Ctrl+K

calculate_bic#

pharmpy.modeling.calculate_bic(model, likelihood, type='mixed', multiple_testing=False, mult_test_p=1, mult_test_e=1)[source]#

Calculate BIC

Different variations of the BIC can be calculated:

  • mixed (default)
    BIC = -2LL + n_random_parameters * log(n_individuals) +
    n_fixed_parameters * log(n_observations)
  • fixed
    BIC = -2LL + n_estimated_parameters * log(n_observations)
  • random
    BIC = -2LL + n_estimated_parameters * log(n_individuals)
  • iiv
    BIC = -2LL + n_estimated_iiv_omega_parameters * log(n_individuals)

If multiple_testing option is set to true an additional penalty will be added:

  • mBIC = BIC + 2*(n_estimated_parameters)*log(n_predictors/n_expected_models)
Parameters:

  • model (Model) – Pharmpy model object

  • likelihood (float) – -2LL to use

  • type ({‘mixed’, ‘fixed’, ‘random’, ‘iiv’}) – Type of BIC to calculate. Default is the mixed effects.

  • multiple_testing (bool) – Whether to use penalty for multiple testing (default is False)

  • mult_test_p (int) – Number of expected models if using type multiple_testing

  • mult_test_e (int) – E value if using type mult_test

Returns:

float – BIC of model fit

Examples

>>> from pharmpy.modeling import *
>>> from pharmpy.tools import load_example_modelfit_results
>>> model = load_example_model("pheno")
>>> results = load_example_modelfit_results("pheno")
>>> ofv = results.ofv
>>> calculate_bic(model, ofv)
611.7071686183284
>>> calculate_bic(model, ofv, type='fixed')
616.536606983396
>>> calculate_bic(model, ofv, type='random')
610.7412809453149
>>> calculate_bic(model, ofv, type='iiv')
594.431131169692
On this page