Pymoo External Connector¶

ParetoBench's benchmark problems can be used in pymoo through the wrapper class PymooProblemWrapper. On this page, we demonstrate running one of pymoo's optimization algorithms against a ParetoBench problem.

In [1]:

from pymoo.algorithms.moo.nsga2 import NSGA2
from pymoo.optimize import minimize

import paretobench as pb
from paretobench.ext.pymoo import PymooProblemWrapper

from pymoo.algorithms.moo.nsga2 import NSGA2 from pymoo.optimize import minimize import paretobench as pb from paretobench.ext.pymoo import PymooProblemWrapper

Using ParetoBench Problems in Pymoo¶

PymooProblemWrapper subclasses pymoo's Problem class and evaluates the ParetoBench problem using vectorization. Data such as number of objectives, variables and their bounds, etc are passed through from the ParetoBench problem.

In [2]:

# Wrap one of ParetoBench's problems for use in pymoo
prob = PymooProblemWrapper.from_line_fmt("ZDT3")
prob

# Wrap one of ParetoBench's problems for use in pymoo prob = PymooProblemWrapper.from_line_fmt("ZDT3") prob

Out[2]:

PymooProblemWrapper(ZDT3 (n=30))

The wrapper may be passed directly to pymoo's algorithms. We will run a short optimization with pymoo's NSGA-II implementation.

In [3]:

# Run a short optimization
res = minimize(prob, NSGA2(pop_size=32), ("n_gen", 64), seed=1, verbose=False)
print(f"Number of nondominated solutions: {len(res.X)}")

# Run a short optimization res = minimize(prob, NSGA2(pop_size=32), ("n_gen", 64), seed=1, verbose=False) print(f"Number of nondominated solutions: {len(res.X)}")

Number of nondominated solutions: 32

In [4]:

# Load the result into a ParetoBench Population and plot it against the true Pareto front
pop = pb.Population(x=res.X, f=res.F)
pop.plot_obj_scatter(problem="ZDT3")

# Load the result into a ParetoBench Population and plot it against the true Pareto front pop = pb.Population(x=res.X, f=res.F) pop.plot_obj_scatter(problem="ZDT3")

Out[4]:

(<Figure size 640x480 with 1 Axes>, <Axes: xlabel='$f_1$', ylabel='$f_2$'>)