Article-22745-Categorical-Hidden-Markov-Model-MQL5/Scripts/ModelSelection.mq5

//+------------------------------------------------------------------+
//|                                               ModelSelection.mq5 |
//|                                                           Eugene |
//|                                             https://www.mql5.com |
//+------------------------------------------------------------------+
#property copyright "Eugene"
#property link      "https://www.mql5.com"
#property version   "1.00"
#include <CategoricalHMM\CategoricalHMM.mqh>
#include <Graphics\Graphic.mqh>
//+------------------------------------------------------------------+
//| Script program start function                                    |
//+------------------------------------------------------------------+
void OnStart()
  {
   int n_samples = 5000;
   int n_states_true = 3;
   int n_emissions = 6;

// --- 1. ИНИЦИАЛИЗАЦИЯ ИСТИННОЙ МОДЕЛИ ДЛЯ ГЕНЕРАЦИИ ---
   matrix tr_true =
     {
        {0.8, 0.1, 0.1},
        {0.05, 0.9, 0.05},
        {0.1, 0.1, 0.8}
     };

   matrix e_true =
     {
        {1.0/6, 1.0/6, 1.0/6, 1.0/6, 1.0/6, 1.0/6},
        {0.1, 0.1, 0.1, 0.1, 0.1, 0.5},
        {0.4, 0.1, 0.1, 0.1, 0.1, 0.2}
     };

   vector y, states_true;
   CCategoricalHMM gen_model(n_states_true, n_emissions, 42); // Seed 42
   gen_model.Sample(n_samples, tr_true, e_true, y, states_true);
//VectorToCSV("Sample5000.csv",y);

// --- РАСЧЕТ ИСТИННОГО LOG-LIKELIHOOD ---
   double true_ll = 0;
   matrix p_true, fs_true, bs_true;
   vector s_true;

// --- Вызываем Inference на истинных параметрах ---
   gen_model.Inference(y, tr_true, e_true, p_true, true_ll, fs_true, bs_true, s_true);

   PrintFormat("==================================================");
   PrintFormat("TRUE MODEL LOG-LIKELIHOOD: %.2f", true_ll);
   PrintFormat("==================================================");

// --- 2. ПОДБОР МОДЕЛИ (MODEL SELECTION) ---
   int ns[] = {1,2, 3, 4, 5, 6, 7};
   int n_fits = 10;
   int n_points = ArraySize(ns);
   double x_vals[], aic_vals[], bic_vals[]; // Массивы для графика
   ArrayResize(x_vals, n_points);
   ArrayResize(aic_vals, n_points);
   ArrayResize(bic_vals, n_points);

   Print("==================================================");
   Print("HMM MODEL SELECTION ");
   PrintFormat("True States: %d | Samples: %d", n_states_true, n_samples);
   Print("==================================================");

   for(int i = 0; i < ArraySize(ns); i++)
     {
      int n_curr = ns[i];
      double best_ll = -DBL_MAX;
      int n_params;

      // --- Пробуем несколько случайных стартов для каждого N ---
      for(int fit = 0; fit < n_fits; fit++)
        {
         CCategoricalHMM model(n_curr, n_emissions, fit);
         n_params = model.GetParams();

         // --- Обучаем (20 итераций для быстрой оценки) ---
         if(!model.Fit(y, 20, 1e-4,false))
            continue;

         // --- Получаем Log-Likelihood ---
         double current_ll = 0;
         matrix p, fs, bs;
         vector s;
         if(model.Inference(y, model.m_TR, model.m_E, p, current_ll, fs, bs, s))
           {
            if(current_ll > best_ll)
               best_ll = current_ll;
           }
        }

      // --- Оценка информационных критериев для лучшей попытки в текущем N ---
      CCategoricalHMM eval_model(n_curr, n_emissions);
      double val_aic = eval_model.AIC(best_ll);
      double val_bic = eval_model.BIC(best_ll, n_samples);

      // --- Сохраняем данные для графика ---
      x_vals[i] = n_curr;
      aic_vals[i] = val_aic;
      bic_vals[i] = val_bic;

      PrintFormat("N=%d | LL: %.2f | AIC: %.2f | BIC: %.2f | n_params:%d",
                  n_curr, best_ll, val_aic, val_bic,n_params);
     }
   Print("==================================================");

// --- ВИЗУАЛИЗАЦИЯ ---
   ChartSetInteger(0, CHART_SHOW, false);

   CGraphic graphic;
   ulong width = ChartGetInteger(0, CHART_WIDTH_IN_PIXELS);
   ulong height = ChartGetInteger(0, CHART_HEIGHT_IN_PIXELS);

   if(ObjectFind(0, "HMM_Selection_Plot") < 0)
      graphic.Create(0, "HMM_Selection_Plot", 0, 0, 0, int(width), int(height));
   else
      graphic.Attach(0, "HMM_Selection_Plot");

   CCurve *curveAIC = graphic.CurveAdd(x_vals, aic_vals, ColorToARGB(clrDodgerBlue, 255), CURVE_POINTS_AND_LINES, "AIC");
   curveAIC.LinesStyle(STYLE_SOLID);

   CCurve *curveBIC = graphic.CurveAdd(x_vals, bic_vals, ColorToARGB(clrLimeGreen, 255), CURVE_POINTS_AND_LINES, "BIC");
   curveBIC.LinesStyle(STYLE_SOLID);

   graphic.BackgroundMainColor(ColorToARGB(clrBlack, 255));
   graphic.BackgroundMainSize(24);
   graphic.BackgroundMain("HMM Model Selection (AIC vs BIC)");
   graphic.XAxis().Name("Number of States");
   graphic.XAxis().NameSize(18);
   graphic.YAxis().Name("AIC/BIC Value (Lower is Better)");
   graphic.YAxis().NameSize(18);
   graphic.CurvePlotAll();
   graphic.Update();
   Sleep(10000);
   ChartSetInteger(0, CHART_SHOW, true);
   graphic.Destroy();
   ChartRedraw(0);
  }
//+------------------------------------------------------------------+