{

/*

public static class LINQExtension

{

public static double Median(this IEnumerable<double>? source)

{

if (!(source?.Any() ?? false))

{

throw new InvalidOperationException("Cannot compute median for a null or empty set.");

}

var sortedList = (from number in source

orderby number

select number).ToList();

int itemIndex = sortedList.Count / 2;

if (sortedList.Count % 2 == 0)

{

// Even number of items.

return (sortedList[itemIndex] + sortedList[itemIndex - 1]) / 2;

}

else

{

// Odd number of items.

return sortedList[itemIndex];

}

}

}*/

public class Gene

{

//3 layer NN

public double[] weight_gene1 { get; set; } //double[num input data * second layer units]

public double[] bias_gene1 { get; set; }

public double[] weight_gene2 { get; set; } //double[second layer units * third layer units]

public double[] bias_gene2 { get; set; }

public int[] num_units { get; set; }

public int[] num_index { get; set; }

public Gene(int[] units, int[] index)

{

var random_generator = new RandomGenerator();

this.num_units = units;

this.num_index = index;

weight_gene1 = new double[units[0] * units[1]];

bias_gene1 = new double[units[1]];

weight_gene2 = new double[units[1] * units[2]];

bias_gene2 = new double[units[2]];

weight_gene1 = random_generator.getRandomArray(units[0] * units[1]);

bias_gene1 = random_generator.getRandomArray(units[1]);

weight_gene2 = random_generator.getRandomArray(units[1] * units[2]);

bias_gene2 = random_generator.getRandomArray(units[2]);

}

}

//for multile middle layer and dic type weights

public class Gene2

{

public List<Dictionary<int, double[]>> weight_gene { get; set; } //weight_gene[layer][output unit][input unit] -> [<inputs units id as key, double[num middle units-1]>, <middle units id as key, double[num middle units-2>, ..]

public List<double[]> bias_gene { get; set; } //bias_gene[layer][output unit] -> [num_unit[1], num_unit[2], .. num_unit[num_layers - 1]]

public int[] num_units { get; set; } //[num_inputs, num_middle, num_middle2... , num_output]

public int[] num_index { get; set; }

public Gene2(int[] units, int[] index)

{

var random_generator = new RandomGenerator();

this.num_units = units;

this.num_index = index;

weight_gene = new List<Dictionary<int, double[]>>();

bias_gene = new List<double[]>();

//initialize weight / bias

for (int i = 1; i < units.Length; i++) //for layers

{

var weight = new Dictionary<int, double[]>();

for (int j = 0; j < units[i]; j++) //for units in a layer

{

weight[j] = random_generator.getRandomArray(units[i - 1]);

}

weight_gene.Add(weight);

var gene = random_generator.getRandomArray(units[i]);

bias_gene.Add(gene);

}

}

}

public class GA

{

public Gene2[] chromos { get; set; }

public List<double> best_eva_log { get; set; }

public double best_eva { get; set; }

public int best_chromo { get; set; }

public List<int> best_chromo_log { get; set; }

public SimAccount best_ac { get; set; }

public List<SimAccount> best_ac_log { get; set; }

public ConcurrentDictionary<int, long> eva_time { get; set; }

public List<int> generation_time_log { get; set; }

public double estimated_time_to_completion { get; set; }

public List<int> best_chromo_gene { get; set; }

public int island_id { get; set; }

private RandomGenerator random_generator { get; set; }

public GA(int island_id)

{

RandomSeed.initialize();

generation_time_log = new List<int>();

estimated_time_to_completion = -1;

best_chromo_log = new List<int>();

best_eva_log = new List<double>();

best_ac_log = new List<SimAccount>();

random_generator = new RandomGenerator();

this.island_id = island_id;

}

public Gene2 readWeights(int island_id, bool multi_sim)

{

var file_name = "";

if (multi_sim)

file_name = @"./log_best_weight_ID-" + island_id.ToString() + ".csv";

else

file_name = @"./best_weight_ID-" + island_id.ToString() + ".csv";

using (StreamReader sr = new StreamReader(file_name, Encoding.UTF8, false))

