Conformalization of Sparse Generalized Linear Models

Authors: Etash Kumar Guha, Eugene Ndiaye, Xiaoming Huo

ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We show how our path-following algorithm accurately approximates conformal prediction sets and illustrate its performance using synthetic and real data examples. and 6. Numerical Experiments Our central claim is twofold. Our method efficiently and accurately generates the homotopy over general loss functions. Our method also efficiently and accurately generates conformal sets over general loss functions. We demonstrate these two claims over different datasets and loss functions.
Researcher Affiliation Collaboration 1College of Computing, Georgia Institute of Technology, Atlanta, GA, USA 2 Apple (Work partly done while at Georgia Tech) 3H. Milton Stewart School of Industrial and Systems Engineering, Georgia Institute of Technology, Atlanta, GA, USA.
Pseudocode Yes Algorithm 1 Full Homotopy Generation and Algorithm 2 Conformal Set Generation
Open Source Code Yes For reproducibility, our implementation is at github.com/Etash Guha/sparse_conformal.
Open Datasets Yes The first three are real datasets sourced from (Pedregosa et al., 2011). The Diabetes dataset is a regression dataset with 20 features and 442 samples. Additionally, we use the well-known regression dataset from (H., 1991) denoted as Friedman1, which has 10 features and 100 samples. We also use the multivariate dataset denoted Friedman2 from (Breiman, 1996), which has 100 samples and 4 features.
Dataset Splits No The paper describes the datasets used (Diabetes, Friedman1, Friedman2, Synthetic) but does not provide explicit details on how these datasets were split into training, validation, and test sets, nor does it mention cross-validation strategies or predefined splits.
Hardware Specification No The paper does not provide specific hardware details (e.g., CPU, GPU models, memory, or cloud instances) used for running the experiments.
Software Dependencies No The paper mentions using 'Proximal Gradient Descent for Lasso Loss and CVXPY for Robust and Asymmetric as Primal Correctors' and cites SKGLM, but it does not provide specific version numbers for any software libraries or dependencies.
Experiment Setup No The paper mentions 'For our experiments, we used α = 0.1' but does not provide a comprehensive set of specific experimental setup details such as learning rates, batch sizes, optimizer settings, or other hyperparameters needed for full reproducibility.