The hardest sudokus (new thread)

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

The hardest sudokus (new thread)

Postby tarek » Fri Jan 09, 2009 4:03 am

For the latest compiled list of potential hardest puzzles (Sep 30 2019) please visit champagne's https://drive.google.com/drive/folders/0B5lH6mGXxWzXTDFRMnVTbGNlZU0 and download ph_1910.zip

The following 2 posts reflect all updates until June 2011

The latest Hardest sudokus Database can be downloaded by following this link: HardestDatabase110626.txt

The following is a "Top 5" list of hardest sudokus according to some popular sudoku rating programs (See next post for more info)

q1 Top 5
Code: Select all
Rating Program: gsf's sudoku q1
Rating: 99529
Poster: eleven
Label: HardestSudokusThread-02085;Discrepancy   
 1 2 . | 4 . . | 3 . . 
 3 . . | . 1 . | . 5 . 
 . . 6 | . . . | 1 . . 
-------+-------+------
 7 . . | . 9 . | . . . 
 . 4 . | 6 . 3 | . . . 
 . . 3 | . . 2 | . . . 
-------+-------+------
 5 . . | . 8 . | 7 . . 
 . . 7 | . . . | . . 5 
 . . . | . . . | . 9 8 

Rating Program: gsf's sudoku q1
Rating: 99495
Poster: eleven
Label: HardestSudokusThread-02023;cigarette
 1 2 . | 3 . . | . . . 
 3 4 . | . . . | 1 . . 
 . . 5 | . . . | . . . 
-------+-------+------
 6 . 2 | 4 . . | 5 . . 
 . . . | . 6 . | . 7 . 
 . . . | . . 8 | . . 6 
-------+-------+------
 . . 4 | 2 . . | 3 . . 
 . . . | . 7 . | . . 9 
 . . . | . . 9 | . 8 . 

Rating Program: gsf's sudoku q1
Rating: 99486
Poster: coloin
Label: HardestSudokusThread-00078;Platinum_Blonde
 . . . | . . . | . 1 2 
 . . . | . . . | . . 3 
 . . 2 | 3 . . | 4 . . 
-------+-------+------
 . . 1 | 8 . . | . . 5 
 . 6 . | . 7 . | 8 . . 
 . . . | . . 9 | . . . 
-------+-------+------
 . . 8 | 5 . . | . . . 
 9 . . | . 4 . | 5 . . 
 4 7 . | . . 6 | . . . 

Rating Program: gsf's sudoku q1
Rating: 99432
Poster: eleven
Label: HardestSudokusThread-00209;Cheese
 . 2 . | . 5 . | 7 . . 
 4 . . | 1 . . | . . 6 
 8 . . | . . 3 | . . . 
-------+-------+------
 2 . . | . . 8 | . . 3 
 . 4 . | . 2 . | 5 . . 
 . . . | 6 . . | . 1 . 
-------+-------+------
 . . 2 | . 9 . | . . . 
 . 9 . | . . . | . . 5 
 7 . 4 | . . . | 9 . . 

Rating Program: gsf's sudoku q1
Rating: 99420
Poster: tarek
Label: HardestSudokusThread-00041;tarx0001;Fata_Morgana
 . . . | . . . | . . 3 
 . . 1 | . . 5 | 6 . . 
 . 9 . | . 4 . | . 7 . 
-------+-------+------
 . . . | . . 9 | . 5 . 
 7 . . | . . . | . . 8 
 . 5 . | 4 . 2 | . . . 
-------+-------+------
 . 8 . | . 2 . | . 9 . 
 . . 3 | 5 . . | 1 . . 
 6 . . | . . . | . . . 

