Recognition of auxiliary filters about particles HOT 3 CLOSED

Hi!
First of all, nice work on thw waste free SMC @hai-dang-dau !
On this, testing for behaviour based on the presence or absence of an attribute is an antipattern an can lead to issues as you highlighted. This should be handled by the inheritance contract, so the real thing to check is if self is an instance of APFMixin (or AuxiliaryPF, depending on logic). The way I would go around it is to delete the isAPF altogether and then replace its use by a direct isinstance check. For instance this

if isinstance(self.fk, AuxiliaryPF):

if you are worried about circular imports, you can also use local imports of AuxiliaryPF

from particles.

@AdrienCorenflos Thanks for your feedback. I do not entirely agree with the type checking solution because that seems to go against Python's "duck typing" philosophy (i.e. any object with the logeta method should work with the implementation, regardless of its type).

My proposition would be to add a boolean-valued useAPF keyword argument in the SMC class. My arguments are:

(1) Even if a FeynmanKac object does include the logeta method, the user may, out of curiosity, prefers running the standard filter instead and sets useAPF=False.

(2) If useAPF is set to True, the implementation shall call fk.logeta without checking. So if the user passes in a FeynmanKac object without logeta method and sets useAPF=True, an AttributeError is therefore automatically raised (which is what we want anyway).

from particles.

I applied the quick fix recommended by Dang. Now things should work as expected.
But it's true that the current design is a bit dodgy, as proved by this "silent bug". I'll think about it.

from particles.