Enumerate Lasso Solutions for Feature Selection

Authors: Satoshi Hara, Takanori Maehara

AAAI 2017 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We confirm our theoretical results also in numerical simulations. Finally, in the gene expression and the text data, we demonstrate that the proposed method can enumerate a wide variety of meaningful feature sets, which are overlooked by the global optima.
Researcher Affiliation Academia Satoshi Hara,1,3 Takanori Maehara2,3 1. National Institute of Informatics, Tokyo, Japan 2. Shizuoka University, Shizuoka, Japan 3. JST, ERATO, Kawarabayashi Large Graph Project
Pseudocode Yes Algorithm 1 Enumeration algorithm
Open Source Code Yes 1The experiment codes are available at https://github.com/ sato9hara/Lasso Variants
Open Datasets Yes We used the thaliana gene expression data used in (Atwell et al. 2010). The 20 Newsgroups 2 is a dataset for text categorization.
Dataset Splits No The paper states: "we randomly split samples into 134 training samples and 33 test samples." and later "The training set comprised n = 1, 168 samples with p = 11, 648 words, whereas the test set consisted of n = 777 samples." It does not explicitly mention a separate validation split or dataset.
Hardware Specification Yes All experiments were conducted on 64-bit Cent OS 6.7 with an Intel Xeon E5-2670 2.6GHz CPU and 512GB RAM.
Software Dependencies Yes All codes were implemented in Python 3.5 with scikit-learn1.
Experiment Setup Yes For the training set, we applied Algorithm 1 with the regularization parameter ρ = 0.1n and the support parameter η = 0.05. Using Algorithm 1, we enumerated top-50 solutions with the regularization parameter ρ = 0.001n and the support parameter η = 4.