SE Top 5
Code: Select all
Rating Program: Nicolas Juillerat's Sudoku explainer 1.2.1
Rating: 11.9  (ER/EP/ED=11.9/11.9/11.3)
Poster: tarek
Label: HardestSudokusThread-00058;Golden_Nugget;pearly6000-1812
 . . . | . . . | . 3 9 
 . . . | . . 1 | . . 5 
 . . 3 | . 5 . | 8 . . 
-------+-------+------
 . . 8 | . 9 . | . . 6 
 . 7 . | . . 2 | . . . 
 1 . . | 4 . . | . . . 
-------+-------+------
 . . 9 | . 8 . | . 5 . 
 . 2 . | . . . | 6 . . 
 4 . . | 7 . . | . . . 

Rating Program: Nicolas Juillerat's Sudoku explainer 1.2.1
Rating: 11.9  (ER/EP/ED=11.9/11.9/9.9)
Poster: eleven
Label: HardestSudokusThread-00208;Kolk
 1 2 . | 3 . . | . . . 
 4 . . | . . . | 3 . . 
 . . 3 | . 5 . | . . . 
-------+-------+------
 . . 4 | 2 . . | 5 . . 
 . . . | . 8 . | . . 9 
 . 6 . | . . 5 | . 7 . 
-------+-------+------
 . . 1 | 5 . . | 2 . . 
 . . . | . 9 . | . 6 . 
 . . . | . . 7 | . . 8 

Rating Program: Nicolas Juillerat's Sudoku explainer 1.2.1
Rating: 11.9  (ER/EP/ED=11.9/11.9/2.6)
Poster: eleven
Label: HardestSudokusThread-02095;Patience
 1 2 . | 3 . . | . . . 
 4 . 5 | . . . | 6 . . 
 . 7 . | . . . | . 2 . 
-------+-------+------
 6 . . | 1 . . | 3 . . 
 . . 4 | 5 3 . | . . . 
 . . . | . . 8 | . . 9 
-------+-------+------
 . . . | 4 5 . | 1 . . 
 . . . | . . . | . 8 . 
 . . . | . . 2 | . . 7 

Rating Program: Nicolas Juillerat's Sudoku explainer 1.2.1
Rating: 11.9  (ER/EP/ED=11.9/1.2/1.2)
Poster: eleven
Label: HardestSudokusThread-00211;Imam_bayildi
 . . 3 | . . 6 | . 8 . 
 . . . | 1 . . | 2 . . 
 . . . | . 7 . | . . 4 
-------+-------+------
 . . 9 | . . 8 | . 6 . 
 . 3 . | . 4 . | . . 1 
 . 7 . | 2 . . | . . . 
-------+-------+------
 3 . . | . . 5 | . . . 
 . . 5 | . . . | 6 . . 
 9 8 . | . . . | . 5 . 

Rating Program: Nicolas Juillerat's Sudoku explainer 1.2.1
Rating: 11.8  (ER/EP/ED=11.8/11.8/11.6)
Poster: eleven
Label: HardestSudokusThread-00212
 1 . . | . . . | . . 9 
 . . 6 | 7 . . | . 2 . 
 . 8 . | . . . | 4 . . 
-------+-------+------
 . . . | . 7 5 | . 3 . 
 . . 5 | . . 2 | . . . 
 . 6 . | 3 . . | . . . 
-------+-------+------
 . 9 . | . . . | 8 . . 
 6 . . | . 4 . | . . 1 
 . . 2 | 5 . . | . 6 .

q2 Top 5
Code: Select all
Rating Program: gsf's sudoku q2
Rating: 99743
Poster: eleven
Label: HardestSudokusThread-00245;Red_Dwarf
 1 2 . | 3 . . | . . 4 
 3 5 . | . . . | 1 . . 
 . . 4 | . . . | . . . 
-------+-------+------
 . . 5 | 4 . . | 2 . . 
 6 . . | . 7 . | . . . 
 . . . | . . 8 | . 9 . 
-------+-------+------
 . . 3 | 1 . . | 5 . . 
 . . . | . . 9 | . 7 . 
 . . . | . 6 . | . . 8 

Rating Program: gsf's sudoku q2
Rating: 99587
Poster: eleven
Label: HardestSudokusThread-02023;cigarette
 1 2 . | 3 . . | . . . 
 3 4 . | . . . | 1 . . 
 . . 5 | . . . | . . . 
-------+-------+------
 6 . 2 | 4 . . | 5 . . 
 . . . | . 6 . | . 7 . 
 . . . | . . 8 | . . 6 
-------+-------+------
 . . 4 | 2 . . | 3 . . 
 . . . | . 7 . | . . 9 
 . . . | . . 9 | . 8 . 

