dawkins

apply evolution strategies to supervised, classification tasks. at every iteration (“generation”), a population of parameter vectors (“genotypes”) is perturbed (“mutated”) and their objective function value (“fitness”) is evaluated.

why evolution strategies (es)?

es is an optimization technique that learns parameters without backpropagation. no gradients are computed - why do things the easy way when you can do it the hard way?

example

if you know scikit, you know the drill.

# iris dataset
iris = learn.datasets.load_dataset('iris')
X = iris.data
y = np.eye(3)[iris.target]
X_train, X_test, y_train, y_test = train_test_split(X, y, test_size=0.33)
# create and fit es model
from Dawkins import Dawkins
d = Dawkins(n_pop=200, n_generations=2000)
d.fit(X_train, y_train)
d.predict(X_test, y_test)