pymoo

Pymoo - Multi-Objective Optimization in Python

Overview

Pymoo is a comprehensive Python framework for optimization with emphasis on multi-objective problems. Solve single and multi-objective optimization using state-of-the-art algorithms (NSGA-II/III, MOEA/D, SPEA2), benchmark problems (ZDT, DTLZ), customizable genetic operators, and multi-criteria decision making methods. Excels at finding trade-off solutions (Pareto fronts) for problems with conflicting objectives. Current stable release: pymoo 0.6.1.6 (November 2025).

Installation

uv pip install pymoo

For reproducible environments, pin a version: uv pip install "pymoo==0.6.1.6".

Dependencies: NumPy (2.x compatible since 0.6.1.3), SciPy, matplotlib (visualization). Autograd is optional for gradient-based features (since 0.6.1.3).

Documentation: https://pymoo.org/ — LLM-friendly index: https://pymoo.org/llms.txt

When to Use This Skill

This skill should be used when:

Core Concepts

The Unified Interface

Pymoo uses a consistent minimize() function for all optimization tasks:

from pymoo.optimize import minimize

result = minimize(
    problem,        # What to optimize
    algorithm,      # How to optimize
    termination,    # When to stop
    seed=1,
    verbose=True
)

Result object contains:

Problem Definition Styles

Pymoo supports three problem definition styles:

Problem Types

Single-objective: One objective to minimize/maximize
Multi-objective: 2-3 conflicting objectives → Pareto front
Many-objective: 4+ objectives → High-dimensional Pareto front
Constrained: Objectives + inequality/equality constraints
Mixed-variable: Continuous, integer, binary, and categorical variables in one problem
Dynamic: Time-varying objectives or constraints

Quick Start Workflows

Workflow 1: Single-Objective Optimization

When: Optimizing one objective function

Steps:

Example:

from pymoo.algorithms.soo.nonconvex.ga import GA
from pymoo.problems import get_problem
from pymoo.optimize import minimize

# Built-in problem
problem = get_problem("rastrigin", n_var=10)

# Configure Genetic Algorithm
algorithm = GA(
    pop_size=100,
    eliminate_duplicates=True
)

# Optimize
result = minimize(
    problem,
    algorithm,
    ('n_gen', 200),
    seed=1,
    verbose=True
)

print(f"Best solution: {result.X}")
print(f"Best objective: {result.F[0]}")

See: scripts/single_objective_example.py for complete example

Workflow 2: Multi-Objective Optimization (2-3 objectives)

When: Optimizing 2-3 conflicting objectives, need Pareto front

Algorithm choice: NSGA-II (standard for bi/tri-objective)

Steps:

Example:

from pymoo.algorithms.moo.nsga2 import NSGA2
from pymoo.problems import get_problem
from pymoo.optimize import minimize
from pymoo.visualization.scatter import Scatter

# Bi-objective benchmark problem
problem = get_problem("zdt1")

# NSGA-II algorithm
algorithm = NSGA2(pop_size=100)

# Optimize
result = minimize(problem, algorithm, ('n_gen', 200), seed=1)

# Visualize Pareto front
plot = Scatter()
plot.add(result.F, label="Obtained Front")
plot.add(problem.pareto_front(), label="True Front", alpha=0.3)
plot.show()

print(f"Found {len(result.F)} Pareto-optimal solutions")

See: scripts/multi_objective_example.py for complete example

Workflow 3: Many-Objective Optimization (4+ objectives)

When: Optimizing 4 or more objectives

Algorithm choice: NSGA-III (designed for many objectives)

Key difference: Must provide reference directions for population guidance

Steps:

Example:

from pymoo.algorithms.moo.nsga3 import NSGA3
from pymoo.problems import get_problem
from pymoo.optimize import minimize
from pymoo.util.ref_dirs import get_reference_directions
from pymoo.visualization.pcp import PCP

# Many-objective problem (5 objectives)
problem = get_problem("dtlz2", n_obj=5)

# Generate reference directions (required for NSGA-III)
ref_dirs = get_reference_directions("das-dennis", n_obj=5, n_partitions=12)

# Configure NSGA-III
algorithm = NSGA3(ref_dirs=ref_dirs)

# Optimize
result = minimize(problem, algorithm, ('n_gen', 300), seed=1)

# Visualize with Parallel Coordinates
plot = PCP(labels=[f"f{i+1}" for i in range(5)])
plot.add(result.F, alpha=0.3)
plot.show()

See: scripts/many_objective_example.py for complete example

Workflow 4: Custom Problem Definition

When: Solving domain-specific optimization problem

Steps:

