Experiments for "Structured Bayesian Gaussian process latent variable model: applications to data-driven dimensionality reduction and high-dimensional inversion"

See Atkinson and Zabaras (2019)

Prerequisites

This repository uses python >=3.5 and structured-gpflow. You can try running the included configure.bash to set things up quickly.

Running the experiments

The main experiment modules are:

• elliptic_forward_sgplvm_separate.py (Forward problem with two-model approach)
• elliptic_forward_sgplvm_joint.py (Forward problem with jointly-trained model)
• elliptic_inverse_sgplvm_separate.py ( Inverse problem with two-model approach)
• elliptic_inverse_sgplvm_joint.py (Inverse problem with jointly-trained model)

bash scripts that help with suggested parameter settings from the paper can be found in the scripts directory.

Data and plotting

Data used for the experiments in the paper is provided in elliptic/data. The MATLAB script for plotting the files outputted by the experiements can be found at ellitpic/matlab/plot_predictions.m.

If you have any questions, please email Steven Atkinson or Nicholas Zabaras