Closed willtebbutt closed 4 years ago
hm. seem wrong
I don't think we need to test for Real
eltype either
should probably just test for Univariate and Multivariate distributions?
but @glennmoy may know better
It seems overly-restrictive
Yeah I think is just a legacy of when it lived in our internal code base. I don't think there was a conscious decision to restrict them so much - at least going by the initial review there was no comment made about it.
I think it's reasonable to replace with <:Univariate
and <:Multivariate
(I assume that's what you mean?) and not worry about the eltype as @nickrobinson251 says.
The
Models.TestUtils
tests currently assume that single-output regressors return vectors ofNormal
distributions, and that multi-output regressors returnVector
s ofMvNormal
s. See here and here resp.@glennmoy @nickrobinson251 why are we doing this? It seems overly-restrictive.