pymoors

Overview

pymoors is the python implementation of the moo-rs project. It's an extension crate (via pyo3) exposing moors algorithms with a Pythonic API

Inspired by the amazing Python project pymoo, pymoors delivers both the speed of Rust and the ease-of-use of Python.

Installation

pip install pymoors

Quickstart

import numpy as np

from pymoors import (
    Nsga2,
    RandomSamplingBinary,
    BitFlipMutation,
    SinglePointBinaryCrossover,
    ExactDuplicatesCleaner,
)
from pymoors.typing import TwoDArray


PROFITS = np.array([2, 3, 6, 1, 4])
QUALITIES = np.array([5, 2, 1, 6, 4])
WEIGHTS = np.array([2, 3, 6, 2, 3])
CAPACITY = 7


def knapsack_fitness(genes: TwoDArray) -> TwoDArray:
    # Calculate total profit
    profit_sum = np.sum(PROFITS * genes, axis=1, keepdims=True)
    # Calculate total quality
    quality_sum = np.sum(QUALITIES * genes, axis=1, keepdims=True)

    # We want to maximize profit and quality,
    # so in pymoors we minimize the negative values
    f1 = -profit_sum
    f2 = -quality_sum
    return np.column_stack([f1, f2])


def knapsack_constraint(genes: TwoDArray) -> TwoDArray:
    # Calculate total weight
    weight_sum = np.sum(WEIGHTS * genes, axis=1, keepdims=True)
    # Inequality constraint: weight_sum <= capacity
    return weight_sum - CAPACITY


algorithm = Nsga2(
    sampler=RandomSamplingBinary(),
    crossover=SinglePointBinaryCrossover(),
    mutation=BitFlipMutation(gene_mutation_rate=0.5),
    fitness=knapsack_fitness,
    constraints_fn=knapsack_constraint,
    duplicates_cleaner=ExactDuplicatesCleaner(),
    num_vars=5,
    num_objectives=1,
    num_constraints=1,
    population_size=32,
    num_offsprings=32,
    num_iterations=10,
    mutation_rate=0.1,
    crossover_rate=0.9,
    keep_infeasible=False,
)

algorithm.run()
pop = algorithm.population
# Get genes
>>> pop.genes
array([[1., 0., 0., 1., 1.],
       [0., 1., 0., 0., 1.],
       [1., 1., 0., 1., 0.],
       ...])
# Get fitness
>>> pop.fitness
array([[ -7., -15.],
       [ -7.,  -6.],
       [ -6., -13.],
       ...])
# Get constraints_fn evaluation
>>> pop.constraints_fn
array([[ 0.],
       [-1.],
       [ 0.],
       ...])
# Get rank
>>> pop.rank
array([0, 1, 1, 2, ...], dtype=uint64)
# Get best individuals
>>> pop.best
[<pymoors.schemas.Individual object at 0x...>]
>>> pop.best[0].genes
array([1., 0., 0., 1., 1.])
>>> pop.best[0].fitness
array([ -7., -15.])
>>> pop.best[0].constraints_fn
array([0.])

In this small example, the algorithm finds a single solution on the Pareto front: selecting the items (A, D, E), with a profit of 7 and a quality of 15. This means there is no other combination that can match or exceed both objectives without exceeding the knapsack capacity (7).

Once the algorithm finishes, it stores a population attribute that contains all the individuals evaluated during the search.

Contributing

Contributions welcome! Please read the contribution guide and open issues or PRs in the relevant crate’s repository

License

This project is licensed under the MIT License.