Ontolearn

Ontolearn is an open-source software library for description logic learning problem. Find more in the Documentation.

Learning algorithms:

• Drill → Neuro-Symbolic Class Expression Learning
• EvoLearner → EvoLearner: Learning Description Logics with Evolutionary Algorithms
• NCES2 → (soon) Neural Class Expression Synthesis in ALCHIQ(D)
• NCES → Neural Class Expression Synthesis
• NERO → (soon) Learning Permutation-Invariant Embeddings for Description Logic Concepts
• CLIP → Learning Concept Lengths Accelerates Concept Learning in ALC
• CELOE → Class Expression Learning for Ontology Engineering
• OCEL → A limited version of CELOE

Installation

pip install ontolearn 

or

git clone https://github.com/dice-group/Ontolearn.git 
python -m venv venv && source venv/bin/activate # for Windows use: .\venv\Scripts\activate 
pip install -r requirements.txt
wget https://files.dice-research.org/projects/Ontolearn/KGs.zip -O ./KGs.zip && unzip KGs.zip

pytest -p no:warnings -x # Running 158 tests takes ~ 3 mins

Description Logic Concept Learning

from ontolearn.concept_learner import CELOE
from ontolearn.knowledge_base import KnowledgeBase
from ontolearn.learning_problem import PosNegLPStandard
from ontolearn.search import EvoLearnerNode
from owlapy.model import OWLClass, OWLClassAssertionAxiom, OWLNamedIndividual, IRI, OWLObjectProperty, OWLObjectPropertyAssertionAxiom
from owlapy.render import DLSyntaxObjectRenderer
# (1) Load a knowledge graph.
kb = KnowledgeBase(path='KGs/father.owl')
# (2) Initialize a learner.
model = CELOE(knowledge_base=kb)
# (3) Define a description logic concept learning problem.
lp = PosNegLPStandard(pos={OWLNamedIndividual(IRI.create("http://example.com/father#stefan")),
                           OWLNamedIndividual(IRI.create("http://example.com/father#markus")),
                           OWLNamedIndividual(IRI.create("http://example.com/father#martin"))},
                      neg={OWLNamedIndividual(IRI.create("http://example.com/father#heinz")),
                           OWLNamedIndividual(IRI.create("http://example.com/father#anna")),
                           OWLNamedIndividual(IRI.create("http://example.com/father#michelle"))})
# (4) Learn description logic concepts best fitting (3).
dl_classifiers=model.fit(learning_problem=lp).best_hypotheses(2)

# (5) Inference over unseen individuals
namespace = 'http://example.com/father#'
# (6) New Individuals
julia = OWLNamedIndividual(IRI.create(namespace, 'julia'))
julian = OWLNamedIndividual(IRI.create(namespace, 'julian'))
thomas = OWLNamedIndividual(IRI.create(namespace, 'thomas'))
# (7) OWLClassAssertionAxiom  about (6)
male = OWLClass(IRI.create(namespace, 'male'))
female = OWLClass(IRI.create(namespace, 'female'))
axiom1 = OWLClassAssertionAxiom(individual=julia, class_expression=female)
axiom2 = OWLClassAssertionAxiom(individual=julian, class_expression=male)
axiom3 = OWLClassAssertionAxiom(individual=thomas, class_expression=male)
# (8) OWLObjectPropertyAssertionAxiom about (6)
has_child = OWLObjectProperty(IRI.create(namespace, 'hasChild'))
# Existing Individuals
anna = OWLNamedIndividual(IRI.create(namespace, 'anna'))
markus = OWLNamedIndividual(IRI.create(namespace, 'markus'))
michelle = OWLNamedIndividual(IRI.create(namespace, 'michelle'))
axiom4 = OWLObjectPropertyAssertionAxiom(subject=thomas, property_=has_child, object_=julian)
axiom5 = OWLObjectPropertyAssertionAxiom(subject=julia, property_=has_child, object_=julian)

# 4. Use loaded class expressions for predictions
predictions = model.predict(individuals=[julia, julian, thomas, anna, markus, michelle],
                            axioms=[axiom1, axiom2, axiom3, axiom4, axiom5],
                            hypotheses=dl_classifiers)
print(predictions)
"""
          (¬female) ⊓ (∃ hasChild.⊤)  male
julia                            0.0   0.0
julian                           0.0   1.0
thomas                           1.0   1.0
anna                             0.0   0.0
markus                           1.0   1.0
michelle                         0.0   0.0
"""

Fore more please refer to the examples folder.

