How to install or update haxeNeuralNetwork

In order to install or update haxeNeuralNetwork, you fist need to have Haxe installed on your machine,
then you may run the following:

haxelib git haxeNeuralNetwork https://github.com/indigoUan/haxeNeuralNetwork.git

How to use haxeNeuralNetwork

Creating a blank NeuralNetwork

Once you include the library in your project, you can create a new network with

var network:neuralnetwork.NeuralNetwork = new neuralnetwork.NeuralNetwork(array);

where array is an array of integers, where each integer represent a layer and its size.
For instance, [2, 3, 3, 1] will have 2 (called "Input Size") input neurons, one 3-neuron layer, another 3-neuron layer, and finally a 1-neuron (called "Output Size") output layer.

Using a NeuralNetwork

Then you can call process on the network instance passing a Float array of with a size equal to Input Size, and it will return a Float array with a size equal to Output Size.

var input:Array<Float> = [ 0.2, 0.75 ]; // if the Input Size is 2, you will want to pass a two-member Float array 
var result:Array<Float> = network.process(input); // if the Output Size is 1, you will get a one-member Float array 

Creating a NeuralNetwork from JSON data

The class neuralnetwork.NeuralNetwork has a public static function named fromJSON which takes as argument one String value containing (hopefully) JSON data for the network's neurons.

var network:neuralnetwork.NeuralNetwork = neuralnetwork.NeuralNetwork.fromJSON(json_string_here);

Creating JSON data from a NeuralNetwork instance

You can convert an existing neuralnetwork.NeuralNetwork instance to a JSON string by calling its toJSON method.

var json:String = network.toJSON();

toJSON may also take one boolean argument (defaulting to false), which tells whether the JSON string should be pretty-printed or not.

Creating a NeuralNetwork by cloning

A neuralnetwork.NeuralNetwork instance has a public method named clone which will return a perfect copy of that instance.

var clone:neuralnetwork.NeuralNetwork = network.clone();

Training

Creating a group of children from one NeuralNetwork with random mutations

As of the 28th July 2024, the package neuralnetwork.training only has the class Neuroevolution with the function createChildren, which takes as arguments parent:NeuralNetwork, offspring:Int, biasMutationThreshold:Float, weightMutationThreshold:Float.
It returns an array of neuralnetwork.NeuralNetwork instances based on the arguments passed.

Arguments

parent is the NeuralNetwork you want to reproduce;
offspring is the amont of NeuralNetworks you want to make;
biasMutationThreshold is the range within which random mutation can affect a neuron's bias;
weightMutationThreshold is the range within which random mutation can affect a neuron's weights;

var children:Array<NeuralNetwork> = neuralnetwork.training.Neuroevolution.createChildren(network, 4, 1, 1.2);