epiworldRcalibrate

A generic framework for calibrating epidemic and agent-based models.

Every calibration method is represented by the same simple S3 object:

calibrator_example <- list(
  pre_process = function(x, ...) { ... },
  run = function(x, ...) { ... },
  post_process = function(x, ...) { ... },
  type = "SIR"
) |>
  structure(class = "calibrator")

All methods are called through:

calibrate(daily_cases, calibrator, ...)

This keeps the epidemic model independent of the calibration method.

R/01_simulation.R     data generation/import
R/02_training.R       training + calibrator definitions
R/03_test_compare.R   calibration + testing + benchmark comparisons

The package currently generates training banks for:

• SIR
• SEIR

For other models such as SIRH/SEIRD, users may provide their own training data or plug in their own simulator.

The package currently demonstrates:

• BiLSTM
• ABC-MCMC
• ABC-SMC
• Nelder-Mead
• Differential Evolution

A user can add any new method by creating another four-field calibrator object; calibrate() itself does not change.

See USER_GUIDE.md for complete examples.

For the complete copy/paste workflow, see STEP_BY_STEP_GUIDE.md.

To actually generate SIR/SEIR training banks, train both BiLSTM models, and produce validation/test results, open:

examples/RUN_SIR_SEIR_BILSTM.Rmd

and Knit it.

At the top of the file:

FULL_RUN <- FALSE

runs a small pipeline test. Change to:

FULL_RUN <- TRUE

for the current full settings (5,000 SIR simulations, 20,000 SEIR simulations, 100 epochs for each model).

The official Getting Started vignette remains lightweight documentation and does not run the expensive full training workflow during package checks.