Trading Complexity for Sparsity in Random Forest Explanations
Authors: Gilles Audemard, Steve Bellart, Louènas Bounia, Frédéric Koriche, Jean-Marie Lagniez, Pierre Marquis5461-5469
AAAI 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Our experiments made on standard benchmarks assess both the runtime complexity of finding abductive explanations and the sparsity of such explanations (i.e., how much parsimonious they are). |
| Researcher Affiliation | Academia | 1Univ. Artois, CNRS, CRIL, F-62300 Lens, France 2Institut Universitaire de France |
| Pseudocode | No | The paper describes algorithms but does not present them in a structured pseudocode block or a clearly labeled algorithm section. |
| Open Source Code | No | The paper provides links to project pages (e.g., "www.cril.univ-artois.fr/expekctation/papers.html") and mentions that 'results obtained on the other datasets are similar and available on line', but does not explicitly state that the source code for their methodology is released or provide a direct link to a code repository. |
| Open Datasets | Yes | We have focused on 15 datasets for binary classification, which are standard benchmarks from the repositories Kaggle (www.kaggle.com), Open ML (www.openml.org), or UCI (archive.ics.uci.edu/ml/). These datasets are compas, placement, recidivism, adult, ad data, mnist38, mnist49, gisette, dexter, dorothea, farm-ads, higgs boson, christine, gina, and bank. |
| Dataset Splits | Yes | For every benchmark b, a 10-fold cross validation process has been achieved: a set of 10 random forests have been computed and evaluated from the labelled instances of b, partitioned into 10 parts. One part was used as the test set and the remaining 9 parts as the training set for generating a forest. |
| Hardware Specification | Yes | Experiments have been conducted on a computer equipped with Intel(R) XEON E5-2637 CPU @ 3.5 GHz and 128 Gib of memory. |
| Software Dependencies | Yes | For this learner, we have used the version 0.23.2 of the Scikit-Learn library (Pedregosa et al. 2011). For computing prime-implicant explanations and min MAJ reasons, we took advantage of the Pysat library (Ignatiev, Morgado, and Marques-Silva 2018) (version 0.1.6.dev15) that provides the implementation of the RC2 PARTIAL MAXSAT solver and an interface to MUSER (Belov and Marques-Silva 2012). |
| Experiment Setup | Yes | The maximal depth of any decision tree in a forest has been bounded at 8. All other hyper-parameters of the learning algorithm have been set to their default value, except the number of trees. |