/informative_prior

Code for "Informative Bayesian Neural Network Priors for Weak Signals"

Primary LanguageJupyter NotebookMIT LicenseMIT

Informative BNNs prior

Code for "Informative Bayesian Neural Network Priors for Weak Signals"

Three demos are included:

  • Feature selection task: feature_selection_demo.ipynb
  • Optimizing the hyper-parameters according to prior PVE: learning_pve_demo.ipynb
  • BNN regression with informative prior: BNN_demo.ipynb

Credits

  • The Energy dataset is from here.
  • The horseshoe prior code is adapted from here.
  • The original paper of Stein gradient estimator is here, and the code is adapted from here.