Rating Program: gsf's sudoku q2
Rating: 99578
Poster: eleven
Label: HardestSudokusThread-02085;Discrepancy   
 1 2 . | 4 . . | 3 . . 
 3 . . | . 1 . | . 5 . 
 . . 6 | . . . | 1 . . 
-------+-------+------
 7 . . | . 9 . | . . . 
 . 4 . | 6 . 3 | . . . 
 . . 3 | . . 2 | . . . 
-------+-------+------
 5 . . | . 8 . | 7 . . 
 . . 7 | . . . | . . 5 
 . . . | . . . | . 9 8 

Rating Program: gsf's sudoku q2
Rating: 99551
Poster: coloin
Label: HardestSudokusThread-00078;Platinum_Blonde
 . . . | . . . | . 1 2 
 . . . | . . . | . . 3 
 . . 2 | 3 . . | 4 . . 
-------+-------+------
 . . 1 | 8 . . | . . 5 
 . 6 . | . 7 . | 8 . . 
 . . . | . . 9 | . . . 
-------+-------+------
 . . 8 | 5 . . | . . . 
 9 . . | . 4 . | 5 . . 
 4 7 . | . . 6 | . . . 

Rating Program: gsf's sudoku q2
Rating: 99516
Poster: eleven
Label: HardestSudokusThread-00209;Cheese
 . 2 . | . 5 . | 7 . . 
 4 . . | 1 . . | . . 6 
 8 . . | . . 3 | . . . 
-------+-------+------
 2 . . | . . 8 | . . 3 
 . 4 . | . 2 . | 5 . . 
 . . . | 6 . . | . 1 . 
-------+-------+------
 . . 2 | . 9 . | . . . 
 . 9 . | . . . | . . 5 
 7 . 4 | . . . | 9 . . 

Suexrat9 Top 5
Code: Select all
Rating Program: dukuso's suexrat9
Rating: 10364
Poster: eleven
Label: HardestSudokusThread-01418;coloin
 . . 3 | . . . | . . . 
 4 . . | . 8 . | . 3 6 
 . . 8 | . . . | 1 . . 
-------+-------+------
 . 4 . | . 6 . | . 7 3 
 . . . | 9 . . | . . . 
 . . . | . . 2 | . . 5 
-------+-------+------
 . . 4 | . 7 . | . 6 8 
 6 . . | . . . | . . . 
 7 . . | 6 . . | 5 . . 

Rating Program: dukuso's suexrat9
Rating: 9968
Poster: eleven
Label: HardestSudokusThread-02087   
 1 . . | . 5 . | . . . 
 . . 7 | . . 9 | . 3 . 
 . . 9 | . . 7 | 5 4 . 
-------+-------+------
 . . 4 | . . 3 | . 7 . 
 . 6 . | . . . | . . . 
 . 9 . | 8 . . | . . . 
-------+-------+------
 . . . | 7 9 . | . 2 . 
 . . . | . . 2 | 4 . 3 
 . . 2 | . . . | . . . 

Rating Program: dukuso's suexrat9
Rating: 9453
Poster: eleven
Label: HardestSudokusThread-01419
 . . 3 | . . . | . . . 
 4 . . | . 8 . | . 3 6 
 . . 8 | . . . | 1 . . 
-------+-------+------
 . 4 . | . 6 . | . 7 3 
 . . . | 9 . . | . . . 
 . . . | . . 2 | . . 5 
-------+-------+------
 . . 4 | . 7 . | . 6 8 
 6 . . | . . 4 | . . . 
 7 . . | . . . | 5 . . 

