Getting started

The home page says what the package is for and shows the shortest fit that runs. The tutorials below go further.

Installation

using Pkg
Pkg.add("DistributionsInference")

Fitting needs a sampler and a chain type as well, and you choose both. The tutorials sample with AdvancedMH or Turing and hold the draws in FlexiChains.

Pkg.add(["AdvancedMH", "Turing", "FlexiChains"])

Two tutorials ask for more. The custom-distribution page finds a maximum-a-posteriori point through Bijectors and Optim, and the composed-tree page needs ComposedDistributions.

Pkg.add(["Bijectors", "Optim", "ComposedDistributions"])

The route through a fit

Every fit here is the same three steps, whatever the distribution and sampler.

Step	What you write	What it gives
Declare	parameter_rows and reconstruct on your type	one row per scalar parameter, and a way back from a flat vector
Score	distribution_to_logdensity or distribution_to_turing	a LogDensityProblems problem, or a DynamicPPL model
Read back	inference_to_distribution or inference_to_distributions	the fitted distribution

Only the first step is written per type, and a distribution from a package that already implements the protocol skips it. A point estimate replaces the last two: optimise_distribution takes the distribution, the data and an optimiser, and returns the fitted distribution.

Tutorials

• Fitting a custom distribution writes the protocol for a hand-rolled type, samples it with AdvancedMH, and finds a maximum-a-posteriori point with an external optimiser.

• Sampling with Turing fits the same distribution as a DynamicPPL model and reads the chain back with the same two calls.

• Fitting a composed distribution runs both routes against a ComposedDistributions tree, partial pooling included. Its extension writes the protocol methods.

Getting help

For usage questions, ask on the Julia Discourse (the SciML or usage categories) or the epinowcast community forum, our home for epidemiological modelling questions. Please use GitHub issues for bug reports and feature requests only.

The layout, navigation, and infrastructure of this site are generated by EpiAwarePackageTools. Its docs cover customising the generated pages and how template sync keeps this repository current.