This is the implementation for the Linear Ordinary Differential Equations Gaussian Processes (LODE-GPs), published in [1].

Installation

First of all, install SageMath. Additional required libraries are GPyTorch and PyTorch.

Then simply installing it via pip install . (or pip install -e . if edits in the code should be automatically recognized) is enough to use it.

Running

MWE.py should run out of the box calling python MWE.py, after switching to the sagemath conda environment using conda activate <sage-env-name>.

Options

By now the following features are implemented:

• Arbitrary homogenuous ODEs with constant coefficients through the A matrix
• Trainable parameters through SageMath FunctionFields (compare for the Heating system or the Bipendulum Parameterized in LODEGP.py)
• SE, Mat52 and Mat32 base kernels

[1] https://proceedings.neurips.cc/paper_files/paper/2022/hash/bcef27c5825d1ed8757290f237b2d851-Abstract-Conference.html