The New Sudoku Players' Forum

Website · by **denis_berthier** » Fri Mar 08, 2024 8:49 am

blue wrote:I have a big list, locally, with ~1.5G puzzles in it.
I extracted all of the T&E(2) puzzles from it, but there were only 780 of them.
Close to half of them were solvable with basics ... due to the bit with naked triples, I think.
I could put that list in a post, if you like ... FWIW

Ah yes, It's the T&E(2) part I'm interested in. Thanks in advance. I wonder if there'll be a B3B.
.

by **1to9only** » Fri Mar 08, 2024 9:05 am

denis_berthier wrote:BTW, do you have any reference to a list of 2,000,000 18c. I copied it from somewhere, but I can't find it again.

http://forum.enjoysudoku.com/minlex-routine-t39261.html#p307828

by **blue** » Fri Mar 08, 2024 9:13 am

The list of 780 T&E(2)'s ...

Hidden Text: Show

Website · by **denis_berthier** » Fri Mar 08, 2024 11:00 am

.
1to9only
Thanks for the reference.

Blue,
Thanks for the list.
There's no B3B puzzle. Here's the detailed BpB classif (it took SHC 11s to compute it):

Hidden Text: Show

.

by **blue** » Fri Mar 08, 2024 1:29 pm

B1B : 496 puzzles
B2B : 284 puzzles

11 seconds ... wow !
SE takes ~8 minutes.

Website · by **denis_berthier** » Fri Mar 08, 2024 2:51 pm

blue wrote:11 seconds ... wow !
SE takes ~8 minutes.

The purpose of SHC was to be faster than SER. I think more optimisations would still be possible.
.

by **JPF** » Fri Mar 08, 2024 5:32 pm

Hi blue,

You wrote:I have a big list, locally, with ~1.5G puzzles in it.

and here 4 years ago :

... my estimate of 1,934,000,000 18C puzzles

do you conduct a comprehensive search for the 18C puzzles, and if so, what is its real progress level?

JPF

by **blue** » Fri Mar 08, 2024 8:52 pm

Hi JPF,

do you conduct a comprehensive search for the 18C puzzles, and if so, what is its real progress level?

No, it isn't comprehensive.
I'll give some details anyway ...

I haven't run the main "search" code, for about a year now.
When I stopped, I was getting ~2.16 million new puzzles per day, with the rate falling disappointingly fast.
It had dropped from ~102000 per hour, to ~90000 per hour, in just 10 days.

At this point the search runs like this:
Each hour ...

For a few seconds: process the new 18's from the previous hour ... {-1,+1} + a few other tricks.
For the rest of the time, run {-1,+1} on 19's looking to produce 18's from non-minimal 19's.
Merge 18's into the master list, looking for new ones to handle in (1)

The falling rate is due to:

The fraction of 18's produced from 19's that are new, is dropping.
It's at ~10% now, even though ~21% of the 18's are still missing.
The new 18's don't "fan out" in {-1,+1} testing, as much as they used to.

At one point I did do a comprehensive search for 18's where (at least) one band/stack has 8 clues or more.
That took (43 days) x (12 cores), and it produced 323,897,702 ED puzzles.

The last breakdown by (B,S) class, with B/S = # of clues in most populated band/stack (B>=S), was:

Code: Select all: +-----------+-----------+----------+------+ | S=6 | S=7 | S=8 | S=9 | +------+-----------+-----------+----------+------+ | B=6 | 59721848 | | | | puzzles : 1527664516 | B=7 | 448837008 | 695207958 | | | grids : 789213426 +------+-----------+-----------+----------+------+ | B=8 | 83546486 | 221554111 | 11057553 | | (this section is complete) | B=9 | 2502213 | 4821125 | 387001 | 2487 | | B=10 | 11482 | 14387 | 817 | 23 | | B=11 | 17 | | | | +------+-----------+-----------+----------+------+

by **JPF** » Fri Mar 08, 2024 10:45 pm

Thank you for this detailed information.
Last year, I also started generating 18C puzzles. I stopped due to a lack of time, CPU resources, and patience. The performance I achieved was far from yours.

Edited after blue's comments, see post below.
So, I have my own file of 102,102,520 minimal 18C puzzles that can be found using this link : https://drive.google.com/file/d/1eglCCE ... sp=sharing
It's a compressed file of 1.09 gigabytes that loads within a reasonable timeframe.
I would be interested to know what proportion is included in your catalog.