{

var data = new List<string>();

var units = new List<int>();

var index = new List<int>();

var bias = new List<double[]>();

var weights = new List<Dictionary<int, double[]>>();

var layer_id_list = new List<int>();

while (true)

{

var line = sr.ReadLine();

data.Add(line);

if (line == null)

break;

else

{

if (line.Contains("units"))

{

var ele = line.Split(',').ToList();

units = ele.GetRange(1, ele.Count - 1).Select(int.Parse).ToList();

}

else if (line.Contains("index"))

{

var ele = line.Split(',').ToList();

index = ele.GetRange(1, ele.Count - 1).Select(int.Parse).ToList();

}

else if (line.Contains("bias"))

{

var ele = line.Split(',').ToList();

var ele_range = ele.GetRange(1, ele.Count - 1).Select(double.Parse).ToList();

bias.Add(ele_range.ToArray());

}

else if (line.Contains("weight")) //weight:0:0,-0.369,0.9373 -> weight:layer:unit

{

var ele = line.Split(',').ToList();

var ele_range = ele.GetRange(1, ele.Count - 1).Select(double.Parse).ToArray();

var layer_id = Convert.ToInt32(ele[0].Split(':')[1]);

var unit_id = Convert.ToInt32(ele[0].Split(':')[2]);

var dic = new Dictionary<int, double[]>();

dic[unit_id] = ele_range;

if (layer_id_list.Contains(layer_id))

weights[layer_id][unit_id] = ele_range;

else

weights.Add(dic);

layer_id_list.Add(layer_id);

}

}

}

data.RemoveAt(data.Count - 1); //remove null

var chrom = new Gene2(units.ToArray(), index.ToArray());

chrom.bias_gene = bias;

chrom.weight_gene = weights;

chrom.num_units = units.ToArray();

chrom.num_index = index.ToArray();

return chrom;

}

}

public SimAccount sim_ga(int from, int to, Gene2 chromo, string title)

{

var sim = new Sim();

var ac = new SimAccount();

ac = sim.sim_ga(from, to, chromo, ac);

Console.WriteLine("pl=" + ac.performance_data.total_pl);

Console.WriteLine("num trade=" + ac.performance_data.num_trade);

Console.WriteLine("num market order=" + ac.performance_data.num_maker_order);

Console.WriteLine("win rate=" + ac.performance_data.win_rate);

Console.WriteLine("sharp_ratio=" + ac.performance_data.sharp_ratio);

Console.WriteLine("num_buy=" + ac.performance_data.num_buy);

Console.WriteLine("num_sell=" + ac.performance_data.num_sell);

Console.WriteLine("buy_pl=" + ac.performance_data.buy_pl_list.Sum());

Console.WriteLine("sell_pl=" + ac.performance_data.sell_pl_list.Sum());

LineChart.DisplayLineChart(ac.total_pl_list, title);

return ac;

}

public SimAccount sim_ga_limit(int from, int to, int max_amount, Gene2 chromo, string title, bool chart)

{

var sim = new Sim();

var ac = new SimAccount();

ac = sim.sim_ga_limit(from, to, max_amount, chromo, ac);

Console.WriteLine("pl=" + ac.performance_data.total_pl);

Console.WriteLine("num trade=" + ac.performance_data.num_trade);

Console.WriteLine("num market order=" + ac.performance_data.num_maker_order);

Console.WriteLine("win rate=" + ac.performance_data.win_rate);

Console.WriteLine("sharp_ratio=" + ac.performance_data.sharp_ratio);

Console.WriteLine("num_buy=" + ac.performance_data.num_buy);

Console.WriteLine("num_sell=" + ac.performance_data.num_sell);

Console.WriteLine("buy_pl=" + ac.performance_data.buy_pl_list.Sum());

Console.WriteLine("sell_pl=" + ac.performance_data.sell_pl_list.Sum());

if (chart)

