a solver. Interesting definition.
Distrust is the basis of science. Sorry for you that you don't know this. Your "results" are much too vague for anyone to take them for granted.

Now, be serious. Do you have any results about the number of puzzles solved by ALS-chains, depending on their length?
denis_berthier
2010 Supporter
 
Posts: 3166
Joined: 19 June 2007
Location: Paris

Re: Sudoku Explainer Ratings and Chain Classes

Postby AnotherLife » Tue Dec 28, 2021 3:25 pm

Stop distorting my words!

To all the participants of the forum: I own the results of my work. I have a right to delete the whole thread if I think that my work is not enough appreciated.
Bogdan
AnotherLife
 
Posts: 172
Joined: 07 January 2021
Location: Moscow, Russia

Re: Sudoku Explainer Ratings and Chain Classes

Postby Hajime » Wed Dec 29, 2021 11:07 am

Bogdan, I appreciate your work and the preliminary results.
In SE are many Chains: Nescio, Cell, Region, Dynamic, Multiple, Nested...
But only the highest rated method determines the SE rating.
That's my objection for all the methods in SE.
A puzzle with 1 Naked Quad is equally rated as one with 5 Naked Quads and 7 Skyscrapers.
User avatar
Hajime
 
Posts: 1003
Joined: 20 April 2018
Location: Netherlands

Re: Sudoku Explainer Ratings and Chain Classes

Postby eleven » Wed Dec 29, 2021 9:40 pm

From a manual solver's view the Sukaku rating is a lot better than Denis'.
But i like Denis' rating, because it says, that 2 of my puzzles are under the 3 hardest ;)
eleven
 
Posts: 2785
Joined: 10 February 2008

Re: Sudoku Explainer Ratings and Chain Classes

Postby denis_berthier » Thu Dec 30, 2021 4:51 am

eleven wrote:From a manual solver's view the Sukaku rating is a lot better than Denis'.

On seeing a reference to a "manual solver", one can be sure in advance the author has no rational argument.
My main rating (the W rating), is indeed much better in the mean than the SER (or any current modification of it) for predicting the difficulty of a puzzle for a manual solver - for the mere reason that whips subsume almost all the popular rules (and this is supported by detailed statistics, not by prejudices).
Note e.g. that none of AnotherLife's criticisms of the SER is valid in case the SER is replaced by the W rating.

eleven wrote:But i like Denis' rating, because it says, that 2 of my puzzles are under the 3 hardest ;)

You're talking of a totally different rating (indeed only classification system): the BpB classification, relevant for the few puzzles in T&E(2). I've always liked your tamagotchi algorithm. My results show it can generate the really hardest puzzles.
The p in the BpB classification is a pure logic, intrinsic property of a puzzle. But it also has an algorithmic interpretation, as the level of look-ahead one must have if trying to solve with a (modified) T&E procedure with look-ahead.
denis_berthier
2010 Supporter
 
Posts: 3166
Joined: 19 June 2007
Location: Paris

Re: Sudoku Explainer Ratings and Chain Classes

Postby Pupp » Mon Jan 03, 2022 12:47 am

In my experience, solvers in general are designed to solve puzzles with the least difficult steps and fewest harder techniques, which seriously is not how humans think.

It's ok to use a solver to rate a puzzle, or get a hint when stuck, but not to use as a road map to a solution.

Analyzing a puzzle just gives you a ballpark idea of how difficult it might be... but it doesn't mean the puzzle might be solvable with easier steps.

Plus Sudoku puzzles are very malleable on what techniques could be used. Give an intricate sudoku puzzle rated 4.0 to a half dozen talented players and they would all solve it differently.
Pupp
 
Posts: 246
Joined: 18 October 2019

Re: Sudoku Explainer Ratings and Chain Classes

Postby AnotherLife » Sun Feb 13, 2022 12:51 pm

In my previous posts, I have tried to convince the community that the contemporary ALS-based methods should be included in a new rating system to give a more relevant estimate of sudoku complexity for a manual solver. Also, I have posted some puzzles with lower ratings that are harder for a human than some puzzles rated SER 9.0. For example,
- this puzzle rated SER 9.0 is solvable by Sue de Coq;
- this puzzle rated SER 9.0 is solvable by an AIC with a group and a basic AIC;
- this puzzle rated SER 7.8 needs Sashimi Franken Jellyfish, Almost Sashimi X-Wing, or kraken oddagons;
- this puzzle rated SER 7.9 needs bivalue oddagons with internals (or a dynamic forcing chain!).

In my opinion, we could leave the old name SER to the dinosaurs of the forum as a common reference, and use instead the name USER (Updated Sudoku Explainer Rating). This name will reflect its user-oriented approach rather than a computer-oriented one. I ask you to express your opinion on the way of the impementation of the new methods and the new name.

P.S.
Hidden Text: Show
I ask Mr Berthier to abstain from any comments in this thread.
Bogdan
AnotherLife
 
Posts: 172
Joined: 07 January 2021
Location: Moscow, Russia

Postby 1to9only » Sun Feb 13, 2022 2:48 pm

AnotherLife wrote: - this puzzle rated SER 9.0 is solvable by Sue de Coq;
- this puzzle rated SER 9.0 is solvable by an AIC with a group and a basic AIC;
- this puzzle rated SER 7.8 needs Sashimi Franken Jellyfish, Almost Sashimi X-Wing, or kraken oddagons;
- this puzzle rated SER 7.9 needs bivalue oddagons with internals (or a dynamic forcing chain!).

The SudokuExplainer version by Keith Corlett (wookie2u2) from here gives these alternate ratings for the 4 puzzles:
Code: Select all
.2.....8...71......96..7.......6...43.9...85..6...13.......5.3..4.2.8...9....41.. ED=4.5/1.1/1.1, 4.5, ALS-XZ, 4.51, ALS-Wing
12..5.....5.7.9.....9.2.5...17.48...3..6..........24.15....4....78...95......781. ED=4.5/1.2/1.2, 4.5, ALS-XZ
..7.6..24.3....16....9........5...48.6.2.8.9.89...4........3....26....8.95..8.6.. ED=4.6/1.2/1.2, 4.62, Finned Franken Jellyfish
.2345.....5..891....8.....4......9..3641......7....4.1..59.8......3...95....4.8.2 ED=4.7/1.2/1.2, 4.72, Kraken type 2: Finned Franken Swordfish

Ratings produced by V6_30.128, the most recent version is V6_30.205!!
1to9only
 
Posts: 3993
Joined: 04 April 2018

Previous

Return to General