This is a minimal implementation of Gaussian processes that runs on CUDA.jl and supports Zygote. Currently, there is not proper way to implement Gaussian processes on Julia in a way that supports both GPUs and differentiation. This package is a minimal attempt to fill the void in the meantime. For a detailed discussion, see this blog post
The main routines are in gp_cuda_utils.jl.
This package is more like a code snippet and not a proper "package".
Pplease copy gp_cuda_utils.jl to your project and use it or modify it to your use.
Currently, the implementation only has two types of kernels: squared exponential and Matern 5/2 both with automatic relevance determination.
Detailed usage can be found in main.jl
As usual, for download the package as:
git clone CUDAGaussianProcessExample.jl
cd CUDAGaussianProcessExample.jlWithin the Julia interpreter, enter:
> . activate
> instantiate
include("main.jl")
main()The results in the blog post are reproducible using benchmark.jl
main.jl contains an example where MAP-II hyperparameter inference is done using Optim.jl on the Boston housing dataset.
Here is an example result:
┌ Info: MAP-II Hyperparameter Optimization Result
│ likelihood_before = -544.3303199616416
│ likelihood_after = -116.86849745187607
│ rmse_before = 0.60338885f0
│ rmse_after = 0.3102568f0
│ lpd_before = -0.8926057396811591
└ lpd_after = -0.16185267732364805
When the Cholesky fails, the current implementation does not throw.
Instead, it spits a -Inf for the likelihood and CUDA.zeros arrays for the gradients.
MIT License
Copyright (c) 2022 Kyurae Kim
Permission is hereby granted, free of charge, to any person obtaining a copy
of this software and associated documentation files (the "Software"), to deal
in the Software without restriction, including without limitation the rights
to use, copy, modify, merge, publish, distribute, sublicense, and/or sell
copies of the Software, and to permit persons to whom the Software is
furnished to do so, subject to the following conditions:
The above copyright notice and this permission notice shall be included in all
copies or substantial portions of the Software.
THE SOFTWARE IS PROVIDED "AS IS", WITHOUT WARRANTY OF ANY KIND, EXPRESS OR
IMPLIED, INCLUDING BUT NOT LIMITED TO THE WARRANTIES OF MERCHANTABILITY,
FITNESS FOR A PARTICULAR PURPOSE AND NONINFRINGEMENT. IN NO EVENT SHALL THE
AUTHORS OR COPYRIGHT HOLDERS BE LIABLE FOR ANY CLAIM, DAMAGES OR OTHER
LIABILITY, WHETHER IN AN ACTION OF CONTRACT, TORT OR OTHERWISE, ARISING FROM,
OUT OF OR IN CONNECTION WITH THE SOFTWARE OR THE USE OR OTHER DEALINGS IN THE
SOFTWARE.