DMGP Documentation

DMGP is a Python library for sparse deep Gaussian processes with GPU acceleration. It is built on top of PyTorch and provides a simple and flexible API for building complex deep GP models. This documentation is for the GitHub Repo.

Installation

To use DMGP, make sure you have PyTorch installed, then install it using pip:

Install from GitHub

$ git clone https://github.com/warrenzha/dmgp.git
$ cd dmgp
$ pip install -r requirements.txt

Install from Package

(.venv) $ pip install -i https://test.pypi.org/simple/ dmgp

Tutorials:

• Deep Gaussian Processes
  • Deep GPs as neural networks
  • Deep GPs with the sparse grid
  • Deep GPs with the additive structure
• Installation
  • Requirements
  • Install from GitHub
  • Install from Package

Examples:

• Basic Usage
• Exact GPs (Regression)

Package Reference

• dmgp.models
• dmgp.kernels
• dmgp.layers

References

• Liang Ding, Rui Tuo, and Shahin Shahrampour. A Sparse Expansion For Deep Gaussian Processes. IISE Transactions (2023): 1-14. Code in MATLAB version.

• Rishabh Agarwal, et al. Neural Additive Models: Interpretable Machine Learning with Neural Nets. Advances in neural information processing systems 34 (2021): 4699-4711.

• Ranganath Krishnan, Pi Esposito and Mahesh Subedar. Bayesian-Torch: Bayesian neural network layers for uncertainty estimation. Code in PyTorch version.