Hyperbolic SVM

Python implementation of hyperbolic SVM, as introduced in [1]. This is a Python adaptation of the official imlementation in Matlab [2]. This is the official code repository from this paper https://www.aclweb.org/anthology/W19-5805/ .

Raison d'être

• Hyperbolic SVM compatible with scikit-learn, i.e., inherits from BaseEstimator, LinearClassifierMixin for an easier integration into scikit-learn pipelines
• Simple matplotlib visualizations of decision boundaries for both Euclidean and hyperbolic SVMs in 2 dimensions
• Integrates seemlessly with evaluation pipeline for knowledge graph embeddings as in [3]

Usage

• python3 train.py ./data/gaussian/data_002.mat --c 1
  • will train Euclidean and hyperbolic SVM on data generated with hyperbolic Gaussian
• see --help of evaluate_embeddings.py, plot_results.py, train.py for further details

Requirements

see requirements.txt

Citation

If you use it and find useful please consider citing this paper https://www.aclweb.org/anthology/W19-5805/

@inproceedings{agibetov-etal-2019-using,
    title = "Using hyperbolic large-margin classifiers for biological link prediction",
    author = "Agibetov, Asan  and
      Dorffner, Georg  and
      Samwald, Matthias",
    booktitle = "Proceedings of the 5th Workshop on Semantic Deep Learning (SemDeep-5)",
    month = "12 " # aug,
    year = "2019",
    address = "Macau, China",
    publisher = "Association for Computational Linguistics",
    url = "https://www.aclweb.org/anthology/W19-5805",
    pages = "26--30",
}

References

• [1] https://arxiv.org/abs/1806.00437
• [2] https://github.com/hhcho/hyplinear
• [3] https://github.com/plumdeq/neuro-kglink