Rating Program: dukuso's suexrat9
Rating: 9195
Poster: eleven
Label: HardestSudokusThread-01629
 . . 3 | . . . | . . . 
 4 . . | . 8 . | . 3 6 
 . . 8 | 3 . . | 1 . . 
-------+-------+------
 . 4 . | . 6 . | . 7 3 
 . . . | 9 . . | . . . 
 . . . | . . 2 | . . 5 
-------+-------+------
 . . 4 | . 7 . | . 6 8 
 6 . . | . . . | . . . 
 7 . . | . . . | 5 . . 

Rating Program: dukuso's suexrat9
Rating: 8946
Poster: coloin
Label: HardestSudokusThread-02061
 . . 3 | . 9 . | . . . 
 4 . . | . 8 . | . 3 6 
 . . 8 | . . . | 1 . . 
-------+-------+------
 . 4 . | . 6 . | . 7 3 
 . . . | 9 . . | . . . 
 . . . | . . 2 | . . . 
-------+-------+------
 . . 4 | . 7 . | . 6 8 
 6 . . | . . . | . . . 
 7 . . | . . . | 5 . 4 

Suexratt Top 5
Code: Select all
Rating Program: dukuso's suexratt
Rating: 5796
Poster: eleven
Label: HardestSudokusThread-01418;coloin
 . . 3 | . . . | . . . 
 4 . . | . 8 . | . 3 6 
 . . 8 | . . . | 1 . . 
-------+-------+------
 . 4 . | . 6 . | . 7 3 
 . . . | 9 . . | . . . 
 . . . | . . 2 | . . 5 
-------+-------+------
 . . 4 | . 7 . | . 6 8 
 6 . . | . . . | . . . 
 7 . . | 6 . . | 5 . . 

Rating Program: dukuso's suexratt
Rating: 5693
Poster: eleven
Label: HardestSudokusThread-01629
 . . 3 | . . . | . . . 
 4 . . | . 8 . | . 3 6 
 . . 8 | 3 . . | 1 . . 
-------+-------+------
 . 4 . | . 6 . | . 7 3 
 . . . | 9 . . | . . . 
 . . . | . . 2 | . . 5 
-------+-------+------
 . . 4 | . 7 . | . 6 8 
 6 . . | . . . | . . . 
 7 . . | . . . | 5 . . 

Rating Program: dukuso's suexratt
Rating: 4969
Poster: eleven
Label: HardestSudokusThread-02087
 1 . . | . 5 . | . . . 
 . . 7 | . . 9 | . 3 . 
 . . 9 | . . 7 | 5 4 . 
-------+-------+------
 . . 4 | . . 3 | . 7 . 
 . 6 . | . . . | . . . 
 . 9 . | 8 . . | . . . 
-------+-------+------
 . . . | 7 9 . | . 2 . 
 . . . | . . 2 | 4 . 3 
 . . 2 | . . . | . . . 

Rating Program: dukuso's suexratt
Rating: 4931
Poster: eleven
Label: HardestSudokusThread-01419
 . . 3 | . . . | . . . 
 4 . . | . 8 . | . 3 6 
 . . 8 | . . . | 1 . . 
-------+-------+------
 . 4 . | . 6 . | . 7 3 
 . . . | 9 . . | . . . 
 . . . | . . 2 | . . 5 
-------+-------+------
 . . 4 | . 7 . | . 6 8 
 6 . . | . . 4 | . . . 
 7 . . | . . . | 5 . . 

