Stochastic Three-Composite Convex Minimization

Authors: Alp Yurtsever, Bang Cong Vu, Volkan Cevher

NeurIPS 2016 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Numerical evidence supports the effectiveness of our method in real-world problems. We present numerical evidence to assess the theoretical convergence guarantees of the proposed algorithm. We provide two numerical examples from Markowitz portfolio optimization and support vector machines.
Researcher Affiliation Academia Laboratory for Information and Inference Systems (LIONS) École Polytechnique Fédérale de Lausanne, Switzerland alp.yurtsever@epfl.ch, bang.vu@epfl.ch, volkan.cevher@epfl.ch
Pseudocode Yes Algorithm 1 Stochastic three-composite minimization algorithm (S3CM)
Open Source Code No The paper does not provide concrete access to source code for the methodology described in this paper.
Open Datasets Yes We use 5 different real portfolio datasets: Dow Jones industrial average (DJIA, with 30 stocks for 507 days), New York stock exchange (NYSE, with 36 stocks for 5651 days), Standard & Poor s 500 (SP500, with 25 stocks for 1276 days), Toronto stock exchange (TSE, with 88 stocks for 1258 days) that are also considered in [4]; and one dataset by Fama and French (FF100, 100 portfolios formed on size and book-to-market, 23,647 days) that is commonly used in financial literature, e.g., [6, 14]. We use UCI machine learning dataset a1a , with d = 1605 datapoints and p = 123 features [8, 18].
Dataset Splits No The paper states "We split all the datasets into test (10%) and train (90%) partitions randomly." but does not explicitly mention a validation dataset split.
Hardware Specification No The paper does not provide specific hardware details (like GPU/CPU models or processor types) used for running its experiments.
Software Dependencies No The paper does not provide specific ancillary software details (e.g., library or solver names with version numbers) needed to replicate the experiment.
Experiment Setup Yes For the deterministic algorithm, we set η = 0.1. ... We simply use γn = γ0/(n + 1) with γ0 = 1 for FF100, and γ0 = 103 for others. ... We start both algorithms from the zero vector. ... We set the desired return as the average return over all assets in the training set, b = mean(aav). and we fix problem parameters C = 1 and σ = 2 2, and we focus on the effects of algorithmic parameters on the convergence behavior. ... We use S3CM with the learning rate γn = γ0/(n + 1) for various values of γ0.