Mind the duality gap: safer rules for the Lasso

Authors: Olivier Fercoq, Alexandre Gramfort, Joseph Salmon

ICML 2015 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental In Section 4, we applied our GAP SAFE rules with a coordinate descent solver for the Lasso problem. Using standard data-sets, we report the time improvement compared to prior safe rules. We implemented the screening rules of Table 1 based on the coordinate descent in Scikit-learn (Pedregosa et al., 2011). Figures 4, 5 and 6 illustrate results on three datasets.
Researcher Affiliation Academia Olivier Fercoq OLIVIER.FERCOQ@TELECOM-PARISTECH.FR Alexandre Gramfort ALEXANDRE.GRAMFORT@TELECOM-PARISTECH.FR Joseph Salmon JOSEPH.SALMON@TELECOM-PARISTECH.FR Institut Mines-T el ecom, T el ecom Paris Tech, CNRS LTCI 46 rue Barrault, 75013, Paris, France
Pseudocode Yes Algorithm 1 Coordinate descent with GAP SAFE rules
Open Source Code No The paper states, 'We implemented the screening rules of Table 1 based on the coordinate descent in Scikit-learn (Pedregosa et al., 2011).', indicating use of an existing library, but does not explicitly state that the authors' own code for the described methodology is being released or provide a link to it.
Open Datasets Yes Using standard data-sets, we report the time improvement compared to prior safe rules. Figure 3 presents the proportion of variables screened by several safe rules on the standard Leukemia dataset. Figures 4, 5 and 6 illustrate results on three datasets. Figure 4 presents results on the dense, small scale, Leukemia dataset. Figure 5 presents results on a medium scale sparse dataset obtained with bag of words features extracted from the 20newsgroup dataset (comp.graphics vs. talk.religion.misc with TF-IDF removing English stop words and words occurring only once or more than 95% of the time). Text feature extraction was done using Scikit-Learn. Figure 6 focuses on the large scale sparse RCV1 (Reuters Corpus Volume 1) dataset, cf. (Schmidt et al., 2013).
Dataset Splits No The paper mentions 'cross-validation' and 'Lasso paths' but does not specify exact percentages, sample counts, or a detailed methodology for dataset splits (e.g., train/validation/test) that would allow for reproduction of the data partitioning.
Hardware Specification No The paper does not provide any specific details about the hardware used for running the experiments, such as GPU models, CPU types, or memory specifications.
Software Dependencies No The paper states: 'We implemented the screening rules of Table 1 based on the coordinate descent in Scikit-learn (Pedregosa et al., 2011).' While 'Scikit-learn' is mentioned, no specific version number is provided for it or any other software component used in the experiments.
Experiment Setup Yes In practice, we perform the dynamic screening tests every f 10 passes through the entire (active) variables. Iterations are stopped when the duality gap is smaller than the target accuracy. For each Lasso path, solutions are obtained for T 100 values of λ s, as detailed in Section 3.3. Remark that the grid used is the default one in both Scikit-Learn and the glmnet R package.