/gprn-notebook

Gaussian Process Regression Networks research notebook

Primary LanguageJupyter NotebookMIT LicenseMIT

Binder

GPRN Notebook

Alexander Von Moll

A research notebook for exploring the usage of bayespy for implementing Gaussian Process Regression Networks.

Tasks

  • Draw graph of all variables, parameters, and hyperparameters for GPRN
    • Gaussian process regression network factor graph
  • Implement GPRN
    • Build GPRN framework in bayespy for learning hyperparameters
      • Translate graph into code
      • Implement update message for $a_j$ as described in the paper
      • Implement update for $\theta_f$
      • Implement update for $\theta_W$
    • Implement MCMC sampling (possibly using elliptical slice sampling)
    • Build a simple API
  • Obtain data used in the paper
  • Test new implementation using data