eleven wrote:I guess, coloin is right, and the 21 can be searched faster.
Suppose, you start with box 5. Then you can fix 4 digits like this:
- Code: Select all
..x...x..
.x...x.x.
x.......x
...Y.1.a.
....2....
.x.3.4...
x.......x
.x.b...x.
..x...x..
With a 5th clue in Y you can fix it either (to be 5), while the rest of the puzzle is unchanged, i.e. you cant have more puzzles to check this way.
But additionally you have one possibility less for cells a and b.
Nice idea anyway.
[Added:] (But this way you would not find 20 clues, where a=b)
I agree that adding a clue we loose a=b and we have some in the output.
So, the right process would be to solve the extended pattern plus the specific case a=b in the 20 clues pattern.
This would be feasible with my code as it is.
Regarding the right way to extend the 21 clues, my code is not that flexible to take benefit of symmetries
Morphing is cancelled if it's not possible to apply all morphs,
At the end, I think the best process would be
step one 8 clues in lines 1 3 5 7 clean morphs
step two add 4 clues in box 5 (not the central clue) and again clean symmetries
at that point the best is done
step 3 (but generating huge files) add missing clues in boxes 1 3 7 9
my solver does not accept adding only 2 clues and computing morphs
here the ratio of morphs seems very low.
In my former test, I added in step 2 the central point, but this generates no morph and just increases the size of files.
champagne