PALM-Mean: An Efficient Spatial Branch-and-Bound Algorithm for Global Optimization of Gaussian Process Posterior Mean Functions

This repository contains the code to reproduce the PALM-Mean algorithm proposed in the paper An Efficient Spatial Branch-and-Bound Algorithm for Global Optimization of Gaussian Process Posterior Mean Functions.

Installation

pip install -r requirements.txt

Running Experiments

• Experiments can be run using the main.py script. Users need to specify the path for the .npy and .json files containing Gaussian processes (GPs) training data and hyperparameter settings. Default files are placed under data.

• Gurobi license is required to solve the lower bounding problem (MIQCP).

• Hyperparameter setting for PALM-Mean can be specified in main.py.

• We currently only support GP posterior mean with squared exponential kernel function. Support of Matern class kernel function will be released in future.

Basic Command

python main.py