The New Sudoku Players' Forum

by **blue** » Sat Jun 29, 2013 1:04 am

Leren wrote:Thus endeth the lesson !!

"A round of applause spreads through the room" ...

by **champagne** » Sat Jun 29, 2013 2:06 am

Leren wrote:
Code: Select all
Champagne wrote: I refer strictly to your following wording 16 unsolved cells containing a maximum of 4 candidates at the intersection of 4 rows and 4 columns

Hi Champagne, I'm not sure what this statement means either, I'll just explain how I've implemented the MSLS search process.

Leren

Hi leren,

I can tell you that I could easily design my own process out of yours and produce my own first 4x4 cells based rank 0 logic.

To keep your wording, it was a vanilla logic, so I have still to check whether I catch solutions using boxes covers, but the design is supposed to do it.

If I want to extract all puzzles having one of these rank 0 logic (I am only interested here to catch one rank 0 logic in the extraction process) I have also to extend the process from 4x4 to 4x5, what I should do to day if what I have in mind is correct.

As I wrote earlier, in your process, we see no condition on the number of candidates in the 4x4 cells matrix, so, when I red in a post from JC Van Hay that such cells should not have more than 4 candidates, I saw one way to speed up the search, that's it.
After several exchanges, including through pm, I think that this is not true, but as it is an easy check, I'll do the test. It's not a key point.

Just to let you know, my design for a 4x4 is the following, with likely some room for optimisation, but basically well adjusted to my infrastructure

Code: Select all: a) create all combinations of 4 rows or columns (126 perms) b) combine one row combination and one columns combination and see if it is a 15 or 16 unknown cells one now for a valid combination 1) collect all digits in the cells (n) 2) try all combinations 3/n-3 4/n-4 digits once for rows columns once for columns rows for each of these groups, do the row cover and the columns cover applying the following process for each digit selected . explore each band/stack containing one or more selected rows or columns . count the boxes and the rows/columns . use the rows columns cover except if the box cover gives less links.

I intend to produce the 4x5 out of the valid 4x4 adding a row or a column.

I am not expecting a strong improvement (but I can be wrong) replacing the 126 x126 perms analysis by a progressive search,but an alternative process can be

explore all 2x2
extend valid 2x2 to 3x3
extend valid 3x3 to 4x4

which is closer to your process.

I should be in a position within one or 2 days to make some performance tests

by **Leren** » Sat Jun 29, 2013 3:42 am

Hi Champagne,

It will be interesting to compare results when you have your process going.

I've tested the 3 digit base puzzles you provided but they all have 4 digit bases as well. Puzzle 498 remains the only one I know of where a 3 digit base is "necessary".

Leren

by **champagne** » Sat Jun 29, 2013 5:47 am

Leren wrote:
I've tested the 3 digit base puzzles you provided but they all have 4 digit bases as well. Puzzle 498 remains the only one I know of where a 3 digit base is "necessary".

Leren

2 tests runs to extract puzzles with rank0 logic are over. I loaded the results
here-> file gr_r0

This gives you the possibility to run more tests. You'll find close to 90 000 puzzles in that bloc.

Leren wrote:
It will be interesting to compare results when you have your process going.

sure but don't forget my code will be oriented toward extraction and will stop at the first rank 0 logic found. Exploring as you all possibilities is not my priority. My to do list is overloaded.

by **champagne** » Sat Jun 29, 2013 9:40 am

Hi again leren,

I finished the coding and started the test on the file containing 50201 puzzles rating 10.3 and more, but not in the potential hardest file.

In all puzzles, the code has seen a cell based rank 0 logic, except in 241.

I guess that most of them are outside the limits of your specifications
Here after the file of puzzles not extracted

Hidden Text: Show

by **JC Van Hay** » Sat Jun 29, 2013 11:37 am

There are very simple continuous nice loops that can be easily transformed into a Rank 0 set of cells (as for example into an ALS-XZ Rule, Double Link - eventually extended outside the 2 units containing them).

The puzzle #10 will certainly be of some interest to Leren as it contains a Rank 0 3x4 array of unsolved cells !

