GPyTorch is a Gaussian process library implemented using PyTorch. GPyTorch is designed for creating scalable, flexible, and modular Gaussian process models with ease.
Internally, GPyTorch differs from many existing approaches to GP inference by performing all inference operations using modern numerical linear algebra techniques like preconditioned conjugate gradients. Implementing a scalable GP method is as simple as providing a matrix multiplication routine with the kernel matrix and its derivative via our LazyVariable interface, or by composing many of our already existing LazyVariables. This allows not only for easy implementation of popular scalable GP techniques, but often also for significantly improved utilization of GPU computing compared to solvers based on the Cholesky decomposition.
GPyTorch provides (1) significant GPU acceleration (through MVM based inference); (2) state-of-the-art implementations of the latest algorithmic advances for scalability and flexibility (SKI/KISS-GP, stochastic Lanczos expansions, LOVE, SKIP, stochastic variational deep kernel learning, ...); (3) easy integration with deep learning frameworks.
Right now, the package is in alpha release, and while we believe that the interface is reasonably stable, things may change. For now, see our numerous examples and tutorials on how to construct all sorts of models in GPyTorch. Here are some examples:
- Simple GP regression and classification
- Multitask GP regression (example here)
- Scalable GP regression and classification using kernel interpolation
- Deep kernel learning for regression
- Large scale stochastic deep kernel learning for classification and regression
If you use GPyTorch, please cite the following papers:
@inproceedings{gardner2018product,
title={Product Kernel Interpolation for Scalable Gaussian Processes},
author={Gardner, Jacob R and Pleiss, Geoff and Wu, Ruihan and Weinberger, Kilian Q and Wilson, Andrew Gordon},
booktitle={AISTATS},
year={2018}
}
@inproceedings{pleiss2018constant,
title={Constant-Time Predictive Distributions for Gaussian Processes},
author={Pleiss, Geoff and Gardner, Jacob R and Weinberger, Kilian Q and Wilson, Andrew Gordon},
booktitle={ICML},
year={2018}
}
The easiest way to install GPyTorch is by installing PyTorch >= 0.4.0 using the appropriate command from here, and then installing
GPyTorch using pip:
pip install git+https://github.com/cornellius-gp/gpytorch.gitTo use packages globally but install GPyTorch as a user-only package, use pip install --user above.
We also provide two conda environment files, environment.yml and environment_cuda90.yml. As an example, to install GPyTorch in a conda environment with cuda support, run:
git clone git+https://github.com/cornellius-gp/gpytorch.git
conda create -f gpytorch/environment_cuda.yml
source activate gpytorch
pip install gpytorch/- For tutorials and examples, check out the examples folder.
- For in-depth documentation, check out our read the docs.
To run the unit tests:
python -m unittestBy default, the random seeds are locked down for some of the tests. If you want to run the tests without locking down the seed, run
UNLOCK_SEED=true python -m unittestPlease lint the code with flake8.
pip install flake8 # if not already installed
flake8GPyTorch is developed at Cornell University by
- Jake Gardner (lead developer)
- Geoff Pleiss (lead developer)
- Kilian Weinberger
- Andrew Gordon Wilson
We would like to thank our other contributors including (but not limited to) Max Balandat, Ruihan Wu, Bram Wallace, Jared Frank.
Development of GPyTorch is supported by funding from the Bill and Melinda Gates Foundation.