A simple discrete genetic algorithm for Python.

Download

To download the latest official release, do

$ pip install dga

To get the git version do

$ git clone https://github.com/msparapa/dga

Documentation and usage

This is a simple genetic algorithm where every design variable is modeled as a discrete bit array. Each bit can take a value of either 0 or 1. Eg. A problem with a single parameter and 3 bits has the bit-string 000 corresponding to the lowest possible state.

To start, an initial population is generated with possible solutions randomly assigned. The cost function is evaluated for every possible solution. The lowest 50% of evaluated solutions are kept, and the highest 50% are removed. Children are then created from the top 50% by mixing the bit-strings of two parents. For example, the children of

Parent 1: 001
Parent 2: 010

may look like

Child 1: 000
Child 2: 011

There is a small chance that a bit may flip when creating the children, called mutation. The process runs again with the new population. The algorithm converges once 90% of all bit-strings in the population are the same or the max generation limit has been reached.