LineChart.DisplayLineChart(ac.total_pl_list, title);

return ac;

}

public SimAccount sim_ga_market_limit(int from, int to, int max_amount, Gene2 chromo, string title, bool chart, double nn_threshold)

{

var sim = new Sim();

var ac = new SimAccount();

ac = sim.sim_ga_market_limit(from, to, max_amount, chromo, ac, nn_threshold, false);

Console.WriteLine("pl=" + ac.performance_data.total_pl);

Console.WriteLine("pl ratio=" + ac.performance_data.total_pl_ratio);

Console.WriteLine("num trade=" + ac.performance_data.num_trade);

Console.WriteLine("num market order=" + ac.performance_data.num_maker_order);

Console.WriteLine("win rate=" + ac.performance_data.win_rate);

Console.WriteLine("sharp_ratio=" + ac.performance_data.sharp_ratio);

Console.WriteLine("num_buy=" + ac.performance_data.num_buy);

Console.WriteLine("num_sell=" + ac.performance_data.num_sell);

Console.WriteLine("buy_pl=" + ac.performance_data.buy_pl_list.Sum());

Console.WriteLine("sell_pl=" + ac.performance_data.sell_pl_list.Sum());

if (chart)

LineChart.DisplayLineChart(ac.total_pl_list, title);

return ac;

}

public SimAccount sim_ga_limit_conti(int from, int to, int max_amount, Gene2 chromo, string title, SimAccount ac, bool chart)

{

var sim = new Sim();

ac = sim.sim_ga_limit(from, to, max_amount, chromo, ac);

Console.WriteLine("pl=" + ac.performance_data.total_pl);

Console.WriteLine("num trade=" + ac.performance_data.num_trade);

Console.WriteLine("num market order=" + ac.performance_data.num_maker_order);

Console.WriteLine("win rate=" + ac.performance_data.win_rate);

Console.WriteLine("sharp_ratio=" + ac.performance_data.sharp_ratio);

Console.WriteLine("num_buy=" + ac.performance_data.num_buy);

Console.WriteLine("num_sell=" + ac.performance_data.num_sell);

Console.WriteLine("buy_pl=" + ac.performance_data.buy_pl_list.Sum());

Console.WriteLine("sell_pl=" + ac.performance_data.sell_pl_list.Sum());

if (chart)

LineChart.DisplayLineChart(ac.total_pl_list, title);

return ac;

}

//複数chromを使ったsimを行い、それらの結果の総合したパフォーマンスを表示する。

public SimAccount sim_ga_multi_chromo(int from, int to, int max_amount, List<Gene2> chromo, string title, bool chart, List<double> nn_threshold)

{

var ac_list = new List<SimAccount>();

for (int i = 0; i < chromo.Count; i++)

{

var sim = new Sim();

var ac = new SimAccount();

ac = sim.sim_ga_market_limit(from, to, max_amount, chromo[i], ac, nn_threshold[i], false);

ac_list.Add(ac);

Console.WriteLine("Chromo-" + i.ToString() + ":");

Console.WriteLine("pl=" + ac.performance_data.total_pl);

Console.WriteLine("num trade=" + ac.performance_data.num_trade);

Console.WriteLine("num market order=" + ac.performance_data.num_maker_order);

Console.WriteLine("win rate=" + ac.performance_data.win_rate);

Console.WriteLine("sharp_ratio=" + ac.performance_data.sharp_ratio);

Console.WriteLine("num_buy=" + ac.performance_data.num_buy);

Console.WriteLine("num_sell=" + ac.performance_data.num_sell);

Console.WriteLine("buy_pl=" + ac.performance_data.buy_pl_list.Sum());

Console.WriteLine("sell_pl=" + ac.performance_data.sell_pl_list.Sum());

}

//各chrom sim結果を平均する。

var ac_master = new SimAccount();

var buy_pl_sum = 0.0;

var sell_pl_sum = 0.0;

for (int i = 0; i < ac_list[0].total_pl_list.Count; i++)

