High-dimensional Additive Gaussian Processes under Monotonicity Constraints

Authors: Andrés López-Lopera, Francois Bachoc, Olivier Roustant

NeurIPS 2022 | Conference PDF | Archive PDF | Plain Text | LLM Run Details

Reproducibility Variable Result LLM Response
Research Type Experimental We demonstrate the performance and scalability of the methodology in several synthetic examples with hundreds of dimensions under monotonicity constraints as well as on a real-world flood application. Section 5 provides the numerical experiments.
Researcher Affiliation Academia Andrés F. López-Lopera CERAMATHS, UPHF 59313 Valenciennes, France andres.lopezlopera@uphf.fr François Bachoc IMT, UMR5219 CNRS Université Paul Sabatier 31062 Toulouse, France Olivier Roustant IMT, UMR5219 CNRS INSA Toulouse 31077 Toulouse, France
Pseudocode Yes Algorithm 1 summarizes the routine of Max Mod. Algorithm 1 Max Mod for additive c GPs
Open Source Code Yes Finally, we provide open-source codes for our full framework. Both R codes and notebooks to reproduce some of the numerical results are available in the Github repository: https://github.com/anfelopera/lineq GPR.
Open Datasets No The database contains a flood study conducted by the French multinational electric utility company EDF in the Vienne river [34]. The flood database is private.
Dataset Splits No As training data, we use random Latin hypercube designs (LHDs). The remaining data are used for testing the c GPs. (No explicit validation set or percentages/counts for splits).
Hardware Specification Yes Experiments throughout this section are executed on an 11th Gen Intel(R) Core(TM) i5-1145G7 2.60GHz 1.50 GHz, 16 Gb RAM.
Software Dependencies Yes Implementations of the additive c GP framework are based on the R package lineq GPR [31]. [31] A. F. López-Lopera, lineq GPR: Gaussian process regression models with linear inequality constraints, 2021, R package version 0.3.0.
Experiment Setup Yes Input parameters: > 0, 0 > 0, d. We set 5 knots per dimension. We fix = (σ2i , i)1 i d = (1, 2). The c GP mean is obtained by averaging 103 HMC samples.