Frequently asked questions

Short answers to the questions this package tends to raise. For a full walkthrough see Composing distributions, and for the verb map see the Concepts page.

Where did censoring go?

ComposedDistributions is the generic composition layer, split out from CensoredDistributions.jl. It composes any Distributions.jl UnivariateDistribution and adds no censoring of its own. If you need primary-event censoring, interval censoring, or truncation of the observation process, use CensoredDistributions.jl, which builds its censored leaves on top of this composition algebra. Right-truncation of a plain leaf is still available through truncated() from Distributions.jl, since any leaf is an ordinary distribution.

What does rand of a Resolve or Compete node return?

A one_of node draws the named event record of the outcome that fired: a NamedTuple keyed by event_names, a positional origin slot then one slot per outcome, with the fired outcome's time present and the othersmissing. The record names which outcome occurred, so it feeds straight back intologpdf.

using ComposedDistributions, Distributions, Random

node = resolve(:death => (Gamma(1.5, 1.0), 0.3), :recover => Gamma(2.0, 1.5))
rand(Xoshiro(1), node)

(event_1 = 0.0, death = 1.8400307468613506, recover = missing)

For the compact (outcome, time) pair pass rand(node; outcome = true); for the marginal time to resolution alone (discarding which outcome fired) sample as_mixture(node).

rand(Xoshiro(1), node; outcome = true),
rand(Xoshiro(1), as_mixture(node))

((:death, 1.8400307468613506), 1.8400307468613506)

I'm migrating from CensoredDistributions' compose: what changed?

CensoredDistributions' compose took varargs pairs, compose(:a => d1, :b => d2). This package's compose takes a NamedTuple, compose((a = d1, b = d2)), as its primary spelling, because a NamedTuple also carries recursively-nested branch names for tables and matrices, where a flat pairs list cannot. The varargs-pairs spelling still works: compose(:a => d1, :b => d2, ...) is a thin convenience method that lowers to the NamedTuple form and returns the identical stack, so existing CensoredDistributions-style call sites keep working unmodified. Prefer the NamedTuple form in new code.

How do I get the total-delay distribution of a chain?

A Sequential chain models each step separately; to collapse it to the single distribution of its origin-to-final gap, use observed_distribution, which convolves the steps into one delay. For two standalone delays that are not part of a tree, ConvolvedDistributions.jl's convolved gives their sum X + Y directly. See Delay chains, additive moments, and collapsing for a worked example.

How do I fix a parameter across branches?

Use shared or tie to make several leaves one free parameter: shared tags a leaf where it is built, tie walks an assembled tree to named leaves and ties them. Either way params_table then inventories the tied occurrences once under the shared tag rather than as separate parameters. See Composing distributions for a worked example.

Why is my logpdf -Inf?

A composed logpdf is -Inf when a value falls outside a leaf's support. The commonest cause is a negative or zero value scored against a positive-support delay, such as a Gamma or LogNormal. When scoring a whole record, check that every value sits in its own leaf's support and that the record is laid out in the tree's event order (see event_names); a value in the wrong slot can land outside the leaf that scores it.

How do I fit a composed distribution to data?

Mark the parameters to estimate by building a leaf with uncertain, or promote a leaf/subtree/whole tree already in place with uncertain(tree, ...) / uncertain(tree), then hand the tree and data to DistributionsInference.jl's distribution_to_logdensity for a LogDensityProblems-conformant posterior, or its distribution_to_turing to wrap it as a DynamicPPL model for direct sampling with Turing.jl. Its composed-tree tutorial covers the full pipeline, including reading a fitted chain back onto the tree.

Is a composed distribution really a Distribution?

Yes. Every composer subtypes Distributions.Distribution, so logpdf, rand, mean, var, cdf (where defined) and the rest of the interface work unchanged, and a composed object drops into any code that expects a distribution.