The New Sudoku Players' Forum

[champagne]

Is this 11.7 not a new puzzle? What is the isomorphism puzzle in ph_1910.zip

It is a new one, now added to my data base

[m_b_metcalf]

This data base is shared to give samples of "potential hardest puzzles".

champagne,
Until now you've understandably restricted your data base to standard sudokus. However, now that tarek has provided us with a way to rate X-sudokus too, do you think it worth adding a section for that variant? In particular, I believe that these three are rated higher than any others to date (see here):

Code: Select all

..1...2.....3....45........6.3....1..5......6..7..........82.....6.9.....8...7.9.   ED=12.0/12.0/7.7 : Xtreme_bronze
..1...2....23....45........4.6....1..5......7..8..........96.....5.3.....9...8.3.   ED=12.0/12.0/7.8 : Xtreme_silver
..1...2....23....44........5.3....6..1......5..6..........78.......9.....7...1.9.   ED=12.0/12.0/8.1 : Xtreme_gold

Best regards,

Mike

[champagne]

Hi Mike,

for sure this makes sense.

till now, I did not follow tarek's efforts to extend the scope of "Sudoku Explainer" and I never worked on sudoku variants.
IMO, this should be a separate data base, one for each variant, and the volunteer to do the job should be personally involved in the variant (and if possible relatively young).

Years ago, tarek produced one of the first data bases of hard puzzles, but each of us has some tools to do it.

[tarek]

Hi Mike & Champagne:

There could be a sperate thread for Hardest Variants based on Explainer or others. I think That Latin Square, SudokuX, Windoku and SudokuP are relatively reasonable. Certainly Latin Square is quicker to search & rate with SE and the Mathematics/Computer Sciences community may have an interest in the hardest Latin Square (QWH)

With Sukaku Explainer I can see at least 2 issues:
1. It is constantly being updated & therefore a rating of "x" today may change in the next release
2. There are several options that control the rating: one is the -B option which controls how close SE is to a depth 1st search The other is the -P option which
controls what SE uses as nested techniques in Dynamic forcing chains + and above. Both of these are by default similar to what Sudoku Explainer 1.2.1 does.

The rating therefore is correct for the version of the release being used under the set of parameters utilized. The -S option would output all the parameters used to assist in referencing all of this.

There are few freely available solvers to compare to out there and although I have checked for errors/glitches/bugs, There are very few that would be able to verify that 11 is actually harder than 10. I'm assuming that - without evidence to the contrary - that Sukaku explainer is doing a good job!

[m_b_metcalf]

the count per clue in the data base is now the following

Code: Select all

clues Count
20      79
21   11005 +44
22   88531 +536
23  393612 +3235
24  709032 +9443
25  449086 +14069
26  483459 +2540

champagne, In these tables there appears to be a cut-off at 26 clues. Are there any high flyers with 29 clues? Have you seen the new 10+ results in the current Patterns Game?

Mike

[champagne]

Hi Mike,
I have no cut off in the data base, I just reported till now in areas deeply searched.
In fact, due to the work of paquita I assume, I have now a significant volume with more clues.
The entire to-day table is

Code: Select all

20   79
21   11093
22   89650
23   366678
24   759480
25   593710
26   853490
27   302100
28   69787
29   3979
30   132
31   15
32   8
33   3
35   1

37 1

In the high side, if I remember properly, the constraint to have a puzzle minimum has been voluntarily omitted.
Here below, the table giving for ratings 11.0 and more the situation for 27 clues and more

Code: Select all

clues er   Count
27   11.4   3
27   11.3   85
27   11.2   78
27   11.1   26182
27   11.0   275622
28   11.2   24
28   11.1   9202
28   11.0   60559
29   11.1   804
29   11.0   3175
30   11.1   32
30   11.0   98
31   11.0   12
32   11.0   3
37   11.2   1

and yes, I follow the games on an irregular basis .

[pjb]

Hi Tarek

You wrote "I'm assuming that - without evidence to the contrary - that Sukaku explainer is doing a good job!", and no one should dispute this, given how it has given us such valuable structure to the daily practice of sudoku. However, it has been mentioned more than once that its methods, some of which are astonishingly complex, could be augmented by newer ones such as exocet, msls to name a couple. For example, the 11.8 puzzle mentioned above in denis_berthier's post is solved by an MSLS and 2 simple xy chains. I imagine it would be a massive undertaking, but is there any move to re-engineer sudoku explainer ?

Cheers, Phil

[tarek]

Hi Phil,

Thanks. I have no objections on adding any new techniques to Sudoku Explainer. As the interface (both GUI and command line) allow you to select/deselect them.

Sukaku explainer is a modified and improved version. Oh sudoku explainer and includes a few more techniques. The core engine is roughly the same.

