pycuga 0.0.14

Python Cuda Genetic Algorithm Package

📦pycuga

Pycuga (PYthon CUda Genetic Algorithm) provides a package for performing island-based genetic algorithm on Python and CUDA.

• Methods for migartion, selection and mutation are implemented already. Users only need to pick the method during initialisation.
• User needs to define evaluation using Cuda code.

Motivation

• When I worked on my previous project on Solving Maximum Satisfiability Problem using CUDA, I realised a lot of code could be reused, which save a lot of development time for solving other optimisation problems using genetic algorithm and CUDA.

Parameters

Methods currently supported
Selection	Elitism, Roulette Wheel
Crossover	One (point), Two (points), Uniform
Mutation	Number

pip install pycuga

p1 = PyCUGA(constArr, chromosomeSize, stringPlaceholder, mutationNumber, selectionMode, crossoverMode)
p1. launchKernel(islandSize, blockSize , chromosomeNo, isTime, timeAllowed, migrationRounds, rounds)

• constArr (numpy array): defines the details of the problem

• chromosomeSize (int): number of variables in a chromosome

• stringPlaceholder (string): defines the evalutaion method and other constants

• mutationNumber (int): /

• selectionMode (string): "elitism" or "roulettewheel"

• crossoverMode (string): "one" or "two" or "uniform"

• islandSize (int): size of an island

• blockSize (int): Cuda block size

• chromosomeNo (int): number of chromosomes

• migrationRounds (int): number of rounds per migration

• isTime (bool): stopped by time or number of rounds

• timeAllowed (int): seconds allowed

• rounds (int): total number of rounds

Examples here.

Colab MaxSat example here.

Limitations

• Use multiples of 32 (for chromosome parameters) to avoid bugs and increase efficiency.
• Island migration is limited to 1 item currently
• Lack unit testing

❗Disclaimer

This is a mini project which I've put quite a lot of time and effort into, but I can't take responsibility for any bugs nor errors.