ScalerByLeo/ScalerBase.mqh

<EFBFBD><EFBFBD>//+------------------------------------------------------------------+
//|                                                   ScalerBase.mqh |
//|                                             Copyright 2025, Leo. |
//|                                             https://www.mql5.com |
//+------------------------------------------------------------------+
#property copyright "Copyright 2025, Leo."
#property link      "https://www.mql5.com/en/users/nique_372"
#property strict

#ifndef SCALER_BY_LEO_GEN_MQH
#define SCALER_BY_LEO_GEN_MQH

#include "..\\MQLArticles\\Utils\\Funciones Array.mqh"
#include "..\\MQLArticles\\Utils\\File.mqh"

//+------------------------------------------------------------------+
//|                         Scaler Base                              |
//+------------------------------------------------------------------+
class ScalerBase : public CLoggerBase
 {
protected:
  string             file_name_out;
  string             prefix_file;
  bool               loaded_scaler;

  bool               use_custom; //Bandera para saber si se usa custom (true) o excluyed (false)
  ulong              start_col;
  ulong              count_cols;
  ulong              excluyed_cols;

  virtual bool       Save() = 0;
  virtual bool       Load(string prefix_name) = 0;

  //--- M<EFBFBD>todos auxiliares
  bool               CheckSizeCustom(const matrix &mtx) const;
  bool               CheckSizeExcluded(const matrix &mtx) const;
  bool               CheckSizeCustom(const vector &v) const;
  bool               CheckSizeExcluded(const vector &v) const;

  //---
  matrix             ExtractMatrixToScale(const matrix &X) const;
  matrix             ReconstructMatrix(const matrix &X_original, const matrix &X_scaled) const;
  vector             ExtractVectorToScale(const vector &X) const;
  vector             ReconstructVector(const vector &X_original, const vector &X_scaled) const;

public:
                     ScalerBase(void);
                    ~ScalerBase(void) {}

  //---
  inline void        SetRangeEscaler(ulong start_col_, ulong count_col_); //Custom
  inline void        SetRangeEscaler(ulong excluyed_cols_ = 1); //Mas simple, el usuario decide cuantas columnas empezando por atras se excluyen

  //---
  inline bool        save(string prefix_name);
  inline bool        load(string prefix_name);

  //---
  virtual matrix     fit_transform(const matrix &X, bool save_data) = 0;
  virtual vector     fit_transform(const vector &X) = 0; //Para vectores no se guarda data
  virtual matrix     transform(const matrix &X,  bool solo_escalar_lo_previsto = false, bool reconstruir = false) = 0;
  virtual vector     transform(const vector &X,  bool solo_escalar_lo_previsto = false, bool reconstruir = false) = 0;
  virtual matrix     inverse_transform(const matrix &X_scaled,  bool solo_escalar_lo_previsto = false, bool reconstruir = false) = 0;
  virtual vector     inverse_transform(const vector &X_scaled,  bool solo_escalar_lo_previsto = false, bool reconstruir = false) = 0;

  // --- M<EFBFBD>todos comunes ---
  virtual inline string  GetOutputFile() const final { return this.file_name_out; }
 };

//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
ScalerBase::ScalerBase(void)
  : start_col(0), count_cols(0), use_custom(false), excluyed_cols(1), file_name_out(NULL), loaded_scaler(false)
 {
 }

//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
inline void ScalerBase::SetRangeEscaler(ulong start_col_, ulong count_col_)
 {
  this.use_custom = true;
  this.count_cols = count_col_;
  this.start_col = start_col_;
  LogInfo(StringFormat("Configurado escalado CUSTOM: columnas %I64u a %I64u (%I64u columnas)", start_col_, start_col_ + count_col_ - 1, count_col_), FUNCION_ACTUAL);
 }

//+------------------------------------------------------------------+
inline void ScalerBase::SetRangeEscaler(ulong excluyed_cols_ = 1)
 {
  this.excluyed_cols = excluyed_cols_;
  this.use_custom = false;
  LogInfo(StringFormat("Configurado escalado EXCLUDED: excluir <00>ltimas %I64u columnas", excluyed_cols_), FUNCION_ACTUAL);
 }

