scan solution grids for 17 clues as of blue

Programs which generate, solve, and analyze Sudoku puzzles

Re: scan solution grids for 17 clues as of blue

Postby champagne » Mon Jul 22, 2019 7:21 pm

New milestone in this project.

Now , bands1 index 0 to 7 and 330 to 415 are closed.
all known 17 came, no new 17 appeared.
Band 8 is half covered,
band 189 is covered at 90%

This is more or less 6% of the total band 3 count.
I'll check carefully in the next days the index 189 status.

Next milestone (in one month??) should cover at minimum bands 1 index 8;9 31

As no fee core joined the project so far, the process remains too slow for my available power.

Meantime, I checked directly that no 17 with 2 clues in a band exists. (looking for 18 clues with a band having 2 clues)
Doing so, I have seen neglected puzzle properties (AFAIK) and a possible new design of the current search.

Implementing and testing such a design is to-day my priority in the limit of the time dedicated to the project. I could get a first feeling of the feasibility in 2 or 3 weeks.

I used the third post in this thread to give information on the new design. This post will be edited when something of interest shows up
champagne
2017 Supporter
 
Posts: 6949
Joined: 02 August 2007
Location: France Brittany

search of 17 with 665 distribution milestone

Postby champagne » Wed Aug 14, 2019 9:34 am

97 bands are now closed or nearly closed for one band(index 0_9 189 and >=330)

up to now, in each closed band
all known puzzles are here
no new 17 came up.

This is about 8% of the total number of solution grids to scan.

Results for the band index 189

Results for the band index 189 are interesting to decipher.
This band was scanned mainly because it has the highest number of expected 17 with distribution 665
It is in the middle of the index. At this point, 88% of the bands 2 to scan are there and 97% of the solution grids to scan have been covered (if all previous bands 1 are closed)
The average number of bands 3 per band 2 is 3.11

The bad news for the future is that the worst cases scanned on my AMD phenom6 with 7 cores fully occupied required about 17 seconds per band 2 (nearly 6 seconds per solution grid)

I don't have benchmarks on the Intel i7 machines, but anyway, this is on the high side of my expectations.

Alternative process

The good news is that the alternative process shows a potential to divide by 2 or more this run time as far as I can see out of the first tests.

The fresh code produced it's first 17 recently. In this code, the band 2 is expanded in a way very close to the band 3 expansion in the current process. Some changes have been made also in the band 3 process to diminish the run time. Some of these accelerators have to be implemented in the second band as well.

In 2 or 3 weeks, all needed tests should have been done and I'll have a better idea of the true potential of this process.

In the new process, many pieces are common to the old one, However, I thought necessary to open a specific repository https://github.com/GPenet/sk_s17

Next steps

Waiting for the results of the alternative process, the scan goes ahead with bands index 10_12 ( more if necessary) and 31 (another band with many expected known 17).
Next milestone will come with the results of the alternative process. This could come this month.
champagne
2017 Supporter
 
Posts: 6949
Joined: 02 August 2007
Location: France Brittany

Previous

Return to Software