As I am reading through this, here are some thoughts.
First, here are some alternate definitions I am noticing that might be
worthwhile.
xor=: ~:/
nOnes=: = i.
dToSudoku=: ,~@%:@# $ (1 = +/"1) * (1 + {.@I."1)
You could replace = with e.~ in nOnes if you wanted to respect a
variable length X
Also, I think I might write sudokuToArcs something like this:
require 'stats'
sudokuToArcs=: 3 :0
N=. #y
B=. %: N
T=. *: N
pairs=. [: ,/ (2 comb N)&{/.&(i.T)
row=. N#i.N
col=. T$i.N
box=. , (] +/ B * ] ) B # B $ i. B
1,.~ ~. /:~ (pairs row),(pairs col),pairs box
)
The line
9 {. D NB. show the domains for the first row of the sudoku.
does not make sense to me.
I started to write a variant on revise, but I ran into two issues:
1. You are modifying the value D
2. The test values are not illustrative
This means that the intent of the code is not apparent to me, and any
suggestions I make would likely be irrelevant.
--
Raul
On Thu, Nov 22, 2012 at 12:57 AM, Michal D. <michal.dobrog...@gmail.com> wrote:
> Hi All,
>
> The time has come for a more minimal version of Arc Consistency that works
> over a list of arcs and does away with an adjacency matrix. The main
> change is that revDom now modifies both domains at the same time. Along
> with this is a random sudoku generator. It's setup to generate 4x4 sudokus
> though you can easily generate 9x9 sudokus as well.
>
> NOTE: arc consistency, in general, will not be able to solve a sudoku
> unless it's easy. What it allows is decreasing the search space during
> back tracking search.
>
> http://pastebin.com/JdECdEbK
>
> Cheers,
>
> Mike
> ----------------------------------------------------------------------
> For information about J forums see http://www.jsoftware.com/forums.htm
----------------------------------------------------------------------
For information about J forums see http://www.jsoftware.com/forums.htm
