Base.show(::ParameterDistribution) error with Gaussian Random Fields

[haakon-e]

To reproduce (from ParameterDistributions test (link):

import GaussianRandomFields
const GRF = GaussianRandomFields
using EnsembleKalmanProcesses.ParameterDistributions
dim = 1
dofs = 10
pts = collect(0:0.01:1)
pkg = GRFJL()
grf = GRF.GaussianRandomField(GRF.CovarianceFunction(dim, GRF.Matern(0.05, 1)), GRF.KarhunenLoeve(dofs), pts)
d5 = GaussianRandomFieldInterface(grf, pkg)
c5 = bounded(-5, 5)
name5 = "grf_in_min5_5"
u5 = ParameterDistribution(d5, c5, name5)

gives this error:

julia> u5 = ParameterDistribution(d5, c5, name5)
Error showing value of type ParameterDistribution{GaussianRandomFieldInterface, Constraint{Bounded}, String}:
ERROR: BoundsError: attempt to access 1-element Vector{Constraint{Bounded}} at index [[1, 2, 3, 4, 5, 6, 7, 8, 9, 10]]
Stacktrace:
   [1-3] ⋮ internal
       @ Base, Unknown
     [4] getindex(A::Vector{Constraint{Bounded}}, I::Vector{Int64})
       @ Base ./abstractarray.jl:1294
     [5] show(io::IOContext{Base.TTY}, distributions::ParameterDistribution{GaussianRandomFieldInterface, Constraint{Bounded}, …})
       @ EnsembleKalmanProcesses.ParameterDistributions ~/Documents/CliMA/EnsembleKalmanProcesses.jl/src/ParameterDistributions.jl:447
  [6-22] ⋮ internal
       @ Base.Multimedia, REPL, AbbreviatedStackTraces, REPL.LineEdit, Unknown
Use `err` to retrieve the full stack trace.

[haakon-e]

Base.show() currently looks like this:

function Base.show(io::IO, distributions::ParameterDistribution)
    n = length(distributions.name)
    out = "ParameterDistribution with $n entries: \n"
    for (i, inds) in enumerate(batch(distributions))
        dist = distributions.distribution[i]
        dist_string = replace("$dist", "\n" => " ")  # hack to remove `\n` from `Parameterized(FullNormal(...))`
        cons = distributions.constraint[inds]
        nam = distributions.name[i]
        out *= "'$(nam)' with $(cons) over distribution $dist_string \n"
    end
    print(io, out)
end

The error seems to be that I try to index into the contraints:

cons = distributions.constraint[inds]

where

inds = [1, 2, 3, 4, 5, 6, 7, 8, 9, 10]
distributions.constraint
1-element Vector{Constraint{Bounded}}:
 Constraint{Bounded} with bounds (-5, 5)

i.e., I am trying to index into a 1-vector with the indices 1:10, which obviously won't work.

@odunbar do you have any insight into GRFs that may be useful to resolve this?
Perhaps we should loop over distributions in some other way for GRFs?
And more generally, what is useful information to be printed about GRFs?

[odunbar]

I think this is resolved by using the keyword option function_parameter_opt=constraint in the call to batch() in the loop. This will return 1 for inds always as required.

More generally I agree that for (1) long VectorOfParameterized distributions, and (2) GaussianRandomFieldInterface distributions there may be more readable summary information that can be provided