Multi-value Rule Sets for Interpretable Classification with Feature-Efficient Representations

Authors: Tong Wang

NeurIPS 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments on synthetic and realworld data demonstrate that MRS models have significantly smaller complexity and fewer features than baseline models while being competitive in predictive accuracy.
Researcher Affiliation Academia Tong Wang Tippie School of Business University of Iowa Iowa City, IA 52242 tong-wang@uiowa.edu
Pseudocode No The paper states: "See the supplementary material for the complete algorithm." This indicates the algorithm details, potentially including pseudocode, are not provided within the main body of the paper itself.
Open Source Code Yes Code: The MRS code is available at https://github.com/wangtongada/MRS.
Open Datasets Yes We then evaluate the performance of MRS on six real-world data sets from law enforcement, healthcare, and demography where interpretability is most desired. The data sets are publicly available at UCI Machine Learning Repository or ICPSR. Table 1: A summary of data sets [...] Juvenile Delinquency [23] [...] Credit card [34] [...] Census [14] [...] Hospital Readmission [27]
Dataset Splits Yes Each data set is partitioned into 75% training and 25% testing. We performed 5-fold cross validation for each method. In each fold, we set aside 20% of data during training for parameter tuning and used a grid search to locate the best set of parameters.
Hardware Specification No The paper does not provide any specific details about the hardware used to run the experiments, such as GPU models, CPU types, or memory specifications.
Software Dependencies No The paper mentions using "R and python packages for the random forest, SBRL, CBA and Ripper [11]" and "the publicly available code for BRS 1", but it does not specify any version numbers for these software dependencies (e.g., Python 3.x, PyTorch 1.x, specific package versions).
Experiment Setup Yes We set entries in θ to 1, α+ = α = 100 and β+ = β = 1. αM, βM control the number of rules and αL, βL control lengths of rules. We set αM, αL to 1 and vary βM, βL. We report in Table 2 the average test error, the average number of conditions in the output model, and the average number of unique features used in each model, computed from the 5 folds.