Implement particles for NamedTuple with array-valued field

[ghost]

Given

dept_id = [1,1,2,2,3,3,4,4,5,5,6,6]
male = [1,0,1,0,1,0,1,0,1,0,1,0]
applications = [825,108,560,25,325,593,417,375,191,393,373,341]
admit = [512,89,353,17,120,202,138,131,53,94,22,24]

we would like a varying intercepts model for each unique department

using Soss

varying_intercepts = @model (dept_id, applications, male) begin
    α ~ For(dept_id) do id
        Normal(0,10)
    end
    β ~ Normal(0,1)
    p = logistic.(α .+ β .* male)
    N = length(male)
    admit ~ For(1:N) do n
        Binomial(applications[n], p[n])
    end
end

errors

post = nuts(varying_intercepts(dept_id = dept_id, applications = applications, male = male), (admit = admit,)) |> particles

with output

(β = 0.00901 ± 0.97, α = ERROR: LoadError: MethodError: no method matching eps(::Type{Array{Float64,1}})
Closest candidates are:
  eps(!Matched::Type{Float16}) at float.jl:739
  eps(!Matched::Type{Float32}) at float.jl:740
  eps(!Matched::Type{Float64}) at float.jl:741
  ...
Stacktrace:
 [1] eps(::Particles{Array{Float64,1},1000}) at /home/oliver/.julia/packages/MonteCarloMeasurements/WZtSS/src/particles.jl:392

We would like this model to output the intercepts for each department.

reprex.zip

[cscherrer]

Thanks @OdE33. This is a current limitation in particles. We'll need to add some more methods to cover this case. I'll edit the issue title to reflect this.

Just to be sure... On my machine it works fine if I remove |> particles. Can you confirm this works for you as well?

[cscherrer]

We'll be changing nuts to dynamicHMC (more descriptive, since we also have AdvancedHMC). But at least the vector-of-vectors case is working now:

julia> post = dynamicHMC(varying_intercepts(dept_id = dept_id, applications = applications, male = male), (admit = admit,)) |> particles
(β = -0.053 ± 1.0, α = Particles{Float64,1000}[0.543 ± 1.0, 1.56 ± 0.26, 0.584 ± 1.0, 0.794 ± 0.41, -0.486 ± 1.0, -0.662 ± 0.084, -0.651 ± 1.0, -0.624 ± 0.11, -0.911 ± 1.0, -1.16 ± 0.12, -2.73 ± 1.0, -2.61 ± 0.21])