Bayesian Nonparametrics Meets Data-Driven Distributionally Robust Optimization

Authors: Nicola Bariletto, Nhat Ho

NeurIPS 2024 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental Finally, we provide insights into the workings of our method by applying it to a variety of tasks based on simulated and real datasets.
Researcher Affiliation Academia Nicola Bariletto Department of Statistics and Data Sciences The University of Texas at Austin Austin, TX 78712 nicola.bariletto@utexas.edu Nhat Ho Department of Statistics and Data Sciences The University of Texas at Austin Austin, TX 78712 minhnhat@utexas.edu
Pseudocode Yes Algorithm 1 in Appendix B details the procedure...
Open Source Code Yes Code to replicate our experiments can be found at the folllowing link: https://github.com/nbariletto/BNP_ for_DRO.
Open Datasets Yes we applied our method to predict diabetes development based on a host of features, as collected in the popular and public Pima Indian Diabetes dataset. ... The Wine Quality dataset [8] and the Liver Disorders dataset [15].
Dataset Splits Yes To test our method, we randomly select 300 training observations and leave out the rest for as a test sample. Then, we randomly split the training data into 15 folds of size 20 and select, via k-fold cross validation, the optimal DP concentration parameter α over a wide grid of values.
Hardware Specification Yes All experiments were performed on a desktop with 12th Gen Intel(R) Core(TM) i9-12900H, 2500 Mhz, 14 Core(s), 20 Logical Processor(s) and 32.0 GB RAM.
Software Dependencies No The paper mentions 'scikit-learn' [31] as a used library, but does not provide specific version numbers for it or any other software dependencies crucial for reproduction.
Experiment Setup Yes Robust Criterion Parameters. For each simulated sample, we run our robust procedure setting the following parameter values: ϕ(t) = β exp(t/β) β, β {1, }, α = a/n for a {1, 2, 5, 10}, and p0 = N(0, I), where the β = setting corresponds to Ridge regression with regularization parameter α (see Proposition 2.1). Finally, we run 300 Monte Carlo simulations to approximate the criterion, and truncate the Multinomial-Dirichlet approximation at T = 50. Stochastic Gradient Descent Parameters We initialize the algorithm at θ = (0, . . . , 0) and set the step size at ηt = 50/(100 + t). The number of passes over data is set after visual inspection of convergence of the criterion value.