The hardest sudokus (new thread)

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

Postby tarek » Fri Jan 16, 2009 2:44 am

The ordering of fields in the database shouldn't be a problem.

The database is actually shouldn't be suitable vieing on forums but should be posted as a single file on a free hosting website (the numbers are managable now but that may change).

If anybody has links or updates to puzzles rating 11.6 according to SE121. These should go the database.

SE12 rated puzzles of 11.2+ are candidates as the version change to 121 brought with it a .2 to .4 addition to the ratings.

Thanks to Chamoagne for his efforts to crack these puzzles using new techniques.

The puzzles in the database are based on rating programs that are freely available and therefore should stay unless a decision was made to exclude a rating program from what is currently being used.

tarek
User avatar
tarek
 
Posts: 2622
Joined: 05 January 2006

Postby champagne » Fri Jan 16, 2009 3:30 am

tarek wrote:The puzzles in the database are based on rating programs that are freely available and therefore should stay unless a decision was made to exclude a rating program from what is currently being used.

tarek


I would just react on that point.

The first puzzle on the list is not requiring any sophisticated or new method. Just ALS/AC's and AIC's nets, not so big compared to others.

Let us assume a "visitor" looking for a challenging puzzle see that thread "hardest puzzle" and understand the most difficult are at the top.

He tries the first and solves it "easily" in one or 2 hours.

First possible reaction

He guys I am the best less than one day to solve the hardest known puzzle
:D:D:D

Second possible reaction

This is not serious I have nothing to do here:(:(:(

Third possible reaction:?::?::?:

Is it necessary to exclude an agreed "rating program" just because he does not integrate a new technic:?:

Is it realistic to build such a data base ignoring the "best practices":?::?:

I will not suggest to use my solver an an additionnal ranking tool. It is for sure free, but I should prepare a stable version. For the time being, it is far to be the case.

champagne
champagne
2017 Supporter
 
Posts: 5680
Joined: 02 August 2007
Location: France Brittany

Postby ronk » Fri Jan 16, 2009 10:43 pm

***Off-topic***

tarek wrote:
ronk wrote:To make more relevant fields stay on-screen for human readers, would you please make [edit: PuzzleC] the last field:?:

BTW IIRC, if you replaced hyphens with underscores, your Pearly6000 entries wouldn't line wrap.
I sense that there is a display problem which isn't present in IE7 ... because I used the [size] tags I managed to shrink the text to esablish 1 record per line (that is true even with 1024*768 display)

Thanks, I obviously have a setup problem. Things haven't been the same since my NEC Multisync died recently.:( FWIW I'm using IE7, a 19" LCD monitor with 1680x1050 native resolution configured for "fill aspect ratio", and video card configured for 1280x960. Didn't know it when I purchased the LCD monitor, but the high native resolution is useless when it comes to Cerfing the net. Most of the sites I frequent don't let one scale the text with [View]/[Text Size], because many fonts are defined absolutely, e.g., "10px". As a result, a lot of text is unreadable using 1400x1050.

[edit: If anyone has a helpful hint, please consider using a pm. hobiwan, thanks for yours; it works well.]

tarek wrote:
ronk wrote:BTW2 I applaud your use of ",," instead of ",0," for empty fields.
is that sarcasm:D , I'll continue with it regardless:D

Just implying I would've done the same; no sarcasm intended.:)

[edit 2: added link)
Last edited by ronk on Sat Jan 17, 2009 7:37 pm, edited 2 times in total.
ronk
2012 Supporter
 
Posts: 4764
Joined: 02 November 2005
Location: Southeastern USA

Postby tarek » Thu Mar 12, 2009 12:20 am

I on 16 Jan 09 wrote:If anybody has links or updates to puzzles rating 11.6 according to SE121. These should go the database.

SE12 rated puzzles of 11.2+ are candidates as the version change to 121 brought with it a .2 to .4 addition to the ratings.
Zero/Zéro/Null responses to date:(

tarek
User avatar
tarek
 
Posts: 2622
Joined: 05 January 2006

Re: The hardest sudokus (new thread)

Postby champagne » Thu May 27, 2010 9:15 am

Hi,

An attempt to re activate the thread on hardest puzzles.

