PyDE

Global optimization using differential evolution in Python [Storn97].

Installation

git clone https://github.com/hpparvi/PyDE.git
cd PyDE
python setup.py install [--user]

Basic usage

Import the class from the package

from pyde.de import DiffEvol

Create a DiffEvol instance

de = DiffEvol(minfun, bounds, npop)

where minfun is the function to be optimized, bounds is an initialization array, and npop is the size of the parameter vector population.

Now, you can run the optimizer ngen generations:

res = de.optimize(ngen=100)

or run the optimizer as a generator:

for res in de(ngen=100):
    do something

Usage with emcee

The PyDE parameter vector population can be used to initialize the affine-invariant MCMC sampler emcee when a simple point estimate of the function minimum (or maximum) is not sufficient:

de = DiffEvol(lnpost, bounds, npop, maximize=True)
de.optimize(ngen)

sampler = emcee.EnsembleSampler(npop, ndim, lnpost)
sampler.run_mcmc(de.population, 1000)

References

API

pyde.de.DiffEvol (minfun, bounds, npop, F=0.5, C=0.5, seed=0, maximize=False)

Parameters