Rating Program: dukuso's suexratt
Rating: 4743
Poster: eleven
Label: HardestSudokusThread-02092
 1 . . | 4 . . | . . 9 
 . 5 6 | . . 9 | . . . 
 . . . | . 1 . | . 6 . 
-------+-------+------
 . 6 . | . . . | 8 . . 
 5 . . | . . 4 | . 9 . 
 9 . . | . . 5 | . 1 . 
-------+-------+------
 . 7 . | . . . | 2 . . 
 6 . . | . . 1 | . 5 . 
 . . . | 3 . . | . . .


tarek
Last edited by tarek on Fri Oct 04, 2019 7:00 pm, edited 15 times in total.
User avatar
tarek
 
Posts: 3762
Joined: 05 January 2006

Postby tarek » Fri Jan 09, 2009 4:04 am

This is to reserve the 2nd post for material related to the 1st post above (links, history, ....)

again, this also should be updated soon & hopefully on a regular basis

Popular downloadable rating programs:
Sudoku Explainer 1.2.1 by Nicolas Juillerat, for the Explainer rating (ER)
serate (a modified Sudoku Explainer 1.2.1, for the Explainer, Pearl & Diamond ratings (ER, EP & ED))
sudoku by gsf, for q1, q2 & other functions
suexrat9 by dukuso, for the suexrat9 rating (sx9)
suexratt by dukuso, for the suexrat9 & suexratt ratings (sx9 & sxt)

Links to rated puzzles on other pages (Last update dates provided when possible):
Ravel's Original Hardest Sudokus thread & RMS top list (updated 02 May 2007)
Ravel's Latest RMS top list (updated Jan 08 2008)
JPF's 10.5+ ER list (updated 15 Apr 2007)
gsf's q1 taxonomy list (? updated Dec 2007)
gsf's q2 taxonomy list (? updated Dec 2007)

Links to pages with information on "Hardest Puzzles":
eleven's sudo-eleven Hardest sudoku page
The making of a gotchi, a simple way to find extreme sudokus thread

Command Line examples under windows for rating a group of puzzles:
Code: Select all
sudoku  -q1 puzzles.txt -o q1.txt (q1 rating)
sudoku  -q2 puzzles.txt -o q2.txt (q1 rating)
suexratt puzzles.txt 1000 2 > Sxtr.txt (for Sx9 & Sxt screening rating, faster)
suexratt puzzles.txt 10000 2 > Sxtr.txt (for Sx9 & Sxt database rating, slower)
java -Xrs -Xmx500m -cp SudokuExplainer.jar diuf.sudoku.test.serate -i puzzles.txt -f %%r/%%p/%%d/%%g/%%e -o serate.tXt (for ER/EP/ED in Sudoku Explainer)


A database of hardest puzzles is maintained with the following suggested filters:
Code: Select all
Entries should have:
Q1  >=99000 OR
Q2  >=99100 OR
SX9 >=4200 OR
SXT >=2000 OR
ER  >=11.6


Preferable Database fields (unknown fields to be left empty):
Code: Select all
Puzzle in line format
Label(s)
Poster(s)
q1
q2
ER
EP
ED
sx9
sxt


tarek
Last edited by tarek on Sun Jul 03, 2011 8:18 pm, edited 3 times in total.
User avatar
tarek
 
Posts: 3762
Joined: 05 January 2006

Postby ttt » Sun Jan 11, 2009 7:39 pm

Hi tarek,
Great works!

For hardest puzzles, I see that there are 4 types of them:
- Type 1: puzzles with symmetries that can solve based on Gurth’s methods and great works recently by eleven, glyn, udosuk here.
- Type 2: puzzles contain SK loops here like Easter Monster…
- Type 3: puzzles contain “floors” like Platinum Blonde, Fanta Morgana, Trompe L’oeil… and can solve as here or here
- Type 4: the rest, some of them contain “almost SK loop” like tarx0075, Silver Plate… that has not yet nice solutions or paths (at least by me:( ), I’m still thinking on them and coly013.

