Ejemplo de uso de numl para generar árboles de decisión mediante aprendizaje automático sobre un corpus de Princesas Disney

MachineLearning.cs

// Require instalar el paquete numl desde NuGet

using System;
using numl;
using numl.Model;
using numl.Supervised.DecisionTree;

namespace MachineLearning
{
	public enum HairColor
	{
		Blonde,
		Brown,
		Red,
		Black,
	}

	public class Princess
	{
		public string Name { get; set; }
		public string Country { get; set; }
		public int Age { get; set; }
		public HairColor HairColor { get; set; }
		public bool Married { get; set; }
		public bool FatherAlive { get; set; }
		public bool MotherAlive { get; set; }
	}

	class Program
	{

		private static readonly Princess[] PRINCESSES =
		{
			new Princess
			{
				Name = "Blancanieves",
				Age = 14,
				Country = "Alemania",
				FatherAlive = false,
				MotherAlive = false,
				HairColor = HairColor.Black,
				Married = true
			},
			new Princess
			{
				Name = "Cenicienta",
				Age = 19,
				Country = "Francia",
				FatherAlive = false,
				MotherAlive = false,
				HairColor = HairColor.Blonde,
				Married = true,
			},
			new Princess
			{
				Name = "Aurora",
				Age = 16,
				Country = "Francia",
				FatherAlive = true,
				MotherAlive = true,
				HairColor = HairColor.Blonde,
				Married = true,
			},
			new Princess
			{
				Name = "Ariel",
				Age = 16,
				Country = "Atlántida",
				FatherAlive = true,
				MotherAlive = false,
				HairColor = HairColor.Red,
				Married = true,
			},
			new Princess
			{
				Name = "Bella",
				Age = 18,
				Country = "Francia",
				FatherAlive = true,
				MotherAlive = false,
				HairColor = HairColor.Brown,
				Married = true,
			},
			new Princess
			{
				Name = "Jazmín",
				Age = 16,
				Country = "Arabia",
				FatherAlive = true,
				MotherAlive = false,
				HairColor = HairColor.Black,
				Married = true,
			},
			new Princess
			{
				Name = "Pocahontas",
				Age = 18,
				Country = "Estados Unidos",
				FatherAlive = true,
				MotherAlive = false,
				HairColor = HairColor.Black,
				Married = true,
			},
			new Princess
			{
				Name = "Mulán",
				Age = 19,
				Country = "China",
				FatherAlive = true,
				MotherAlive = true,
				HairColor = HairColor.Black,
				Married = true,
			},
			new Princess
			{
				Name = "Tiana",
				Age = 19,
				Country = "Maldonia",
				FatherAlive = false,
				MotherAlive = true,
				HairColor = HairColor.Black,
				Married = true,
			},
			new Princess
			{
				Name = "Rapunzel",
				Age = 18,
				Country = "Alemania",
				FatherAlive = true,
				MotherAlive = true,
				HairColor = HairColor.Blonde,
				Married = true,
			},
			new Princess
			{
				Name = "Mérida",
				Age = 16,
				Country = "Escocia",
				FatherAlive = true,
				MotherAlive = true,
				HairColor = HairColor.Red,
				Married = false,
			},
			new Princess
			{
				Name = "Anna",
				Age = 18,
				Country = "Noruega",
				FatherAlive = false,
				MotherAlive = false,
				HairColor = HairColor.Blonde,
				Married = true,
			},
			new Princess
			{
				Name = "Elsa",
				Age = 21,
				Country = "Noruega",
				FatherAlive = false,
				MotherAlive = false,
				HairColor = HairColor.Blonde,
				Married = false,
			},
		};

		static void Main(string[] args)
		{
			Console.Out.WriteLine("Prediciendo matrimonios...");
			PredictMarriage();

			Console.Out.WriteLine("Prediciendo muertes paternas...");
			PredictFatherDecease();

			Console.ReadKey();
		}

		private static void PredictMarriage()
		{
			// Creamos un descriptor para ver qué necesita una Princesa Disney para conseguir marido
			var marriedDescriptor = Descriptor.For<Princess>()
				.With(x => x.Age)
				.WithString(property: x => x.Country, asEnum: true, splitType: StringSplitType.Word)
				.With(x => x.FatherAlive)
				.With(x => x.MotherAlive)
				.With(x => x.HairColor)
				.Learn(x => x.Married);

			// Preparamos un generador de árboles de decisión...
			var generator = new DecisionTreeGenerator(marriedDescriptor);
			// ... que detecte cuando Married = true
			generator.SetHint(true);

			// Y lo construimos a partir de los ejemplos
			var model = Learner.Learn(PRINCESSES, 0.8, 1000, generator);

			// Veamos qué pinta tiene el árbol que hemos generado
			Console.Out.WriteLine("Árbol de decisión para bodas reales:");
			Console.Out.WriteLine(model);

			// Vamos a comprobar si una (hipotética) princesa Celia conseguiría casarse
			var celia = new Princess
			{
				Name = "Celia",
				Age = 16,
				Country = "Noruega",
				FatherAlive = true,
				MotherAlive = true,
				HairColor = HairColor.Blonde,
			};

			// Juguemos a los adivinos con Celia
			var willGetMarried = model.Model.Predict(celia).Married;
			Console.Out.WriteLine("¿Se casará Celia? {0}", willGetMarried ? "Sí" : "No");
			// Sí, se casará. Las princesas noruegas de menos de 19 años se casan
		}

		private static void PredictFatherDecease()
		{
			// Creamos un descriptor para ver en qué condiciones muere el padre de una princesa
			var orphanDescriptor = Descriptor.For<Princess>()
				.With(x => x.Age)
				.WithString(property: x => x.Country, asEnum: true, splitType: StringSplitType.Word)
				.With(x => x.Married)
				.With(x => x.MotherAlive)
				.Learn(x => x.FatherAlive);

			// Preparamos un generador de árboles de decisión...
			var generator = new DecisionTreeGenerator(orphanDescriptor);
			// ... que detecte cuando FatherAlive = true
			generator.SetHint(false);

			var model = Learner.Learn(PRINCESSES, 0.8, 1000, generator);

			// Veamos qué pinta tiene el árbol que hemos generado
			Console.Out.WriteLine("Árbol de decisión para muertes paternas:");
			Console.Out.WriteLine(model);

			// ¿Estará vivo el padre de Rosa?
			var rosa = new Princess
			{
				Name = "Rosa",
				Age = 23,
				Country = "Francia",
				MotherAlive = true,
				HairColor = HairColor.Blonde,
				Married = true,
			};


			// Juguemos a los adivinos con Rosa
			var isFatherDead = !model.Model.Predict(rosa).FatherAlive;
			Console.Out.WriteLine("¿Ha muerto el padre de Rosa? {0}", isFatherDead ? "Sí" : "No");
			// Estará vivo. Los padres de princesas francesas cuya madre está viva suelen estar vivos
		}
	}
}

Tremendo ejemplo! Metía numl en la lista "awesome" https://github.com/quozd/awesome-dotnet#machine-learning

La guinda sería recopilar ejemplos reales del uso de numl, más potentes y de uso real, serviría de mucho para los paquetes como yo.

Offtopic- Otra cosa, con http://lmgtfy.com/ saldrían miles de referencias, pero cuál es su elección de única referencia de todas en la que indique cómo conectar VS 2012 o VS 2013 con github ? Es decir, empezar a crear proyectos con VS 2013 y tenerlos en github abiertos a la comunidad.

Gracias.