What PyFlux Needs in the Future

(aka the endless list of doom)

Models

• Sparse Gaussian Processes
• Gaussian Process State Space Models
• Online parameter estimation
• Markov blankets for models (for BBVI)
• Cyclical components for state space models
• Multivariate GAS models
• Copula models
• Kernel regression for time series
• GP-kernel search methods (automatically constructs appropriate kernel)
• Smoothed state estimates for GAS models

Inference

• Overdispersed BBVI
• Sequential Monte Carlo

Other

• Interval Reform (use Bayesian intervals or bootstrapped intervals)
• OOS validation framework for models (and model comparison)

Library Design

• Cythonization

Your contributions are very welcome!