Celer: a Fast Solver for the Lasso with Dual Extrapolation

Authors: Mathurin MASSIAS, Alexandre Gramfort, Joseph Salmon

ICML 2018 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Thanks to our new dual point construction, we show significant computational speedups on multiple real-world problems.
Researcher Affiliation Academia 1INRIA, Universit e Paris-Saclay 2LTCI, T el ecom Paris Tech, Universit e Paris-Saclay.
Pseudocode Yes Algorithm 1 CYCLIC CD WITH DUAL EXTRAPOLATION, Algorithm 2 DYKSTRA S ALTERNATING PROJECTION, Algorithm 3 DYKSTRA FOR THE LASSO DUAL, Algorithm 4 CELER
Open Source Code Yes The implementation5 is done in Python and Cython (Behnel et al., 2011). 5 https://github.com/mathurinm/celer
Open Datasets Yes The dataset for this experiment is the Finance/E2006-log1p dataset (publicly available from LIBSVM8), preprocessed as follows: ... 8 http://www.csie.ntu.edu.tw/ cjlin/ libsvmtools/datasets/.
Dataset Splits No The paper mentions using specific datasets but does not explicitly provide details about training, validation, or test dataset splits (e.g., percentages, sample counts, or specific pre-defined splits for reproduction).
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory specifications, or cloud computing instances) used for running its experiments.
Software Dependencies No The implementation is done in Python and Cython (Behnel et al., 2011). The paper mentions software used but does not provide specific version numbers for Python, Cython, or any other libraries or dependencies, which are necessary for full reproducibility.
Experiment Setup Yes The algorithm is run without warm start (β0 = 0p) for λ = λmax/20, and the values of θt accel and θt res are monitored. The solver stops once a duality gap of 10^-6 is reached. The initial WS size is set to p1 = 100, except when an initialization β0 = 0p is provided (e.g., for path/sequential computations, see Section 6.3), in which case we set p1 = |Sβ0|.