//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
bool ScalerBase::CheckSizeCustom(const matrix &mtx) const
 {
  if(start_col >= mtx.Cols())
   {
    LogError(StringFormat("Columna de inicio %I64u >= total columnas %I64u", start_col, mtx.Cols()), FUNCION_ACTUAL);
    return false;
   }

  if(start_col + count_cols > mtx.Cols())
   {
    LogError(StringFormat("Rango [%I64u:%I64u] excede columnas disponibles %I64u", start_col, start_col + count_cols - 1, mtx.Cols()), FUNCION_ACTUAL);
    return false;
   }

  return true;
 }

//+------------------------------------------------------------------+
bool ScalerBase::CheckSizeExcluded(const matrix &mtx) const
 {
  if(mtx.Cols() < excluyed_cols)
   {
    LogError(StringFormat("Columnas a excluir %I64u >= total columnas %I64u", excluyed_cols, mtx.Cols()), FUNCION_ACTUAL);
    return false;
   }

  return true;
 }


//+------------------------------------------------------------------+
bool ScalerBase::CheckSizeCustom(const vector &v) const
 {
  if(start_col >= v.Size())
   {
    LogError(StringFormat("Columna de inicio %I64u >= tama<00>o total del vector%I64u", start_col, v.Size()), FUNCION_ACTUAL);
    return false;
   }

  if(start_col + count_cols > v.Size())
   {
    LogError(StringFormat("Rango [%I64u:%I64u] excede el tama<00>o del vector %I64u", start_col, start_col + count_cols - 1, v.Size()), FUNCION_ACTUAL);
    return false;
   }

  return true;
 }

//+------------------------------------------------------------------+
bool ScalerBase::CheckSizeExcluded(const vector &v) const
 {
  if(v.Size() < excluyed_cols)
   {
    LogError(StringFormat("Columnas a excluir %I64u >= tama<00>o del vector: %I64u", excluyed_cols, v.Size()), FUNCION_ACTUAL);
    return false;
   }

  return true;
 }


//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
matrix ScalerBase::ExtractMatrixToScale(const matrix &X) const
 {
  matrix result;

  if(use_custom)
   {
    if(X.Cols() == count_cols)
      return X;

    result.Init(X.Rows(), count_cols);

    for(ulong row = 0; row < X.Rows(); row++)
      for(ulong col = 0; col < count_cols; col++)
        result[row][col] = X[row][start_col + col];
   }
  else
   {
    if(excluyed_cols == 0)
      return X;

    ulong cols_to_scale = X.Cols() - excluyed_cols;
    result.Init(X.Rows(), cols_to_scale);

    for(ulong row = 0; row < X.Rows(); row++)
      for(ulong col = 0; col < cols_to_scale; col++)
        result[row][col] = X[row][col];
   }

  return result;
 }

//+------------------------------------------------------------------+
matrix ScalerBase::ReconstructMatrix(const matrix &X_original, const matrix &X_scaled) const
 {
  if(X_original.Rows() == X_scaled.Rows() && X_original.Cols() == X_scaled.Cols())
    return X_scaled;

  matrix result = X_original; // Copia completa

  if(use_custom)
   {
    for(ulong row = 0; row < X_original.Rows(); row++)
      for(ulong col = 0; col < count_cols; col++)
        result[row][start_col + col] = X_scaled[row][col];
   }
  else
   {
    for(ulong row = 0; row < X_original.Rows(); row++)
      for(ulong col = 0; col < X_scaled.Cols(); col++)
        result[row][col] = X_scaled[row][col];
   }

  return result;
 }

//+------------------------------------------------------------------+
vector ScalerBase::ExtractVectorToScale(const vector &X) const
 {
  vector result;

  if(use_custom)
   {
    if(X.Size() == count_cols)
      return X;

    // Extraer rango espec<EFBFBD>fico
    result.Resize(count_cols);
    for(ulong i = 0; i < count_cols; i++)
      result[i] = X[start_col + i];
   }
  else
   {
    if(excluyed_cols == 0)
      return X;

