Thomas Viehmann, tv@lernapparat.de
Note: This stems from before the Tensor/Variable merge, so it is really old. I recently needed bits and ported those to PyTorch 1.4, but there are rough edges w.r.t. dimensionality of quantities and there will be unported bits.
I felt Bayesian the other day, so I ported (a tiny part of) the excellent GPFlow library to Pytorch.
I have kept the structure close to GPFlow. Most of the great ideas are from the GPFlow team, errors are my own.
The functionality is demonstrated by the following notebooks adapted from gpflow:
- Classification example
- Regression example
- Markov Chain Monte Carlo for non-gaussian likelihoods
- Upper bound for variational inference
- SVGP with minibatch training
- Multi-class classification
Todo:
- GPLVM