GENO -- GENeric Optimization for Classical Machine Learning

Authors: Soeren Laue, Matthias Mitterreiter, Joachim Giesen

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

Reproducibility Variable Result LLM Response
Research Type Experimental We conducted a number of experiments to show the wide applicability and efficiency of our approach. For the experiments we have chosen classical problems that come with established well-engineered solvers like logistic regression or elastic net regression, but also problems and algorithms that have been published at Neur IPS and ICML only within the last few years. The experiments cover smooth unconstrained problems as well as constrained, and non-smooth problems. To prevent a bias towards GENO, we always used the original code for the competing methods and followed the experimental setup in the papers where these methods have been introduced. We ran the experiments on standard data sets from the LIBSVM data set repository, and, in some cases, on synthetic data sets on which competing methods had been evaluated in the corresponding papers.
Researcher Affiliation Collaboration Sören Laue Friedrich-Schiller-Universität Jena & Data Assessment Solutions Gmb H soeren.laue@uni-jena.de Matthias Mitterreiter Friedrich-Schiller-Universität Jena Germany matthias.mitterreiter@uni-jena.de Joachim Giesen Friedrich-Schiller-Universität Jena Germany joachim.giesen@uni-jena.de
Pseudocode Yes Algorithm 1 Augmented Lagrangian Algorithm 1: input: instance of Problem 1 2: output: approximate solution x Rn, λ Rp, µ Rm 0 3: initialize x0 = 0, λ0 = 0, µ0 = 0, and ρ = 1 4: repeat 5: xk+1 := argminx Lρ(x, λk, µk) 6: λk+1 := λk + ρh(xk+1) 7: µk+1 := µk + ρg(xk+1) + 8: update ρ 9: until convergence 10: return xk, λk, µk
Open Source Code Yes An interface to the GENO framework can be found at http://www.geno-project.org.
Open Datasets Yes We ran the experiments on standard data sets from the LIBSVM data set repository, and, in some cases, on synthetic data sets on which competing methods had been evaluated in the corresponding papers.
Dataset Splits No The paper uses "standard data sets" and refers to "covtype.binary" and "rcv1_test.binary" (which imply splits) in figure labels, but it does not explicitly describe the training, validation, or test dataset splits or percentages in the text, nor does it cite a source that defines these splits in the context of their usage.
Hardware Specification Yes Our test machine was equipped with an eight-core Intel Xeon CPU E5-2643 and 256GB RAM.
Software Dependencies Yes We used Python 3.6, along with Num Py 1.16, Sci Py 1.2, and scikit-learn 0.20. In some cases the original code of the competing methods was written and run in MATLAB R2019.
Experiment Setup Yes For our experiments we set the regularization parameter λ = 10−4 and used real world data sets that are commonly used in experiments involving logistic regression.