Changelog

Source: NEWS.md

EpiAwareR (development version)

New Features

• Added posterior S3 methods for epiaware_fit objects: as_draws_array(), as_draws_df(), as_draws_matrix(), as_draws_rvars(), and as_draws_list(). This enables direct integration with bayesplot and loo packages (#20).

Improvements

• Julia global state is now cleaned up after inference. .generate_quantities() receives data explicitly rather than relying on implicit Julia globals (#27).

Documentation

• Added @param and @return documentation to all internal helper functions (#26).

EpiAwareR 0.1.1

New Features

• Added Pathfinder initialisation for NUTS sampling, improving convergence for complex models.

• Increased default target_acceptance to 0.9 for more robust sampling.

Improvements

• Plotting methods (plot.epiaware_fit()) now fully functional with Rt trajectories, case predictions, and posterior distributions - no longer placeholders.

• Uses EpiAware’s generated_observables() for Rt and infections extraction, providing more reliable posterior summaries.

• MCMC internal parameters (e.g., n_steps, step_size) now filtered from diagnostics output for cleaner summaries.

Documentation

• Enhanced Mishra et al. case study vignette with comparison plots and parameters matching the original preprint exactly.

Bug Fixes

• Fixed fallback path in .julia_chains_to_draws() when DataFrames.jl is not available. The function now correctly uses Array(chains.value) to extract the 3D array (iterations × parameters × chains) instead of Array(chains) which only returns a 2D array. This resolves the “incorrect number of dimensions” error reported in #5. Thanks to @owenjonesuob for the diagnosis and suggested fix.

EpiAwareR 0.1.0 (MVP Release)

Initial MVP release of EpiAwareR providing R interface to Julia-based EpiAware framework.

Features

Core Components

• Latent Models
  • AR(): Autoregressive processes of arbitrary order for time-varying parameters
  • Generic access to other models (MA, ARMA, ARIMA) via epiaware_call()
• Infection Models
  • Renewal(): Renewal equation with customizable generation time distribution
  • Support for fixed and uncertain generation time distributions
• Observation Models
  • NegativeBinomialError(): Overdispersed count observations
  • LatentDelay(): Reporting delay wrapper enabling hierarchical composition
• Model Composition
  • EpiProblem(): Compose latent, infection, and observation models into complete epidemiological model
  • Automatic validation of component compatibility

Inference

• Bayesian Inference
  • fit(): Fit models to data using MCMC sampling
  • nuts_sampler(): Configure NUTS (No-U-Turn Sampler) for efficient inference
  • Integration with Turing.jl PPL backend
• Results and Diagnostics
  • S3 methods: print(), summary(), plot()
  • Convergence diagnostics (Rhat, ESS)
  • Posterior predictive distributions
  • Generated quantities (Rt trajectories, infections, predictions)

Distributions

• Prior specification functions:
  • norm(): Normal distribution
  • truncnorm(): Truncated normal
  • halfnorm(): Half-normal
  • gamma_dist(): Gamma distribution
  • lognorm(): Log-normal distribution
  • exponential(): Exponential distribution

Extensibility

• Tiered Access Pattern
  • Tier 1: Explicit R wrappers for common components (AR, Renewal, NegativeBinomialError)
  • Tier 2: Generic epiaware_call() for accessing newer Julia features
  • Tier 3: Direct JuliaCall access for power users

Setup and Configuration

• epiaware_setup_julia(): Automated Julia installation and configuration
• epiaware_available(): Check Julia/EpiAware availability
• Graceful degradation when Julia unavailable

Documentation

• Comprehensive README with quick start example
• Getting Started vignette with Mishra et al. (2020) case study
• Complete function reference with roxygen2 documentation
• Contributing guidelines

Testing

• Unit tests for all components using testthat
• Integration tests for end-to-end workflows
• Tests skip gracefully when Julia unavailable

Case Studies

• Mishra et al. (2020): COVID-19 in South Korea
  • AR(2) latent process
  • Renewal infection model
  • Negative binomial observations
  • Demonstrates compositional approach

Known Limitations

• Plotting methods currently return placeholder plots
• Generated quantities extraction needs full implementation
• Limited to NUTS inference (other samplers available via epiaware_call())
• Some advanced EpiAware features only accessible via generic wrapper

Future Plans

• Enhanced plotting with actual Rt and case trajectory visualization
• Additional explicit wrappers for MA, ARMA, ARIMA models
• Day-of-week effect observation models
• Additional case studies and vignettes
• Integration with other R epidemiological packages
• CRAN submission