The New Sudoku Players' Forum

[ronk]

Interestingly the results from Glenn's solver will vary depending on which version of the puzzle is used. 41826 is the toughest of the 47621 puzzles I downloaded (with SE rating 9.1).

meet my nemesis (order dependent constraints/methods/techniques)
[...]
its really tough to wash out the for(i=0;i<81;i++) and for(j=0;j<9;j++) coding bias

I thought yours was a "batch solver", where such bias wouldn't occur.

[gsf]

Interestingly the results from Glenn's solver will vary depending on which version of the puzzle is used. 41826 is the toughest of the 47621 puzzles I downloaded (with SE rating 9.1).

meet my nemesis (order dependent constraints/methods/techniques)
[...]
its really tough to wash out the for(i=0;i<81;i++) and for(j=0;j<9;j++) coding bias

I thought yours was a "batch solver", where such bias wouldn't occur.

background:
batching is enabled by the -B option, off by default
with -B enabled, at each ply, each constraint determines the possible placements/eliminations
but does not apply them until all for the current constraint have been identified

the K constraint applies singles multiple times (to extend the net via individual chains)
and I think the bias has crept in between the multiple applications
its probably not a 5 minute fix

[ronk]

batching is enabled by the -B option, off by default

dang, that's another thing I knew at one time and forgot
this forgetfulness is starting to get worrisome

[gsf]

I did some empirical analysis on order dependencies in my solver
it fingers the XYK constraints as the culprits
these are the chain related constraints
I already had code in place to mitigate order dependencies, but its not sufficient
batching (-B) removes the order dependencies for the FNBTHW constraints (empirically)
not sure at the moment why XYK are order dependent