Simultaneous Safe Screening of Features and Samples in Doubly Sparse Modeling
Authors: Atsushi Shibagaki, Masayuki Karasuyama, Kohei Hatano, Ichiro Takeuchi
ICML 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details
| Reproducibility Variable | Result | LLM Response |
|---|---|---|
| Research Type | Experimental | We first theoretically investigate the synergy effect, and then illustrate the practical advantage through intensive numerical experiments for problems with large numbers of features and samples. 6. Numerical experiments We demonstrate the advantage of simultaneous safe screening through numerical experiments. |
| Researcher Affiliation | Academia | Nagoya Institute of Technology, Nagoya, 466-8555, Japan Kyushu University, Fukuoka, 819-0395, Japan |
| Pseudocode | No | The paper contains mathematical formulations and theorems but does not include any structured pseudocode or algorithm blocks. |
| Open Source Code | Yes | We wrote the code in C++. The code is available on https://github.com/takeuchi-lab/s3fs. |
| Open Datasets | Yes | Table 1 summarizes the datasets used in the experiments. We picked up four datasets whose numbers of features and samples are both large from libsvm dataset repository (Chang & Lin, 2011). |
| Dataset Splits | No | The paper lists datasets but does not provide specific details on how they were split into training, validation, or test sets for the experiments. |
| Hardware Specification | Yes | All the computations were conducted by using a single core of an Intel Xeon CPU E5-2643 v2 (3.50GHz), 64GB MEM. |
| Software Dependencies | No | The paper states 'We wrote the code in C++' and mentions using 'Proximal Stochastic Dual Coordinate Ascent (SDCA)' and 'Stochastic Primal Dual Coordinate (SPDC)' but does not provide specific version numbers for these or any other software libraries or compilers. |
| Experiment Setup | Yes | We set λmax := Z 1 , and considered problems with various values of the penalty parameter λ between λmax and 10 4λmax. The parameter in the smoothed hinge loss γ is set to be 0.5. whenever possible, we used dynamic safe screening strategies (Bonnefoy et al., 2014) in which safe screening rules are evaluated every time the duality gap Gλ( ˆw, ˆα) was 0.1 times smaller than before. |