Beispiel #1
0
        public static double[] TrainNeuralNetworkRoulette(TrainingExample[] trainingSet, uint[] config, Func <int, int> rng, double threshold)
        {
            Genetics.Individual[] individuals = new int[30].Select(_ => {
                int nFeatures          = (int)Neural.NWeightsFromConfig(config);
                Genetics.Gene[] genesA = Genetics.Generate(rng, 48334).Select(x => new Genetics.Gene((short)x)).Take(nFeatures).ToArray();
                Genetics.Gene[] genesB = Genetics.Generate(rng, 94353).Select(x => new Genetics.Gene((short)x)).Take(nFeatures).ToArray();
                Genetics.Chromosome cA = new Genetics.Chromosome(genesA);
                Genetics.Chromosome cB = new Genetics.Chromosome(genesB);
                return(new Genetics.Individual(cA, cB));
            }).ToArray();

            Func <Genetics.Individual, double> individualCost = individual => {
                double[]     wData = individual.Express(Genetics.DefaultGeneExpression);
                double[][][] wsss  = Neural.FoldExpression(wData, config);
                return(Cost(trainingSet, input => Neural.Network(Neural.Sigmoid, wsss, input).ToArray()));
            };

            Func <Genetics.Gene, Genetics.Gene> mutator = Genetics.CreateDefaultGeneMutator(546794);

            int runs = 0;

            Tuple <Genetics.Individual, double>[] currentGen =
                individuals.Select(ind => Tuple.Create(ind, individualCost(ind))).ToArray();

            Genetics.Individual solution = null;
            while (solution == null)
            {
                runs++;

                {
                    int        _rn  = 9479238;
                    Func <int> _rng = () => {
                        _rn = Genetics.Hash(_rn);
                        return(_rn);
                    };

                    Random     r     = new Random();
                    Func <int> __rng = () => {
                        return(r.Next());
                    };

                    currentGen = new bool[currentGen.Count()].AsParallel().Select(_ => {
                        Genetics.Individual mommy = Roulette(currentGen, __rng);
                        Genetics.Individual daddy = Roulette(currentGen, __rng);
                        //Genetics.Individual daddy;
                        //for (int i = 0; ; i++) {
                        //    daddy = Roulette(currentGen, __rng);
                        //    if (mommy != daddy) break;
                        //}
                        Genetics.Individual child = mommy.Mate(__rng, daddy, mutator);
                        return(Tuple.Create(child, individualCost(child)));
                    }).OrderBy(c => c.Item2).ToArray();
                }

                for (int i = 0; i < Math.Min(currentGen.Length, 10); i++)
                {
                    Console.Write($"{string.Format("{0:0.000}", currentGen[i].Item2).PadRight(10)}");
                }
                Console.WriteLine();



                solution = currentGen[0].Item2 <= threshold ? currentGen[0].Item1 : null;
            }

            Console.WriteLine($"runs: {runs}");
            return(solution.Express(Genetics.DefaultGeneExpression));
        }
Beispiel #2
0
        public static double[] TrainNeuralNetworkSelectiveBreeding(TrainingExample[] trainingSet, uint[] config, Func <int, int> rng, double threshold)
        {
            Genetics.Individual[] individuals = new int[30].Select(_ => {
                int nFeatures          = (int)Neural.NWeightsFromConfig(config);
                Genetics.Gene[] genesA = Genetics.Generate(rng, 48334).Select(x => new Genetics.Gene((short)x)).Take(nFeatures).ToArray();
                Genetics.Gene[] genesB = Genetics.Generate(rng, 94353).Select(x => new Genetics.Gene((short)x)).Take(nFeatures).ToArray();
                Genetics.Chromosome cA = new Genetics.Chromosome(genesA);
                Genetics.Chromosome cB = new Genetics.Chromosome(genesB);
                return(new Genetics.Individual(cA, cB));
            }).ToArray();

            Func <Genetics.Individual, double> individualCost = individual => {
                double[]     wData = individual.Express(Genetics.DefaultGeneExpression);
                double[][][] wsss  = Neural.FoldExpression(wData, config);
                return(Cost(trainingSet, input => Neural.Network(Neural.Sigmoid, wsss, input).ToArray()));
            };

            Func <Genetics.Gene, Genetics.Gene> mutator = Genetics.CreateDefaultGeneMutator(546794);

            Tuple <Genetics.Individual, double>[] currentGen =
                individuals.Select(ind => Tuple.Create(ind, individualCost(ind))).OrderBy(c => c.Item2).ToArray();


