The New Sudoku Players' Forum

[champagne]

It looks lke this would come from an uninitialized area acting in very specific conditions.

very small chances in that case to put the finger on the bug.

If this remains an isolated case, we can accept it. No chance anyway to have it fixed.

good idea about uninitialized data
I also wonder about 32 vs 64 bit arch (I run on 64 bit x86)

we have the java source
and we know its failing in the 9.7-9.8 code
and the source has already been modified for the patterns game
so the chances to fix it are > 0

I have a 32 bits processor (2 including the home server). OS is VISTA in both cases

I run the puzzle on the home server with the same result.

I did not "reload" the program but having to different batches giving mismatching ratings, I think the problem is not there.
I am using the version for windows. The file to download seems relatively old, so i shoul have that one on my computer.

I have no skill in Java, but if any comparative test can help you, don't hesitate to ask me.

champagne

PS I have also an old computer running under windows xp. I will test the puzzle on it.

[champagne]

results of the test for the puzzle alone under windows xp 32 bits

9.8 9.8 9.3

A suspicion of a problem under Vista

champagne

[Pat]

i had been using gsf's 304,533 ( and got 9.8 )

now went back to his web-site -- and got a different version !! 304,777 -- which gives 9.7

[Pat]

just to clarify --
rated just the 1 puzzle
ran both JARs on same computer ( Windows XP )

[gsf]

i had been using gsf's 304,533 ( and got 9.8 )

now went back to his web-site -- and got a different version !! 304,777 -- which gives 9.7

I just ran both versions and got 9.8
it may be a floating point aberration in in the jvm

Code: Select all

(linux.i386) java -version
java version "1.6.0_06"
Java(TM) SE Runtime Environment (build 1.6.0_06-b02)
Java HotSpot(TM) Client VM (build 10.0-b22, mixed mode, sharing)


we really need a C implementation of the SE rating code

[Pat]

on this computer (Windows XP) --


java -version
java version "1.6.0_17"
Java(TM) SE Runtime Environment (build 1.6.0_17-b04)
Java HotSpot(TM) Client VM (build 14.3-b01, mixed mode, sharing)

"Java 6 Update 20 is ready to install" -- should i try it?

[Pat]

i had been using gsf's 304,533 ( and got 9.8 )

now went back to his web-site -- and got a different version !! 304,777 -- which gives 9.7

just did an experiment using lksudoku's new version

rating just the 1 puzzle

2..3....5.....473.....9..1......1.9.....7.4...6.8....34.5......6..5......79.....8

i got

9.7/9.7/9.3

but rating it as first and 3rd in a set of 4 puzzles

2..3....5.....473.....9..1......1.9.....7.4...6.8....34.5......6..5......79.....8
6..4....2.....573.....9..1......1.9.....7.5...4.8....34.5......8..2......63.....8
2..3....5.....473.....9..1......1.9.....7.4...6.8....34.5......6..5......79.....8
2..3....6.....473.....9..1......1.4.....7.9...6.2....55.9......8..5......32.....8

it becomes 9.8

9.8/9.8/9.3
9.7/9.7/9.3
9.8/9.8/9.3
9.7/9.7/9.3

[ronk]

lksudoku's fix of a uniqueness bug shouldn't be directly related to that. I just ran the same sequence of puzzles using the new Explainer serate v1.2.2 and got ED=9.7/9.7/9.3 for all four. Ditto using serate v1.2.1.

'java -showversion' printed ...

java version "1.6.0_21"
Java(TM) SE Runtime Environment (build 1.6.0_21-b07)
Java HotSpot(TM) 64-Bit Server VM (build 17.0-b17, mixed mode)

I'm inclined to believe gsf's conjecture that this is due to a java floating point anomaly.