I intend to publish several files giving the list of these puzzles included in my database having specific properties.

I have still some work to do.

Meantime, I come back to the "Silver Plate" pattern that was discussed in the missing posts.

Coloin made a lot of search on that pattern and here is the list of puzzles with a SE rating over 10.5 already in my database

Code: Select all
100006000050700003009020000040500300000090050000000074070800040600000800002001000 tarekdb   Silver_Plate   ED=11.5/11.5/10.1   
000000069000700050006050800003060005700002000010400000005080090020001000400000300 col0906   4651      ED=11.3/1.2/1.2      
000001030000000065003050900008090003070002000100400000006080050020700000400000800 col 08 05   1192      ED=11.2/1.2/1.2      
000000083070000005005030900009050008000200600100004000006090050200700000040001000 colefg   ole006     ED=11.2/1.2/1.2      
000001030000000065003050900008090006070002000100400000009080050020700000400000800 col 08 05   698      ED=11.0/1.2/1.2      
100000500080400060003007000040080000000094080000600002007000100060900020500000003 col 08 05   1410      ED=10.9/10.9/3.4   
000001030000000065006050900008030006070002000100400000009080050020700000400000800 col 08 05   1156      ED=10.8/1.2/1.2      
003000500080400060100000003000094020090008000000200006700000100060900080005070000 coly   863         ED=10.7/1.2/1.2      
000001030000000085003050900008090003100002000070400000005080090020000600400700000 col 08 05   382      ED=10.6/10.6/9.2   
090400700008000030000000024030200070000050900100006000500010000040700002006008000 colBF3   1061      ED=10.6/1.2/1.2      
000700030000000086003090500005060008010002000700400000006030050400001000020000900 cola2   1017         ED=10.6/1.2/1.2      
003000500080400060100000007000094020090008000000200006700000100040900080005030000 cola2   1731         ED=10.6/1.2/1.2      
000500030000000069008060700009080003500002000010400000006030070020001000400000800 cola2   3412         ED=10.6/1.2/1.2      
030000009000000065005090800008050006000002100700400000009080050400007000020300000 colBF1   2263      ED=10.6/1.2/1.2      
003000500080400060100070000090004000000098040000200006700000100060900080005000003 col0906   23308      ED=10.6/1.2/1.2      
000001030000000089003050200008090003100002000070400000005080090020000600400700000 col 08 05   592      ED=10.5/10.5/9.2   

After hours of exploration of that pattern, I can only add 4 more, The task is not finished, but I doubt I'll find new ones.

Code: Select all
98.7.....7.....8....4.3.....1......2..84..9.......5.6...79..3......2..5......6..1 GP H1 2   ED=10.9/10.9/10.6
98.7.....7.....9....4.3.....1......2..84..7.......5.6...93..4......2..5......6..1 GP H1 3   ED=10.6/10.6/9.5
98.7.....7.....8....4.3.....1......2..73..4.......5.6...84..7......2..1......6..5 GP H1 4   ED=11.2/11.2/3.4
98.7.....7.....8....4.3.....1......5..94..3.......2.6...83..7......6..2......5..1 GP H1 6   ED=11.1/1.2/1.2



I don't know how coloin found the famous "Silver Plate", but I guess he applied a kind of -n +m process.
If this is true, he would surely be prepared after so many years to disclose the seed.

champagne
champagne
2017 Supporter
 
Posts: 5680
Joined: 02 August 2007
Location: France Brittany

Re: The hardest sudokus (new thread)

Postby m_b_metcalf » Thu May 27, 2010 11:40 am

champagne wrote: After hours of exploration of that pattern, I can only add 4 more, The task is not finished, but I doubt I'll find new ones.