Something’s wrong or missing…?
BTW, I hope I have time to solve “all” your Pearly6000:D

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

Postby ronk » Sun Jan 11, 2009 11:08 pm

tarek wrote:Popular downloadable rating programs:
[...]
Links to rated puzzles
[...]
Preferable Database fields (unknown fields to be left as 0):
Code: Select all
Puzzle in line format
Puzzle in row normal minlex line format
Label(s)
Poster(s)
Date: yyyymmddhhmmss format
q1
q2
ER
EP
ED
sx9
sxt

tarek, thanks for a very nice directory to ratings programs and the ratings of difficult puzzles. As to the database:

1) Is the overall format to be csv -- comma separated values?

2) Is a "c14n" (row normal minlex canonicalization) field actually useful? When prospecting for new difficult puzzles, one needs to be sure the same algorithm is applied to prospective puzzles and their c14n library. Doesn't this mean virtually everyone will build their own c14n library anyway?

3) I now see some of those long labels, like tarek071223170000-170, included hours, minutes, and seconds.:( Is such time precision really necessary?
ronk
2012 Supporter
 
Posts: 4764
Joined: 02 November 2005
Location: Southeastern USA

Postby gsf » Mon Jan 12, 2009 2:33 am

ronk wrote:3) I now see some of those long labels, like tarek071223170000-170, included hours, minutes, and seconds.:( Is such time precision really necessary?

yes, nice work -- sorry that the depth of g.r.emlin's todo list pushed this work back
I thought about labels a bit -- how about userYYMMDDcc where ccc is a counter that starts at 01 each day
gsf
2014 Supporter
 
Posts: 7306
Joined: 21 September 2005
Location: NJ USA

Postby tarek » Mon Jan 12, 2009 6:38 am

Thanks for the kind words everybody, still some work to do though (most of what I posted was already done)

ronk wrote:1) Is the overall format to be csv -- comma separated values?
The idea is to keep a database somewhere for organization purposes. Posting puzzles to me using whatever sperator shouldn't be a problem ... for general posting, the csv format is widely acceptable

ronk wrote:2) Is a "c14n" (row normal minlex canonicalization) field actually useful? When prospecting for new difficult puzzles, one needs to be sure the same algorithm is applied to prospective puzzles and their c14n library. Doesn't this mean virtually everyone will build their own c14n library anyway?
This is to make sure (For me at least) that the puzzles are unique in the database. Most of the fields can be filled electronically through gsf & dukuso's programs....

That means that filling all fields is preferable but not manadatory ... The manadatory fields should be:
1. Puzzle in line format .
2. ER (not EP or ED) [because it is time consuming for me to do]

ronk wrote:3) I now see some of those long labels, like tarek071223170000-170, included hours, minutes, and seconds.:( Is such time precision really necessary?
This is to sort puzzle chronologically. I agree that hhmmss is a bit excessive. gsf's suggestion is appropriate (YYMMDDcccc) .... All of this is simple for new submissions ... The problem is finding old ones (maybe g.r.emlin could help) ... If it is too much of a problem we can omit it:(

tarek
User avatar
tarek
 
Posts: 3762
Joined: 05 January 2006

Postby tarek » Mon Jan 12, 2009 8:17 am

ttt wrote:For hardest puzzles, I see that there are 4 types of them:
.
.
.
I'm looking to update the above post with some form of reference to these topics ... Thanks
ttt wrote:BTW, I hope I have time to solve “all” your Pearly6000:D
Good luck, you'll need it

tarek
User avatar
tarek
 
Posts: 3762
Joined: 05 January 2006

Postby briturner » Tue Jan 13, 2009 1:04 am

greetings,

I have just recently started generating sudokus, but nowhere near these in difficulty level (but I am still learning).

Can you add 1 more piece of information to the second post. The command line (where applicable) used to produce the output (understanding gsf's command line options can be a daunting task). You say convert to row normal minlex line format, and provide q1 and q2, but now how to produce them.