#10 98.76....7..5..9....4..9...6...9.3...2.....4...8......3....65.....35..94....7...11;5;10.5;1.2;1.2;fl; 35679;tr; 14;
corrected as
#10 98.76....7..5..9....4..9...6...9.3...2.....4...8......3....65.....35..94....7...1;12;5;10.5;1.2;1.2;fl; 35679;tr; 14;

[In copy-pasting a puzzle of the list, the last digit has to be dropped !]

I am pretty sure that a little(?) extension of his code will unravel all the Rank 0 set of cells in those puzzles.

by **Leren** » Sat Jun 29, 2013 11:50 am

Ch

ampagne wrote: Here after the file of puzzles not extracted

Hi Champagne, could you check that last list of puzzles you posted. There appears to be a spurious blank or an extra digit in them.

When I paste these puzzles into my solver the last digit doesn't appear.

Leren

by **champagne** » Sat Jun 29, 2013 12:08 pm

Leren wrote:Ch
ampagne wrote: Here after the file of puzzles not extracted

Hi Champagne, could you check that last list of puzzles you posted. There appears to be a spurious blank or an extra digit in them.

When I paste these puzzles into my solver the last digit doesn't appear.

Leren

sorry, but a semi-colon is missing after the last digit of the puzzle.
what appears as erased is the the start of the previous qualification of the puzzle (position 82)
1 if a row r0 was detected
2 if a column r0 was detected
4 if a "X" r0 was detected

by **champagne** » Sat Jun 29, 2013 3:09 pm

JC Van Hay wrote:
I am pretty sure that a little(?) extension of his code will unravel all the Rank 0 set of cells in those puzzles.

hi,

first of all, thanks to JC, the previous run was bugged and after having fixed it, I have about 1000 puzzles not extracted in the file.

A question to JC :

I am running (as leren I guess) a search 4x4 and 4x5 with at most one given and a shift of digits 3/4 n-3/n-4

based on your experience, what are the extension of specs to do in priority

. more than one given (easy, just increasing the runtime)
. 3x4 (as suggested in your post) not to hard but again more runtime
. why not 3x5 same remark

anything else ? assuming we stick to a base (truths) containing exclusively cells

I don't consider a 5x5 search supposed to be complementary to a 4x4

EDIT:
I am running a test with the new code for ratings in the area 9.0. First impression, it goes fast, but in that area we are losing most of the puzzles to extract. Reversely, the current process is very very slow in that area. I
t would be logic to find in majority simple rank 0 for such ratings.

by **ronk** » Sat Jun 29, 2013 6:30 pm

champagne wrote:In all puzzles, the code has seen a cell based rank 0 logic, except in 241.

What is puzzle #241?

by **champagne** » Sat Jun 29, 2013 8:09 pm

ronk wrote:
champagne wrote:In all puzzles, the code has seen a cell based rank 0 logic, except in 241.

What is puzzle #241?

the list of the 241 puzzles immediately follow the words except in 241

by **ronk** » Sat Jun 29, 2013 9:54 pm

champagne wrote:
ronk wrote:
champagne wrote:In all puzzles, the code has seen a cell based rank 0 logic, except in 241.
What is puzzle #241?

the list of the 241 puzzles immediately follow the words except in 241

That interpretation never crossed my mind.

by **ronk** » Sun Jun 30, 2013 2:18 am

champagne wrote:... cell based rank 0 logic, except in 241 [ronk edit: puzzles].

I guess that most of them are outside the limits of your specifications

The first 0-rank pattern I found (for the first of the 241 puzzles) is comprised of column truths rather than cell truths.

