PSyKE

Platform for Symbolic Knowledge Extraction
(description)

PSyKE (a Platform for Symbolic Knowledge Extraction) provides general-purpose support to symbolic knowledge extraction from different sorts of black-box predictors via many extraction algorithms.

PSyKE in a nutshell

PSyKE (Platform for Symbolic Knowledge Extraction) is a library for extracting symbolic knowledge, in the form of logic rule lists, out of sub-symbolic predictors. It offers a general-purpose API for knowledge extraction, together with a number of algorithms implementing it, for both classification and regression problems. The extracted knowledge is either a Prolog theory (a list of Horn clauses) or an OWL ontology containing SWRL rules.

PSyKE is designed around the notion of extractor: any object capable of extracting a logic theory out of a trained sub-symbolic classifier or regressor, which acts as an oracle. An extractor offers two methods: extract, which returns the logic theory for a given dataset, and predict, which predicts a value through the extracted rules in place of the original predictor.

The supported extraction algorithms are CART, for both classification and regression; REAL and Trepan, for classification; ITER, GridEx and GridREx, for regression. The PEDRO algorithm, included in PSyKE, tunes the hyper-parameters of GridEx and GridREx. Work is ongoing to extend PSyKE to explainable clustering, and to make its algorithms more general-purpose.

PSyKE relies on 2ppy (tuProlog in Python) for logic support, which in turn is based on the 2P-Kt logic ecosystem.

Downloads

Logo