    // Extraer todas excepto las <EFBFBD>ltimas N
    ulong size_to_scale = X.Size() - excluyed_cols;
    result.Resize(size_to_scale);
    for(ulong i = 0; i < size_to_scale; i++)
      result[i] = X[i];
   }

  return result;
 }

//+------------------------------------------------------------------+
vector ScalerBase::ReconstructVector(const vector &X_original, const vector &X_scaled) const
 {
  if(X_original.Size() == X_scaled.Size())
    return X_scaled;

  vector result = X_original;
  if(use_custom)
   {
    // Reemplazar rango espec<EFBFBD>fico
    for(ulong i = 0; i < count_cols; i++)
      result[start_col + i] = X_scaled[i];
   }
  else
   {
    // Reemplazar todas excepto las <EFBFBD>ltimas N
    for(ulong i = 0; i < X_scaled.Size(); i++)
      result[i] = X_scaled[i];
   }

  return result;
 }

//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
inline bool ScalerBase::load(string prefix_name)
 {
  loaded_scaler = true;
  return this.Load(prefix_name);
 }

//+------------------------------------------------------------------+
inline bool ScalerBase::save(string prefix_name)
 {
  this.file_name_out = prefix_name + this.prefix_file;
  return this.Save();
 }

//+------------------------------------------------------------------+
//|               Standardization Scaler                             |
//+------------------------------------------------------------------+
class StandardizationScaler : public ScalerBase
 {
protected:
  vector             mean, std;
  bool               Save() override;
  bool               Load(string prefix_name) override;

public:
                     StandardizationScaler() : ScalerBase() { this.prefix_file = "_mean_std.csv"; }

  matrix             fit_transform(const matrix &X, bool save_data) override;
  vector             fit_transform(const vector &X) override; //Para vectores no se guarda data
  matrix             transform(const matrix &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
  vector             transform(const vector &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
  matrix             inverse_transform(const matrix &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
  vector             inverse_transform(const vector &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
 };

//+------------------------------------------------------------------+
bool StandardizationScaler::Save()
 {
  FileDelete(this.file_name_out);
  ResetLastError();

  int handle = FileOpen(this.file_name_out, FILE_WRITE | FILE_CSV | FILE_COMMON);
  if(handle == INVALID_HANDLE)
   {
    LogFatalError(StringFormat("Invalid handle Err= %d >> Filename= %s", GetLastError(), this.file_name_out), FUNCION_ACTUAL);
    return false;
   }

  FileWrite(handle, vector_to_string(this.mean));
  FileWrite(handle, vector_to_string(this.std));
  FileWrite(handle, count_cols);
  FileWrite(handle, start_col);
  FileWrite(handle, excluyed_cols);
  FileWrite(handle, (int)use_custom);

  FileClose(handle);
  return true;
 }

//+------------------------------------------------------------------+
bool StandardizationScaler::Load(string prefix_name)
 {
  this.file_name_out = prefix_name + this.prefix_file;

  ResetLastError();
  int handle = FileOpen(file_name_out, FILE_READ | FILE_CSV | FILE_COMMON, "\n");
  if(handle == INVALID_HANDLE)
   {
    LogFatalError(StringFormat("Invalid handle Err= %d >> Filename= %s", GetLastError(), this.file_name_out), FUNCION_ACTUAL);
    return false;
   }

  this.mean = string_to_vector(FileReadString(handle));
  this.std = string_to_vector(FileReadString(handle));
  this.count_cols = (ulong)StringToInteger(FileReadString(handle));
  this.start_col = (ulong)StringToInteger(FileReadString(handle));
  this.excluyed_cols = (ulong)StringToInteger(FileReadString(handle));
  this.use_custom = (bool)StringToInteger(FileReadString(handle));

  FileClose(handle);
  return true;
 }

//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
vector StandardizationScaler::fit_transform(const vector &X)
 {
  if(loaded_scaler)
   {
    LogWarning("Este es un escalador cargado >> no es necesario ajustarlo a los nuevos datos, llame a otra instancia de una clase", FUNCION_ACTUAL);
    return transform(X);
   }

//---
  if(use_custom)
   {
    if(!CheckSizeCustom(X))
      return X;
   }
  else
   {
    if(!CheckSizeExcluded(X))
      return X;
   }

