Evolutionary Competition

compete the "evolution" and evolve the "competition"!

Generating Benchmark by GP

Generate benchmark functions to differ a pair of evolutionary algorithms based on genetic programming algorithm.

Requirements

• Python 3.9 (tested, other version should also work)
• deap
• scipy
• matplotlib
• pygraphviz

Installation

Install Python and Graphviz, then install the library in this repo.

pip install git+https://github.com/Y1fanHE/evolutionary-competition.git

Usage

Create a tuple for DE.

from evocomp.evo import de_rand_1_bin

de = (
    de_rand_1_bin,                # de
    dict(n_eval = 2000,           # parameters
         n_pop  = 20,
         F      = 0.5,
         CR     = 0.5),
    "DE"                          # alias
)

Create a tuple for PSO.

from evocomp.evo import particle_swarm_optimization

pso = (
    particle_swarm_optimization,  # pso
    dict(n_eval = 2000,           # parameters
         n_pop  = 20,
         w1     = 0.5,
         w2     = 0.5,
         c      = 0.1),
    "PSO"                         # alias
)

Create a Competition using wasserstein_distance.

from evocomp.comp import Competition
from evocomp.comp.metrics import wasserstein_distance

comp = Competition(
    metric      = wasserstein_distance, # distance metric
    algorithm1  = de,                   # two eas
    algorithm2  = pso,
    lower_bound = -5,                   # bounds of search space
    upper_bound = 5,
    dimension   = 2,                    # dimension of search space
    repetition  = 1,                    # repetition of ea runs
    mode        = "differ"              # mode: differ/match
)

Evolve the Competition by simple EA.

from evocomp.comp.gp import simple

comp.evolve(                            # gp parameters
    method          = simple,
    population_size = 10,
    max_generation  = 6,
    tournament_size = 7,
    crossover_rate  = 0.8,
    mutation_rate   = 0.1,
    seed            = 1000,
    verbose         = True
)

Plot the results.

comp.plot_space(target="out.png")       # contor plot
comp.plot_tree(target="out_tree.png")   # tree plot
comp.save(target="out.sol")             # text file

Also see example.py for a multiprocessing example.

python -W ignore example.py

The Generated Benchmark Functions

Check the following functions generated by GP!

$$\frac{\cos{x}}{x-y-\cos{5}}$$

$$-(y+xy+\sin(\sin{y}))$$

$$\frac{6xy^2\log(\cos\frac{y}{8})(\frac{x}{9}-\sin{x})}{\cos(\sin(\sin{x})(\frac{2}{7+y}))}$$

$$\sin(7)xy\log(\log{x^2}-\sin{y})$$

Genetic Programming

• random search
• simple
• map-elites

Evolutionary Algorithms

• random search
• differential evolution
• particle swarm optimization
• cuckoo search

Distance Metrics

• wasserstein distance

Todo

• write README file
• check high dimension cases
• add multiprocessing
• implement map-elite gp
• add more eas
• add new metrics