Skip to contents

Update Priors from Posteriors for Staged Estimation

Source: R/update_priors_from_posteriors.R
update_priors_from_posteriors.Rd

Merges posterior distributions from a completed calibration stage into a complete priors object, producing a new priors object suitable as input to the next calibration stage.

Usage

update_priors_from_posteriors(priors, posteriors, verbose = TRUE)

Arguments

priors

Complete priors object (list) or path to a priors JSON file.

posteriors

Posteriors object (list) or path to a posteriors JSON file (as produced by calc_model_posterior_distributions).

verbose

Logical. Print merge diagnostics (default TRUE).

Value

A complete priors object (list) with identical top-level structure to the input priors. Posterior fits replace their corresponding prior entries; all other entries are preserved unchanged.

Details

The posteriors object produced by calc_model_posterior_distributions is pruned: it only contains parameters that were sampled and successfully fitted. Parameters that were frozen (not sampled) are removed entirely. A valid priors object for sample_parameters must be complete — it must have entries for every parameter that could be sampled. This function merges the two: posterior fits replace their corresponding priors, everything else is preserved from the original priors.

Merge rules (per parameter):

  • Posterior has a fitted distribution (beta, gamma, lognormal, normal, truncnorm, uniform, gompertz): replace the prior.

  • Posterior has "fixed" distribution: replace the prior. The parameter had zero posterior variance and will return a constant when sampled.

  • Posterior has "failed" distribution: keep the original prior. The fitting failed, so the prior is more informative than a failure marker.

  • Parameter absent from posteriors: keep the original prior. It was not sampled in this stage.

The function also strips any residual fitted-diagnostic metadata (e.g. fitted_mode, fitted_ci) from posterior parameter entries, keeping only the canonical distribution fields that sample_from_prior reads.

Validation

Before returning, the function checks that:

  • The output has metadata, parameters_global, and parameters_location top-level slots.

  • Every parameter entry has distribution (character) and parameters (list) fields.

  • No "failed" distribution markers remain (they are replaced with original priors).

  • Global parameter count matches the original priors.

  • Location-specific parameter groups and their location counts match.

Examples

if (FALSE) { # \dontrun{
# After a completed calibration stage
priors <- jsonlite::read_json("1_inputs/priors.json")
posteriors <- jsonlite::read_json("2_calibration/posterior/posteriors.json")

# Merge posteriors into priors for the next stage
updated_priors <- update_priors_from_posteriors(priors, posteriors)

# Use in next calibration stage
run_MOSAIC(priors = updated_priors, ...)
} # }