How to cite and reference prior work

ptgp borrows heavily from existing GP libraries (GPflow, GPJax, GPyTorch) and from the papers behind the methods. One of the goals of the library is to make GPs more accessible to practitioners, and part of that is making it easy for users to learn more about the modeling choices they're making.

This is especially important now that AI coding tools are generating code that uses these methods. Users should be able to trace back from the code to the underlying research.

What's the best way to surface references and citations?

Some options:

• Docstrings on objectives/models — e.g. collapsed_elbo could cite Titsias (2009), FourierFeatures1D could cite Hensman, Durrande, Solin (2017). Short and close to where the user is working.
• A references page in documentation — a central bibliography that docstrings can point to. Keeps docstrings clean but adds a layer of indirection.
• Inline comments in the code — e.g. "# See Eq. 3.12 in Rasmussen & Williams (2006)". Useful for implementers but not visible to users who only read docs.
• A "learn more" pattern — each model/objective/kernel class has a references attribute or a section in its docstring with a short reading list.

These aren't mutually exclusive. What combination would be most useful?

Start by reviewing the docstrings and documentation for objectives and models such as collapsed_elbo and FourierFeatures1D, then compare how the issue's proposed citation patterns would surface research to users. Done means agreeing on a citation approach and documenting the resulting references or documentation structure.