Add Generation

Some more complex algorithms are not yet implemented.

src/Generation.cpp

@@ -0,0 +1,97 @@
+#include "Generation.hpp"
+
+
+
+size_t Generation::size = 500;
+size_t Generation::living = Generation::size*0.25;
+Fitness* Generation::fitness;
+std::mt19937_64* Generation::re;
+
+
+Generation::Generation()
+{
+	this->individuals.reserve(Generation::size);
+}
+
+
+void Generation::updateFitness()
+{
+	for (auto individual : this->individuals) {
+		if (!individual.fitnessEvaluated) {
+			individual.fitness = Generation::fitness->of(individual.chromosome);
+			individual.fitnessEvaluated = true;
+		}
+	}
+}
+
+
+void Generation::cull(CullType algorithm)
+{
+	// sort by fitness
+	std::sort(
+		this->individuals.begin(),
+		this->individuals.end(),
+		[](Generation::Individual& a, Generation::Individual& b) -> bool {
+			return a.fitness < b.fitness;
+		}
+	);
+	
+	// apply correct culling algorithm
+	switch (algorithm) {
+		case Generation::CUTOFF: this->cullCutoff(); break;
+		case Generation::ROULETTE: this->cullRoulette(); break;
+	}
+}
+
+
+void Generation::reproduce(float crossover)
+{
+	// sort by fitness/available
+	std::sort(
+		this->individuals.begin(),
+		this->individuals.end(),
+		[](Generation::Individual& a, Generation::Individual& b) -> bool {
+			if (a.alive && b.alive) {
+				return a.fitness < b.fitness;
+			} else {
+				return a.alive;
+			}
+		}
+	);
+	
+	// reproduce according to parameters
+	std::uniform_int_distribution<> alivedist(0, Generation::living-1);
+	std::uniform_real_distribution<> crossoverdist(0, 1);
+	auto it = this->individuals.begin() + Generation::living;
+	for (; it!=this->individuals.end(); ++it) {
+		auto father = this->individuals.begin() + alivedist(*Generation::re);
+		
+		if (crossoverdist(*Generation::re) <= crossover) {
+			auto mother = this->individuals.begin() + alivedist(*Generation::re);
+			it->chromosome = Chromosome(father->chromosome, mother->chromosome);
+		} else {
+			it->chromosome = father->chromosome;
+		}
+		
+		it->alive = true;
+		it->fitness = 0;
+		it->fitnessEvaluated = false;
+		
+		it->chromosome.mutate();
+	}
+}
+
+
+void Generation::cullCutoff()
+{
+	auto it = this->individuals.begin() + Generation::living;
+	for (; it!=this->individuals.end(); ++it) {
+		it->alive = false;
+	}
+}
+
+
+void Generation::cullRoulette()
+{
+	// TODO: implement this
+}

src/Generation.hpp

@@ -0,0 +1,57 @@
+#pragma once
+
+#include <random>
+#include <vector>
+#include "Chromosome.hpp"
+#include "Fitness.hpp"
+
+
+
+class Generation
+{
+public:
+	static size_t size;
+	static size_t living;
+	static Fitness* fitness;
+	static std::mt19937_64* re;
+	
+	Generation();
+	
+	enum CullType
+	{
+		CUTOFF,
+		ROULETTE
+	};
+	
+	void updateFitness();
+	void cull(CullType algorithm=Generation::ROULETTE);
+	void reproduce(float crossover=0.5);  // how often a child is the result of two parents,
+	                                      // and not simply mutation.
+	
+	struct Individual
+	{
+// 		Individual(Chromosome& father, Chromosome& mother) : chromosome(father, mother) {}
+		
+		Chromosome chromosome;
+		unsigned long long fitness = 0;
+		bool fitnessEvaluated = false;
+		bool alive = true;
+	};
+	
+	std::vector<Individual> individuals;
+	
+private:
+	// all of the culling algorithms can assume:
+	//  - sorted array
+	//  - alive flag set to true on all individuals
+	void cullCutoff();
+	void cullRoulette();
+	
+	// pool of Chromosomes
+	// sort pool?
+	// min/max/median
+	// extract stats
+	// calculate fitness
+	// cull/selection
+	// render full generation? No
+};