Decision Trees with Short Explainable Rules

Authors: Victor Feitosa Souza, Ferdinando Cicalese, Eduardo Laber, Marco Molinaro

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental experiments with 20 real datasets show that our algorithm has accuracy competitive with CART while producing trees that allow for much simpler explanations.
Researcher Affiliation Collaboration Victor F. C. Souza Departamento de Informática Pontifical Catholic University of Rio de Janeiro Rio de Janeiro, RJ Brazil vfsouza@inf.puc-rio.br Ferdinando Cicalese Department of Computer Science University of Verona Verona Italy ferdinando.cicalese@univr.it Eduardo Sany Laber Departamento de Informática Pontifical Catholic University of Rio de Janeiro Rio de Janeiro, RJ Brazil laber@inf.puc-rio.br Marco Molinaro Microsoft Research & Pontifical Catholic University of Rio de Janeiro mmolinaro@microsoft.com
Pseudocode Yes Algorithm 1 SER-DT (S : set of objects)
Open Source Code Yes In the supplementary material we include the url for our anonymous repository.
Open Datasets Yes We considered the 20 datasets that appear on Column 1 of Table 1 (see Appendix E for their main characteristics). For all of them, 70% of the examples were used for training and the remaining 30% for testing. We use public datasets, we are citing all of them
Dataset Splits No For all of them, 70% of the examples were used for training and the remaining 30% for testing.
Hardware Specification Yes We give the specification of the PC employed in our experiments in the supplementary material.
Software Dependencies No No specific software dependencies with version numbers are explicitly mentioned in the main text or the provided ethics checklist.
Experiment Setup Yes Let Factor Expl be a hyper-parameter in the range [0, 1]. In terms of stopping rules, we do not expand a leaf ν if it is either located at depth 6 or if there is no test τ for which Gini(τ, ν) is smaller than Gini(Examples that reach ν).