Re: [eclipse-clp-users] negate constraints

From: Marco Gavanelli <marco.gavanelli_at_unife.it>
Date: Thu, 08 Apr 2010 18:18:24 +0200
Dear Olivier,

Oliver Shycle wrote:
> Dear all,
> 
> thanks to all your help and advices I successfully set up my user defined
> reified constraints which no behave as I expect. The Constraint Library
> Manual however left some unanswered questions.
> On page 22 the manual says:
> - Enforcing constraints: It seems to be better to unify the truth reified
> values to variables to get active constraints. Is this correct?

What do you exactly mean here? Do you mean: "I impose a reified
constraint. Now, should I immediately impose that the corresponding
boolean is true or false? E.g.:

	B#= X#>Y,
	labeling([B]),
	(other constraints)
	labeling(...other variables ...)

"

If this is the question, the answer is no.
You should first impose all the constraints, then you perform search:

	B#= X#>Y,
	(other constraints)
	labeling(...all variables, including B ...)

> What are
> advantages and disadvantages of passive and active constraints?

It is not a matter of advantages or disadvantages: in that page the
manual says how the system works. The constraint remains passive (i.e.,
it can do nothing) until:
- either B becomes bound
- or the solver is able to infer that the original constraint (in the
example, X#>Y) is true
- or the solver is able to infer that the original constraint is false

> - "In the event that the reified variable becomes bound to 0 then the
> constraint should actively propagate its negation." How is this active
> propagation ensured by my constraints? What do I have to regard?

You have to write your propagation scheme for both the positive and the
negative part of the constraint. If you use Propia, you have to write a
predicate that considers all the possibilities: when the constraint is
true, and when it is false.

E.g., you want to write a new constraint that is true when the absolute
value of X is greater than 10, and you want to do it with Propia. So,
you start writing a predicate:

abs10(X):- X #>  10.
abs10(X):- X #< -10.

Now you want it to work as a constraint, so you call it with

	abs10(X) infers most.

Then you want to use it as a reified constraint, so you have to add an
argument B, and you have to write a predicate with two arguments:

abs10(X,1):- X #>  10.
abs10(X,1):- X #< -10.
abs10(X,0):- X #=< 10, X #>= -10.

Notice that last clause: you have to write code that explains what
happens when the constraint "absolute value of X is greater than 10" is
false. This will "actively propagate its negation": you write code for
this case.

Finally, you embed everything in a new predicate that invokes abs10 as a
Propia constraint:

abs10_constr(X,B):- abs10(X,B) infers most.

In this way you can combine it with other reified constraints, as in

	abs10_constr(X) or X #< 5.

> - same for "actively propagate its normal semantics"

The normal semantics, in the example, is "absolute value of X is greater
than 10". I.e., you have to write code that explains what happens when
this sentence is true. In the example there are two clauses: one when X
is greater than 10, and one in which X is less than -10.

I am not sure I understand your problem.
Is your problem: "I have found a nice constraint C, but I want to negate
it. I have two options:
1. I write from scratch a new constraint, called notC
2. I simply write neg(C) and it magically works"

Option 2 does not exist, unless you take a very simple constraint (one 
of the constraints that are already reified).
You only have option 1. Then, if you want, you can use your code to 
write C also as a reified constraint.

Best,
Marco

-- 
Marco Gavanelli, Ph.D. in Computer Science
Dept of Engineering
University of Ferrara
Tel/Fax  +39-0532-97-4833
http://www.ing.unife.it/docenti/MarcoGavanelli/
Received on Thu Apr 08 2010 - 16:18:35 CEST

This archive was generated by hypermail 2.2.0 : Thu Feb 02 2012 - 02:31:58 CET