ProbabilityDistributions

API is: Distributions are structs, with unnormalized logdensity and ∂logdensity! methods.

logdensity(Noram{(true,false,true)}(), y, μ, σ)
∂logdensity!((∂y, ∂μ, ∂σ) Noram{(true,false,true)}(), y, μ, σ)

logdensity returns the target density in an unreduced form. ∂logdensity! returns a tuple (target,(∂y, ∂μ, ∂σ))

The true/false specifies whether the corresponding parameter is to be treated as a constant or as an unknown for the purpose of normalization. ∂logdensity! will calculate gradients for each unknown. If the corresponding entry, e.g. ∂y is nothing, it will return the calculated gradient. Otherwise, it will return Zero().