Better data types

[ayushpatnaikgit]

Instead of Prior_Cauchy Prior_Gauss, etc

We should have

julia> abstract type Prior end

julia> struct Cauchy <: Prior end

julia> struct Gauss <: Prior end

Later on, you can have a function(::Prior), and you won't have to define it for every kind of Prior like function(::Prior_Cauchy), function(::Prior_Gauss), etc.

[ayushpatnaikgit]

Similarly, with Link functions also. We should have a parent and child system.

[ShouvikGhosh2048]

The Distributions library has a Laplace function which we use - renaming Prior_Laplace to Laplace will causeCRRao's Laplace to take preference within our library. We can use Distributions.Laplace to specify which one to use within CRRao, but users using both Distributions and CRRao will have to specify which Laplace to use. (Also TDist)

[ShouvikGhosh2048]

Similarly, with Link functions also. We should have a parent and child system.

We have a parent and child system for Link.

[sourish-cmi]

The Distributions library has a Laplace function which we use - renaming Prior_Laplace to Laplace will causeCRRao's Laplace to take preference within our library. We can use Distributions.Laplace to specify which one to use within CRRao, but users using both Distributions and CRRao will have to specify which Laplace to use. (Also TDist)

We are using Laplace in Turing. I think Turing is using Distributions only

@model LinearRegression(X, y) = begin
        p = size(X, 2)

        #priors
        a0 = 0.1
        b0 = 0.1
        v ~ InverseGamma(h, h)
        σ ~ InverseGamma(a0, b0)
        #α ~ Laplace(0, σ * v)
        β ~ filldist(Laplace(0, σ * v), p)

        #likelihood
        #y ~ MvNormal(α .+ X * β, σ)
        y ~ MvNormal(X * β, σ)
    end