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

[Storn97]

Storn, R., Price, K., Journal of Global Optimization 11: 341--359, 1997

API

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

Parameters

minfun:	Function to be minimized.
bounds:	Parameter space bounds as [npar,2] array.
npop:	Size of the parameter vector population.
F:	Difference amplification factor. Values between 0.5-0.8 are good in most cases.
C:	Cross-over probability. Use 0.9 to test for fast convergence, and smaller values (~0.1) for a more elaborate search.
seed:	Random seed.
maximize:	An optional switch telling whether we want maximize or minimize the function. Defaults to minimization.