Parefs’ MOO algorithms#

Paref provides a series of ready to use generic (mainly minimization algorithms) and problem tailored (i.e. targeting certain properties) MOO algorithms implemented in the ParefMOO interface.

In order to apply any MOO algorithm (except ExpressSearch which only needs the maximum number of iterations) you need to initialize (or implement your own) StoppingCriteria indicating when to stop the optimization process and call the MOO algorithm to it and the black-box function (bbf).

from paref.moo_algorithms.multi_dimensional.find_edge_points import FindEdgePoints
from paref.moo_algorithms.stopping_criteria.max_iterations_reached import MaxIterationsReached

moo = FindEdgePoints()
stopping_criteria = MaxIterationsReached(max_iterations=10)
moo(blackbox_function=bbf, stopping_criteria=stopping_criteria)

All MOO algorithms are found within the paref.moo_algorithms module. You can access the Pareto points fitting the properties best by calling the best_fits attribute. Currently, Paref includes implementations of the following MOO algorithms (illustrated by their corresponding property):

Property

Graphic

Algorithm

Supported target space dimension

Note

Being minima of the objectives and a real trade-off

Minimal Search

ExpressSearch.minimal_search

All

This is a great initial search

Being in the center of the Pareto front

Minimal Search

ExpressSearch.search_for_best_real_trade_off (after ExpressSearch.search_for_minima for all components)

All

Reflecting a user defined priority of the objectives

Priority

ExpressSearch.priority_search (after ExpressSearch.minimal_search)

All

Higher priority results in smaller target values in that component

Being minima of the objectives

Edge points

Find1ParetoPoints for all components or ExpressSearch.search_for_minima for some component

All

Being edge points for all objective

Edge points

FindEdgePoints

All

Edge points equal the minima in some objective only in 2 dimensions

Filling the gaps of the Pareto front

Fill gaps of Pareto front

FillGapsOfParetoFront2D

2

Being evenly distributed

Evenly Scanned

FillGapsOfParetoFront2D (after FindEdgePoints)

2

Finding a minimum with minimum number of iterations

GPRMinimizer

1

Apply to expensive bbf

Finding a minimum with high number of iterations (for cheap bbf)

DifferentiallEvolutionMinimizer

1

Apply to cheap bbf
Show Source