Orthogonal Additive Gaussian Processes

This is the code repo for the paper Additive Gaussian Processes Revisited (https://arxiv.org/pdf/2206.09861.pdf)

Getting Started

Installation

Clone the repository (https://github.com/amzn/orthogonal-additive-gaussian-processes) and install the package with pip install -e .. The package is tested with Python 3.7. The main dependency is gpflow and we relied on gpflow == 2.2.1, where in particular implements the posteriors module.

Tests

Run pytest to run the tests in the tests folder.

Key Components

• Kernels:

  • ortho_binary_kernel.py implements the constrained binary kernel

  • ortho_categorical_kernel.py implements the constrained coregional kernel for categorical variables

  • ortho_rbf_kernel.py implements the constrained squared exponential (SE) kernel for continuous variables

  • oak_kernel.py multiples and adds kernels over feature dimensions using Newton Girard method

• Measures:

  • input_measures.py implements Uniform measure, (mixture of) Gaussian measure, empirical measure for input distributions

• Normalising Flow:

  • normalising_flow.py implements normalising flows to transform input densities into Gaussian random variables

• Model API:

  • model_utils.py is the model API for model inference, prediction and plotting, and Sobol calculations

• Utilities:

  • utils.py contains utility functions
  • plotting_utils.py contains utility functions for plotting

Usage

Data

UCI benchmark data are saved in the ./data directory. They are obtained from https://github.com/duvenaud/additive-gps/blob/master/data/. Run ./data/download_data.py to download all the datasets.

Examples

Example tutorials and scripts are in the ./example directory.

UCI:

• Contains training scripts for UCI regression and classification benchmark datasets. See ./examples/uci/README_UCI.md for details.

Security

See CONTRIBUTING for more information.

License

This project is licensed under the Apache-2.0 License.