Can't apply Bijectors.ordered to TDist(), says it's constrained

[boogiewoogit]

The following fails with the error ArgumentError: ordered transform is currently only supported for unconstrained distributions.:

using Turing

Bijectors.ordered(filldist(TDist(1),2))

However, the t-distribution is supported on the real line and thus in fact unconstrained in the sense of this package.

[Red-Portal]

Could you file this to Bijectors.jl?

[Red-Portal]

The reason seems to be that TDist's support result in the wrong type:

julia> bijector(TDist(1))
Bijectors.TruncatedBijector{Float64, Float64}(-Inf, Inf)

On a side note, Distributions also has MvTDist for this purpose, but this doesn't work either, which I think it should. So this is also a bug.

[Red-Portal]

In the meantime, you can bypass the check by calling the constructor directly:

julia> Bijectors.transformed(filldist(TDist(1), 2), Bijectors.OrderedBijector())

MultivariateTransformed{Product{Continuous, TDist{Float64}, FillArrays.Fill{TDist{Float64}, 1, Tuple{Base.OneTo{Int64}}}}, Bijectors.OrderedBijector}(
dist: Product{Continuous, TDist{Float64}, FillArrays.Fill{TDist{Float64}, 1, Tuple{Base.OneTo{Int64}}}}(v=Fill(TDist{Float64}(ν=1.0), 2))
transform: Bijectors.OrderedBijector()
)

But please report this to Bijector when you can since there seem to be at least two to three bugs here!

[boogiewoogit]

Thanks a lot for sharing the method to bypass!

I have filed the issue at Bijectors.jl: TuringLang/Bijectors.jl#287.