Evolutionary Strategies made simple!
Evolutionary Strategies made simple!
Use evopy to easily optimize a vector of floats in Python.
All you need to use evopy is Python 3! Run this command to fetch evopy from PyPI:
pip install evopy
Then you can import EvoPy like this:
from evopy import EvoPy
Let’s say we wanted to find the optimum of a parabola, without using exact methods from calculus! With Evopy, this is as easy as writing the following two lines:
evopy = EvoPy(lambda x: pow(x, 2), 1)best_coordinates = evopy.run()
The main ingredient here is the fitness function (the lambda). This can also be a normal function reference, just make sure that it accepts a float or an array of floats and outputs a single float. The other ingredient is the 1 at the end of the first line: This is the dimensionality of the inputs that you expect in your fitness function. best_coordinates will contain an array with a single element, which is the best x value the algorithm could find in the default number of generations.
If the previous example seemed too simple to you, we can also look at the optimum of a more complex, two-dimensional function, like the Rastrigin function. We don’t have to modify much in our previous code snippet to get this to work:
evopy = EvoPy(lambda X: 5 + sum([(x**2 - 5 * np.cos(2 * np.pi * x)) for x in X]),2,generations=1000,population_size=100)best_coordinates = evopy.run()
Compared to the first example, we have interchanged the fitness function for a more complex one, set the dimensionality to 2, and given the algorithm more time to find an optimum by setting a higher generation and individual count than the default.
For more detailed information on evopy’s functionality, have a look at the docs!
Clone this repository and fetch all dependencies from within the cloned directory:
pip install -r requirements.dev.txt
Run all tests with:
nosetests
To check your code style, run:
pylint evopy
To measure your code coverage, run:
nosetests --with-coverage --cover-package=evopy --cover-html --cover-branches --cover-erase