Perceptron for C# and Unity, based on PerceptronJS and the Perceptron Model.

Perceptron.cs

using UnityEngine;

public class Perceptron {

	private float accuracy = 0;
	private int samples = 0;
	private int[][] xtrain;
	private int[] ytrain;
	private int epochs;
	private float learnrate;
	private float bias = 0;
	private float[] weights;

	public Perceptron(int[][] x, int[] y, int e = 1000, float l = 0.1f) {
		this.accuracy = 0;
		this.samples = 0;
		xtrain = x;
		ytrain = y;

		this.epochs = e;
		this.learnrate = l;

		this.bias = 0;
		this.weights = new float[x[0].Length];

		for (var n = 0; n < x[0].Length; n++) {
			this.weights [n] = this.random();
		}
	}

	private float current_accuracy()
	{
		return this.accuracy / this.samples;
	}

	private float random()
	{
		return Random.value * 2 - 1;
	}

	private int activation(float n)
	{
		return n < 0 ? 0 : 1;
	}

	private int predict(int[] input)
	{
		var total = this.bias;
		for (int w = 0; w < weights.Length; w++) {
			total += input [w] * weights [w];
		}
		return this.activation (total);
	}

	public void fit()
	{
		for (int a = 0; a < this.epochs; a++) {
			for (int b = 0; b < this.xtrain.Length; b++) {
				var prediction = this.predict (this.xtrain [b]);

				Debug.Log ("Expected: " + this.ytrain [b] + " Model Output: " + prediction);

				if (this.ytrain [b] == prediction) {
					this.accuracy += 1;
				} else {
					this.accuracy -= 1;
				}
				this.samples++;

				var loss = this.ytrain[b] - prediction;

				for (int w = 0; w < this.weights.Length; w++) {
					this.weights [w] += loss * this.xtrain [b] [w] * this.learnrate;
				}

				this.bias += loss * this.learnrate;
			}
				
			Debug.Log ("Accuracy: " + this.current_accuracy ());
		}
	}
}

TestPerceptron.cs

using UnityEngine;

public class TestPerceptron : MonoBehaviour {

	private void Awake()
	{
		int[][] x = new int[3][];
		// 111
		x[0] = new int[3];
		x [0] [0] = 1;
		x [0] [1] = 1;
		x [0] [2] = 1;
		// 000
		x[1] = new int[3];
		x [1] [0] = 0;
		x [1] [1] = 0;
		x [1] [2] = 0;
		// 101
		x[2] = new int[3];
		x [2] [0] = 1;
		x [2] [1] = 0;
		x [2] [2] = 1;

		int[] y = { 1, 0, 1 };
		var p = new Perceptron (x, y, 10, 0.1f);
		p.fit ();
	}
}