Code: Select all
98.7.....7.....8....4.3.....1......2..84..9.......5.6...79..3......2..5......6..1 GP H1 2   ED=10.9/10.9/10.6
98.7.....7.....9....4.3.....1......2..84..7.......5.6...93..4......2..5......6..1 GP H1 3   ED=10.6/10.6/9.5
98.7.....7.....8....4.3.....1......2..73..4.......5.6...84..7......2..1......6..5 GP H1 4   ED=11.2/11.2/3.4
98.7.....7.....8....4.3.....1......5..94..3.......2.6...83..7......6..2......5..1 GP H1 6   ED=11.1/1.2/1.2


If I've understood you correctly, does this count as a new one?
Code: Select all
12.3.....3.....2....4.5.....6......7..24..1.......8.9...35..4......7..8......9..6 ED=10.6/10.6/3.4

Regards,

Mike Metcalf
User avatar
m_b_metcalf
2017 Supporter
 
Posts: 8369
Joined: 15 May 2006
Location: Berlin

Re: The hardest sudokus (new thread)

Postby champagne » Thu May 27, 2010 12:33 pm

m_b_metcalf wrote:If I've understood you correctly, does this count as a new one?
Code: Select all
12.3.....3.....2....4.5.....6......7..24..1.......8.9...35..4......7..8......9..6 ED=10.6/10.6/3.4

Regards,

Mike Metcalf



Yes it does,

As most (if not all) of the puzzles in that lot, it has an exocet pattern. I enter it in my database.

champagne
champagne
2017 Supporter
 
Posts: 5680
Joined: 02 August 2007
Location: France Brittany

Re: The hardest sudokus (new thread)

Postby m_b_metcalf » Thu May 27, 2010 1:22 pm

champagne wrote:Yes it does

Then allow me to introduce his big brother:
Code: Select all
12.3.....3.....2....4.5.....6......7..24..3.......8.9...15..4......7..8......9..6 ED=10.9/10.9/3.4

Regards,

Mike Metcalf
User avatar
m_b_metcalf
2017 Supporter
 
Posts: 8369
Joined: 15 May 2006
Location: Berlin

Re: The hardest sudokus (new thread)

Postby dukuso » Thu May 27, 2010 1:54 pm

hmm, suexratt doesn't create randomness of good quality ?
but suexrat9 does ?

it's possible, I might check that, use another random number generator, another seed

------edit-------------
well, looks to me that coly013 just varies a lot in difficulty, depending on what
isomorph is being used - so use a larger sample like suexratt coly013 10000 or such
to get more reliable values

---------edit---------
OK, I see now. It gives different values for isomorphic puzzles.
So maybe not all isomorphs are created with same probability
for suexrat9 and suexratt

-----------edit--------------
suexrat9 and suexratt don't actually create isomorphic puzzles but rather
solve the puzzles by randomly selecting rows or colums in the exact-cover
matrix when there are multiple choices. I was assuming this gives
the same sort of randomization than creating random isomorphs,
but it seems that this is not always true.
So you can try suexrat with several isomorphs and then take
the average for a etter rating.
I'll try to make a new version with randomly selecting isomorphs
Last edited by dukuso on Thu May 27, 2010 4:37 pm, edited 3 times in total.
dukuso
 
Posts: 479
Joined: 25 June 2005

Re: The hardest sudokus (new thread)

Postby champagne » Thu May 27, 2010 2:16 pm

m_b_metcalf wrote:Then allow me to introduce his big brother:
Code: Select all
12.3.....3.....2....4.5.....6......7..24..3.......8.9...15..4......7..8......9..6 ED=10.9/10.9/3.4

Regards,

Mike Metcalf


I got that one as well, unhappilly, it is an isomorph of col 08 05 1410 above.

champagne
champagne
2017 Supporter
 
Posts: 5680
Joined: 02 August 2007
Location: France Brittany

Re: The hardest sudokus (new thread)

Postby ttt » Thu May 27, 2010 5:33 pm

Hi dukuso,
At present, I don’t think coly013 (hardest puzzle based on Suexrat9) or tarx0134 (hardest puzzle based on Suexratt) here that is still on hardest puzzles list... (at least for me :D). It’s not too hard for the human to solve them with few hours... :D

ttt
ttt
 
Posts: 185
Joined: 20 October 2006
Location: vietnam

Re: The hardest sudokus (new thread)

Postby dukuso » Thu May 27, 2010 6:24 pm

humans don't really count, they are inferior to computers ;-)
Many puzzles that are hard for humans are easy for computers.
First you must fix the solving strategy, then you can rate the puzzles.
And I think it's reasonable to choose the best known strategy here.

