Practical and Efficient Rashomon Set Sampling for Model Interpretability

Sichao Li, Amanda S Barnard, Quanling Deng
Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, PMLR 300:262-270, 2026.

Abstract

Explaining a single model can be misleading when many near-optimal models (a \emph{Rashomon set}) yield different feature attributions. We frame this as a Rashomon set sampling problem and propose two practical axioms that any Rashomon sampler should satisfy: \emph{generalizability} (meaning it must accept arbitrary reference models and loss functions) and \emph{Implementation Sparsity} (meaning it should return a small, attribution-diverse subset of valid models). These two axioms are not satisfied by most known attribution methods, which we consider to be a fundamental weakness. Building on these axioms, we propose an $\epsilon$-subgradient-based sampling framework and quantify effectiveness with \emph{Search Efficiency Ratio} (SER) and \emph{Functional Explanation Range} (FER). Experiments on a synthetic quadratic task and five real-world datasets show that our sampler achieves comparable or higher FER with up to $\sim100\times$ fewer models than exhaustive baselines such as TreeFARMS, while remaining agnostic to model class and loss. Even when the reference model is sub-optimal in practice, the resulting attributions align with ground truth and accepted domain knowledge.

Cite this Paper

BibTeX
@InProceedings{pmlr-v300-li26a, title = { Practical and Efficient Rashomon Set Sampling for Model Interpretability }, author = {Li, Sichao and Barnard, Amanda S and Deng, Quanling}, booktitle = {Proceedings of The 29th International Conference on Artificial Intelligence and Statistics}, pages = {262--270}, year = {2026}, editor = {Khan, Emtiyaz and Li, Yingzhen and Solin, Arno and Ramdas, Aaditya}, volume = {300}, series = {Proceedings of Machine Learning Research}, month = {02--05 May}, publisher = {PMLR}, pdf = {https://raw.githubusercontent.com/mlresearch/v300/main/assets/li26a/li26a.pdf}, url = {https://proceedings.mlr.press/v300/li26a.html}, abstract = { Explaining a single model can be misleading when many near-optimal models (a \emph{Rashomon set}) yield different feature attributions. We frame this as a Rashomon set sampling problem and propose two practical axioms that any Rashomon sampler should satisfy: \emph{generalizability} (meaning it must accept arbitrary reference models and loss functions) and \emph{Implementation Sparsity} (meaning it should return a small, attribution-diverse subset of valid models). These two axioms are not satisfied by most known attribution methods, which we consider to be a fundamental weakness. Building on these axioms, we propose an $\epsilon$-subgradient-based sampling framework and quantify effectiveness with \emph{Search Efficiency Ratio} (SER) and \emph{Functional Explanation Range} (FER). Experiments on a synthetic quadratic task and five real-world datasets show that our sampler achieves comparable or higher FER with up to $\sim100\times$ fewer models than exhaustive baselines such as TreeFARMS, while remaining agnostic to model class and loss. Even when the reference model is sub-optimal in practice, the resulting attributions align with ground truth and accepted domain knowledge. } }
Endnote
%0 Conference Paper %T Practical and Efficient Rashomon Set Sampling for Model Interpretability %A Sichao Li %A Amanda S Barnard %A Quanling Deng %B Proceedings of The 29th International Conference on Artificial Intelligence and Statistics %C Proceedings of Machine Learning Research %D 2026 %E Emtiyaz Khan %E Yingzhen Li %E Arno Solin %E Aaditya Ramdas %F pmlr-v300-li26a %I PMLR %P 262--270 %U https://proceedings.mlr.press/v300/li26a.html %V 300 %X Explaining a single model can be misleading when many near-optimal models (a \emph{Rashomon set}) yield different feature attributions. We frame this as a Rashomon set sampling problem and propose two practical axioms that any Rashomon sampler should satisfy: \emph{generalizability} (meaning it must accept arbitrary reference models and loss functions) and \emph{Implementation Sparsity} (meaning it should return a small, attribution-diverse subset of valid models). These two axioms are not satisfied by most known attribution methods, which we consider to be a fundamental weakness. Building on these axioms, we propose an $\epsilon$-subgradient-based sampling framework and quantify effectiveness with \emph{Search Efficiency Ratio} (SER) and \emph{Functional Explanation Range} (FER). Experiments on a synthetic quadratic task and five real-world datasets show that our sampler achieves comparable or higher FER with up to $\sim100\times$ fewer models than exhaustive baselines such as TreeFARMS, while remaining agnostic to model class and loss. Even when the reference model is sub-optimal in practice, the resulting attributions align with ground truth and accepted domain knowledge.
APA
Li, S., Barnard, A.S. & Deng, Q.. (2026). Practical and Efficient Rashomon Set Sampling for Model Interpretability . Proceedings of The 29th International Conference on Artificial Intelligence and Statistics, in Proceedings of Machine Learning Research 300:262-270 Available from https://proceedings.mlr.press/v300/li26a.html.

Related Material