{

ac_master.total_pl_list.Add(ac_list[0].total_pl_list[i]);

ac_master.total_pl_ratio_list.Add(ac_list[0].total_pl_ratio_list[i]);

}

for (int i = 0; i < chromo.Count; i++)

{

ac_master.performance_data.num_trade += ac_list[i].performance_data.num_trade;

ac_master.performance_data.num_buy += ac_list[i].performance_data.num_buy;

ac_master.performance_data.num_sell += ac_list[i].performance_data.num_sell;

ac_master.performance_data.num_win += ac_list[i].performance_data.num_win;

ac_master.performance_data.total_pl += ac_list[i].performance_data.total_pl;

ac_master.performance_data.total_pl_ratio += ac_list[i].performance_data.total_pl_ratio;

buy_pl_sum += ac_list[i].performance_data.buy_pl_list.Sum();

sell_pl_sum += ac_list[i].performance_data.sell_pl_list.Sum();

for (int j = 0; j < ac_list[i].total_pl_list.Count; j++)

{

if (i > 0)

{

ac_master.total_pl_list[j] += ac_list[i].total_pl_list[j];

ac_master.total_pl_ratio_list[j] += ac_list[i].total_pl_ratio_list[j];

}

}

}

for (int i = 0; i < ac_list[0].total_pl_list.Count; i++)

{

ac_master.total_pl_list[i] = ac_master.total_pl_list[i] / Convert.ToDouble(ac_list.Count);

ac_master.total_pl_ratio_list[i] = ac_master.total_pl_ratio_list[i] / Convert.ToDouble(ac_list.Count);

}

Console.WriteLine("");

Console.WriteLine("Master Results:");

Console.WriteLine("pl=" + ac_master.performance_data.total_pl / Convert.ToDouble(ac_list.Count));

Console.WriteLine("num trade=" + ac_master.performance_data.num_trade / Convert.ToDouble(ac_list.Count));

Console.WriteLine("num market order=" + ac_master.performance_data.num_maker_order);

if (ac_master.performance_data.num_trade > 0)

Console.WriteLine("win rate=" + ac_master.performance_data.num_win / ac_master.performance_data.num_trade);

else

Console.WriteLine("win rate=" + "0");

Console.WriteLine("num_buy=" + ac_master.performance_data.num_buy);

Console.WriteLine("num_sell=" + ac_master.performance_data.num_sell);

Console.WriteLine("buy_pl=" + buy_pl_sum);

Console.WriteLine("sell_pl=" + sell_pl_sum);

if (chart)

LineChart.DisplayLineChart(ac_master.total_pl_list, title);

return ac_master;

}

public void start_island_win_ga(int from, int to, List<int[]> sim_windows, int num_chromos, int generation_ind, int[] units, double mutation_rate, double nn_threshold, int[] index)

{

if (generation_ind == 0)

generate_chromos(num_chromos, units, index);

var eva_dic = new ConcurrentDictionary<int, double>();

var ac_dic = new ConcurrentDictionary<int, SimAccount>();

var option = new ParallelOptions();

option.MaxDegreeOfParallelism = System.Environment.ProcessorCount;

/*

Parallel.For(0, chromos.Length, option, j =>

{

(double total_pl, SimAccount ac) res = evaluation(from, to, max_amount, j, chromos[j], sim_type, nn_threshold, index);

eva_dic.GetOrAdd(j, res.total_pl);

ac_dic.GetOrAdd(j, res.ac);

});

*/

//Console.WriteLine("island No."+island_id.ToString() + ", eva time="+sw.Elapsed.Seconds.ToString());

for (int k = 0; k < chromos.Length; k++)

{

(double total_pl, SimAccount ac) res = evaluationSimWin(from, to, sim_windows, k, chromos[k], nn_threshold);

eva_dic.GetOrAdd(k, res.total_pl);

ac_dic.GetOrAdd(k, res.ac);

}

//check best eva

check_best_eva(eva_dic, ac_dic);