Benchmark Results

# To download learning problems. # Benchmark learners on the Family benchmark dataset with benchmark learning problems.
wget https://files.dice-research.org/projects/Ontolearn/LPs.zip -O ./LPs.zip && unzip LPs.zip

# To download learning problems and benchmark learners on the Family benchmark dataset with benchmark learning problems.
python examples/concept_learning_evaluation.py --lps LPs/Family/lps.json --kb KGs/Family/family-benchmark_rich_background.owl --max_runtime 60 --report family_results.csv  && python -c 'import pandas as pd; print(pd.read_csv("family_results.csv", index_col=0).to_markdown(floatfmt=".3f"))'

python examples/concept_learning_cv_evaluation.py --path_of_nces_embeddings NCESData/mutagenesis/embeddings/ConEx_entity_embeddings.csv --path_of_clip_embeddings CLIPData/mutagenesis/embeddings/ConEx_entity_embeddings.csv --folds 10 --kb KGs/Mutagenesis/mutagenesis.owl --lps LPs/Mutagenesis/lps.json --max_runtime 60 --report mutagenesis_results.csv && python -c 'import pandas as pd; print(pd.read_csv("mutagenesis_results.csv", index_col=0).to_markdown(floatfmt=".3f"))'

python examples/concept_learning_cv_evaluation.py --path_of_nces_embeddings NCESData/carcinogenesis/embeddings/ConEx_entity_embeddings.csv --path_of_clip_embeddings CLIPData/carcinogenesis/embeddings/ConEx_entity_embeddings.csv --folds 10 --kb KGs/Carcinogenesis/carcinogenesis.owl --lps LPs/Carcinogenesis/lps.json --max_runtime 60 --report carcinogenesis_results.csv && python -c 'import pandas as pd; print(pd.read_csv("carcinogenesis_results.csv", index_col=0).to_markdown(floatfmt=".3f"))'

Deployment

pip install gradio # (check `pip show gradio` first)

Available models to deploy: EvoLearner, NCES, CELOE and OCEL. To deploy EvoLearner on the Family knowledge graph:

python deploy_cl.py --model evolearner --path_knowledge_base KGs/Family/family-benchmark_rich_background.owl

Run the help command to see the description on this script usage:

python deploy_cl.py --help

Citing

Currently, we are working on our manuscript describing our framework. If you find our work useful in your research, please consider citing the respective paper:

# DRILL
@inproceedings{demir2023drill,
  author = {Demir, Caglar and Ngomo, Axel-Cyrille Ngonga},
  booktitle = {The 32nd International Joint Conference on Artificial Intelligence, IJCAI 2023},
  title = {Neuro-Symbolic Class Expression Learning},
  url = {https://www.ijcai.org/proceedings/2023/0403.pdf},
 year={2023}
}

# NCES2
@inproceedings{kouagou2023nces2,
author={Kouagou, N'Dah Jean and Heindorf, Stefan and Demir, Caglar and Ngonga Ngomo, Axel-Cyrille},
title={Neural Class Expression Synthesis in ALCHIQ(D)},
url = {https://papers.dice-research.org/2023/ECML_NCES2/NCES2_public.pdf},
booktitle={Machine Learning and Knowledge Discovery in Databases},
year={2023},
publisher={Springer Nature Switzerland},
address="Cham"
}

# NCES
@inproceedings{kouagou2023neural,
  title={Neural class expression synthesis},
  author={Kouagou, N’Dah Jean and Heindorf, Stefan and Demir, Caglar and Ngonga Ngomo, Axel-Cyrille},
  booktitle={European Semantic Web Conference},
  pages={209--226},
  year={2023},
  publisher={Springer Nature Switzerland}
}

# EvoLearner
@inproceedings{heindorf2022evolearner,
  title={Evolearner: Learning description logics with evolutionary algorithms},
  author={Heindorf, Stefan and Bl{\"u}baum, Lukas and D{\"u}sterhus, Nick and Werner, Till and Golani, Varun Nandkumar and Demir, Caglar and Ngonga Ngomo, Axel-Cyrille},
  booktitle={Proceedings of the ACM Web Conference 2022},
  pages={818--828},
  year={2022}
}


# CLIP
@inproceedings{kouagou2022learning,
  title={Learning Concept Lengths Accelerates Concept Learning in ALC},
  author={Kouagou, N’Dah Jean and Heindorf, Stefan and Demir, Caglar and Ngonga Ngomo, Axel-Cyrille},
  booktitle={European Semantic Web Conference},
  pages={236--252},
  year={2022},
  publisher={Springer Nature Switzerland}
}

In case you have any question, please contact: onto-learn@lists.uni-paderborn.de