This is a heavily modified fork of OpenKE. This repository contains implementations of XKE-PRED and XKE-TRUE, as used in the paper:
- Arthur Colombini Gusmão, Alvaro Henrique Chaim Correia, Glauber De Bona, and Fabio Gagliardi Cozman. Interpreting Embedding Models of Knowledge Bases: A Pedagogical Approach. Proceedings of the ICML Workshop on Human Interpretability in Machine Learning, pp. 79-86, 2018. Link
Backward compatibility for past commits is not to be expected.
This code was built from this commit of OpenKE, which is distributed under the MIT License. All code developed thereafter is distributed under the terms of the GNU General Public License, version 3 (or, at your choosing, any later version of that license). You can find the text of that license here. This code also has PRA as a dependency, which is also licensed under the GNU General Public License.
Examples for running each stage of both XKE-PRED and XKE-TRUE can be found in the examples/
directory:
examples/emb_grid_search/
Learn embedding models.examples/extract_features
Extract features.examples/explain
Learn an interpretable classifier (that uses features as input) and get its explanations.