//roulette selection

var selected_chro_ind_list = roulette_selection(eva_dic);

//cross over

crossover(selected_chro_ind_list, 0.3);

//mutation

mutation(mutation_rate, -10, 10);

write_best_chromo();

eva_dic = null;

ac_dic = null;

}

public void start_island_ga(int from, int to, int max_amount, int num_chromos, int generation_ind, int[] units, double mutation_rate, int sim_type, double nn_threshold, int[] index)

{

if (generation_ind == 0)

generate_chromos(num_chromos, units, index);

var eva_dic = new ConcurrentDictionary<int, double>();

var ac_dic = new ConcurrentDictionary<int, SimAccount>();

var option = new ParallelOptions();

option.MaxDegreeOfParallelism = System.Environment.ProcessorCount;

/*

Parallel.For(0, chromos.Length, option, j =>

{

(double total_pl, SimAccount ac) res = evaluation(from, to, max_amount, j, chromos[j], sim_type, nn_threshold, index);

eva_dic.GetOrAdd(j, res.total_pl);

ac_dic.GetOrAdd(j, res.ac);

});

*/

//Console.WriteLine("island No."+island_id.ToString() + ", eva time="+sw.Elapsed.Seconds.ToString());

for (int k = 0; k < chromos.Length; k++)

{

(double total_pl, SimAccount ac) res = evaluation(from, to, max_amount, k, chromos[k], sim_type, nn_threshold);

eva_dic.GetOrAdd(k, res.total_pl);

ac_dic.GetOrAdd(k, res.ac);

}

//check best eva

check_best_eva(eva_dic, ac_dic);

//roulette selection

var selected_chro_ind_list = roulette_selection(eva_dic);

//cross over

crossover(selected_chro_ind_list, 0.3);

//mutation

mutation(mutation_rate, -1, 1);

//inversion mutation

if (generation_ind % 5 == 0)

inversion_mutation();

write_best_chromo();

eva_dic = null;

ac_dic = null;

}

public void start_ga(int from, int to, int max_amount, int num_chromos, int num_generations, int[] units, double mutation_rate, bool display_info, int sim_type, double nn_threshold, int[] index)

{

//initialize chromos

Console.WriteLine("started GA");

generate_chromos(num_chromos, units, index);

for (int i = 0; i < num_generations; i++)

{

Stopwatch generationWatch = new Stopwatch();

generationWatch.Start();

//evaluation chromos

var eva_dic = new ConcurrentDictionary<int, double>();

var ac_dic = new ConcurrentDictionary<int, SimAccount>();

var option = new ParallelOptions();

option.MaxDegreeOfParallelism = System.Environment.ProcessorCount;

Parallel.For(0, chromos.Length, option, j =>

{

(double total_pl, SimAccount ac) res = evaluation(from, to, max_amount, j, chromos[j], sim_type, nn_threshold);

eva_dic.GetOrAdd(j, res.total_pl);

ac_dic.GetOrAdd(j, res.ac);

});

/*

for (int k =0; k<chromos.Length; k++)

{

(double total_pl, SimAccount ac) res = evaluation(from, to, k, chromos[k]);

eva_dic.GetOrAdd(k, res.total_pl);

ac_dic.GetOrAdd(k, res.ac);

}*/

//check best eva

check_best_eva(eva_dic, ac_dic);

//roulette selection

var selected_chro_ind_list = roulette_selection(eva_dic);

//cross over

crossover(selected_chro_ind_list, 0.3);

//mutation

mutation(mutation_rate, -1, 1);

generationWatch.Stop();

generation_time_log.Add(generationWatch.Elapsed.Seconds);

calc_time_to_complete_from_generation_time(i, num_generations);

if (display_info)

display_generation(i, generationWatch);

write_best_chromo();

}

Console.WriteLine("Completed GA.");

}

private void generate_chromos(int num_chrom, int[] num_units_layer, int[] index)

