pyflux
pyflux copied to clipboard
GP-NARX improvements
Decided to do more work on this before releasing the beta:
- [x] Fix the log-likelihood term
- [x] Implement iterative GP
- [ ] Implement PITC pre-conditioning
- [ ] Refactor to allow custom covariance matrices
- [ ] Custom tolerance bounds for iterative GP
Some changes to the plan. Now going for sparse GP instead of iterative implementation. Steps to finish this work off are to:
- Finish testing of inducing point model with different inference options
- Finish cythonization of new kernel functions
- Reformat standard gaussian process regression
- Make sure everything is well documented !