Also, a couple questions:
- Are ratings only applied to normal minlex version, since swapping numbers or rotating can cause large variations in ratings (depending on the program)?
- Are you only collecting the absolute most difficult (thus only a q1 rating of 99486+ should be submitted), or just really hard (maybe all with a q1 rating of 99000+)? If more than just the absolute hardest, what is the thresholds?
- The absolute hardest are posted, but are the runner-up in the database available someplace?

thank you
brit
briturner
 
Posts: 3
Joined: 12 January 2009

Postby champagne » Tue Jan 13, 2009 2:15 am

ttt wrote:For hardest puzzles, I see that there are 4 types of them:
- Type 1: puzzles with symmetries that can solve based on Gurth’s methods and great works recently by eleven, glyn, udosuk here.
- Type 2: puzzles contain SK loops here like Easter Monster…
- Type 3: puzzles contain “floors” like Platinum Blonde, Fanta Morgana, Trompe L’oeil… and can solve as here or here
- Type 4: the rest, some of them contain “almost SK loop” like tarx0075, Silver Plate… that has not yet nice solutions or paths (at least by me:( ), I’m still thinking on them and coly013.


Hi ttt,

I would say that differently.

Ranking using the tools described here give a good idea of the overall difficulty except if a special tool can ease the path.

For the list of identified specificities, I would globally agree :

1) "symmetry"
2) SK loop
3) Most of those having the "bi bi" pattern

I am more restrictive than you on the wording for point 3. Allan Barker model can solve any puzzle if you increase the number of floors. On the other side, I don't know f Silver Plate will solved easily although it has the "bi bi" pattern.

Puzzles having nearly the SK loop could enter that field, but nothing clear up to now.

I can not exclude that a puzzle have both 2) and 3), but the test I made on Easter Monster shows some partial "bi bi", not a full picture.

champagne
champagne
2017 Supporter
 
Posts: 7332
Joined: 02 August 2007
Location: France Brittany

Postby tarek » Tue Jan 13, 2009 6:03 am

briturner wrote:Can you add 1 more piece of information to the second post. The command line (where applicable) used to produce the output (understanding gsf's command line options can be a daunting task). You say convert to row normal minlex line format, and provide q1 and q2, but now how to produce them.
That should be easy
briturner wrote:- Are ratings only applied to normal minlex version, since swapping numbers or rotating can cause large variations in ratings (depending on the program)?
ratings are applied to the original puzzle
briturner wrote:- Are you only collecting the absolute most difficult (thus only a q1 rating of 99486+ should be submitted), or just really hard (maybe all with a q1 rating of 99000+)? If more than just the absolute hardest, what is the thresholds?
I think that a cut-off is best. (I will look into it, mostly top 50 in each category based on my pearly6000 collection)
briturner wrote:- The absolute hardest are posted, but are the runner-up in the database available someplace?
None that specific ... The database would hopefully have them

When the database is up & running ... I could post the file renamed using the date posted on public site, each update would have a different name to reflect the update
User avatar
tarek
 
Posts: 3762
Joined: 05 January 2006

Postby 999_Springs » Tue Jan 13, 2009 7:37 am

Has the #77 from the top1465 completely disappeared out of sight from everything regarding the hardest puzzles?
999_Springs
 
Posts: 591
Joined: 27 January 2007
Location: In the toilet, flushing down springs, one by one.

Postby tarek » Tue Jan 13, 2009 7:42 am

Suggested database filter:
Code: Select all
Entries should have:
Q1  >=99000 OR
Q2  >=99100 OR
SX9 >=4200 OR
SXT >=2000 OR
ER  >=11.6
These figures were based on my personal collection.

999_Springs wrote:Has the #77 from the top1465 completely disappeared out of sight from everything regarding the hardest puzzles
If the filter above was applied the #77 would have disappeared long long ago .... Many entries generated in the patterns game are much more difficult and still below the standards of the hardest.
User avatar
tarek
 
Posts: 3762
Joined: 05 January 2006

Postby briturner » Tue Jan 13, 2009 1:45 pm

Greetings,

As a followup question, what do you do about variations with better scores?

