Using Benson’s Algorithm for Regularization Parameter Tracking

Authors: Joachim Giesen, Sӧren Laue, Andreas Lӧhne, Christopher Schneider3689-3696

AAAI 2019 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Experiments for the Elastic Net on real world data sets demonstrate the effectiveness of Benson s algorithm for regularization parameter tracking.
Researcher Affiliation Academia Joachim Giesen, S oren Laue, Andreas L ohne Friedrich-Schiller-Universit at Jena Faculty of Mathematics and Computer Science Ernst-Abbe-Platz 2 07743 Jena, Germany Christopher Schneider Ernst-Abbe-Hochschule Jena Fachbereich Grundlagenwissenschaften Carl-Zeiss-Promenade 2 07745 Jena, Germany
Pseudocode Yes Algorithm 1 Benson Algorithm
Open Source Code No The paper mentions using
Open Datasets Yes For our experiments, we use the following data sets, which are well-known from the literature: ALLAML with m = 7,129 features and n = 72 instances, arcene with m = 10,000 and n = 200, GLI-85 with m = 22,283 and n = 85, GLIOMA with m = 4,434 and n = 50, Prostate-GE with m = 5,966 and n = 102, SMK-CAN-187 with m = 19,993 and n = 187, Carcinom with m = 9,182 and n = 174, 14cancer with m = 16,063 and n = 198.
Dataset Splits No For all data sets, 70% of the data have been used for training, and 30% have been hold out for testing. There is no explicit mention of a validation split percentage or how it was used.
Hardware Specification No The paper does not provide specific hardware details (e.g., GPU/CPU models, memory) used for running its experiments.
Software Dependencies Yes In our implementation of Algorithm 1, we used Gurobi (Gurobi Optimization 2016) for solving the scalarized problems (Pw). Facet enumeration is done by bensolve tools (Ciripoi, L ohne, and Weißing 2018).
Experiment Setup Yes Second, we compute a fine-mesh solution with Grid Search by solving (EN ) for all α, β {0, 0.01, 0.02, . . . , 1}. We then run Benson s algorithm with approximation errors ε = 0.1 (for the first six data sets) and ε = 1 (for the last two data sets), resp., depending on the scale of the objective function values. We fixed the direction parameter c = (1, . . . , 1)T.