GPyConform update: now supports both Transductive and Inductive CP for GPR

[h.papad...@frederick.ac.cy]

Dear colleagues,

I wanted to share a major update to GPyConform, my open-source Python package that extends GPyTorch with Conformal Prediction for Gaussian Process Regression.

In addition to the original Transductive/Full CP support for Exact GPs, GPyConform now also includes Inductive/Split CP functionality  (current release: v0.2.0.post1):

• Transductive (Full) CP for Exact GPs via ExactGPCP

• Inductive (Split) CP for any GPyTorch regression model via GPRICPWrapper

• A model-agnostic InductiveConformalRegressor (can work beyond GPyTorch if you provide predictive mean/variance)

• Support for symmetric and asymmetric prediction intervals in both framework versions

• Various utilities and packaging/docs improvements

I also wrote a book chapter describing the package and its functionality:
https://doi.org/10.1007/978-3-032-15120-9_20

GPyConform is available via PyPI and conda-forge, and the code/docs are here:
GitHub Repository: https://github.com/harrisp/GPyConform
Documentation: https://gpyconform.readthedocs.io/en/latest/

As always, I look forward to any feedback, should you or some of your colleagues/students try it out.

Warm regards,

Harris