Optimize core processes for large scale CEL

[alkidbaci]

Some processes can be optimized to achieve a better performance overall when using a CEL algo.

E.g. this method in triple_store.py retrieves object properties every time its is called:

def get_object_properties_for_ind(self, ind: OWLNamedIndividual, direct: bool = True) \
        -> Iterable[OWLObjectProperty]:
    properties = set(self.get_object_properties())
    yield from (pe for pe in self.reasoner.ind_object_properties(ind, direct) if pe in properties)

Solution: we can cache the properties and reuse them.

The method ind_object_properties also retrieves all object properties every time its called. The already retrieved object properties can be passed to this method and lower the time overhead.

I believe after a closer inspection more potential optimization can be figured out.

[Demirrr]

I am unsure about this particular optimization.

The purpose of using the triple store within CEL is to decouple the learning process from the data store and access.

Given that most databases are optimized for the data storage and access, such caching should be already implemented in a database.

Yet, I do believe that we have plenty of room in ontolearn to improve the scalability further.
For example, we can optionally use auch caching in a reasoner that already stores the data in memory. @Louis-Mozart did some work on using caching algorithms to reduce the total runtime of a CEL algorithm. Perhaps, we can start thinking of integrating his work dev branch.

[alkidbaci]

Well, this is not only for the triple store, its also valid for the local KB. I don't mean any sophisticated cache here, but basically just storing the object properties in a class attribute and check for its existence the next time we want to access the object properties. At least in the triplestore scenario we assume that the object properties are not altered, hence its safe to do this without loss of information.

If the concern is how much memory storing these object properties will consume, then we can make it optional and the user can decide depending on what works best for him, or as of your suggestion, we can just expect the user to have the dataset optimized for data storage/access and just continue with our lives, undisturbed ^^

[Demirrr]

Sure. We can use native python caching : https://docs.python.org/3/library/functools.html