mhsa 1.1.0

MHSA - Metaheuristics Similarity Analyzer

MHSA - Metaheuristics Similarity Analyzer

This repository contains the source code of the experiments in the paper Measuring the similarity of metaheuristic search strategies with machine learning models. MHSA provides an alternative way to compare and analyze metaheuristic-search strategies with the help of machine learning.

Usage

To use MHSA for similarity analysis we first have to define the gene spaces which will be used by the genetic algorithm. First key of the gene space dictionary must correspond with the class name of the the algorithm which must be implemented in the NiaPy micro-framework. In this case we chose BatAlgorithm and ParticleSwarmAlgorithm.

PSA_gene_spaces = {
    "ParticleSwarmAlgorithm": {
        "c1": {"low": 0.01, "high": 2.5, "step": 0.01},
        "c2": {"low": 0.01, "high": 2.5, "step": 0.01},
        "w": {"low": 0.0, "high": 1.0, "step": 0.01},
    }
}
BA_gene_spaces = {
    "BatAlgorithm": {
        "loudness": {"low": 0.01, "high": 1.0, "step": 0.01},
        "pulse_rate": {"low": 0.01, "high": 1.0, "step": 0.01},
        "alpha": {"low": 0.9, "high": 1.0, "step": 0.001},
        "gamma": {"low": 0.0, "high": 1.0, "step": 0.01},
    }
}

We also have to chose diversity metrics which will be used as the basis of the analysis and the optimization problem the metaheuristics are going to solve.

from mhsa.diversity_metrics.population_diversity.dpc import DPC
from mhsa.diversity_metrics.population_diversity.fdc import FDC
from mhsa.diversity_metrics.population_diversity.pfsd import PFSD
from mhsa.diversity_metrics.population_diversity.pfm import PFM
from mhsa.diversity_metrics.individual_diversity.idt import IDT
from mhsa.diversity_metrics.individual_diversity.isi import ISI
from mhsa.diversity_metrics.individual_diversity.ifm import IFM
from mhsa.diversity_metrics.individual_diversity.ifiqr import IFIQR
from mhsa.problems.schwefel import Schwefel

OPTIMIZATION_PROBLEM = Schwefel(dimension=20)

POP_DIVERSITY_METRICS = [
    DPC(OPTIMIZATION_PROBLEM),
    FDC(OPTIMIZATION_PROBLEM, [420.968746], True),
    PFSD(),
    PFM(),
]
INDIV_DIVERSITY_METRICS = [
    IDT(),
    ISI(),
    IFM(),
    IFIQR(),
]

In the next step we have to instantiate the MetaGA class which uses the GA class from the PyGAD library. At this point we decide which of the algorithms will be analyzed and which will be the "reference" or "target" algorithm. The gene space of the analyzed algorithm gets assigned to the gene_spaces argument of the MetaGA.

from mhsa.tools.meta_ga import MetaGA, MetaGAFitnessFunction

meta_ga = MetaGA(
    fitness_function_type=MetaGAFitnessFunction.TARGET_PERFORMANCE_SIMILARITY,
    ga_generations=20,
    ga_solutions_per_pop=15,
    ga_percent_parents_mating=60,
    ga_parent_selection_type="tournament",
    ga_k_tournament=2,
    ga_crossover_type="uniform",
    ga_mutation_type="random",
    ga_crossover_probability=0.9,
    ga_mutation_num_genes=1,
    ga_keep_elitism=1,
    gene_spaces=BA_gene_spaces,
    pop_size=30,
    max_evals=10000,
    num_runs=30,
    problem=OPTIMIZATION_PROBLEM,
    pop_diversity_metrics=POP_DIVERSITY_METRICS,
    indiv_diversity_metrics=INDIV_DIVERSITY_METRICS,
)

In the last step we have to instantiate the MetaheuristicSimilarityAnalyzer class and pass it the configured MetaGA instance and the gene space of the target algorithm. Then we simply call the run_similarity_analysis method to start the analysis.

from mhsa.tools.metaheuristics_similarity_analyzer import MetaheuristicsSimilarityAnalyzer

mhsa = MetaheuristicsSimilarityAnalyzer(meta_ga=meta_ga, target_gene_space=PSA_gene_spaces)

mhsa.run_similarity_analysis(
    num_comparisons=10
    get_info=True,
    generate_dataset=True,
    calculate_similarity_metrics=True,
    export=True,
)

After the analysis we can choose to export results of the analysis as a .pdf and/or .tex file with export_results_to_latex method or access them trough the MetaheuristicsSimilarityAnalyzer class instance.

mhsa.export_results_to_latex(generate_pdf=True)

For more information check out examples.

This project depends on

NiaPy Python microframework

PyGAD Python library