Loo cv to Bayesian Estimation of Differential Equations (challenge)

General Usage
,
11

For ArviZ.jl, you’ll need to compute the likelihoods using Turing and then package them into an array for ArviZ. There’s an example of how to do this in Quickstart · ArviZ.jl, but I’ve adapted it here:

julia> using ArviZ

julia> log_likelihood = let
           log_likelihood = pointwise_loglikelihoods(
               model, MCMCChains.get_sections(chain, :parameters)
           )
           names = sort(collect(keys(log_likelihood)); by=x->parse(Int, split(x, ",")[2][1:end-1]))
           log_likelihood_data = getindex.(Ref(log_likelihood), names)
           (; data=cat(log_likelihood_data...; dims=3))
       end;

julia> idata = from_mcmcchains(chain; log_likelihood)
InferenceData with groups:
  > posterior
  > log_likelihood
  > sample_stats

julia> loo(idata)
1×9 DataFrame
 Row │ elpd_loo  se       p_loo    n_samples  n_data_points  warning  ⋯
     │ Float64   Float64  Float64  Int64      Int64          Bool     ⋯
─────┼─────────────────────────────────────────────────────────────────
   1 │ -411.174  9.93684  105.107       3000            101    false  ⋯
2 Likes