//---
  vector X_to_scaled = ExtractVectorToScale(X);

//---
  double mean_val = X_to_scaled.Mean();
  double std_val = X_to_scaled.Std();

  if(std_val < 1e-9)
    std_val = 1.0;

//---
  for(ulong i = 0; i < X_to_scaled.Size(); i++)
    X_to_scaled[i] = (X_to_scaled[i] - mean_val) / std_val;

  return ReconstructVector(X, X_to_scaled);
 }

//+------------------------------------------------------------------+
matrix StandardizationScaler::fit_transform(const matrix &X, bool save_data)
 {
  if(loaded_scaler)
   {
    LogWarning("Este es un escalador cargado >> no es necesario ajustarlo a los nuevos datos, llame a otra instancia de una clase", FUNCION_ACTUAL);
    return transform(X);
   }

  LogInfo(StringFormat("Numero de columnas de entrada: %I64u", X.Cols()), FUNCION_ACTUAL);

//---
  if(use_custom)
   {
    if(!CheckSizeCustom(X))
      return X;
   }
  else
   {
    if(!CheckSizeExcluded(X))
      return X;
   }

//---
  matrix X_to_scale = ExtractMatrixToScale(X);
  LogInfo(StringFormat("Columnas a escalar: %I64u", X_to_scale.Cols()), FUNCION_ACTUAL);

  vector mean_cts(X_to_scale.Cols());
  vector std_cts(X_to_scale.Cols());


  for(ulong i = 0; i < X_to_scale.Cols(); i++)
   {
    mean_cts[i] = X_to_scale.Col(i).Mean();
    std_cts[i] = X_to_scale.Col(i).Std();

    // Evitar divisi<EFBFBD>n por cero
    if(std_cts[i] < 1e-9)
     {
      LogWarning(StringFormat("Columna %I64u tiene std muy peque<00>o (%.2e), usando 1.0", i, std_cts[i]), FUNCION_ACTUAL);
      std_cts[i] = 1.0;
     }
   }

//---
  matrix X_scaled(X_to_scale.Rows(), X_to_scale.Cols());
  for(ulong row = 0; row < X_to_scale.Rows(); row++)
    for(ulong col = 0; col < X_to_scale.Cols(); col++)
      X_scaled[row][col] = (X_to_scale[row][col] - mean_cts[col]) / std_cts[col];


  if(save_data)
   {
    this.mean = mean_cts;
    this.std = std_cts;
   }

//--- Aqui siempre se reconstruye
  return ReconstructMatrix(X, X_scaled);
 }

//+------------------------------------------------------------------+
matrix StandardizationScaler::transform(const matrix &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Primero llame a fit_transform o load() antes de transform", FUNCION_ACTUAL);
    return X;
   }

//---
  if(use_custom)
   {
    if(!CheckSizeCustom(X))
      return X;
   }
  else
   {
    if(!CheckSizeExcluded(X))
      return X;
   }

//---
  matrix X_to_scale = solo_escalar_lo_previsto ? ExtractMatrixToScale(X) : X;

  if(X_to_scale.Cols() != this.mean.Size())
   {
    LogError(StringFormat("Columnas a escalar %I64u != columnas entrenadas %u", X_to_scale.Cols(), this.mean.Size()), FUNCION_ACTUAL);
    return X;
   }

//---
  matrix X_scaled(X_to_scale.Rows(), X_to_scale.Cols());
  for(ulong row = 0; row < X_to_scale.Rows(); row++)
    for(ulong col = 0; col < X_to_scale.Cols(); col++)
      X_scaled[row][col] = (X_to_scale[row][col] - this.mean[col]) / this.std[col];

//---
  if(reconstruir)
    return ReconstructMatrix(X, X_scaled);
  else
    return X_scaled;
 }

//+------------------------------------------------------------------+
vector StandardizationScaler::transform(const vector &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Primero llame a fit_transform o load() antes de transform", FUNCION_ACTUAL);
    return X;
   }

//---
  vector X_to_scale = solo_escalar_lo_previsto ? ExtractVectorToScale(X) : X;

  if(X_to_scale.Size() != this.mean.Size())
   {
    LogError(StringFormat("Elementos a escalar %I64 != elementos entrenados %u", X_to_scale.Size(), this.mean.Size()), FUNCION_ACTUAL);
    return X;
   }