Unconstrained example:

from pymoo.core.problem import ElementwiseProblem
import numpy as np

class MyProblem(ElementwiseProblem):
    def __init__(self):
        super().__init__(
            n_var=2,              # Number of variables
            n_obj=2,              # Number of objectives
            xl=np.array([0, 0]),  # Lower bounds
            xu=np.array([5, 5])   # Upper bounds
        )

    def _evaluate(self, x, out, *args, **kwargs):
        # Define objectives
        f1 = x[0]**2 + x[1]**2
        f2 = (x[0]-1)**2 + (x[1]-1)**2

        out["F"] = [f1, f2]

Constrained example:

class ConstrainedProblem(ElementwiseProblem):
    def __init__(self):
        super().__init__(
            n_var=2,
            n_obj=2,
            n_ieq_constr=2,        # Inequality constraints
            n_eq_constr=1,         # Equality constraints
            xl=np.array([0, 0]),
            xu=np.array([5, 5])
        )

    def _evaluate(self, x, out, *args, **kwargs):
        # Objectives
        out["F"] = [f1, f2]

        # Inequality constraints (g <= 0)
        out["G"] = [g1, g2]

        # Equality constraints (h = 0)
        out["H"] = [h1]

Constraint formulation rules:

See: scripts/custom_problem_example.py for complete examples

Workflow 5: Constraint Handling

When: Problem has feasibility constraints

Approach options:

1. Feasibility First (Default - Recommended)

from pymoo.algorithms.moo.nsga2 import NSGA2

# Works automatically with constrained problems
algorithm = NSGA2(pop_size=100)
result = minimize(problem, algorithm, termination)

# Check feasibility
feasible = result.CV[:, 0] == 0  # CV = constraint violation
print(f"Feasible solutions: {np.sum(feasible)}")

2. Penalty Method

from pymoo.constraints.as_penalty import ConstraintsAsPenalty

# Wrap problem to convert constraints to penalties
problem_penalized = ConstraintsAsPenalty(problem, penalty=1e6)

3. Constraint as Objective

from pymoo.constraints.as_obj import ConstraintsAsObjective

# Treat constraint violation as additional objective
problem_with_cv = ConstraintsAsObjective(problem)

4. Specialized Algorithms

from pymoo.algorithms.soo.nonconvex.sres import SRES

# SRES has built-in constraint handling
algorithm = SRES()

See: references/constraints_mcdm.md for comprehensive constraint handling guide

Workflow 6: Decision Making from Pareto Front

When: Have Pareto front, need to select preferred solution(s)

Steps:

Example using Pseudo-Weights:

from pymoo.mcdm.pseudo_weights import PseudoWeights
import numpy as np

# After obtaining result from multi-objective optimization
# Normalize objectives
F_norm = (result.F - result.F.min(axis=0)) / (result.F.max(axis=0) - result.F.min(axis=0))

# Define preferences (must sum to 1)
weights = np.array([0.3, 0.7])  # 30% f1, 70% f2

# Apply decision making
dm = PseudoWeights(weights)
selected_idx = dm.do(F_norm)

# Get selected solution
best_solution = result.X[selected_idx]
best_objectives = result.F[selected_idx]

print(f"Selected solution: {best_solution}")
print(f"Objective values: {best_objectives}")

Other MCDM methods:

See:

Workflow 7: Visualization

Choose visualization based on number of objectives:

2 objectives: Scatter Plot

from pymoo.visualization.scatter import Scatter

plot = Scatter(title="Bi-objective Results")
plot.add(result.F, color="blue", alpha=0.7)
plot.show()

3 objectives: 3D Scatter

plot = Scatter(title="Tri-objective Results")
plot.add(result.F)  # Automatically renders in 3D
plot.show()

4+ objectives: Parallel Coordinate Plot

from pymoo.visualization.pcp import PCP

plot = PCP(
    labels=[f"f{i+1}" for i in range(n_obj)],
    normalize_each_axis=True
)
plot.add(result.F, alpha=0.3)
plot.show()

Solution comparison: Petal Diagram

from pymoo.visualization.petal import Petal

plot = Petal(
    bounds=[result.F.min(axis=0), result.F.max(axis=0)],
    labels=["Cost", "Weight", "Efficiency"]
)
plot.add(solution_A, label="Design A")
plot.add(solution_B, label="Design B")
plot.show()

See: references/visualization.md for all visualization types and usage

Workflow 8: Parallel Evaluation

When: Each _evaluate call is expensive (simulations, ML models, external solvers)

Approach: Pass an elementwise_runner to ElementwiseProblem using StarmapParallelization or JoblibParallelization.

