SimpleNeuralNetwork

This is a simple neural network implementation in Ruby.

This gem does not include any learning implementations (back-prop, etc).

gem install simple_neural_network

The following code implements the above neural network.

require 'simple_neural_network'

network = SimpleNeuralNetwork::Network.new

network.create_layer(neurons: 5)
network.create_layer(neurons: 7)
network.create_layer(neurons: 7)
network.create_layer(neurons: 4)

Neural networks are initialized with random edge weights and zero-valued neuron biases. Edge/bias initialization is configurable by setting the initialization lambdas.

The following code runs an input set against the network.

 network.run([0.5, 0.4, 0.8, 0, 0.9])
 => [0.2257, 0.7488, 0.1016, 0.9935]

network.layers[layer_index].neurons[neuron_index].bias = new_neuron_bias

In SimpleNeuralNetwork, edges point forward and edge weights stored as an array of integers on a Neuron object. To access the edge pointing from a neuron to the n'th neuron in the next layer:

network.layers[layer_index].neurons[neuron_index].edges[n] = new_edge_weight

Improvements and bugs are listed as issues in the gem repository.