The New Sudoku Players' Forum

[ronk]

Sorry, i dont even know, what EP and ED stands for. One of them is the rating of the first step, if i remember right.

My Explainer version just gives me the rating and the techniques used.

[Added:] Ah, i see, there is an explanation in the Patterns game intro. So whats the command line for serate you would like ?

Congratulations eleven. Your puzzle with its ER/EP/ED rating:

Code: Select all

+-------+-------+-------+
| . . 9 | . . . | 4 . . |
| . 7 . | 3 . . | . 2 . |
| 8 . . | . 6 . | . . 7 |
+-------+-------+-------+
| 1 . . | 8 . . | . . 6 |
| . . . | . 1 . | . 7 . |
| . . . | . 5 6 | . . . |
+-------+-------+-------+
| 3 . . | . . 5 | . . 1 |
| . 4 . | . . . | . 9 . |
| . . 2 | . . . | 7 . . |
+-------+-------+-------+

eleven's ED=11.7/11.7/11.6

The rating may apparentedly be affected by the BUG recently fixed by lksudoku here.

[ronk]

I had a quick look at the path proposed by the solver for that puzzle.

No SK loop, no pure EXOCET, but something very close (I voluntarily don't say where).

Would it be this or something else?

Hidden Text: Show

____

Code: Select all

eleven's ED=11.7/11.7/11.6
..9...4...7.3...2.8...6...71..8....6....1..7.....56...3....5..1.4.....9...2...7..

At the outset:
 56(2)     1356-2  9      | 1257     278       1278    | 4       13568  358
 56(4)     7       156-4  | 3        489       1489    | 15689   2      58(9)
 8         135(2)  135(4) | 15(249)  6         -1(249) | 135(9)  135    7
--------------------------+----------------------------+-------------------------
 1         2359    3457   | 8        23479     23479   | 2359    345    6
 -56(249)  235689  34568  | (249)    1         2349    | 23589   7      358(249)
 (2479)    2389    3478   | (2479)   5         6       | 12389   1348   38(249)
--------------------------+----------------------------+-------------------------
 3         68(9)   68(7)  | 6(2479)  -8(2479)  5       | 68(2)   68(4)  1
 56(7)     4       1568-7 | 1267     2378      12378   | 23568   9      358(2)
 56(9)     1568-9  2      | 1469     3489      13489   | 7       34568  358(4)

16 Truths = {249R3 2479R7 2479C1 249C9 56N4}
26 Links = {249r5 2479r6 2479c4 56n1 37n4 7n5 3n6 56n9 2b19 4b19 7b7 9b37}
8 Eliminations --> r5c1<>56, r1c2<>2, r2c3<>4, r3c6<>1, r7c5<>8, r8c3<>7, r9c2<>9

[eleven]

Thanks Ron,

so i will use the fixed version. Hope i can generate some more interesting puzzles for your solvers. For manual players they are just boring hard.

btw if the Patterns game results should be more interesting for manual players, the reverse of your rating for puzzles with the same ER would be better. Manual players like the hardest step with as few candidates as posssible.

[champagne]

I had a quick look at the path proposed by the solver for that puzzle.

No SK loop, no pure EXOCET, but something very close (I voluntarily don't say where).

Would it be this or something else?

My solver does not extract so complex SLG's, but the weak point is the same, r56c4 working as a "nearly EXOCET".

Regarding the SLG, it has just seen the potentiel using rookeries 2479. Your SLG (using no AUD) cover nearly all the potential.

I am travelling to day so I'll stop working on sudoku for a while, but I assume that my solver path has AURs
It starts with scenario analysis based on r56c4.

champagne

[champagne]

Thanks Ron,

so i will use the fixed version. Hope i can generate some more interesting puzzles for your solvers. For manual players they are just boring hard.

btw if the Patterns game results should be more interesting for manual players, the reverse of your rating for puzzles with the same ER would be better. Manual players like the hardest step with as few candidates as posssible.

It's not so difficult to extract puzzles manual players could like to solve, but this can not be done using serate output.

serate rates in the same way all puzzles having the same "hardest point".
What you are looking for has, for my solver, a short print compared to the "hardest point". So, based on the print size, it is easy to do it.

The final questions are:

who is interested in getting such puzzles
where should they be posted.

champagne

[Pat]

so i will use the fixed version

yes, of course lksudoku's fixed version is better


but when comparing ratings --
we should all be using the same version


so, are we all prepared to change over to lksudoku's newest version ?

[eleven]

The final questions are:

who is interested in getting such puzzles
where should they be posted.

Good question.
I have offered two newspapers hand selected puzzles with a nice pattern and solution path (not hard, max. UR), but i did not even get an answer (they continued to bring those boring naked triple puzzles as "tricky").

[eleven]

Your puzzle with its ER/EP/ED rating:

Code: Select all

+-------+-------+-------+
| . . 9 | . . . | 4 . . |
| . 7 . | 3 . . | . 2 . |
| 8 . . | . 6 . | . . 7 |
+-------+-------+-------+
| 1 . . | 8 . . | . . 6 |
| . . . | . 1 . | . 7 . |
| . . . | . 5 6 | . . . |
+-------+-------+-------+
| 3 . . | . . 5 | . . 1 |
| . 4 . | . . . | . 9 . |
| . . 2 | . . . | 7 . . |
+-------+-------+-------+

eleven's ED=11.7/11.7/11.6

The rating may apparentedly be affected by the BUG recently fixed by lksudoku here.

