ArviZ.jl

Documentation for ArviZ.jl ↗

FlexiChains contains an extension that allows you to convert a FlexiChain into an InferenceObjects.InferenceData (InferenceObjects.jl is one of the sublibraries of ArviZ.jl, and is re-exported by ArviZ).

InferenceObjects.convert_to_inference_data Function

InferenceObjects.convert_to_inference_data(
    chain::FlexiChain{<:TKey}; group::Symbol = posterior, kwargs...
) where {TKey}

Convert a FlexiChain to an InferenceObjects.InferenceData object.

The group keyword argument specifies the group name for the chain's parameters. The chain's extras are assigned to :sample_stats if group is :posterior, and to :sample_stats_<group> otherwise.

Other keyword arguments are passed to InferenceObjects.convert_to_dataset.

source

For example:

using InferenceObjects, FlexiChains, DynamicPPL, Distributions

@model function f(y)
    x ~ Normal()
    y ~ Normal(x)
end
model = f(1.0)

chn = FlexiChains._make_prior_chain(model, 100, 2)
idata = InferenceObjects.convert_to_inference_data(chn)

InferenceData

posterior

┌ 100×2 Dataset ┐
├───────────────┴─────────────────────────────────────── dims ┐
  ↓ draw Sampled{Int64} 1:100 ForwardOrdered Regular Points,
  → chain Sampled{Int64} 1:2 ForwardOrdered Regular Points
├───────────────────────────────────────────────────── layers ┤
  :x eltype: Float64 dims: draw, chain size: 100×2
├─────────────────────────────────────────────────── metadata ┤
  Dict{String, Any} with 1 entry:
  "created_at" => "2026-09-01T08:11:20.339"
└─────────────────────────────────────────────────────────────┘

sample_stats

┌ 100×2 Dataset ┐
├───────────────┴───────────────────────────────────────── dims ┐
  ↓ draw Sampled{Int64} 1:100 ForwardOrdered Regular Points,
  → chain Sampled{Int64} 1:2 ForwardOrdered Regular Points
├─────────────────────────────────────────────────────── layers ┤
  :logprior      eltype: Float64 dims: draw, chain size: 100×2
  :loglikelihood eltype: Float64 dims: draw, chain size: 100×2
  :logjoint      eltype: Float64 dims: draw, chain size: 100×2
├───────────────────────────────────────────────────── metadata ┤
  Dict{String, Any} with 1 entry:
  "created_at" => "2026-09-01T08:11:20.934"
└───────────────────────────────────────────────────────────────┘

You can combine multiple InferenceData objects with merge:

llike_chn = DynamicPPL.pointwise_loglikelihoods(model, chn)
idata2 = InferenceObjects.convert_to_inference_data(llike_chn; group=:log_likelihood)
idata_merged = merge(idata, idata2)

InferenceData

posterior

┌ 100×2 Dataset ┐
├───────────────┴─────────────────────────────────────── dims ┐
  ↓ draw Sampled{Int64} 1:100 ForwardOrdered Regular Points,
  → chain Sampled{Int64} 1:2 ForwardOrdered Regular Points
├───────────────────────────────────────────────────── layers ┤
  :x eltype: Float64 dims: draw, chain size: 100×2
├─────────────────────────────────────────────────── metadata ┤
  Dict{String, Any} with 1 entry:
  "created_at" => "2026-09-01T08:11:20.339"
└─────────────────────────────────────────────────────────────┘

log_likelihood

┌ 100×2 Dataset ┐
├───────────────┴─────────────────────────────────────── dims ┐
  ↓ draw Sampled{Int64} 1:100 ForwardOrdered Regular Points,
  → chain Sampled{Int64} 1:2 ForwardOrdered Regular Points
├───────────────────────────────────────────────────── layers ┤
  :y eltype: Float64 dims: draw, chain size: 100×2
├─────────────────────────────────────────────────── metadata ┤
  Dict{String, Any} with 1 entry:
  "created_at" => "2026-09-01T08:11:23.294"
└─────────────────────────────────────────────────────────────┘

sample_stats

┌ 100×2 Dataset ┐
├───────────────┴───────────────────────────────────────── dims ┐
  ↓ draw Sampled{Int64} 1:100 ForwardOrdered Regular Points,
  → chain Sampled{Int64} 1:2 ForwardOrdered Regular Points
├─────────────────────────────────────────────────────── layers ┤
  :logprior      eltype: Float64 dims: draw, chain size: 100×2
  :loglikelihood eltype: Float64 dims: draw, chain size: 100×2
  :logjoint      eltype: Float64 dims: draw, chain size: 100×2
├───────────────────────────────────────────────────── metadata ┤
  Dict{String, Any} with 1 entry:
  "created_at" => "2026-09-01T08:11:20.934"
└───────────────────────────────────────────────────────────────┘

From here you can use the full functionality of ArviZ.jl, which includes various plotting and analysis tools: please see the ArviZ.jl documentation for more info.