Assisted specification

Example of the estimation of several versions of the model using assisted specification algorithm. The catalog of specifications is defined in Specification of a catalog of models . Compared to 21a. Assisted specification, the number of specifications exceeds the maximum limit, so a heuristic is applied. See Bierlaire and Ortelli, 2023 for a detailed description of the use of the assisted specification algorithm.

Michel Bierlaire, EPFL Sat Jun 28 2025, 12:25:12

import biogeme.biogeme_logging as blog
from biogeme.assisted import AssistedSpecification
from biogeme.catalog import count_number_of_specifications
from biogeme.multiobjectives import aic_bic_dimension
from biogeme.results_processing import compile_estimation_results

from plot_b22b_multiple_models_spec import PARETO_FILE_NAME, the_biogeme

logger = blog.get_screen_logger(blog.INFO)
logger.info('Example b22multiple_models')

nbr = count_number_of_specifications(the_biogeme.log_like)
if nbr is None:
    print('There are too many possible specifications to be enumerated')
else:
    print(f'There are {nbr} possible specifications')

Creation of the object capturing the assisted specification algorithm. Its constructor takes three arguments:

• the biogeme object containing the specifications and the database,

• an object defining the objectives to minimize. Here, we use three objectives: AIC, BIC and number of parameters.

• the name of the file where the estimated are saved, and organized into a Pareto set.

assisted_specification = AssistedSpecification(
    biogeme_object=the_biogeme,
    multi_objectives=aic_bic_dimension,
    pareto_file_name=PARETO_FILE_NAME,
)

The algorithm is run.

non_dominated_models = assisted_specification.run()

summary, description = compile_estimation_results(
    non_dominated_models, use_short_names=True
)
print(summary)

Explanation of the short names of the model.

for k, v in description.items():
    if k != v:
        print(f'{k}: {v}')