Neural network with genetic algorithm

Implementation of neural network with genetic algorithm.

The program takes 8 parametars and they all have to be specified when running the program:

--train: file to train the NN
--test: file to test the NN
--n: layers of neural network split with 's' eg. NN with 2 hidden layer 4 and 5 would be '4s5s'
--popsize: size of population
--elitism: how many parents we want to add to next generation
--p: probability of mutation for elements of a chromosome
--K: standard deviation in mutation fuction
--iters: number of iterations to run

Example:

foo@bar:~$ python solution.py --train ./files/sine_train.txt --test ./files/sine_test.txt --nn 5s --popsize 10 --elitism 1 --p 0.1 --K 0.1 --iter 10000

It also writes out error of the best individual in a population every 2000 iterations.

The only package that is needed to run the program is numpy.

tprhat/neural_genalg

Neural network with genetic algorithm