A Fortiori Case-Based Reasoning: From Theory to Data
Authors: Wijnand van Woerkom, Davide Grossi, Henry Prakken, Bart Verheij
JAIR 2024 | Venue PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We use this implementation to evaluate the model on a series of datasets. Through this analysis, we characterize the types of datasets that are more, or less, suitable to be described by the theory. |
| Researcher Affiliation | Academia | Wijnand van Woerkom EMAIL Department of Information and Computing Sciences Utrecht University, The Netherlands Davide Grossi EMAIL Bernoulli Institute for Maths, CS and AI University of Groningen, The Netherlands Henry Prakken EMAIL Department of Information and Computing Sciences Utrecht University, The Netherlands Bart Verheij EMAIL Bernoulli Institute for Maths, CS and AI University of Groningen, The Netherlands |
| Pseudocode | Yes | Algorithm 1: Completing a case base using Z3. |
| Open Source Code | Yes | The full implementation is available online.4 https://github.com/wijnanduu/AFCBR. |
| Open Datasets | Yes | We begin by repeating the experiment by Prakken and Ratsma (2022, Section 6) on the Churn,3 Mushroom (Schlimmer, 1981), and Admission datasets (Acharya et al., 2019). For our second evaluation of the model we use the compas recidivism dataset, published by Angwin et al. (2016) We consider the fictional welfare benefit domain first introduced by Bench-Capon (1993), and several variations on this dataset. In addition, we look at data on a real legal setting, namely the domain of Dutch tort law (Verheij, 2017). |
| Dataset Splits | No | The paper does not provide explicit training/test/validation splits. The experiments involve evaluating the fit of the 'a fortiori model' to entire datasets, rather than training a machine learning model on splits and testing its generalization. |
| Hardware Specification | No | The paper mentions using Python and the Z3 SMT solver but does not specify any particular hardware (e.g., GPU, CPU models, or memory specifications) used for running the experiments. |
| Software Dependencies | No | The paper mentions using Python, the SMT solver Z3, and the scikit-learn implementation. However, it does not provide specific version numbers for any of these software components. |
| Experiment Setup | Yes | We find appropriate values for the β parameters using the scikit-learn implementation of a maximum likelihood estimation with default parameters (Pedregosa et al., 2011), and after this is done we can simply put c(xi) := βi. |