[lksudoku]

i had been using gsf's 304,533 ( and got 9.8 )

now went back to his web-site -- and got a different version !! 304,777 -- which gives 9.7

just did an experiment using lksudoku's new version

rating just the 1 puzzle

2..3....5.....473.....9..1......1.9.....7.4...6.8....34.5......6..5......79.....8

i got

9.7/9.7/9.3

but rating it as first and 3rd in a set of 4 puzzles

2..3....5.....473.....9..1......1.9.....7.4...6.8....34.5......6..5......79.....8
6..4....2.....573.....9..1......1.9.....7.5...4.8....34.5......8..2......63.....8
2..3....5.....473.....9..1......1.9.....7.4...6.8....34.5......6..5......79.....8
2..3....6.....473.....9..1......1.4.....7.9...6.2....55.9......8..5......32.....8

it becomes 9.8

9.8/9.8/9.3
9.7/9.7/9.3
9.8/9.8/9.3
9.7/9.7/9.3

I have now found and fixed your described bug, the bug is not a floating point problem nor is it an uninitialized memory variable

The bug is caused due to the uncertain ordering of a HashMap, a HashMap in java does not guarantee that the order of the keys will remain the same for every run. For the puzzle above

2..3....5.....473.....9..1......1.9.....7.4...6.8....34.5......6..5......79.....8

When solving, there is a potential which can be removed by two different sets of Dynamic Contradiction Forcing Chains
When SE can remove a potential using Dynamic Contradiction Forcing Chains, it removes it after the first such contradiction is found. However, due to HashMap ordering of the hints used by the chains, sometimes it finds one of the contradiction chains first, and sometimes the other

I have created a fix for the problem by sorting the hints used during the finding of the contradiction chains, after this hints sorting, anytime the above puzzle is rated, it finds the same contradiction chains first, and thus it produces the same rating for every different run

The fixed version can be found at http://sites.google.com/site/sefixshare/files/FIXED2SudokuExplainer.jar

This version will always give the same result rating for the above puzzle and every puzzle with this hashset ordering contradiction chains rate changing bug

I numbered this version as 1.2.3.0

[GouinJP]

Hi

Code: Select all

java -Xrs -Xmx500m -cp SudokuExplainer.jar diuf.sudoku.test.serate --format=%r/%p/%d --input=test.txt --output=ratings.txt

What can I change in this command line to have the puzzle AND the rating?

Thank you

JPS

[champagne]

Hi

Code: Select all

java -Xrs -Xmx500m -cp SudokuExplainer.jar diuf.sudoku.test.serate --format=%r/%p/%d --input=test.txt --output=ratings.txt

What can I change in this command line to have the puzzle AND the rating?

Thank you

JPS

I use that command

-f"%g ED=%r/%p/%d"

instead of your

--format=%r/%p/%d

This should be what you are looking for

[Pat]

%g ( as shown by champagne )

i tend to send the output to a CSV:

Code: Select all

java -Xrs -Xmx500m -cp SudokuExplainer.jar diuf.sudoku.test.serate --format=%g,%r,%p,%d --input=puzzles.TXT --output=ratings.CSV


[GouinJP]

Hi

I found this.

000001002003004560070060030000200090009006005460098000020000900037600008500080010

ED skfr =4.7/2.0/2.0

ED SExplainer=7.1/2.0/2.0

If it can help.

Best regards

JPS
.

[champagne]

Hi

I found this.

000001002003004560070060030000200090009006005460098000020000900037600008500080010

ED skfr =4.7/2.0/2.0

ED SExplainer=7.1/2.0/2.0

If it can help.

Best regards

JPS
.

skfr and SE don't have exactly the same priority in multiple UR handling. This is a classical consequence (I did not check that specific example).

The most common deviation in that area is a +0.1 in skfr rating due to the so-called "K rule" not reproduced in skfr.

[Pat]

.76...1..98.7.......1.....2.9.5....8.....46......2..7.7...6.3.......5.1...38....9
4.6 = skFR
9.0 = Sudoku Explainer (Nicolas Juillerat): Dynamic Region Forcing Chain
+4.4
23 cells given ( Layout 212 minus r3c1 )