Skip to content

A framework for single/multi-objective optimization with metaheuristics

License

Notifications You must be signed in to change notification settings

king-follow/jMetalPy

 
 

Repository files navigation

jMetalPy

CI PyPI Python version DOI PyPI License Code style: black

A paper introducing jMetalPy is available at: https://doi.org/10.1016/j.swevo.2019.100598

Table of Contents

Installation

You can install the latest version of jMetalPy with pip,

pip install jmetalpy  # or "jmetalpy[distributed]"
Notes on installing with pip

jMetalPy includes features for parallel and distributed computing based on pySpark and Dask.

These (extra) dependencies are not automatically installed when running pip, which only comprises the core functionality of the framework (enough for most users):

pip install jmetalpy

This is the equivalent of running:

pip install "jmetalpy[core]"

Other supported commands are listed next:

pip install "jmetalpy[dev]"  # Install requirements for development
pip install "jmetalpy[distributed]"  # Install requirements for parallel/distributed computing
pip install "jmetalpy[complete]"  # Install all requirements

Hello, world! 👋

Examples of configuring and running all the included algorithms are located in the documentation.

from jmetal.algorithm.multiobjective import NSGAII
from jmetal.operator import SBXCrossover, PolynomialMutation
from jmetal.problem import ZDT1
from jmetal.util.termination_criterion import StoppingByEvaluations

problem = ZDT1()

algorithm = NSGAII(
    problem=problem,
    population_size=100,
    offspring_population_size=100,
    mutation=PolynomialMutation(probability=1.0 / problem.number_of_variables, distribution_index=20),
    crossover=SBXCrossover(probability=1.0, distribution_index=20),
    termination_criterion=StoppingByEvaluations(max_evaluations=25000)
)

algorithm.run()

We can then proceed to explore the results:

from jmetal.util.solution import get_non_dominated_solutions, print_function_values_to_file, \ 
    print_variables_to_file

front = get_non_dominated_solutions(algorithm.get_result())

# save to files
print_function_values_to_file(front, 'FUN.NSGAII.ZDT1')
print_variables_to_file(front, 'VAR.NSGAII.ZDT1')

Or visualize the Pareto front approximation produced by the algorithm:

from jmetal.lab.visualization import Plot

plot_front = Plot(title='Pareto front approximation', axis_labels=['x', 'y'])
plot_front.plot(front, label='NSGAII-ZDT1', filename='NSGAII-ZDT1', format='png')

Pareto front approximation

Features

The current release of jMetalPy (v1.5.7) contains the following components:

  • Algorithms: local search, genetic algorithm, evolution strategy, simulated annealing, random search, NSGA-II, NSGA-III, SMPSO, OMOPSO, MOEA/D, MOEA/D-DRA, MOEA/D-IEpsilon, GDE3, SPEA2, HYPE, IBEA. Preference articulation-based algorithms (G-NSGA-II, G-GDE3, G-SPEA2, SMPSO/RP); Dynamic versions of NSGA-II, SMPSO, and GDE3.
  • Parallel computing based on Apache Spark and Dask.
  • Benchmark problems: ZDT1-6, DTLZ1-2, FDA, LZ09, LIR-CMOP, unconstrained (Kursawe, Fonseca, Schaffer, Viennet2), constrained (Srinivas, Tanaka).
  • Encodings: real, binary, permutations.
  • Operators: selection (binary tournament, ranking and crowding distance, random, nary random, best solution), crossover (single-point, SBX), mutation (bit-blip, polynomial, uniform, random).
  • Quality indicators: hypervolume, additive epsilon, GD, IGD.
  • Pareto front approximation plotting in real-time, static or interactive.
  • Experiment class for performing studies either alone or alongside jMetal.
  • Pairwise and multiple hypothesis testing for statistical analysis, including several frequentist and Bayesian testing methods, critical distance plots and posterior diagrams.
Scatter plot 2D Scatter plot 3D
Parallel coordinates Interactive chord plot

Changelog

  • [v1.5.7] Use of linters for catching errors and formatters to fix style, minor bug fixes.
  • [v1.5.6] Removed warnings when using Python 3.8.
  • [v1.5.5] Minor bug fixes.
  • [v1.5.4] Refactored quality indicators to accept numpy array as input parameter.
  • [v1.5.4] Added CompositeSolution class to support mixed combinatorial problems. #69

License

This project is licensed under the terms of the MIT - see the LICENSE file for details.

About

A framework for single/multi-objective optimization with metaheuristics

Resources

License

Stars

Watchers

Forks

Packages

No packages published

Languages

  • Python 99.9%
  • Makefile 0.1%