Hm, there seems to be still a problem or did i something wrong ?
I rerated the canonicalized version (first of many puzzles):

Code: Select all

java -Xrs -Xmx500m -cp FIXED2SudokuExplainer.jar diuf.sudoku.test.serate -f%g;%r;%p;%d;%e -itest.txt -otestr.txt
1....6.8....7..1........5.6..9.4.....7.2...3.8....76..3....1..5.4.9.......2.7....;11.8;11.8;11.6;56m34s


[ronk]

Hm, there seems to be still a problem or did i something wrong ?
I rerated the canonicalized version (first of many puzzles):

Code: Select all

java -Xrs -Xmx500m -cp FIXED2SudokuExplainer.jar diuf.sudoku.test.serate -f%g;%r;%p;%d;%e -itest.txt -otestr.txt
1....6.8....7..1........5.6..9.4.....7.2...3.8....76..3....1..5.4.9.......2.7....;11.8;11.8;11.6;56m34s


Is "FIXED2" a typo? For this bug, the "FIXED3" version is required. But even then, it's still possible for some morphs to give different ratings. Always rating the row minlex canonicalization ("c14n") would get around this problem. It's been suggested a few times for the Patterns Game, but not yet adopted.

I'll check the row minlex rating later. Right now my CPU is choking on the current patterns game.

[eleven]

Is "FIXED2" a typo?

This was the version i found from the link you gave, where this problem should be fixed.

I have searched the neighbourhood of the known hardest now for a week and i found over 30 more 11+ puzzles (but no harder one). This was a test for a "hardest sudokus gotchi", not bad.
But the hard work now would be to find new areas with extremely hard puzzles. With a random starting set it probably would take weeks to get the firrst 11+.

[ronk]

Is "FIXED2" a typo?

This was the version i found from the link you gave, where this problem should be fixed.

Sorry. I posted that link because it had the best description of the bug ... but I assumed from lksudoku's post here that FIXED1, FIXED2 and FIXED3 addressed three bugs in the sequence listed there. I'm sure you know the saying about the word "assume."

Suggest you get FIXED3 from here. That it fixes all three bugs is a safer assumption.

[tarek]

Suggest you get FIXED3 from here. That it fixes all three bugs is a safer assumption.

I have been away for a while but I'm looking to update the Hardest database ....
Is everybody happy to use Fixed3 for SE rating ?
I will use gsf sudoko's row-normal minlex version for rating at all fronts.
I will try to target next weekend to do some work.

tarek

[champagne]

Suggest you get FIXED3 from here. That it fixes all three bugs is a safer assumption.

Is everybody happy to use Fixed3 for SE rating ?
tarek

Let me answer in my own way.

1) Sudoku Explainer is the referee for the pattern game.
Al players must use the same version as gremlin in that game.

2) Out of the pattern game, SE rating is not a key issue.
As SE gives a "well known" ??? rating, it's good to have it for "hardest puzzles", although this is the weakest part of the program.
For such ratings, it's better to use a version "as clean as possible", but the rating problems encountered in SE are not so common.

3) An attempt is made to produce a C++ version of SE much faster. Working on that issue, plasson and I have found other bugs in the rating that shoul be fixed as well.

I use already the FIXED3 version of SE in my tests to produce the C++ version of SE, but I'll stick for sure to the common one for the game.

champagne

[eleven]

I have been away for a while but I'm looking to update the Hardest database ....

Here are the 11.5+ puzzles, which i found in {-1+1} clusters of the known ones.
The first one also got 11.8 with FIXED3 rating of the canonicalized puzzle.

Code: Select all

11.8 11.8 11.6  1....6.8....7..1........5.6..9.4.....7.2...3.8....76..3....1..5.4.9.......2.7....
11.7 11.7 11.2  ....567.....1.....6..37......4.....13...6.9...8.....2.5....73....2.....8.1.....4.
11.7  1.2  1.2  1...5...9.......3..79........7.1...65....42..6..8.......5.6...7.....2.4....3..8..
11.5 11.5 11.2  ....567.....1.9...6..37......4.....13...6.9...8.....2.5....73....2.....8.1.....4.
11.5 11.5 11.2  1.......7.2.4...6...3...5...4.2.9.......46.9....5.......7...1...8.9...2.5.......3
11.5 11.5 10.6  .2...67......8.1.2...2..4.6..85......1...2..79...3....3..........59......4...16..
11.5  1.2  1.2  ..3..6...4.....2...9..7..1....59........1...8.6...7.5..1.7...9...2.....38.....4..

I didn't come far with looking for hard puzzles from randoms. However i found 11.0's and 11.1's with unfamiliar patterns. One example:

Code: Select all

 +-------+-------+-------+
 | . 2 . | . 5 . | 7 . . |
 | 4 . . | . . 9 | 2 . . |
 | . . . | . . . | . . 5 |
 +-------+-------+-------+
 | . . . | . . . | . 7 4 |
 | . 9 . | 6 . . | . . . |
 | 8 . 4 | . . 1 | 9 . . |
 +-------+-------+-------+
 | 3 . . | . 1 . | 4 . . |
 | . 4 . | . . 3 | . 2 8 |
 | . . 8 | . . . | 3 . . |
 +-------+-------+-------+
 24 clues, 11.1/11.1/9.8

[tarek]

Thanks champagne & eleven,

I was under the impression that FIXED3 was a C++ version of SE ... It was actually a bug fix JAVA version (thanks to iksudoku) ....

That means that ultra hard sudokus would still need long rating times ...

I'm therefore (for SE ratings) will rely on the poster's ratings when adding to the database and would check for isomorphs & other (q1,q2,....) ratings.

When a quicker rating version comes out hopefully ... or when I'm extremely motivated to use the JAVA version , I will attempt rating the entire collection to verify the posted ratings.

tarek