/laplax

Bayesian Neural Networks with Laplace Approximation

Primary LanguagePython

Automatic Differentiation Laplace Approximation (ADLaplace)

  • All prior (unconstrained) distributions should return a single log prob value.
  • All prior (unconstrained) distributions should have at most 1d array as their sample.

To Do

  • Replace get_likelihood_params with get_likelihood.