The project had 4 people involved at some stage in the start about 6 months ago. At the moment is only yours truly. All of the people involved including me are doing this out their own time with no return except bringing these improvement to everybody.

I’ve managed to introduce a few techniques which are primarily targeting medium to advanced solvers, but nothing as exotic as Exocet Or MSLS. Furthermore even advanced fish and ALS are not even fully implemented. So there plenty of room for improvement in techniques. Sadly, intention and good intentions are on one side, the reality is that there is no free time or motivation from programmers to develop this on Sukaku explainer. I hope that this would change.

Anybody wishing to join the project is more than welcome the madness that is programming sudoku explainer.

[ssxsssxs]

Code: Select all

120300000340010000005006000200000500010000020007008009000500607000030040000007900  ED=11.7/11.7/2.6

[champagne]

Hi Tarek

You wrote "I'm assuming that - without evidence to the contrary - that Sukaku explainer is doing a good job!", and no one should dispute this, given how it has given us such valuable structure to the daily practice of sudoku. However, it has been mentioned more than once that its methods, some of which are astonishingly complex, could be augmented by newer ones such as exocet, msls to name a couple. For example, the 11.8 puzzle mentioned above in denis_berthier's post is solved by an MSLS and 2 simple xy chains. I imagine it would be a massive undertaking, but is there any move to re-engineer sudoku explainer ?

Cheers, Phil

Hi phil and tarek,

some late reactions to this post.
First of all, as I already wrote to tarek, as I have no skill in JAVA, I can not contribute that much to the job.


Introducing new solving rules in the field of "exotic patterns" is something that I have done in my solver as others.

Willing to do this in the rating machine techniques, you have to face many problems here some of them:

what is the intrinsic difficulty of the identification of a given exotic pattern.
What is the list of pattern searched (for example, I am using generic rules to produce exocets and eliminations derived from an exocet, David developped a deep analysis "on pattern" for the Junior Exocets, including derived effects of the abi loop).

Another fact is that exotic patterns and more precisely exocets don't produce always many eliminations at the start.

If I wanted to go in this direction, I would select first patterns known as having a strong effect in terms of eliminations

JExocet 3 digits, often leading to one pair active
conjugated JExocets
SKloop

Regarding big rank 0 logics, the selection of patterns to study is still tougher. I am currently re visiting my old code and I have to cut the search in specific sub fields to cover the main cases seen here and there. (Phil, this is a topic on which we could share our recent experiences)

Just random thoughts so far

[champagne]

Code: Select all

120300000340010000005006000200000500010000020007008009000500607000030040000007900  ED=11.7/11.7/2.6

confirmed new in the data base

[m_b_metcalf]

In the dying days of the Patterns Game, two old records have been smashed. The current Table of highest rating by clue count is now as shown.

Code: Select all

  #   ER   EP   ED    Game         Dealer       Player
clues
 19  10.4 10.4  3.4    381         coloin  m_b_metcalf
 20  11.8 11.8  3.4    169    m_b_metcalf          gsf
 21  11.8 11.8 10.7    349    m_b_metcalf     1to9only  (overalll highest)
 22  11.7  1.2  1.2    193         coloin    champagne
 23  11.4 11.4 10.8    148       Mauricio          gsf
 24  11.3 11.3 10.6    209      champagne  m_b_metcalf  (also in Game 212)
 25  11.3 11.3 11.1    309       Patrice       papy999
 26  11.1 11.1 10.4    302       Patrice       papy999
 27  10.7 10.7  8.9    150   m_b_metcalf        eleven
 28  10.5 10.5  9.2    155      Mauricio     champagne
 29  10.4 10.4  9.9    375   m_b_metcalf           jpf


Regards,

Mike

[Edited to add a third new record (19 clues).]

[dxSudoku]

Those are God-level puzzles!

[tarek]

This scam bot "vondess" needs to be removed from this forum before further polluting threads!

[mith]

A couple 11.7s I found today:

Code: Select all

.....1..2..3....4..2..5.6......6...5.7..2.1..8..9.......48......5...67..9......3. q2=98904 FNBP C21.m/M2.3.56862  ED=11.7/11.7/2.6
.....1..2....3..4...56..7.......2..1.5.7..8..7...4.....69.....7.8.....3.5..9..6.. q2=98842 FNBP C22.m/M2.3.34992  ED=11.7/1.2/1.2

Should be new in the database, unless I have massively screwed something up in my checking (which is entirely possible).

And one hard one by q2:

Code: Select all

.....1..2....3..4...56..7.......2.3..5....8..9...4...1.6..1.....87..6...5..7..6.. q2=99385 FNBP C22.m/M3.608.874  ED=11.4/11.4/4.2