Example (thread pool):

from multiprocessing.pool import ThreadPool
from pymoo.algorithms.soo.nonconvex.ga import GA
from pymoo.core.problem import ElementwiseProblem
from pymoo.optimize import minimize
from pymoo.parallelization.starmap import StarmapParallelization

class MyProblem(ElementwiseProblem):
    def __init__(self, elementwise_runner=None, **kwargs):
        super().__init__(
            n_var=10, n_obj=1, xl=-5, xu=5,
            elementwise_runner=elementwise_runner, **kwargs,
        )

    def _evaluate(self, x, out, *args, **kwargs):
        out["F"] = (x ** 2).sum()  # Replace with expensive evaluation

pool = ThreadPool(4)
runner = StarmapParallelization(pool.starmap)
problem = MyProblem(elementwise_runner=runner)

result = minimize(problem, GA(), ("n_gen", 50), seed=1)
pool.close()

See: references/parallelization.md for process pools, joblib, and pickling notes

Workflow 9: Mixed-Variable Optimization

When: Decision variables include continuous, integer, binary, and/or categorical types

Approach: Define a vars dict with typed variables; use MixedVariableGA (SOO) or add MOO survival.

Example:

from pymoo.core.problem import ElementwiseProblem
from pymoo.core.variable import Real, Integer, Choice, Binary
from pymoo.core.mixed import MixedVariableGA
from pymoo.optimize import minimize

class MixedProblem(ElementwiseProblem):
    def __init__(self, **kwargs):
        vars = {
            "b": Binary(),
            "x": Choice(options=["nothing", "multiply"]),
            "y": Integer(bounds=(0, 2)),
            "z": Real(bounds=(0, 5)),
        }
        super().__init__(vars=vars, n_obj=1, **kwargs)

    def _evaluate(self, X, out, *args, **kwargs):
        b, x, z, y = X["b"], X["x"], X["z"], X["y"]
        f = z + y
        if b:
            f = 100 * f
        if x == "multiply":
            f = 10 * f
        out["F"] = f

algorithm = MixedVariableGA(pop_size=20)
result = minimize(MixedProblem(), algorithm, ("n_evals", 1000), seed=1)

For multi-objective mixed-variable problems, use MixedVariableGA(pop_size=20, survival=RankAndCrowdingSurvival()). For single-objective mixed search, pymoo also wraps Optuna via pymoo.algorithms.soo.nonconvex.optuna.Optuna.

See: references/algorithms.md for MixedVariableGA and Optuna details

Algorithm Selection Guide

Single-Objective Problems

Multi-Objective Problems (2-3 objectives)

Many-Objective Problems (4+ objectives)

Constrained Problems

See: references/algorithms.md for comprehensive algorithm reference

Benchmark Problems

Quick problem access:

from pymoo.problems import get_problem

# Single-objective
problem = get_problem("rastrigin", n_var=10)
problem = get_problem("rosenbrock", n_var=10)

# Multi-objective
problem = get_problem("zdt1")        # Convex front
problem = get_problem("zdt2")        # Non-convex front
problem = get_problem("zdt3")        # Disconnected front

# Many-objective
problem = get_problem("dtlz2", n_obj=5, n_var=12)
problem = get_problem("dtlz7", n_obj=4)

See: references/problems.md for complete test problem reference

Genetic Operator Customization

Standard operator configuration:

from pymoo.algorithms.soo.nonconvex.ga import GA
from pymoo.operators.crossover.sbx import SBX
from pymoo.operators.mutation.pm import PM

algorithm = GA(
    pop_size=100,
    crossover=SBX(prob=0.9, eta=15),
    mutation=PM(eta=20),
    eliminate_duplicates=True
)

Operator selection by variable type:

Continuous variables:

Binary variables:

Permutations (TSP, scheduling):

See: references/operators.md for comprehensive operator reference

Performance and Troubleshooting

Common issues and solutions:

Problem: Algorithm not converging

Problem: Poor Pareto front distribution

Problem: Few feasible solutions

Problem: High computational cost

Best practices:

Resources

This skill includes comprehensive reference documentation and executable examples:

references/

Search patterns for references:

scripts/

Run examples:

python3 scripts/single_objective_example.py
python3 scripts/multi_objective_example.py
python3 scripts/many_objective_example.py
python3 scripts/custom_problem_example.py
python3 scripts/decision_making_example.py

Additional Notes

Common patterns: