T-Processes Subclass of Gaussian-Processes

[conradstevens]

I am implementing T-Processes (TP)s In addition to Gaussian Processes (GP)s.

Due to the great similarities in structure and functionality I have made a new class TProcessRegressor a subclass of GaussianProcessRegressor. However, this will result in duplicate code.

If it where up to me, there would be a class _stochasticProcessRegressor that GPs and TPs would both inherit from. This would minimize duplicate code and make integrating other stochastic processes (eg Gamma Processes) more straight forward. However, this architecture change would complicate an eventual pull request.

I am interested to hear peoples thoughts on implementing TPs and the pros and cons of the two architectures.

T-Process Background:
https://arxiv.org/abs/1402.4306