ropt-pymoo 0.2.0

A pymoo backend for the ropt robust optimization library

A pymoo optimizer plugin for ropt

This package installs a plugin for the ropt robust optimization package, providing access to algorithms from the pymoo optimization package.

ropt-pymoo is developed by the Netherlands Organisation for Applied Scientific Research (TNO). All files in this repository are released under the GNU General Public License v3.0 (a copy is provided in the LICENSE file).

Dependencies

This code has been tested with Python versions 3.8-3.12.

The plugin requires the pymoo optimizer.

Installation

From PyPI:

pip install ropt-pymoo

Usage

An optimization by ropt using the plugin works mostly as any other optimization run (see also the ropt documentation). However, there are a few things to consider:

1. Gradients are not used, as pymoo does not seem to support passing user-defined gradients. Hence, any specifications relating to gradient calulcations in ropt are ignored.
2. Some standard optimization parameters that can be specified in the optimization section are ignored, specifically:
   • max_iterations
   • tolerance
3. The initial values of the variables are ignored, since pymoo generally does not use them. In ropt you still need to specify them, since the size of the vector determines the number of variables. Setting it to a vector of zero values is fine.
4. Linear and non-linear constraints are both supported. Linear constraints are not supported directly, but are internally converted to non-linear constraints.
5. The algorithm and its options are specified using a syntax closely following the pymoo manual. For instance, rather than just giving an algorithm name, you have to specify the full qualified name of the corresponding object as found in the pymoo.algorithms module. For instance to specify the GA algorithm, use: soo.nonconvex.ga.GA.
6. The algorithms itself are entirely configured via the options field in the optimization section of the ropt configuration object. Also in this case, the syntax follows the pymoo manual. See the section below for more information.

Configuring an algorithm.

Configuration of any of the pymoo algorithms is done via the options field in the ropt configuration object. For instance, consider this example for starting a GA optimization from the pymoo manual, with a penalty constraint added:

from pymoo.algorithms.soo.nonconvex.ga import GA
from pymoo.operators.crossover.sbx import SBX
from pymoo.operators.mutation.pm import PM
from pymoo.operators.repair.rounding import RoundingRepair
from pymoo.operators.sampling.rnd import IntegerRandomSampling
from pymoo.optimize import minimize
from pymoo.constraints.as_penalty import ConstraintsAsPenalty

method = GA(
    pop_size=20,
    sampling=IntegerRandomSampling(),
    crossover=SBX(prob=1.0, eta=3.0, vtype=float, repair=RoundingRepair()),
    mutation=PM(prob=1.0, eta=3.0, vtype=float, repair=RoundingRepair()),
    eliminate_duplicates=True,
)

res = minimize(ConstraintsAsPenalty(
    problem, penalty=100.0),
    method,
    termination=('n_gen', 40),
    seed=1234,
)

To run the equivalent optimization, we need to specify the method and the termination in the options field. We also need to specify the constraints object, and a seed. To do this the different objects are specified with their parameters in a nested dictionary that will be parsed into equivalent code. For this example we need to pass a nested dict, for clarity displayed as yml here:

parameters:  # The parameters of the GA object:
    pop_size: 20
    sampling:  # The sampling parameter is an object, specify its full path in pymoo:
    object: operators.sampling.rnd.IntegerRandomSampling
    crossover:  # Also an object:
    object: operators.crossover.sbx.SBX
    parameters:  # Specify the parameters passed to the crossover ojbect:
        prob: 1.0
        eta: 3.0
        vtype: float
        repair:  # A repair object, passed to the crossover object:
        object: operators.repair.rounding.RoundingRepair
    mutation:  # An object:
    object: operators.mutation.pm.PM
    parameters:  # And its parameters:
        prob: 1.0
        eta: 3.0
        vtype: float
        repair:  # A repair object, passed to the mutation object:
        object: operators.repair.rounding.RoundingRepair
    eliminate_duplicates: True
termination:  # Specification of the termination object:
    name: max_gen.MaximumGenerationTermination
    parameters:
        n_max_gen: 10
# Alternative specification for the termination, following pymoo practice:
# "termination": ("n_iter", 30)
constraints:  # Specification of the constraint object:
    name: as_penalty.ConstraintsAsPenalty
    parameters:
        penalty: 100.0
seed: 1234  # The seed that is passed to the minimize function:

Development

The ropt-pymoo source distribution can be found on GitHub. To install from source, enter the ropt-pymoo distribution directory and execute:

pip install .

Running the tests

To run the test suite, install the necessary dependencies and execute pytest:

pip install .[test]
pytest