GPflowPILCO

A modern implementation of PILCO based on GPflow. This package focuses on model-based reinforcement learning with Gaussian processes. Alongside GPflowSampling, this code is intended as a companion for the paper Pathwise Conditioning of Gaussian Processes.

Package content

Module	Description
components	Miscellania such as objective functions and state encoders
dynamics	Generic classes for working with dynamical systems
envs	Various enviroments based on Gym
loops	High-level algorithm suites, such as variants of PILCO
models	Gaussian processes and associated methods
moment_matching	Compute the moments of, e.g., nonlinear functions of Gaussian rvs
utils	Assorted utility methods

Installation

git clone git@github.com:j-wilson/GPflowPILCO.git
cd GPflowPILCO
pip install .

Note that this package relies on GPflowSampling, which should be installed first.

Examples

Code for experiments discussed in Section 7.4 of Pathwise Conditioning of Gaussian Processes is provided in examples/cartpole_swingup.

Citing Us

If our work helps you in a way that you feel warrants reference, please cite the following paper:

@article{wilson2021pathwise,
    title={Pathwise Conditioning of Gaussian Processes},
    author={James T. Wilson
            and Viacheslav Borovitskiy
            and Alexander Terenin
            and Peter Mostowsky
            and Marc Peter Deisenroth},
    booktitle={Journal of Machine Learning Research},
    year={2021},
    url={https://arxiv.org/abs/2002.09309}
}