Research map / Learning theory and algorithms / Prediction and decision losses
Generalized Linear Rule Models
Dennis Wei, Sanjeeb Dash, Tian Gao, Oktay Gunluk
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| add higher-degree rules using column generation instead of restricting the model to singleton rules (ablation) | GLRM · Regression datasets | R^2: LRR(N) advantages over LR1(N) are generally larger in regression; MEPS: 16.4 -> 16.5 (without numerical features), 16.6 | Mixed |
| add higher-degree rules using column generation (ablation) | GLRM · Classification | Brier score: musk: 2.53 E-2 -> 1.37 E-2; pima: 1.63 E-1 -> 1.66 E-1 | Mixed |
| include original numerical features without binarization (ablation) | GLRM · Classification datasets | Brier score: banknote LR1: 2.58 E-3 -> LR1N: 7.43 E-5; heart LR1: 1.33 E-1 -> LR1N: 1.34 E-1 | Mixed |
| include original numerical features without binarization (ablation) | GLRM · Regression datasets | R^2: california LR1: 70.2 -> LR1N: 72.7; wine LR1: 32.3 -> LR1N: 32.9 | Improved |
| use column generation for general rules rather than singleton rules (ablation) | GLRM · Musk | Brier score: 2.53 E-2 -> 1.37 E-2 | Improved |
| use beam search with more than one parent conjunction (ablation) | Column-generation heuristic · Rule search | search benefit | No clear effect |
| allow the heuristic to return multiple best columns instead of one (ablation) | Column-generation heuristic · Rule search | sparsity | Hurt |
| terminate each column-generation search early after finding a negative objective value (ablation) | Column-generation heuristic · Rule search | column-generation iterations | Hurt |
| use higher-degree generated rules rather than first-degree rules (ablation) | Regression rule model · Regression datasets | R^2 | Improved |
| use column generation with an integer-programming subproblem instead of the heuristic (ablation) | LRR · Classification trade-offs | Brier score | Mixed |
Closest papers by meaning
- Learning Certifiably Optimal Rule Lists for Categorical Data (JMLR 2018 · similarity 0.84)
- Approximate and Exact Enumeration of Rule Models (AAAI 2018 · similarity 0.83)
- Learning Interpretable Rules for Scalable Data Representation and Classification (TPAMI 2023 · similarity 0.83)
- Learning Interpretable Decision Rule Sets: A Submodular Optimization Approach (NeurIPS 2021 · similarity 0.83)
- Consistent Sufficient Explanations and Minimal Local Rules for explaining the decision of any classifier or regressor (NeurIPS 2022 · similarity 0.83)
- GRANDE: Gradient-Based Decision Tree Ensembles for Tabular Data (ICLR 2024 · similarity 0.81)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
