Subsample Ridge Ensembles: Equivalences and Generalized Cross-Validation
Authors: Jin-Hong Du, Pratik Patil, Arun K. Kuchibhotla
ICML 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | Figure 2 shows both the GCV estimate and the asymptotic risk for the full ridge ensemble. We observe a close match of the theoretical curves and the GCV estimates. and In Figure 3, we numerically compare the optimal subsampled ridgeless ensemble with the optimal ridge predictor to verify Corollary 3.2. As we can see, their theoretical curves exactly match, and the empirical estimates in finite samples are also close to their asymptotic limits. and We compare tuning subsample size in the full ridgeless ensemble with tuning the ridge parameter on the full data in a real-world data example from multiomics. |
| Researcher Affiliation | Academia | 1Department of Statistics and Data Science, Carnegie Mellon University, Pittsburgh, PA 15213, USA. 2Department of Statistics, University of California, Berkeley, CA 94720, USA. |
| Pseudocode | No | The paper describes methods using mathematical formulations and textual descriptions but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | No | The paper does not provide any explicit statement or link indicating the availability of open-source code for the described methodology. |
| Open Datasets | Yes | This single-cell CITE-seq dataset from Hao et al. (2021) consists of 50,781 human peripheral blood mononuclear cells (PBMCs)... |
| Dataset Splits | Yes | We randomly hold out half of the cells in each cell type as a test set. |
| Hardware Specification | No | The paper does not specify any particular hardware components (e.g., GPU models, CPU models, memory sizes) used for running the experiments. |
| Software Dependencies | No | The paper mentions software like the 'glmnet' package but does not provide specific version numbers for any software dependencies required for reproducibility. |
| Experiment Setup | Yes | For the former, we search over the grid of 25 k s from nν to n spaced evenly on the log scale, with ν = 0.5 and sample size n ranges from 516 to 7864 for different cell types. For the latter, we search over the grid of 100 λ s from 10 2 to 102 spaced evenly on log scale. |