[Date Prev][Date Next][Thread Prev][Thread Next][Date Index][Thread Index]
Re: Problem with CLP
Lars Riis Olsen
Re: Problem with CLP
Tue, 23 Sep 2003 19:35:48 +0200
Mozilla/5.0 (Windows; U; Windows NT 5.0; en-US; rv:1.5b) Gecko/20030901 Thunderbird/0.2
Fergus Henderson wrote:
> In general, a CLP system can give three kinds of results to a query:
> 1. the query is definitely satisfiable
> 2. the query is possibly satisfiable, subject to some constraints
> 3. the query is definitely unsatisfiable
> These three cases correspond to "yes", "maybe", and "no", respectively.
> A good CLP system will distinguish between cases 1 and 2 in its
> for example by printing out a representation of the remaining unresolved
> constraints for case 2. However, many CLP systems -- including GNU
> Prolog -- fail to distinguish these cases. GNU Prolog just prints out
> the variable bindings, and some information about the maximum possible
> ranges of constrained variables, but it does not print out any remaining
> unresolved constraints or indeed give any indication that there might
> be unresolved constraints, and instead of printing out "maybe" for
> it misleadingly prints "yes".
> This is IMHO a significant flaw in the GNU Prolog user interface.
> > From the query it should be clear that it is indeed not satisfiable.
> > Furthermore to recheck this I tried to further constraint the value of
> > the variable X to be eg. 0. In this case it comes up with the correct
> > answer (no). The same happens if X is instead constrained to 1,2,3...
> > and so forth.
> Once you bind the variables, the unresolved constraints can be evaluated.
> Then the query results will be either in category 1 or 3, and will match
> what you expect.
So if I am understanding you correctly there is no way of determining
whether a "yes" result returned by GNU Prolog is correct. Maybe except
from binding one of the variables with each of its possible values to
check whether any of them returns "yes" (which obviuosly is immensely
ugly and not at all feasable for large ranges.). Is this correct a
correct interpretation of your previous post?.
In that case, do you know of any free constraint satisfiability
package(s), with a C interface, that does not have this limitation?.