Minimal Subset Optimizer

Overview

SetGA is a small Python library designed to extract a minimal subset from a given set, optimizing a given (set of) objective(s). Based on the DEAP library.

Features

• Subset Extraction: Automatically identifies and extracts a minimal subset from the input set optimizing given objective(s).
• Multi-objective Optimization: Optimizes the subset based on a provided objective or a set of objectives.
• Highly Customizable: Allows for customization of the crossover, mutation, and selection functions (as long as they are contained in the DEAP library).

Installation

pip3 install setga

Arguments

set_size - size of the the set to optimize

eval_ind - fitness function (can compute multiple fitnesses, value or a list of values expected as an output)

stats_by - number of the objective to base statistic on (0 if num of genes, 1 if first user defined etc..)

Optionally:

pop_size - num of solution on every island

num_gen - maximal number of generations

num_islands - number of islands

min_max - defines if user defined objectives should be minimized or maximized. Tuple of -1 (minimization) and 1 (maximization). Should have the same lenght as the fitness output.

mutation - type of mutation

crossover - type of crossover

mutation_rate - probability of an index to be selected for mutation

crossover_rate - robability of an index to be selected for crossover

weights - weights for weighted bit-flip mutation or weighted uniform crossover

selection - type of selection

ref_points - custom reference points for NSGA3

verbose - if logbook should be printed

stop_after - number of generations without a change in the size of selected subset across all islands to stop after

create_individual_funct - custom function to create the solutions for first generation

Usage

This package is used by gatai, you might want to have a look.

Getting optimized solutions and pareto front:

# Example Usage
from setga import utils, select_subset

pop,logbook,gens,logbook, best_sols = select.run_minimizer(num_of_elements_in_the_set, fitness_function, stats_by, stats_names_list,
                  mutation_rate = 0.001, crossover_rate = 0.02,
                  pop_size = 150, num_gen = num_generations, num_islands = 8, mutation = "bit_flip",
                  crossover = "uniform",
                  selection = "SPEA2", frac_init_not_removed = 0.005)

Genetic operators

One can choose from all mutations and crossover DEAP provides for binary EA, where a the uniform crossover and bit-flip mutation operators have been further optimized. It s also possible to provide own genetic operators by passing a function as the mutation or crossover operator. One can also provide a list of mutation of length of the number of islands, if different mutations should be used for different islands. We provide also weighted bit-flip mutation and weighted uniform crossover, where the probability of a index being mutated (crossed over) is proportional to its weight that is passed as the optional argument "weight".

Crossover operators:

"uniform", "onepoint","twopoint","partialy_matched","ordered","uniform_partialy_matched","weighted"

Mutation operators:

"bit-flip","inversion", "weighted"

Selection

One can choose from all multi-objective selection operators DEAP provides. For NSGA3, uniform ref. points are used by default, but other can be specified.

Selection operators:

"SPEA2","NSGA2","NSGA3"

Contributing

Contributions to this project are welcome. If you have any ideas for improvements, new features, or bug fixes, please submit a pull request. For major changes, please open an issue to discuss the proposed modifications.

License

This project is licensed under the MIT License. Feel free to use and modify the code according to the terms of this license.