Interface

BayesFlux.NetworkPriorType
abstract type NetworkPrior end

All priors over the network parameters are subtypes of NetworkPrior.

Required Fields

  • num_params_hyper: Number of hyper priors (only those that should be inferred)
  • nc::NetConstructor: NetConstructor

Required implementations

Each NetworkPrior must be callable in the form of

(np::NetworkPrior)(θnet, θhyper)
  • θnet is a vector of network parameters
  • θhyper is a vector of hyper parameters

The return value must be the logprior including all hyper-priors.

Each NetworkPrior must also implement a sample function returning a vector θnet of network parameters drawn from the prior.

sample_prior(np::NetworkPrior)(rng::AbstractRNG)