The New Sudoku Players' Forum

[champagne]

Code: Select all

..1...2..
.3...4.5.
2.......1
.....6.3.
....1....
.5.7.8...
8.......9
.7.3...4.
..2...1.. ED=8.8/1.2/1.2

[champagne]

::: comment :::

::: adjustment 2012-03-11+08:31:00-0000 champagne dup-puzzle-tie 0 :::

I guess my 9.7 is a morph of the 9.8 from patrice that was missing in my cache

just a chance that it came within the delay authorised


champagne

[dobrichev]

::: comment :::

I have a vague recollection that rating this puzzle all those years ago took 2.5 hours at 1.6GHz. Does anyone have any idea why it should take so much time now, at 2GHz?

Thanks for any hints.

Mike Metcalf

It could be due to a different
- combination of CPU architecture, chipset, and version of Java machine
- morph of the rated puzzle.

I posted some observations on this pattern in the"related" thread.

[Patrice]

Code: Select all

. . 3 . . . 5 . .
. 8 . . . 1 . 4 .
5 . . . . . . . 2
. . . . . 8 . 1 .
. . . . 2 . . . .
. 7 . 3 . 5 . . .
6 . . . . . . . 5
. 1 . 4 . . . 7 .
. . 2 . . . 3 . .   ED=8.3/1.2/1.2

Regards

[champagne]

Code: Select all

..1...2..
.3...1.4.
5.......6
.....2.1.
....6....
.4.5.7...
7.......8
.9.4...3.
..2...7.. ED=7.1/1.2/1.2

last in my cache and I see no chance to get one more

[champagne]

::: comment :::

mladen wanted to know how I found the 11.6

I have just to say I have been lucky.

I started several parallel tracks to generate a first lot of seeds, including a +-5 search on the first puzzle.

All failed except 2 puzzles coming out of the +-5 process.

I then started a +-6 search on the 2 puzzles and got some results including the 11.6 and 11.8

As the process is relatively long, glen found the 11.8 before I got it

champagne

[Patrice]

Code: Select all

. . 1 . . . 8 . .
. 3 . . . 9 . 5 .
8 . . . . . . . 2
. . . . . 3 . 4 .
. . . . 1 . . . .
. 8 . 9 . 5 . . .
7 . . . . . . . 6
. 9 . 8 . . . 3 .
. . 5 . . . 9 . .   ED=6.6/1.2/1.2

Regards

[champagne]

::: comment :::

we should have received the notification for the end

I made a test for a full scan of the puzzle.

I unhappily filtered puzzles starting with a single, but I did not find any new puzzle with ED 3.4

it's not 100% sure (I have other options in that search), but I bet Patrice has a data base containing all puzzles valid in that pattern (112 if I am right)

Which means that the vicinity process can catch in some conditions all puzzles.

champagne

[dobrichev]

my collection reached the same limit of 112.

[Patrice]

::: comment :::

I found 112 puzzles !
I can send you this collection, or post it here, if you want to verify if we have the same ones

basically i started with a several hours delay, found rapidly 2 puzzles in the vicinity of the initial one.
Then I needed a {-6+6} scan to find new ones. The others came in several hours (4 processors 3 Ghz, vicinity search written in C++)
On Sunday afternoon after 12 hours without no new ones, I stoped everything.

Regards
Patrice

[m_b_metcalf]

::: comment :::

I made a test for a full scan of the puzzle.

I unhappily filtered puzzles starting with a single, ...

Well, with patrice running with 4x3GHz, gsf with 8x3.4, and champagne on a 64-bit 6-core i7, I can see I'm in real trouble with my puny 2x2. Exhaustive searches are just not feasible. However, in this game I of course used the 11.8 as an intial seed. After making some vicinity searches, I started an ab initio search for pearls and found the 11.6 very quickly. In fact, I found it again in a subsequent run. The real problem was finding the initial 1.5 deal! Luckily, an ab initio run for low scores yielded the 2.3 quite quickly, giving 1.5 and 4.4 on vicinity searches. Further quite long runs failed to find any further low scores, or even to repeat the ones I already had. So it seems that my being able to start the game at all was fortuitous. Just searching for anything at all seems to find mainly 9.2 - 9.4. I suppose the rating frequencies in your collection of 112 would confirm that?

Regards,

Mike Metcalf

[dobrichev]

I can contribute with a matrix of pattern-preserving hamming distances between the puzzles. I have to do some additional coding to do it, so if there is interest let me know and I'll do it in few hours.

[champagne]

I can contribute with a matrix of pattern-preserving hamming distances between the puzzles. I have to do some additional coding to do it, so if there is interest let me know and I'll do it in few hours.

Hi mladen,

this seems to be the continuation of another post, but I can't figure out which one
may-be the content needs more explanations

champagne

[champagne]

::: comment :::

:

Well, with patrice running with 4x3GHz, gsf with 8x3.4, and champagne on a 64-bit 6-core i7, I can see I'm in real trouble with my puny 2x2.

I suppose the rating frequencies in your collection of 112 would confirm that?

Regards,

Mike Metcalf

Hi mike,

I know that some are working with limited power, so I am not especially proud if I win a game toward them.
More power allows to do much more, especially in the search of hardest, so I needed that power;

In game 169, one indication: 94 puzzles in my data base (110) are rating 9.0 and more. They seem to be equally spread in several ED starting at 10.7

champagne

[gsf]

::: comment :::

I unhappily filtered puzzles starting with a single, ...

Well, with patrice running with 4x3GHz, gsf with 8x3.4, and champagne on a 64-bit 6-core i7, I can see I'm in real trouble with my puny 2x2.

I pulled back to 5x3.4, but those use very old strategies and the java se,
and are not very productive compared to other players who have upgraded their processes

I manually add one more 3.4Ghz that does an exhaustive search on the pattern minus 5 or 6 clues
that one is productive but it rides on the back of giants right now

seeing other approaches is good and will hopefully lead to new hybrid strategies

we can probably continue in the Patterns game strategies thread