RenewalExamples

A monorepo of renewal-equation epidemic (Rt estimation) modelling examples, exploring different inference approaches and languages: Julia/Turing, the Birch probabilistic programming language, and JAX. Each subproject below is self-contained with its own dependencies and README.

Contents

RenewalLayers/

A Julia package of Lux.jl layers/cells for renewal-equation epidemic models:

• UnknownGI / FixedGI — generation-interval distribution types (src/gis.jl)
• RenewalCell — a recurrent renewal-equation cell (src/steps.jl)
• delay_conv — a causal delay-convolution layer (src/conv.jl)

scripts/ has three runnable Literate.jl-style examples: generation-interval inference via Turing/NUTS with an AD-backend comparison, a basic RenewalCell simulation, and a renewal model with observation delay.

RenewalBirch/

An EpiNow2-style Rt renewal model implemented in the Birch probabilistic programming language, with an epinow_birch() R front-end that slots Birch's SMC into EpiNow2's own API and plotting. Two inference engines: a self-organizing random-walk + particle filter, and a proper SMC² (with PMMH rejuvenation) for static-parameter inference.

smc-jax/

A JAX-based exploration of SMC/particle-filter alternatives to NUTS refits of pyrenew-style renewal models: a bootstrap/Liu-West particle filter and an SMC² (with UKF/EKF inner filters), across five model variants (A–E) of increasing structural realism, plus rolling-origin forecasting.

Packages

Julia (RenewalLayers/Project.toml)

Package	Purpose
CensoredDistributions	interval-censored/truncated distribution utilities for generation-interval construction
Distributions	probability distribution types
Enzyme	automatic differentiation backend
Lux	the neural-network-style layer framework RenewalCell/delay_conv are built on
LuxCore	core layer abstractions used by Lux
Mooncake	automatic differentiation backend
NNlib	neural-network primitives (convolutions) used by delay_conv
Random	standard library RNG
ReverseDiff	automatic differentiation backend
StatsPlots	plotting distributions
Turing	probabilistic programming / MCMC (NUTS) for parameter inference

Python (smc-jax/pyproject.toml)

Package	Purpose
blackjax	sampling library (SMC, MCMC) in JAX
cuthbert	Gaussian/Kalman-filtering state-space library, backs the library EKF (ekf_cuthbert)
jax	array computing and automatic differentiation
matplotlib	plotting
numpyro	probabilistic programming in JAX
pfjax	particle filtering in JAX
properscoring	proper scoring rules (e.g. CRPS) for forecast evaluation
pyrenew	reference renewal-equation epi model (CDC), used for infection/feedback dynamics

R / Birch (RenewalBirch/)

Package	Purpose
Birch	probabilistic programming language/compiler providing the SMC/SMC² inference engines (install via brew install lawmurray/all/birch)
EpiNow2	reference Rt estimation package; reused for example data, PMF discretisation, and plotting
jsonlite	JSON I/O for passing data between R and Birch
ggplot2	plotting
patchwork	combining ggplot2 panels
data.table	fast data wrangling

License

MIT