{

chromos = new Gene2[num_chrom];

for (int i = 0; i < num_chrom; i++)

chromos[i] = new Gene2(num_units_layer, index);

}

private (double, SimAccount) evaluation(int from, int to, int max_amount, int chro_id, Gene2 chro, int sim_type, double nn_threshold)

{

var ac = new SimAccount();

var sim = new Sim();

//ac = sim.sim_ga(from, to, chro, ac);

if (sim_type == 0)

ac = sim.sim_ga_limit(from, to, max_amount, chro, ac);

else if (sim_type == 1)

ac = sim.sim_ga_market_limit(from, to, max_amount, chro, ac, nn_threshold, true);

else

Console.WriteLine("GA-evaluation: Invalid Sim Type!");

//var sm = calcSquareError(ac.total_pl_ratio_list, ac.performance_data.num_trade);

//var eva = ac.performance_data.total_pl * Math.Sqrt(ac.performance_data.num_buy * ac.performance_data.num_sell) / sm;

//var eva = ac.performance_data.sharp_ratio * Math.Sqrt(Math.Sqrt(ac.performance_data.num_buy * ac.performance_data.num_sell));

var eva = ac.performance_data.total_pl * Math.Sqrt(ac.performance_data.num_trade);

//if (ac.performance_data.buy_pl_list.Sum() <= 0 || ac.performance_data.sell_pl_list.Sum() <= 0)

// eva = 0;

if (eva.ToString().Contains("N"))

eva = 0;

return (eva, ac);

}

private (double, SimAccount) evaluationSimWin(int from, int to, List<int[]> sim_windows, int chro_id, Gene2 chro, double nn_threshold)

{

var ac = new SimAccount();

var sim_win = new WinSim();

ac = sim_win.sim_win_market(from, to, sim_windows, chro, ac, nn_threshold);

var eva = ac.performance_data.num_trade > 0 ? ac.performance_data.total_pl / Math.Sqrt(Convert.ToDouble(ac.performance_data.num_trade)) : 0;

return (eva, ac);

}

private double calcSquareError(List<double> data, int num_trade)

{

var res = 0.0;

if (num_trade > 0)

{

//calc stright line

var line = new List<double>();

line.Add(data[0]);

var change = (data[data.Count - 1] - data[0]) / Convert.ToDouble(data.Count);

for (int i = 0; i < data.Count; i++)

line.Add(data[i] + change);

//calc square error

for (int i = 0; i < data.Count; i++)

res += Math.Pow(data[i] - line[i], 2.0);

}

else

res = 1.0;

if (res == 0)

res = 1.0;

return res;

}

private void check_best_eva(ConcurrentDictionary<int, double> eva, ConcurrentDictionary<int, SimAccount> ac)

{

var max_eva = -99999999.0;

var eva_key = eva.Keys.ToArray();

int best_eva_key = -1;

foreach (var k in eva_key)

{

if (eva[k] > max_eva)

{

max_eva = eva[k];

best_eva_key = k;

}

}

//best_ac_log.Add(ac[best_eva_key]); may cause memory leake

best_ac = ac[best_eva_key];

best_chromo = best_eva_key;

best_chromo_log.Add(best_eva_key);

best_eva = max_eva;

best_eva_log.Add(max_eva);

}

/*eva.valueにminを加算して、合計値を10000に置き換えてそれぞれの値を計算。

*/

private List<int> roulette_selection(ConcurrentDictionary<int, double> eva)

{

var selected_chro_ind = new List<int>();

List<int> roulette_board = new List<int>();

//全部の値が同じときは同じ割合でroulette boardを作る

var flg_same = true;

for (int i = 1; i < eva.Count; i++)

{

if (eva[0] != eva[i])

{

flg_same = false;

break;

}

}

if (flg_same)

{

var ave_val = Convert.ToInt32(Math.Round(10000.0 / eva.Count));

for (int i = 0; i < eva.Count; i++)

roulette_board.Add((i + 1) * ave_val);

