Skip to content

Identifiability Analysis

Identifiability Analysis (IA) ensures that identified parameters can be trusted, i.e. have a small uncertainty.

Prerequisites

  • A completed parameter identification run (best-fit parameters computed).
  • param_id_output directory available for the model and dataset.

The Laplace Approximation

The Laplace approximation makes an approximation of your parameter posterior distribution, assuming it is gaussian. This uses the Hessian of the log-likelihood with respect to the parameters.

IA in Circulatory_Autogen

IA is run following a parameter identification run.

Configuration for user_inputs.yaml

To run IA, you need to set 

do_ia: True

and add a specific ia_options block to your user_inputs.yaml configuration file: 

ia_options: 
    method: 'Laplace'
Currently, the available options for the method are 'Laplace'.

Running identifiability analysis

You can run IA as part of parameter identification by setting do_ia: True and running:

./run_param_id.sh <NUM_CORES>

Or run it separately after parameter identification completes:

./run_identifiability_analysis.sh

Expected outcome

Laplace approximation results are saved in your param_id_output directory alongside parameter identification outputs.

Troubleshooting

  • If IA fails with missing files, confirm that parameter identification finished successfully and produced best_param_vals.npy and related outputs.