Feedforward Likelihoods

FeedforwardNormal(nc::NetConstructor{T, F}, prior_σ::D) where {T, F, D<:Distribution}

Use a Gaussian/Normal likelihood for a Feedforward architecture with a single output.

Assumes is a single output. Thus, the last layer must have output size one.

Arguments

• nc: obtained using destruct
• prior_σ: a prior distribution for the standard deviation

FeedforwardTDist(nc::NetConstructor{T, F}, prior_σ::D, ν::T) where {T, F, D}

Use a Student-T likelihood for a Feedforward architecture with a single output and known degress of freedom.

Assumes is a single output. Thus, the last layer must have output size one.

Arguments

• nc: obtained using destruct
• prior_σ: a prior distribution for the standard deviation
• ν: degrees of freedom