script for optimization

optimize.py

def run_optimization(importer, optimizer, output_folder, yaml_file, algorithm_name, num_starts, parallel):
    """
    Run the optimization
    Parameters
    ----------
    importer:
        PEtab model importer
    optimizer:
        name of the optimizer
    output_folder:
        location into which results of optimization should be saved
    yaml_file:
        model description in YAML format
    algorithm_name:
        human readable name of the optimizer used for simulation of model
    num_starts:
        number of starts for multi-start local optimization
    parallel:
        boolean flag indicating if parallelization should be used
    """
    pypesto.logging.log_to_console(logging.INFO)

    # objective
    objective = importer.create_objective(guess_steadystate=False)
    objective.amici_solver.setAbsoluteTolerance(1e-8)
    objective.amici_solver.setRelativeTolerance(1e-6)
    # Change mode of steady states
    # objective.amici_model.setSteadyStateSensitivityMode(1)
    objective.amici_solver.setSensitivityMethodPreequilibration(2)
    objective.amici_solver.setSensitivityMethod(2)
    # Request an approximation to the Hessian
    # objective.get_hess = '2-point'

    # problem, and set options for optimization
    problem = importer.create_problem(objective)
    option = pypesto.optimize.OptimizeOptions(allow_failed_starts=True)

    # for Chen benchmark model, solver settings have to be adapted to make the model work
    if re.match("Chen", yaml_file):
        objective.amici_solver.setMaxSteps(int(2e5))
        objective.amici_solver.setInterpolationType(amici.InterpolationType_polynomial)
        objective.amici_solver.setSensitivityMethod(amici.SensitivityMethod.adjoint)


    # in any case create an engine for the optimization runs
    engine = pypesto.engine.SingleCoreEngine() 

    # if parallelism is demanded (over the multi starts) create MPI engine
    if parallel:
        from pypesto.engine.mpi_pool import MPIPoolEngine
        engine = pypesto.engine.mpi_pool.MPIPoolEngine()

    # specify where and hwo to store history of optimization runs
    history_name = output_folder + f'{yaml_file}/' + f'{algorithm_name}/' + f'history/history_{yaml_file}_{algorithm_name}_{date.today()}' + '_{id}.csv'
    history_options = pypesto.HistoryOptions(trace_record=True, storage_file=history_name)

    # eventually run the multi-start optimization
    np.random.seed(num_starts)
    result = pypesto.optimize.minimize(problem,
                                       n_starts=num_starts,
                                       optimizer=optimizer,
                                       engine=engine,
                                       options=option,
                                       history_options=history_options)
    # return the problem and the results (history)
    return problem, result