Upper bounds for Model-Free Row-Sparse Principal Component Analysis

Authors: Guanyi Wang, Santanu Dey

ICML 2020 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical results on both artificial and real cases are reported to demonstrate the advantages of our method.
Researcher Affiliation Academia 1H. Milton Stewart School of Industrial & Systems Engineering, Georgia Institute of Technology, Atlanta, USA. Correspondence to: Guanyi Wang <gwang93@gatech.edu>, Santanu Dey <santanu.dey@isye.gatech.edu>.
Pseudocode Yes Algorithm 1 Local Search Method
Open Source Code No The paper does not provide any concrete access to source code for the methodology described.
Open Datasets No The paper mentions 'The first two biological data sets (Eisen-1, Eisen-2) [d 300] are collected from ?. The Colon cancer data set [d = 500] is from ?. The Lymphoma data set [d = 500] is from ?. The final instance two real instances is collected from Reddit [d = 1000, 2000].' However, the citations are missing or generic (e.g., '?'), and no specific links or repository information are provided for public access. Artificial instances are generated, not from public datasets.
Dataset Splits No The paper describes the datasets used but does not provide specific details on how the data was split into training, validation, or test sets for reproducibility.
Hardware Specification Yes All numerical experiments are implemented on Mac Book Pro13 with 2GHz Intel Core i5 CPU and 8GB 1867MHz LPDDR3 Memory.
Software Dependencies Yes The SOCIP-impl model was solved using Gurobi 7.0.2.
Experiment Setup Yes Given φ a threshold parameter, and the size of |J+|, set the revised piecewise linear upper approximation (PLA ) constraints as, (g, ξ, η) : j vi, (j, i) [d] [r] gji = N ℓ= N SOS-II ji, (j, i) J+ [r] ξji = N ℓ= N SOS-II Thus the implemented version of SOCIP is j J+(λj φ) r i=1 ξji s + rφ s.t V C , (g, ξ, η) PLA (s-var), (sparse), (cut), (sym-1), (sym-2) (SOCIP-impl)