Evolutionary Optimization

An implementation of two evolutionary algorithms:

• simple -- uses best mean from previous generation (single sample) with a constant, diagonal covariance,
• cma -- wrapper around cma.CMAEvolutionStrategy from pycma.

Four 2D optimization problems taken from DEAP Benchmarks are solved -- sphere, schwefel, h1 & rastrigin.

The Rastrigin Optimization Problem

CMAES

SimpleSolver

Use

$ python main.py h1 cma
$ python main.py rastrigin simple

Plot 3D surfaces for all problems -- saves into ./figs/*.png:

$ python plotting.py

Setup

Python 3.6.8 or above:

$ pip install -r requirements.txt

References

A Visual Guide to Evolution Strategies

DEAP Benchmarks

pycma