Predicate function signature

[lmores]

At the momento a predicate function is defined as PredicateFunction = Callable[[Any], Iterable[str]].
Could we change it to PredicateFunction = Callable[[Any], set[str]]?

If I am not wrong (but maybe I am) we actually never care about the order of the items inside the output of a predicate.
With this change in place we cold carry out a number of optimizations, e.g. in training.py this code:

for predicate in predicates:
    coverage = frozenset(
        i
        for i, (record_1, record_2) in enumerate(pairs)
        if (set(predicate(record_1)) & set(predicate(record_2, target=True)))
    )

would become

for predicate in predicates:
    coverage = frozenset(
        i
        for i, (record_1, record_2) in enumerate(pairs)
        if (predicate(record_1) & predicate(record_2, target=True))
    )

(The above snippet could be optimized even further if we just check if the intersection is non-empty instead of building it completely.)

Let me know!

[lmores]

@fgregg any thoughts on this?

[fgregg]

i think this is a smart idea. we implicitly rely on predicates returning unique collections and it would be good to make that explicit