}

else

{

List<double> vals = new List<double>();

var min = eva.Values.Min();

for (int i = 0; i < eva.Count; i++)

vals.Add(eva[i] - min);//evaのkeyが0-count-1までの連続値になっていることが前提

List<double> con_vals = new List<double>();

var sumv = vals.Sum();

var tmp_val = 0;

foreach (var v in vals)

{

tmp_val += Convert.ToInt32(Math.Round(10000 * v / sumv));

roulette_board.Add(tmp_val);

}

}

Random rnd = new Random(DateTime.Now.Millisecond);

for (int i = 0; i < chromos.Count(); i++)

{

if (i == best_chromo)

{

selected_chro_ind.Add(-1); //best chromoはroulette selectしなくて良い

}

else

{

var selected = rnd.Next(0, roulette_board.Last() + 1);

if (selected <= roulette_board[0])

selected_chro_ind.Add(0);

else

{

for (int j = 1; j < roulette_board.Count; j++)

{

if (selected > roulette_board[j - 1] && selected <= roulette_board[j])

selected_chro_ind.Add(j);

}

}

if (selected_chro_ind.Last() == i) //選択したidが自身のidと同じときはやり直し

{

i--;

selected_chro_ind.RemoveAt(selected_chro_ind.Count - 1);

}

}

}

if (selected_chro_ind.Count != chromos.Count())

Console.WriteLine("selected ind is not matched with num chromo in roulette selection!");

return selected_chro_ind;

}

private void mutation(double mutation_ratio, int random_weight_min, int random_weight_max)

{

Random rnd = new Random(DateTime.Now.Millisecond);

for (int i = 0; i < chromos.Count(); i++)

{

if (i != best_chromo)

{

for (int j = 0; j < chromos[i].bias_gene.Count; j++)

{

for (int k = 0; k < chromos[i].bias_gene[j].Length; k++)

chromos[i].bias_gene[j][k] = rnd.NextDouble() > (1 - mutation_ratio) ? random_generator.getRandomArrayRange(random_weight_min, random_weight_max) : chromos[i].bias_gene[j][k];

}

for (int j = 0; j < chromos[i].weight_gene.Count; j++)

{

for (int k = 0; k < chromos[i].weight_gene[j].Count; k++)

{

for (int l = 0; l < chromos[i].weight_gene[j][k].Length; l++)

chromos[i].weight_gene[j][k][l] = rnd.NextDouble() > (1 - mutation_ratio) ? random_generator.getRandomArrayRange(random_weight_min, random_weight_max) : chromos[i].weight_gene[j][k][l];

}

}

}

}

}

/*

* ランダムに選択した染色体をbest chromoの全ウェイトを反転（*-1）した値に変換する。

*/

private void inversion_mutation()

{

var selected_chro = best_chromo;

while (selected_chro == best_chromo)

selected_chro = RandomSeed.rnd.Next(chromos.Length);

for (int i = 0; i < chromos[selected_chro].bias_gene.Count; i++)

{

for (int j = 0; j < chromos[selected_chro].bias_gene[i].Length; j++)

{

chromos[selected_chro].bias_gene[i][j] = chromos[best_chromo].bias_gene[i][j] * -1.0;

}

}

for (int i = 0; i < chromos[selected_chro].weight_gene.Count; i++)

{

for (int j = 0; j < chromos[selected_chro].weight_gene[i].Count; j++)

{

for (int k=0; k<chromos[selected_chro].weight_gene[i][j].Length; k++)

chromos[selected_chro].weight_gene[i][j][k] = chromos[best_chromo].weight_gene[i][j][k] * -1.0;

}

}

}

//reset chromos with random weigths except

public void resetChromos()

{

for (int i = 0; i < chromos.Length; i++)

chromos[i] = new Gene2(chromos[i].num_units, chromos[i].num_index);

}

private void crossover(List<int> selected, double cross_over_ratio)

{

var rnd = new Random(DateTime.Now.Millisecond);

var new_chromos = new Gene2[chromos.Count()];

