Hi,
My earlier post about min and argmin throwing an exception prompted me
to remember a discussion I had with Joachim Schimpf some time ago about
the semantics of the sequence constraint:
sequence(home, x, s, q, l, u)
if I remember correctly, he thought the constraint should not fail if
the number of variables in x is smaller than q, but larger than l, and
the number of values in x is between l and u. He actually was using the
constraint on some problem instances that had this property.
I did not initially remember that the discussion was about sequence, so
I first looked at the count constraint, and found a similar behaviour
with the IRT_LQ and IRT_LE in
count(home, x, y, IRT_LQ, z)
the constraint fails if the size of x is smaller than than z, even
though the relationship is actually true - the number of times the
variables in x is y is less than z.
I had expected the constraint to succeed in this case, and the atmost
constraint in ic does succeed in such cases (ic restrict z to be an
integer for atmost). I can see that the count fails here because z is
constrained to 0..|x|, for all relationships.
Does it make sense for these constraints to succeed in such cases when
the relationship (number of times the value(s) occur) is satisfied, even
though the number of variables is smaller than the limit/sequence
length? I am uncertain for the sequence case, but for count, succeeding
makes more sense to me.
If not succeeding, should an exception be raised instead of failing?
This will alert the user to the fact that there are too few variables
for the constraint, rather than the number of values not satisfying the
requirements.
Should the behaviour be documented in the MPL? I don't think it is
mentioned there at the moment.
Cheers,
Kish
_______________________________________________
Gecode users mailing list
[email protected]
https://www.gecode.org/mailman/listinfo/gecode-users
