Directly modifying problems

[haz]

Is there any reason why lang.predicates.append(some_predicate_object) shouldn't work? I'm confused as to why it doesn't...

[miquelramirez]

Hi @haz ,

back at the coal mine after a long weekend....

Could you please provide a snippet (over here or via gist) that illustrates the crash? exception? that you're getting?

Regards,

Miquel.

[haz]

That's just it... Nothing happens. I'm perplexed. The list of predicates is immutable somehow?

[miquelramirez]

Hi @haz,

okay, let me match your perplexity with my perplexity then when I understood what the problem you have is.

lang.predicates is a property, defined as follows in src/tarski/fol.py

    @property
    def predicates(self):
        return list(self._predicates.values())

so of course adding a new predicate object to what clearly is a temporary object isn't going to pan out. The correct interface to define predicates is the method predicate

    def predicate(self, name: str, *args):
        self._check_name_not_defined(name, self._predicates, err.DuplicatePredicateDefinition)

        types = [self._retrieve_sort(a) for a in args]  # Convert possible strings into Sort objects
        predicate = Predicate(name, self, *types)
        self._predicates[name] = predicate
        self._global_index[name] = predicate
        return predicate

so you write

foo = lang.predicate('foo', L.Object, L.Object)

to define a binary predicate where arguments of the predicate are to be constants of the built-in type Object. This is the way to define predicates in every test and documentation we have written like this one. As becomes apparent from the implementation of the FirstOrderLanguage.predicate() method, there's some significant bookkeeping that needs to be done in order to ensure that the definition of the new predicate is consistent with 1) that of other predicates and 2) the sorts defined to arrange the objects in the domain theory (or "first order language" as we chose to call it).

Miquel.

[haz]

Aha! Hadn't considered it being generated on the fly. Was so confused...

We worked around it as you suggested, but for the record, we were trying to merge domains that come with the exact same sorts (types and objects are unified, but predicates differ). Attempt was to just add the predicates of one domain, to the other, but we ended up creating them each from scratch.

Mystery solved, thanks!

[miquelramirez]

You need to copy the data from theory to theory. Theory merging is an interesting use case.

[haz]

It'll drop as an ICAPS demo and/or workshop paper. You're gonna like this one for sure ;)