//deep copy chromos

for (int i = 0; i < new_chromos.Length; i++)

new_chromos[i] = new Gene2(chromos[0].num_units, chromos[0].num_index);

for (int i = 0; i < new_chromos.Length; i++)

{

for (int j = 0; j < chromos[i].bias_gene.Count; j++)

{

for (int k = 0; k < chromos[i].bias_gene[j].Length; k++)

new_chromos[i].bias_gene[j][k] = chromos[i].bias_gene[j][k];

}

for (int j = 0; j < chromos[i].weight_gene.Count; j++)

{

for (int k = 0; k < chromos[i].weight_gene[j].Count; k++)

{

for (int l = 0; l < chromos[i].weight_gene[j][k].Length; l++)

new_chromos[i].weight_gene[j][k][l] = chromos[i].weight_gene[j][k][l];

}

}

}

for (int i = 0; i < chromos.Count(); i++)

{

if (i != best_chromo)

{

//bias1/2, weight1/2からそれぞれからランダムにratio %のweightを選択して交配

for (int j = 0; j < chromos[i].weight_gene.Count; j++)

{

for (int k = 0; k < chromos[i].weight_gene[j].Count; k++)

{

if (rnd.NextDouble() > (1 - cross_over_ratio))

{

new_chromos[i].weight_gene[j][k] = chromos[selected[i]].weight_gene[j][k];

new_chromos[i].bias_gene[j] = chromos[selected[i]].bias_gene[j];

}

else

{

new_chromos[i].weight_gene[j][k] = chromos[i].weight_gene[j][k];

new_chromos[i].bias_gene[j] = chromos[i].bias_gene[j];

}

}

}

}

}

chromos = new Gene2[chromos.Count()];

new_chromos.CopyTo(chromos, 0);

}

private void display_generation(int generation, Stopwatch watch)

{

Console.WriteLine("Generation No." + generation.ToString() + " : " + " Best Chromo ID=" + best_chromo.ToString() + ", Estimated completion hour=" + estimated_time_to_completion.ToString() + ", Best eva=" + best_eva.ToString() + ", time elapsed:" + watch.Elapsed.Minutes.ToString());

Console.WriteLine("Best num trade=" + best_ac.performance_data.num_trade.ToString() + " : " + "Best win rate=" + best_ac.performance_data.win_rate.ToString() + " : " + "Best total pl=" + best_ac.performance_data.total_pl.ToString() + " : " + "Best sharp ratio=" + best_ac.performance_data.sharp_ratio.ToString());

Console.WriteLine("---------------------------------------------------------------------------");

}

private void calc_time_to_complete_from_generation_time(int generation, int num_generations)

{

if (generation_time_log.Count() > 0)

{

estimated_time_to_completion = Math.Round((generation_time_log.Average() * (num_generations - generation)) / 3600, 2);

}

}

private void write_best_chromo()

{

//Console.WriteLine("Writing Best Chromo...");

using (StreamWriter sw = new StreamWriter(@"./best_weight_ID-" + island_id.ToString() + ".csv", false, Encoding.UTF8))

{

//units

var units = "units," + string.Join(",", chromos[best_chromo].num_units);

sw.WriteLine(units);

//index

var index = "index," + string.Join(",", chromos[best_chromo].num_index);

sw.WriteLine(index);

//bias

for (int i = 0; i < chromos[best_chromo].bias_gene.Count; i++)

{

var bias = "bias" + i.ToString() + "," + string.Join(",", chromos[best_chromo].bias_gene[i]);

sw.WriteLine(bias);

}

//weight

for (int i = 0; i < chromos[best_chromo].weight_gene.Count; i++)

{

foreach (var key in chromos[best_chromo].weight_gene[i].Keys)

{

var weights = "weight:" + i.ToString() + ":" + key.ToString() + "," + string.Join(",", chromos[best_chromo].weight_gene[i][key]);

sw.WriteLine(weights);

}

}

}

//Console.WriteLine("Completed write best chromo.");

}

}

}