SudokuWiki.org Feedback, Page 4

Saturday 6-Mar-2010

... by: Oliver Buckley, London

Load Sudoku: CLICK TO LOAD

This puzzle is one grid (top right) from today's Samurai in The Times.
At this stage in solving the puzzle, your Solution Count records three viable solutions. Your main program eliminates two of these by referring to Uniqueness (Type 1), however this is incorrect. The uniqueness referred to cannot eliminate any solution, as both the other two solutions are equally valid and in fact clues from other grids in the Samurai confirm that one of the solutions your Uniqueness strategy eliminates, is in fact the correct one. I suspect there is a bug in the Uniqueness process.

Andrew Stuart writes (6-Mar-2010):

The test for the number of solutions is separate from the solver. The solver assumes uniqueness and will give an unpredictable result when non-unique. All works by design.

Tuesday 26-Jan-2010

... by: sss, US

Have you changed the print format? Until today, it printed the remaining numbers small on one line (usually), now they take up most of the cell and I have no room to write my numbers. Is there any way to get the original format back?

Thanks!

Andrew Stuart writes (26-Jan-2010):

Yes I had changed the solver, see
/Whats_New

But didn’t realise it had a knock on effect on the printing. Fixed now. Thanks for the alert

Tuesday 26-Jan-2010

... by: Bob Kukla, Texas USA

It doesn't work this morning -- it saves but doesn't step forward or reload.

Something I'm doing wrong? Hope not -- I use it every day.

Bob

Andrew Stuart writes (26-Jan-2010):

Clear your cache and reload. I've made a new version which I'm quite excited about. "Chaining strategies that report chains now display those chains as a new graphical element on the big board"

Its possible the problem is the new graphics library but I build the new version in Firefox - you using Firefox?

Best regards

Friday 15-Jan-2010

... by: Richard C. Keech, M.D., California, USA

Just verified that your Sudoku Solver works with Konqueror, a Linux browser.

Andrew Stuart writes (24-Jan-2010):

Awesome, that’s good news :)

Tried hard to be as widely browser compatible as possible

Sunday 10-Jan-2010

... by: Doug, Austin, TX USA

I have enjoyed your sudoku solver and it helped me become quite good at solving advanced puzzles. I recently decided to try kenken and was encouraged that you have a 6x6 solver. Unfortunately it doesn't know how to solve most of the puzzles I throw at it. Are you planning to continue further development on your kenken solver? Thanks.

Andrew Stuart writes (24-Jan-2010):

Yes. But I need examples to improve it. I can make the puzzles myself and I do solve > 99% of them so I didn't think I was very far off. Would it be possible to send me your next unsolveable, say a screen shot?

Sunday 10-Jan-2010

... by: Harold Binley, England

J F Crook's solution

It has taken me many attempts and a long time to understand J F Crook's paper and your responses, (Mainly on the former as his concepts were difficult to grasp).

In your response you are dismissive of the Trial and Error approach. But looking at the accepted strategies several of these effectively do this. For example: Simple Colouring is where you select a number and cell (relatively) at random and follow through the links. If a cell can see other cells with the two colours the selcted number is eliminated from that cell. Surely this is effectively trial and error, except that there is some logic behind the elimination?

Andrew Stuart writes (24-Jan-2010):

This is a very interesting area and its quite philosophical. I would distinguish between a) the 'pattern' that is found that a strategy uses to eliminate candidates (logically) and b) finding that pattern in the first place. All the documented strategies assume you can find the opportunity and they are all logical apart from the last two (which I place under the heading 'trial and error'). However, there is little help for the solver in how to find them apart from simple rules like - look for bi value (2 candidates in cell) and bi-location (two candidates in a unit) situations (since these lead to chains which are the building blocks of many strategies).

When searching for opportunities to apply strategies your path is determined by all the dead ends and in a sense these are the 'errors'. A computer has to look at every single dead end because it doesn't have a mind. But in order for a human or a computer not to take years solving a puzzle, we use logical optimizations. Even your 'random' search is not really random (trial) because you intelligently select what seem to be the best opportunities first.

Where I draw the line - because the whole 'trial and error' vs determinism - is blurry - is whether a strategy changes the board to get an answer. Nishio and Bowmans bingo do this, as does Crook fundamentally. By this I mean "If I remove/add a number here - what is the consequence?". A 'pattern' based strategy is, imho, is on the other side of the 'trial and error' divide. It asks the question "because X or Y is present/absent, what is consequence of that?"

I hope that stab at rephrasing helps and I've not repeated myself.

Wednesday 6-Jan-2010

... by: Jerry, Virginia, USA

Load Sudoku: CLICK TO LOAD

Andrew,
Here is a case where your solver misses a hidden unique rectangle. At this stage, your solver eliminates the 4 from H3, but fails to eliminate the four from H4. The 46's in J3-J4 plus the strong link on 6's in column 3 mean that a 4 in H4 would lead to the puzzle having two solutions. Perhaps having eliminated the 4 in H3, the pattern no longer looks like a hidden unique rectangle, even though it is.


+---------------+------------------+--------------+
|    3   9    2 |      7    46   8 |   46   5   1 |
|    7   6    5 |      2    14 134 |   34   8   9 |
|    8   4    1 |   3569   569  35 |   36   7   2 |
+---------------+------------------+--------------+
|    6   7    3 |     45    45   9 |    1   2   8 |
|    1   5    9 |     38    28  23 |    7   6   4 |
|   24  28   48 |      1     7   6 |    9   3   5 |
+---------------+------------------+--------------+
|  249 128  478 |    489     3  47 |    5  14   6 |
|   49  18 4678 |  45689 45689 457 |    2  14   3 |
|    5   3   46 |     46    12  12 |    8   9   7 |
+---------------+------------------+--------------+

The original position of this puzzle is this

This puzzle grades at 889, which is the highest grade I have seen for a puzzle that did not require any Evil techniques to solve. I think it is the hardest puzzle I have solved without using bifurcation.
- Jerry

Andrew Stuart writes (24-Jan-2010):

Very interesting and you are correct. I'd dub this a type 2b version of a Hidden UR. It’s the same principle as Type 2 but the floor is across two boxes. I've tested it against my 27k of 2010 stock and it appears twice as often as the Type 1 and three times as often as Type 2. I've added a section to Hidden Unique Rectangles with credit.

Might be a day or so before I can update the solver online.