External reasoners never used during fitting for reasoner-based concept learners

[Louis-Mozart]

When initializing a reasoner-based concept learner with an external reasoner, it always uses the default reasoner at fitting time. For instance, if I initialize CELOE (Or any order) as follows:

neural_owl_reasoner = TripleStoreNeuralReasoner(path_of_kb=args.kb, gamma=0.9)

celoe = CELOE(knowledge_base=args.kb, reasoner=neural_owl_reasoner, quality_func=F1(), max_runtime=args.max_runtime)

Then at the fitting time, celoe.fit(lp, verbose=False), the neural_owl_reasoner is never used, but at predictions time, predictions_celoe = celoe.predict(individuals=list(typed_pos | typed_neg), hypotheses=hypotheses) , that is only when the owl reasoner is used and if we predict directly doing pred_celoe = celoe.fit(lp).best_hypotheses(n=1), the owl reasoner is even never used neither at fitting time nor at prediction time.

[Demirrr]

While I was looking at this issue, I have also realised that we cannot using a reasoner within a KnowlgeBase

kb = KnowledgeBase(path="/home/cdemir/Desktop/Softwares/Ontolearn/KGs/Family/father.owl",
                   reasoner=SyncReasoner(ontology="/home/cdemir/Desktop/Softwares/Ontolearn/KGs/Family/father.owl", reasoner="HermiT")

leads to KeyError exception at in the ClassHierarchy(reasoner)

self.data_property_hierarchy) = init_hierarchy_instances(self.reasoner,
  File "/home/cdemir/Desktop/Softwares/Ontolearn/ontolearn/utils/static_funcs.py", line 156, in init_hierarchy_instances
    class_hierarchy = ClassHierarchy(reasoner)
  File "/home/cdemir/anaconda3/envs/venv/lib/python3.10/site-packages/owlapy/owl_hierarchy.py", line 286, in __init__
    super().__init__(OWLClass, arg)
  File "/home/cdemir/anaconda3/envs/venv/lib/python3.10/site-packages/owlapy/owl_hierarchy.py", line 51, in __init__
    self._init(hier_down_gen)
  File "/home/cdemir/anaconda3/envs/venv/lib/python3.10/site-packages/owlapy/owl_hierarchy.py", line 124, in _init
    _children_transitive(self._children_map_trans, ent=ent, seen_set=set())
  File "/home/cdemir/anaconda3/envs/venv/lib/python3.10/site-packages/owlapy/owl_hierarchy.py", line 397, in _children_transitive
    _children_transitive(hier_trans, sub_ent, seen_set | {ent})
  File "/home/cdemir/anaconda3/envs/venv/lib/python3.10/site-packages/owlapy/owl_hierarchy.py", line 394, in _children_transitive
    sub_classes_ent = frozenset(hier_trans[ent])
KeyError: OWLClass(IRI('http://www.w3.org/2002/07/owl#', 'Nothing'))

Two observations:

1. reasoner cannot be used within KnowledgeBase
2. Passing a knowledge base instance and a reasoner instance into a learener is confusing. This should not be possible

[Louis-Mozart]

So do you mean it is not possible to use an external reasoner on those concept learners?

[Demirrr]

TDL and Drill can be used with a triple store.
as shown below

from ontolearn.learners import TDL, Drill
from ontolearn.triple_store import TripleStore
from ontolearn.learning_problem import PosNegLPStandard
from owlapy.owl_individual import OWLNamedIndividual
from owlapy import owl_expression_to_sparql, owl_expression_to_dl
from ontolearn.utils.static_funcs import save_owl_class_expressions
# (1) Initialize Triplestore
kb = TripleStore(url="https://dbpedia.data.dice-research.org/sparql")
# (3) Initialize a learner.
model = Drill(knowledge_base=kb) #  or  TDL(knowledge_base=kb)
# (4) Define a description logic concept learning problem.
lp = PosNegLPStandard(pos={OWLNamedIndividual("http://dbpedia.org/resource/Angela_Merkel")},
                      neg={OWLNamedIndividual("http://dbpedia.org/resource/Barack_Obama")})
# (5) Learn description logic concepts best fitting (4).
h = model.fit(learning_problem=lp).best_hypotheses()
print(h)
print(owl_expression_to_dl(h))
print(owl_expression_to_sparql(expression=h))
save_owl_class_expressions(expressions=h,path="#owl_prediction")

By following this example, you can replace
kb = TripleStore(url="https://dbpedia.data.dice-research.org/sparql")
with
neural_owl_reasoner = TripleStoreNeuralReasoner(path_of_kb=args.kb, gamma=0.9). This should work.

You are passing a reasoner and kb together. As pointed above, this should not be even possible. Yet, the current code allows it.

Currently, a reasoner can be used within Drill and TDL.

[Louis-Mozart]

Neither Drill nor TDL works when I replace the kb with the neural_owl_reasoner I don't if that worked for you. See error message

-------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------
concept_learning_drill_train.py 121 <module>
start(parser.parse_args())

concept_learning_drill_train.py 73 start
pred_drill = drill.fit(train_lp).best_hypotheses()

drill.py 323 fit
[i for i in chain.from_iterable((self.kb.get_types(ind, direct=True) for ind in learning_problem.pos))])

drill.py 323 <listcomp>
[i for i in chain.from_iterable((self.kb.get_types(ind, direct=True) for ind in learning_problem.pos))])

drill.py 323 <genexpr>
[i for i in chain.from_iterable((self.kb.get_types(ind, direct=True) for ind in learning_problem.pos))])

AttributeError:
'TripleStoreNeuralReasoner' object has no attribute 'get_types'

[Demirrr]

This error means that TripleStoreNeuralReasoner does not implement get_types funciton that is implemented in TripleStore.

[alkidbaci]

TripleStoreNeuralReasoner should be updated into implementing AbstractOWLReasoner (and all the methods that come with it). This is due to recent triplestore-related changes.