For example, take coly013. Number substitution, swapping rows and col, and rotation can produce the same logical puzzle with higher scores:
Code: Select all
suexratt test.txt 10000 2
rating:   6610 ,    3169 , ....9..5..1.....3...23..7....45...7.8.....2.......64...9..1.....8..6......54....7
rating:   6986 ,    3149 , 3.....9.6.4.2...8.....6.....5.8...2...9...3.7.....7....1..42..........1.5.81.....
rating:   6571 ,    3365 , .....2..5..67..4.......9..8.7..9....6..4..7...1.....8..6.3..1..3.......24....5...
rating:   6101 ,    2768 , 1..5.....2......3...4.6.1....6..7.....8.....94...8.2.......9..7.4..1.6.......5..3
totalnodes:262707078

Do you just note the higher score, or make a separate entry? I also noted that even changing the puzzles before the puzzle under test can alter the rating (at least with suexratt). I can show the 2nd puzzle rating to at least 7000, and the 3rd to 3399 for the second score.

I do not know how to run Q1 or Q2 scores, or min-lex versions, but I assume gsf's sudoku suffers from the same issues.

thanks
brit
briturner
 
Posts: 3
Joined: 12 January 2009

Postby gsf » Tue Jan 13, 2009 3:28 pm

briturner wrote:As a followup question, what do you do about variations with better scores?

For example, take coly013. Number substitution, swapping rows and col, and rotation can produce the same logical puzzle with higher scores:
Code: Select all
suexratt test.txt 10000 2
rating:   6610 ,    3169 , ....9..5..1.....3...23..7....45...7.8.....2.......64...9..1.....8..6......54....7
rating:   6986 ,    3149 , 3.....9.6.4.2...8.....6.....5.8...2...9...3.7.....7....1..42..........1.5.81.....
rating:   6571 ,    3365 , .....2..5..67..4.......9..8.7..9....6..4..7...1.....8..6.3..1..3.......24....5...
rating:   6101 ,    2768 , 1..5.....2......3...4.6.1....6..7.....8.....94...8.2.......9..7.4..1.6.......5..3
totalnodes:262707078

Do you just note the higher score, or make a separate entry? I also noted that even changing the puzzles before the puzzle under test can alter the rating (at least with suexratt). I can show the 2nd puzzle rating to at least 7000, and the 3rd to 3399 for the second score.

I do not know how to run Q1 or Q2 scores, or min-lex versions, but I assume gsf's sudoku suffers from the same issues.

suexrat by definition is an average over n equivalent permutations of the input puzzle
because of this it can be sensitive to the original input puzzle orientation

the -q2 rating of my solver was designed to be resistant to input puzzle permutations
this command line runs the q2 rating on 100 permutations of the first puzzle above
Code: Select all
sudoku -q2 -J100 -f%r ....9..5..1.....3...23..7....45...7.8.....2.......64...9..1.....8..6......54....7

one run (each run starts with a different pseudo random seed) it produced this range of ratings
Code: Select all
   6 98818
  21 98819
  23 98820
  29 98821
  19 98822
   3 98823

-q2 sensitivity on input puzzle orientation tends to be in the noise
gsf
2014 Supporter
 
Posts: 7306
Joined: 21 September 2005
Location: NJ USA

Postby ronk » Tue Jan 13, 2009 4:54 pm

gsf wrote:suexrat by definition is an average over n equivalent permutations of the input puzzle
because of this it can be sensitive to the original input puzzle orientation

Ignoring digit swaps and the transpose (diagonal reflection), there are only 46,656 permutations,so I sure would have expected an average of 10,000 permutations to show better results.

Evidently digit swaps and diagonal reflection are significant factors for suexrat.
ronk
2012 Supporter
 
Posts: 4764
Joined: 02 November 2005
Location: Southeastern USA

Next

Return to General