CG-AI

This is Criz Games' Artificial Intelligence Library for Unity.

It is not the fastest library, but fast enough for my purposes. It is probably better to use Unity's ML-Agents.

Installation

In the Unity package manager, click on the plus sign > Add package from Git URL..
https://github.com/CrizGames/CG-AI.git

Neural Networks

When you want to make a simple neural network, then using SequentialNet is the right way to go.

SequentialNet net = new SequentialNet(
      new Layers.Dense(6, Activations.Identity),
      new Layers.Dense(8, Activations.ReLU),
      new Layers.Dense(8, Activations.Sigmoid),
      new Layers.Dense(4, Activations.Softmax));
net.Init();

As you can see, the library has a number of activation functions.

Custom Layer

Currently, there is only the Dense layer, but you can add your own layer easily. Just inherit from the Layer class.

public class CustomLayer : Layer
{
    public override void Init(int neurons, Layer previousLayer, Func<float[], bool, float[]> activationFunc, bool onlyPositiveWeights, float initWeightsRange)
    {
        // Initialize layer
    }

    public override void Process()
    {
        // Feed forward
    }
}

Custom Activation Function

Activation functions can also be added easily. Just create a function with float[] and bool as parameters and float[] as return type.

public static float[] CustomActivationFunction(float[] layer, bool derivative = false)
{
    if (derivative)
    {
        float[] derivatives = new float[layer.Length];

        // Do stuff if derivative

        return derivatives;
    }

    // Do stuff

    return layer;
}