Code: Select all: 001 of 241 original: 98.7..6..7..6..5......84...6..5..9...35.............2.5...7..69..79....1.....57.. ;4;3;10.9;1.2;1.2;fl; 679;tr; 7; After some opening singles and an x-wing:(9)c23\r69 => r6c56<>9: 98.75.6..7..6..5...56.84.976..5..9...35.....6.......255...7..69..79...51.....57.. +-------------------------+------------------------+---------------------+ | 9 8 1234 | 7 5 123 | 6 134 234 | | 7 124 1234 | 6 1239 1239 | 5 1348 2348 | | 123 5 6 | 123 8 4 | 123 9 7 | +-------------------------+------------------------+---------------------+ | 6 124(7) 1248 | 5 1234 1238(7) | 9 1348(7) 348 | | 1248 3 5 | 1248 1249 1289(7) | 148 148(7) 6 | | 148 -14(79) 1489 | 1348 1346 -138(67) | 1348 2 5 | +-------------------------+------------------------+---------------------+ | 5 124 12348 | 12348 7 1238 | 2348 6 9 | | 2348 24(6) 7 | 9 234-6 238(6) | 2348 5 1 | | 12348 -124(69) 123489 | 12348 12346 5 | 7 348 2348 | +-------------------------+------------------------+---------------------+ 6 Truths = {6C26 7C268 9C2} 6 Links = {6r8 7r45 69n2 6n6} 9 Eliminations --> r6c6<>138, r9c2<>124, r6c2<>14, r8c5<>6

by **Leren** » Sun Jun 30, 2013 4:00 am

The first 2 all cell truths solutions I have found in Champagne's 241 puzzles list are:

1. 98.76....7..5..9....4..9...6...9.3...2.....4...8......3....65.....35..94....7...1 No 10 in the list

Code: Select all: *--------------------------------------------------------------------------------* | 9 8 1235 | 7 6 123 | 4 125 235 | | 7 136 1236 | 5 1248-3 12348 | 9 1268 2368 | |*125 356-1 4 |*128 *1238 9 |*1278 567-128 3567-28 | |--------------------------+--------------------------+--------------------------| | 6 1457 157 | 1248 9 12458 | 3 1258 258 | |*15 2 39 |*168 *138 357-18 |*178 4 5679-8 | |*145 39 8 |*1246 *1234 357-124 |*127 567-12 5679-2 | |--------------------------+--------------------------+--------------------------| | 3 1479 1279 | 12489 1248 6 | 5 278 278 | | 128 167 1267 | 3 5 128 | 268-7 9 4 | | 248-5 4569 2569 | 2489 7 248 | 268 3 1 | *--------------------------------------------------------------------------------*

MSLS 1 : Base 1248; r356 c1457 : 12 Links; 128r3 18r5 124r6 ; 5c1 6c4 3c5 7c7: 18 Eliminations as marked on the PM

The additional MSLS feature shown by this example is the use of a 3 x 4 grid.

2. 98.76....75..4.8....3..8....7.8..4....9.7..3........2..6...75.....65...2...4....1 No 15 in the list

Code: Select all: *--------------------------------------------------------------------------------* | 9 8 *124 | 7 6 *1235 | 123 *145 *345 | | 7 5 *126 | 1239 4 *1239 | 8 *169 *369 | | 1246 124 3 | 1259 129 8 | 12679 4567-19 4567-9 | |--------------------------+--------------------------+--------------------------| | 123-56 7 *1256 | 8 1239 *123569 | 4 *1569 *569 | | 124568 124 9 | 125 7 456-12 | 16 3 568 | | 134568 134 4568-1 | 1359 139 456-139 | 1679 2 5678-9 | |--------------------------+--------------------------+--------------------------| | 123-48 6 *1248 | 1239 1239-8 *7 | 5 *489 *349 | | 1348 1349 478-1 | 6 5 #139 | 379 478-9 2 | | 2358 239 578-2 | 4 2389 #239 | 3679 678-9 1 | *--------------------------------------------------------------------------------*

MSLS 1 : Base 4568; r1247 c3689 + r8c6 r9c6 : 17 Links; 45r1 6r2 56r4 48r7 ; 12c3 1239c6 19c8 39c9: 19 Eliminations as marked on the PM

The additional MSLS feature shown by this example is the use of additional cells r89c6. There is also a given in the basic 4 x 4 grid, so the truth count is 16 - 1 + 2 = 17.

Leren

by **champagne** » Sun Jun 30, 2013 6:38 am

Hi leren,

I made my own work opening the specification and found

4 puzzles with a 3x4 r0 logic
3 puzzles with 2 given

I have no addition of cell in that process, but I see how to do in a similar way to what I am doing in the other patterns.

The negative point is that now the run time is very similar to what I have in my current process, around half a second per puzzle (for that lot)

The New Sudoku Players' Forum

Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume

Re: Exotic patterns a resume