The New Sudoku Players' Forum

[Hajime]

The march 2023 version of SiSeSuSo.exe is now available at
https://sites.google.com/site/sisesuso .

• Cosmetic changes
• Bug in BatchIN fixed
• Help texts updated.

[Hajime]

The april 2023 version of SiSeSuSo.exe is now available at
https://sites.google.com/site/sisesuso .

• Manual adding/deleting givens/candidates implemented.
• User actions are added to logging and also Undo function is available.

[Hajime]

The aug 2023 version of SiSeSuSo.exe is now available at
https://sites.google.com/site/sisesuso .

Changes version 3 Aug 2023, 2023-8

• AIC expanded with Type 2 chains.
  
• Also (dis)continuous Nice Loops implemented but not extensively tested.
  By default not activated, but you can activate in Settings.
  Your experiences with Nice Loops are highly appreciated.

[Hajime]

The sep 2023 version of SiSeSuSo.exe is now available at
https://sites.google.com/site/sisesuso .

Changes version 1 Sep 2023, 2023-9

• Method XYZwing was always omitted during puzzle solving. Fixed.

[urhegyi]

why can't SiSeSuSo find this WXYZ wing?

[Hajime]

Good question. This is a WXYZ type C. I don't know why it is not found.
It should be found in the 2023 versions of SiSeSuSo. I will dive into it. Will take some time...

See here what should be done. Section (STUVW)XYZ Wing

[Hajime]

Found the 'bug' . SiSeSuSo discovers only WXYZ type C with 2 cells in the intersection of a line and a box.
And only 1 cell in each wing. So 4 cells in total. The above example has 2 cells in one wing.
It will take a while to repair.

[urhegyi]

Another example that can't be found. I think for the same reason:

[Hajime]

No, this is a type B. Something wrong somewhere else in the code, but no time to figure it out...

[Hajime]

In puzzle WXYZ_20231016 cell r8c9 has candidates 48. I have 348. Why is the 3 candidate gone?
Busy fixing the WXYZ Type C.
This is a new reason, why SiSeSuSo can not find the WXYZ Type C...
What solver do you use?

[urhegyi]

In puzzle WXYZ_20231016 cell r8c9 has candidates 48. I have 348. Why is the 3 candidate gone?
Busy fixing the WXYZ Type C.
This is a new reason, why SiSeSuSo can not find the WXYZ Type C...
What solver do you use?

Your own solver has already eliminated the 3 from R8C9:

Code: Select all

[1,1] r4c3=7 Hidden Single in col 3
[2,1] Naked/Hidden Pairs,Triplets,Quads  | NQuin (24789)b3e25789 => (-2489)r1c9 (-2479)r2c7
[3,1] Pointing, Claiming  | (9)r7,b9 => (-9)r8c7 (-9)r8c9 | (8)b5,c4 => (-8)r8c4 (-8)r9c4 | (7)b9,c7 => (-7)r3c7 | (9)b3,c8 => (-9)r5c8 (-9)r6c8 (-9)r7c8
[4,1] 2-String-Kite,SkyScraper,TurbotCrane  |
2-String Kite (3)r3c5=r3c1-r2c2=r8c2 => (-3)r8c5
[5,1] X-chain [links<=15]  |
X-chain [5] (3)r7c9=r7c5-r3c5=r3c1-r2c2=r8c2 => (-3)r8c9
[6,1] Alternating Inference Chain (AIC)  |
AIC Type 2 [7] (3)3-r3c5=3=r7c5-3-r7c9=(3-2)r9c9=2=r9c7-(2=4)r3c7 => (-4)r3c5
[7,2] r1c5=4 Hidden Single in col 5


[Hajime]

Found the 2 bugs in WXYZ-method and repaired them.

• Type B was too restrictive
• Type C was with only 1 cell in each wing, is now with only 1 Z-cell in each wing.

Still not "grouped" and because it is a multiple recursive subroutine it is possible that a (STU)VWXYZ is found earlier than a WXYZ.

I am busy with more manual functionality like

• add clues or eliminating candidates/clues in a chosen cell
• highlight all cells with a chosen clue

I expect to release a next version in November.

[Hajime]

The Nov 2023 version of SiSeSuSo.exe is now available at
https://sites.google.com/site/sisesuso .

Changes version 5 nov 2023, 2023-11

• Bugs in WXYZ method repaired
  
• Manual solving extended with highlighting cells with certain clue.
  And Undo button works as expected.

[urhegyi]

I did some experiments with your latest version. Nice work.
I found a curiosity with the following sudoku:

Code: Select all

1.....9......3...4..2..8.5....7..3..5.9.....1.8..2..6...69..8......5...73....1.2.

Your newest version needed a forcing net for it(version 11-2023)while the pattern game version(version 09-2022) solved it with TUVWXYZ-wings, an x-chain, xy-chain and AIC's, so no forcing nets needed?
What has changed?

Edit:
Check Disc nice loops too and augment Aic length from 9 to 15 and it solves without forcing net.

Code: Select all

Summary eliminated candidates per Method And per Sudoku

Method   \  Sudoku |   SER |     1 Total/Methods
                   |-------|-----------------
Solved Clue elims  |     0 |    62    62/0
Naked Singles      |   0.1 |    46    46/46
Hidden Singles     |   0.2 |    30    30/12
Naked Single       |   2.5 |     2     2/2
Naked Pair         |     3 |     6     6/5
Naked Triple       |   3.6 |    13    13/7
Naked Quad         |     5 |    17    17/8
Hidden Quad        |   5.4 |     2     2/1
Pointing/Claiming  |   2.8 |    15    15/12
TUVWXYZ Wing  [B7] |   7.5 |     1     1/1
X-chain  [5 links] |   5.1 |     1     1/1
XY-chain [9 links] |     5 |     1     1/1
AIC-chain[5 links] |   5.5 |     3     3/3
AIC-chain[7 links] |   6.1 |     9     9/7
AIC-chain[11links] |   7.2 |     2     2/1
                   |-------|-----------------
Eliminated Cand's  |   210 |   210
Sum(SER * Cand's)  | 321.6 | 321.6

Initial Candidates :   210
Maximum SER rating :   7.5 <- Approach
Labour rating      : 321.6 <- Experimental rating
Time needed        : 00:00:56.383
SiSeSuSo Solver and Generator (version 2023-11) 

[Hajime]

Yes, found a possible explanation. AIC-code is changed: it is huge extended.
And formerly it counts in cells, now in links. AIC chain of 8 cells is in this example a chain of 11 links (3 in bivalue cell).
If the AIC-chain max length was 10 cells (in 2022-versions) and now 9 links (in 2023 versions), it will not be found.

Why did I restrict the AIC length to the limited 9 links in stead of, say, 15 links?
That is because AIC method is extended to find more candidate eliminations. How?

If in some house are only 2 candidates 5 and it is not the first cell then it must be the second cell.
If the second cell has candidates 123456 and it must be a 5 then this cell cannot hold 12346.
Each of those not-possible candidates *) is examined in the rest of a possible AIC-chain (not X or XY chain).
That extends the search tree tremendous. From 6 seconds in the version 2022-x to 1 minute in 2023-11.

What to do:
- limit the length of the chain?
- limit the AIC search without those not-possible candidates *)

I choose the first, but maybe this is not wise...