CalcGP: Numerical Calculus via Gaussian Process Regression
Introduction
CalcGP is a Gaussian Process Regression framework built as an alternative for numerical integration of gradients and differentiation of scalar functions. The package is based on the autograd framework JAX.
CalcGP is intended to be used as a regression framework for scalar functions and gradients of scalar functions. One can
- directly fit a scalar function from observations,
- "integrate" the gradient of a scalar function,
- "differentiate" a scalar function in order to get its gradient.
Examples for all these cases can be found in ./examples/
. There is a 1D example that shows all three use cases, a 2D example on how to handle higher dimensional data, and an example that shows a sparse model for large datasets.
Installation
Download the package from github via
git clone https://github.com/LukasEin/calcgp.git
Then, to install the package directly, run
python3 setup.py install --user
or to install it in a conda environment run
conda create -n myenv python=3.8
conda activate myenv
python3 setup.py install
direcly in the newly created ./calcgp/
folder.