Neuralnet

Simple, powerfull, robust and easy to use neural network implementation on ruby.
The neural network is meant to be used along a custom made genetic algorithm.
Your workflow to train the neuralnets should be something like this:

• Create 2 (or more) neuralnets and mix them till a (big) population is made
• Pass each neuralnet to a fitness function that represents how well a neuralnet solves your problem
• Mix the neuralnets who have the best score
• Repeat
• When finished save the neuralnet to a file

Usage

Creating & Configuring

require 'neuralnet'

n = NeuralNet.new do |config|
  config.inputs = 3  # required
  config.outputs = 4 # required
  config.hidden = 23 # optional, default value is the average between inputs and outputs
  config.type = :sse # optional, default value is :sse
end

n is a neuralnet object but can't process inputs right now. Before any processing you have to load values to it. Both loading from a file or loading random values will work.

Loading and saving to a file

WIP

Loading random values

n = NeuralNet.new do |config|
  config.inputs = 2
  config.outputs = 1
end

n.random

Mixing two neuralnets to create a population

population_size = 500
neurals = Array.new(population_size) do
  neuralnet1.mix(neuralnet2)
end

neuralnet1 and neuralnet2 are neuralnets with same inputs and outputs.

Processing inputs

the inputs have to be passed as an array.

n = NeuralNet.new do |config|
  config.inputs = 3
  config.outputs = 1
end
n.random
n.process([0.1,0.2,0.3])

TO DO

• Write tests
• Add reproduction to NeuralNets
• Find a better name
• set-up travis CI
• Code loading and saving system
• Write documentation (WIP)
• Publish to rubygems