swimpy.optimization

A collection of Evolutionary Multiobjective Optimization Algorithms.

The algorithm implementation is provided by the evoalgos package. This module implements them for SWIM.

The algorithms are described here: https://ls11-www.cs.tu-dortmund.de/people/swessing/evoalgos/doc/algo.html

To run the algorithm either via SLURM (sbatch jobfiles) or python native multiprocessing, import any of the algorithms in your settings.py file and execute them as a project method. Refer to the docs of the __call__ method for help.

class swimpy.optimization._EvoalgosSwimProblem(project)

The evoalgos.Problem to be run with any of the algorithms provided.

__call__(parameters=None, objectives=None, population_size=10, max_generations=10, output=None, restart=False, test=None, prefix=None, keep_clones=False, **kwargs)

Run the optimisation algorithm.

Parameters

• parameters (dict of {'name': (lower, upper)}) – Dictionary of parameter names and upper and lower boundaries as values (by default) or any other values to determine their initial value if the create_start_population method is overridden. This is only a keyword parameter to allow setting it via the settings file.

• objectives (list of strings or dict of strings) – Performance indicators passed to project.run. Only strings are accepted. Pass a dict to redefine names.

• population_size (int) – Population size the algorithm is run with. By default this is also the num_offspring argument to the algorithm.

• max_generation (int) – Maximum number of iterations the algorithm is run with.

• output (string path, optional) – Output .csv file with saved parameters and performances for each generation and individual (default: [<prefix>_]<algorithm>_populations.csv).

• test (bool | None) – Only run tests before running algorithm if True, dont run them if False and run them before running the algorithm if None (default).

• prefix (str, optional) – A prefix to use for run tags and project clones. Defaults to algorithm name.

• restart (bool) – Attempt to restart the algorithm if output exists.

• keep_clones (bool) – Do not remove the project clones when completed and run the final population in them.

• kwargs – Any overriding parameter parsed to the algorithm (for details see the algorithm descriptions at: https://ls11-www.cs.tu-dortmund.de/people/swessing/evoalgos/doc/algo.html)

Returns

• <browser.Run> – A run instance with the mean objective values as indicators, the mean parameter values of the last generation and the output file attached.

• Settings

• ——–

• All keyword arguements (<plugin>_<kwarg> = value)

algorithm = None

Default algorithm

batch_evaluate(individuals)

Evaluate/run a batch of individuals/projects in parallel.

Parameters

individuals (list of <evoalgos.Individual>s) –

create_individual(parameters, **kwargs)

Create a single evoalgos individual with parameters and bounds.

create_start_population()

Create the initial population out of the parameter boundaries.

Returns

Return type

list of evoalgos.individual.SBXIndividual s

mean_generation_time()

Calculate the average time taken for a generation to finish.

Returns

Return type

datetime.timedelta

nanval = 2e+31

Value to use if nan is returned by objective function

observe_population(ea, initial=False)

Evoalgos function to write the population to the output file.

The function only takes the algorithm instance as first argument and doesnt return anything.

output_attribute_columns = ['generation', 'id_number', 'clone', 'birthgeneration']

Attributes of individuals written to output

restart_population()

Read output and return the last generation as population.

run_tests(quiet=False)

Execute a series of tests before running the algorithm.

Tests:

• creates clone

• calls self.set_parameters

• runs clone

• checks if returned run contains the same number of indicators as self.indicators

• checks if retrived_objectives return same number of values as self.objectives

set_parameters(clonedproject, parameters)

Default parameter setting in method for convenient overriding.

Parameters

• clonedproject (<swimpy.Project>) – Instance of the cloned project.

• parameters (dict) – Parameters to set.

time_safty_factor = 3

factor to multiply the average runtime for job time out estimation