robustGP

WORK IN PROGRESS

Example

See the demo notebook for an example of usage

Overview

This package has been implemented in order to be able to test and prototype sequential methods based on adaptive designs and Gaussian Processes.

The name may soon be changed to something like Adaptivemethods or SURpy.

It relies on

• numpy for array manipulation
• scipy for optimization
• scikit-learn for the creation and manipulation of Gaussian Processes

Other backend can be implemented, especially for GPs

Usage

The class AdaptiveStrategy in the file SURmodel.py is the main interface for runnning this kind of experiments, and is used to define the methods related to GP (fit, add points to design, evaluate true underlying function etc)

To run an AdaptiveStrategy, one needs first to define the corresponding Enrichment.

This Enrichment may take several forms

• OneStep enrichment, which are based on Acquisition functions which are optimized to select the new points
• Sampling based enrichment, (AKMCSEnrichment)