added neural ontology manager

examples/neural_reasoner_retrieval.py

@@ -41,6 +41,7 @@
 
 """
 from owlapy.owl_neural_reasoners.owl_neural_reasoner import OWLNeuralReasoner
+from owlapy.owl_neural_reasoners.neural_ontology_manager import NeuralOntologyManager
 from owlapy.owl_reasoner import StructuralReasoner
 from owlapy.owl_ontology_manager import OntologyManager
 from owlapy.utils import concept_reducer, concept_reducer_properties, jaccard_similarity, f1_set_similarity
@@ -77,9 +78,13 @@ def execute(args):
     symbolic_kb = StructuralReasoner(ontology=OntologyManager().load_ontology(path=args.path_kg))
     # (2) Initialize Neural OWL Reasoner.
     if args.path_kge_model:
-        neural_owl_reasoner = OWLNeuralReasoner(path_neural_embedding=args.path_kge_model, gamma=args.gamma)
+        neural_manager = NeuralOntologyManager()
+        neural_manager.load_neural_embedding(path=args.path_kge_model)
+        neural_owl_reasoner = OWLNeuralReasoner(ontology=neural_manager, gamma=args.gamma)
     else:
-        neural_owl_reasoner = OWLNeuralReasoner(path_of_kb=args.path_kg, gamma=args.gamma)
+        neural_manager = NeuralOntologyManager()
+        neural_manager.load_ontology(path=args.path_kg)
+        neural_owl_reasoner = OWLNeuralReasoner(ontology=neural_manager, gamma=args.gamma)
     # Fix the random seed.
     random.seed(args.seed)
     ###################################################################

owlapy/owl_neural_reasoners/abstract.py

@@ -1,9 +1,14 @@
 from abc import ABC, abstractmethod
-from typing import List, Tuple, Generator
+from typing import List, Tuple, Generator, Union
 from owlapy.class_expression import OWLClassExpression
+from owlapy.owl_neural_reasoners.neural_ontology_manager import NeuralOntologyManager
+from owlapy.owl_ontology import AbstractOWLOntology
 
 class AbstractNeuralReasoner(ABC):
     """Abstract class for Neural Reasoners that operate on OWL Class Expressions using embeddings."""
+    @abstractmethod
+    def __init__(self, ontology: Union[NeuralOntologyManager, AbstractOWLOntology], **kwargs):
+        pass 
 
     @abstractmethod
     def predict(self, h: str = None, r: str = None, t: str = None) -> List[Tuple[str, float]]:
@@ -33,4 +38,5 @@ def object_properties_in_signature(self) -> List:
     @abstractmethod
     def data_properties_in_signature(self) -> List:
         """Retrieve all data properties in the knowledge base."""
-        pass
+        pass
+    #TODO LF: Maybe we can define more methods or remove some of them -> depending on concrete implementations.

owlapy/owl_neural_reasoners/neural_ontology_manager.py

@@ -0,0 +1,28 @@
+from owlapy.owl_ontology import Ontology
+from owlapy.owl_ontology_manager import OntologyManager
+from dicee.knowledge_graph_embeddings import KGE
+
+# LF: for now functions same as normal OntologyManager (+ can return path)
+class NeuralOntologyManager(OntologyManager):
+	def __init__(self, world_store=None):
+		super().__init__(world_store) 
+
+	def load_ontology(self, path: str = None) -> Ontology:
+		self._path = path
+		return super().load_ontology(path)
+
+	def load_neural_embedding(self, path: str = None):
+		self._path_neural_embedding = path	
+		return KGE(path)
+
+	def get_path(self) -> str:
+		if hasattr(self, '_path'):
+			return self._path
+		else:
+			return None
+
+	def get_path_neural_embedding(self) -> str:
+		if hasattr(self, '_path_neural_embedding'):
+			return self._path_neural_embedding
+		else:
+			return None

owlapy/owl_neural_reasoners/owl_neural_reasoner.py

@@ -14,15 +14,19 @@
 from collections import Counter, OrderedDict
 from functools import lru_cache
 from owlapy.owl_neural_reasoners.abstract import AbstractNeuralReasoner
+from owlapy.owl_ontology import Ontology
+from owlapy.owl_neural_reasoners.neural_ontology_manager import NeuralOntologyManager
 
 # TODO:
 def is_valid_entity(text_input: str):
     return True if "/" in text_input else False
 
 class OWLNeuralReasoner(AbstractNeuralReasoner):
     """ OWL Neural Reasoner uses a neural link predictor to retrieve instances of an OWL Class Expression"""
-    def __init__(self, path_of_kb: str = None, path_neural_embedding: str = None, gamma: float = 0.25, max_cache_size: int = 2**20):
+    def __init__(self, ontology: NeuralOntologyManager, gamma: float = 0.25, max_cache_size: int = 2**20):
         assert gamma is None or 0 <= gamma <= 1, "Confidence threshold (gamma) must be in the range [0, 1]."
+        path_of_kb = ontology.get_path()
+        path_neural_embedding = ontology.get_path_neural_embedding( )
         self.gamma = gamma
         self._prediction_cache = OrderedDict()
         self._max_cache_size = max_cache_size

tests/test_neural_reasoner_accuracy.py

@@ -1,4 +1,5 @@
 from owlapy.owl_neural_reasoners.owl_neural_reasoner import OWLNeuralReasoner
+from owlapy.owl_neural_reasoners.neural_ontology_manager import NeuralOntologyManager
 from owlapy.owl_reasoner import StructuralReasoner
 from owlapy.owl_ontology_manager import OntologyManager
 from owlapy.utils import concept_reducer, concept_reducer_properties, jaccard_similarity, f1_set_similarity
@@ -31,12 +32,15 @@ def test_retrieval_performance():
     # (1) Initialize knowledge base.
     symbolic_kb = StructuralReasoner(ontology=OntologyManager().load_ontology(path=path_kg))
 
-    # (2) Initialize Neural OWL Reasoner.
+    # (2) Initialize Neural OWL Reasoner. 
     if path_kge_model:
-        neural_owl_reasoner = OWLNeuralReasoner(path_neural_embedding=path_kge_model, gamma=gamma)
+        neural_manager = NeuralOntologyManager()
+        neural_manager.load_neural_embedding(path=path_kge_model)
+        neural_owl_reasoner = OWLNeuralReasoner(ontology=neural_manager, gamma=gamma)
     else:
-        neural_owl_reasoner = OWLNeuralReasoner(path_of_kb=path_kg, gamma=gamma)
-
+        neural_manager = NeuralOntologyManager()
+        neural_manager.load_ontology(path=path_kg)
+        neural_owl_reasoner = OWLNeuralReasoner(ontology=neural_manager, gamma=gamma)
     # Fix the random seed.
     random.seed(seed)