JPF

by **champagne** » Sat Mar 09, 2024 5:12 am

Hi Blue,

blue wrote:At one point I did do a comprehensive search for 18's where (at least) one band/stack has 8 clues or more.
That took (43 days) x (12 cores), and it produced 323,897,702 ED puzzles.

This looks like a variant of the 17 clues search pass1 giving around 7.2 18s per second,
Having a ratio 5 472 730 538 / 1 934 000 000 = 2.83 ED grids per 18, this leads if I am right to something between

an average of 20 ED solution grids processed per second and per core, using this ratio
an average of 123 ED solution grids processed per second and per core if all ED solution grids have been processed

It seems, from your post, that the 123 / second is the correct number; a remarkable value

Do you have an estimate of the DLL mean run time per solution grid?

by **coloin** » Sat Mar 09, 2024 1:30 pm

blue wrote:.....At one point I did do a comprehensive search for 18's where (at least) one band/stack has 8 clues or more.
That took (43 days) x (12 cores), and it produced 323,897,702 ED puzzles.

very good and am sure Jim would be pleased at this effort !

so have you found all puzzles with clues in band of .......{558},{468},{378},, {459],[369],,, {4410},{3510} !?

and like the 17C search the next level with max 7 clues in a band is a league above in difficulty.

The 666 search can be split in to only a few clue patterns including , 222222222,321213123,321222123,and 330033303 and 420042204 and perhaps 411141114.
I found many [12000] of that last [411141114] pattern ... as ever its not exaustive...

Code: Select all: +---+---+---+ |..1|...|...| |.2.|...|...| |.34|5..|..6| +---+---+---+ |7..|..8|...| |...|.4.|..2| |...|.19|...| +---+---+---+ |...|...|91.| |...|...|5.4| |6..|7..|...| +---+---+---+

All 111 puzzles with 288 pattern have been found
There are no puzzles with 510105051

by **champagne** » Sat Mar 09, 2024 2:40 pm

Hi coloin,

coloin wrote:All 111 puzzles with 288 pattern have been found

I lost my own file and the link to this topic. can you help me

by **coloin** » Sat Mar 09, 2024 3:35 pm

Link to the 111 puzzles here

by **blue** » Sat Mar 09, 2024 8:30 pm

JPF wrote:So, I have my own file of 102,955,877 18C puzzles that can be found using this link : https://drive.google.com/file/d/11Viv4aEYtLjSvP2G8yJ1hZYwPjlP6R8l/view?usp=sharing.
It's a compressed file of 1.09 gigabytes that loads within a reasonable timeframe.
I would be interested to know what proportion is included in your catalog.

Your list includes 853357 non-minimal puzzles (out of 3142658), and mine only includes minimals.
After that, there are only 28777 that aren't in my list.

---

champagne wrote:This looks like a variant of the 17 clues search pass1 giving around 7.2 18s per second,
Having a ratio 5 472 730 538 / 1 934 000 000 = 2.83 ED grids per 18, this leads if I am right to something between

an average of 20 ED solution grids processed per second and per core, using this ratio
an average of 123 ED solution grids processed per second and per core if all ED solution grids have been processed

Yes, yes, and yes.

Do you have an estimate of the DLL mean run time per solution grid?

Well, it's ~5 sec per grid at 4.0GHz, but I think you probably knew that (at some point).
Did you mean the mean time for a DLL that's been hacked to only do the "<= 10 clues in bands 1&2" part of the search ?

---

Colin wrote:so have you found all puzzles with clues in band of .......{558},{468},{378},, {459],[369],,, {4410},{3510} !?

Yes, and 17 (3,4,11)'s !

and like the 17C search the next level with max 7 clues in a band is a league above in difficulty.

Too true, unfortunately.

by **champagne** » Sat Mar 09, 2024 9:52 pm

Hi Blue,

blue wrote:Well, it's ~5 sec per grid at 4.0GHz, but I think you probably knew that (at some point).
Did you mean the mean time for a DLL that's been hacked to only do the "<= 10 clues in bands 1&2" part of the search ?

Your answer is in line with my expectation. And yes, I had in mind the enumeration mode without any limit in the number of clues in bands 1+2.

blue wrote:an average of 123 ED solution grids processed per second and per core if all ED solution grids have been processed[

Yes

This an amazing performance. Far from the pass1 (may be 30 times faster) run on my side. I did not reach your performance so far for the 18 enumeration, but we are in the same order of magnitude.

The New Sudoku Players' Forum

17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update

Re: 17-clue and 18-clue Sudoku update