OKRidge: Scalable Optimal k-Sparse Ridge Regression
Authors: Jiachang Liu, Sam Rosen, Chudi Zhong, Cynthia Rudin
NeurIPS 2023 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | 4 Experiments We test the effectiveness of our OKRidge on synthetic benchmarks and sparse identification of nonlinear dynamical systems (SINDy)[19]. |
| Researcher Affiliation | Academia | Jiachang Liu, Sam Rosen, Chudi Zhong, Cynthia Rudin Duke University {jiachang.liu, sam.rosen, chudi.zhong}@duke.edu, cynthia@cs.duke.edu |
| Pseudocode | Yes | In Appendix E, we provide visual illustrations of Bn B and beam search as well as the complete pseudocodes of our algorithms. |
| Open Source Code | Yes | Implementations of OKRidge discussed in this paper are available at https://github.com/jiachangliu/OKRidge. |
| Open Datasets | No | The paper uses synthetic data generated according to a described process, and refers to dynamical systems (Lorenz, Hopf, MHD) via the Py SINDy library [27, 42] and a model paper [24], but does not provide direct access links or formal citations for the specific datasets used in their experiments. |
| Dataset Splits | Yes | Model selection was performed via cross-validation with training data encompassing the first 2/3rds of a trajectory and the final third used for validation. |
| Hardware Specification | Yes | All experiments were run on the 10x Tensor EX TS2-673917-DPN Intel Xeon Gold 6226 Processor, 2.7Ghz. We set the memory limit to be 100GB. |
| Software Dependencies | Yes | The Gurobi version is 10.0, which can be installed through conda (https://anaconda.org/gurobi/ gurobi). We used the Academic Site License. We implemented the perspective formulations and relaxed convex optimal perspective formulations in MOSEK. The MOSEK version is 10.0, which can be installed through conda (https://anaconda.org/MOSEK/mosek). |
| Experiment Setup | Yes | We set a 1-hour time limit and an optimality gap of relative tolerance 10 4. We use a value of 0.001 for λ2. For both MIOSR and our method OKRidge, the ridge regression hyperparameter choices are λ2 {10 5, 10 3, 10 2, 0.05, 0.2}, and the sparsity level hyperparameter choices are k {1, 2, 3, 4, 5}. As in the experiments of MIOSR [9], we set the time limit for each optimization to be 30 seconds for both MIOSR and our method OKRidge. |