It could make sense to introduce another, "human" rating, that only
allows strategies that are used by humans.
But I have no experience how to do this.
dukuso
 
Posts: 479
Joined: 25 June 2005

Re: The hardest sudokus (new thread)

Postby champagne » Thu May 27, 2010 10:04 pm

dukuso wrote:humans don't really count, they are inferior to computers ;-)
Many puzzles that are hard for humans are easy for computers.
First you must fix the solving strategy, then you can rate the puzzles.
And I think it's reasonable to choose the best known strategy here.

It could make sense to introduce another, "human" rating, that only
allows strategies that are used by humans.
But I have no experience how to do this.



I have to confess I disagree with the first statement.

For sure, computers can find paths in experienced areas in a very short time. So what !!

Resulting solutions are often boring and indigestible.

Reversely Human can produce nice solutions on which we can built new strategies.

For me, a good solver must provide a solution not to far from the "best practice" and that, I agree, is not easy to do.

I am still investing heavily in improving my solver. The main target is to shorten paths approaching what human are doing. In many case, the runtime is longer than in the previous solution. Reversely, and very often for the top list of Sudoku Explainer rating, the improvement is fantastic.

At the end, I hope I will be in a position to extract puzzles humans will solve with big pleasure.

champagne
champagne
2017 Supporter
 
Posts: 5680
Joined: 02 August 2007
Location: France Brittany

Re: The hardest sudokus (new thread)

Postby coloin » Fri May 28, 2010 10:04 am

champagne wrote:I don't know how coloin found the famous "Silver Plate", but I guess he applied a kind of -n +m process.
If this is true, he would surely be prepared after so many years to disclose the seed.


Well, it was a repetitive {-2+2} process. Selecting the hardest - with higher suexratt or gsf -q2 - and repeating.

However I have no idea which seed it came from. The name i hope doesnt offend [s'il vous plait !]

It is good to have the master [dukuso] back after all these years.

The variations with isomorphs has been discussed in this thread - i hadnt questioned its selection process - but obviously this is where its erring !

The search for the most balanced / hardest puzzle perhaps with 22 clues [many more puzzles] might be challenging. I set my program to do this last month when the forum was down. I will get champagne to have a look at them when the run is complete !

Regards

C
coloin
 
Posts: 1637
Joined: 05 May 2005

Re: The hardest sudokus (new thread)

Postby champagne » Fri May 28, 2010 10:41 am

coloin wrote:Well, it was a repetitive {-2+2} process. Selecting the hardest - with higher suexratt or gsf -q2 - and repeating.

However I have no idea which seed it came from.
.

The variations with isomorphs has been discussed in this thread

The search for the most balanced / hardest puzzle perhaps with 22 clues [many more puzzles] might be challenging. I set my program to do this last month when the forum was down. I will get champagne to have a look at them when the run is complete !

Regards

C




Hi coloin,

I have not been surprised reading your post.

I started an nearly "exhaustive scan" on the "Silver Plate" to have a feeling of the value of that strategy to find "hard puzzles". After weeks of works, I only generated, as you can see a tiny number of puzzles not yet found by you, and significantly easier.

Having a look at your file, it seems clear that turning aroung in your -2 +2 process, you found many puzzles which are in fact morphs of the "siver Plate" pattern.

I am wondering whether these puzzles could be found appying the -2 +2 process without changing the pattern. (I will do the test, I have coded that option).


Regarding the 22 clues or more, the field is widely open. In my database of puzzles competing to be in the "hardest family", I have 77000 puzzles with 21 clues ; 2500 with 22clues and nearly nothing above.

champagne

PS: One key issue in the process described here is to find fresh seeds. Just repeating what you have done would give poor results.
champagne
2017 Supporter
 
Posts: 5680
Joined: 02 August 2007
Location: France Brittany

PreviousNext

Return to General