            Tuple <Genetics.Individual, double>[] solutions =
                new Tuple <Genetics.Individual, double> [0];

            int runs = 0;

            while (solutions.Length == 0)
            {
                runs++;

                Tuple <Genetics.Individual, double> uber1 = currentGen[0];
                Tuple <Genetics.Individual, double> uber2 = currentGen[1];

                {
                    int        _rn  = 9479238;
                    Func <int> _rng = () => {
                        _rn = Genetics.Hash(_rn);
                        return(_rn);
                    };
                    Random     r     = new Random();
                    Func <int> __rng = () => {
                        return(r.Next());
                    };
                    currentGen = currentGen.AsParallel().Select(c => {
                        Genetics.Individual child = uber1.Item1.Mate(__rng, uber2.Item1, mutator);
                        return(Tuple.Create(child, individualCost(child)));
                    }).OrderBy(c => c.Item2).ToArray();
                }

                for (int i = 0; i < Math.Min(currentGen.Length, 10); i++)
                {
                    Console.Write($"{string.Format("{0:0.000}", currentGen[i].Item2).PadRight(10)}");
                }
                Console.WriteLine();

                solutions = currentGen.Where(ind => ind.Item2 <= threshold).ToArray();
            }

            Console.WriteLine($"runs: {runs}");
            return(solutions.OrderBy(solution => solution.Item2).FirstOrDefault()?.Item1.Express(Genetics.DefaultGeneExpression));
        }
Beispiel #3
0
        public static double[] TrainNeuralNetworkRuleofTwo(TrainingExample[] trainingSet, uint[] config, Func <int, int> rng, double threshold)
        {
            Genetics.Individual[] individuals = new int[3].Select(_ => {
                int nFeatures          = (int)Neural.NWeightsFromConfig(config);
                Genetics.Gene[] genesA = Genetics.Generate(rng, 48334).Select(x => new Genetics.Gene((short)x)).Take(nFeatures).ToArray();
                Genetics.Gene[] genesB = Genetics.Generate(rng, 94353).Select(x => new Genetics.Gene((short)x)).Take(nFeatures).ToArray();
                Genetics.Chromosome cA = new Genetics.Chromosome(genesA);
                Genetics.Chromosome cB = new Genetics.Chromosome(genesB);
                return(new Genetics.Individual(cA, cB));
            }).ToArray();

            Func <Genetics.Individual, double> individualCost = individual => {
                double[]     wData = individual.Express(Genetics.DefaultGeneExpression);
                double[][][] wsss  = Neural.FoldExpression(wData, config);
                return(Cost(trainingSet, input => Neural.Network(Neural.Sigmoid, wsss, input).ToArray()));
            };

            Func <Genetics.Gene, Genetics.Gene> mutator = Genetics.CreateDefaultGeneMutator(546794);


            Tuple <Genetics.Individual, double> daddy = Tuple.Create(individuals[0], individualCost(individuals[0]));
            Tuple <Genetics.Individual, double> mommy = Tuple.Create(individuals[1], individualCost(individuals[1]));
            Tuple <Genetics.Individual, double> child = Tuple.Create(individuals[2], individualCost(individuals[2]));

            int runs = 0;

            while (child.Item2 >= threshold)
            {
                runs++;

                if (daddy.Item2 > mommy.Item2)   // daddy worse than mommy?
                {
                    daddy = daddy.Item2 > child.Item2 ? child : daddy;
                }
                else     // mommy worse than daddy?
                {
                    mommy = mommy.Item2 > child.Item2 ? child : mommy;
                }

                {
                    int        _rn  = 9479238;
                    Func <int> _rng = () => {
                        _rn = Genetics.Hash(_rn);
                        return(_rn);
                    };
                    Random     r     = new Random();
                    Func <int> __rng = () => {
                        return(r.Next());
                    };


                    Genetics.Individual c = mommy.Item1.Mate(__rng, daddy.Item1, mutator);
                    child = Tuple.Create(c, individualCost(c));
                }
                Console.Write($"mommy: {string.Format("{0:0.000}", mommy.Item2)};   ");
                Console.Write($"daddy: {string.Format("{0:0.000}", daddy.Item2)};   ");
                Console.Write($"child: {string.Format("{0:0.000}", child.Item2)};");
                Console.WriteLine();
            }

            Console.WriteLine($"runs: {runs}");
            return(child.Item1.Express(Genetics.DefaultGeneExpression));
        }