//---
  vector X_scaled(X_to_scale.Size());
  for(ulong i = 0; i < X_to_scale.Size(); i++)
    X_scaled[i] = (X_to_scale[i] - this.mean[i]) / this.std[i];

//---
  if(reconstruir)
    return ReconstructVector(X, X_scaled);
  else
    return X_scaled;
 }

//+------------------------------------------------------------------+
matrix StandardizationScaler::inverse_transform(const matrix &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Escalador no entrenado. Llame primero a fit_transform", FUNCION_ACTUAL);
    return X_scaled;
   }

//---
  matrix X_to_unscale = (solo_escalar_lo_previsto) ? ExtractMatrixToScale(X_scaled) : X_scaled;

//---
  if(X_to_unscale.Cols() != this.mean.Size())
   {
    LogError(StringFormat("Columnas escaladas %I64u != columnas entrenadas %u", X_to_unscale.Cols(), this.mean.Size()), FUNCION_ACTUAL);

    return X_scaled;
   }

//---
  matrix X_unscaled(X_to_unscale.Rows(), X_to_unscale.Cols());
  for(ulong row = 0; row < X_to_unscale.Rows(); row++)
    for(ulong col = 0; col < X_to_unscale.Cols(); col++)
      X_unscaled[row][col] = X_to_unscale[row][col] * this.std[col] + this.mean[col];


//---
  if(reconstruir)
    return ReconstructMatrix(X_scaled, X_unscaled);
  else
    return X_unscaled;
 }

//+------------------------------------------------------------------+
vector StandardizationScaler::inverse_transform(const vector &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Escalador no entrenado. Llame primero a fit_transform", FUNCION_ACTUAL);
    return X_scaled;
   }

//---
  vector X_to_unscale;

  if(solo_escalar_lo_previsto)
    X_to_unscale = ExtractVectorToScale(X_scaled);
  else
    X_to_unscale = X_scaled;

//---
  if(X_to_unscale.Size() != this.mean.Size())
   {
    LogError(StringFormat("Elementos escalados %I64u != elementos entrenados %u", X_to_unscale.Size(), this.mean.Size()), FUNCION_ACTUAL);
    return X_scaled;
   }

//---
  vector X_unscaled(X_to_unscale.Size());
  for(ulong i = 0; i < X_to_unscale.Size(); i++)
   {
    X_unscaled[i] = X_to_unscale[i] * this.std[i] + this.mean[i];
   }

//---
  if(reconstruir)
    return ReconstructVector(X_scaled, X_unscaled);
  else
    return X_unscaled;
 }
//+------------------------------------------------------------------+
//|                    MaxMin Scaler                                 |
//+------------------------------------------------------------------+
class MaxMinScaler : public ScalerBase
 {
protected:
  vector             min_vals, max_vals;

  bool               Save() override;
  bool               Load(string prefix_name) override;

public:
                     MaxMinScaler() : ScalerBase() { this.prefix_file = "_min_max.csv"; }

  vector             fit_transform(const vector &X) override; //Para vectores no se guarda data
  matrix             fit_transform(const matrix &X, bool save_data) override;
  matrix             transform(const matrix &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
  vector             transform(const vector &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
  matrix             inverse_transform(const matrix &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
  vector             inverse_transform(const vector &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
 };

//+------------------------------------------------------------------+
bool MaxMinScaler::Save()
 {
  FileDelete(this.file_name_out);
  ResetLastError();

  int handle = FileOpen(this.file_name_out, FILE_WRITE | FILE_CSV | FILE_COMMON);
  if(handle == INVALID_HANDLE)
   {
    LogFatalError(StringFormat("Invalid handle Err= %d - Filename= %s", GetLastError(), this.file_name_out), FUNCION_ACTUAL);
    return false;
   }

  FileWrite(handle, vector_to_string(this.min_vals));
  FileWrite(handle, vector_to_string(this.max_vals));
  FileWrite(handle, count_cols);
  FileWrite(handle, start_col);
  FileWrite(handle, excluyed_cols);
  FileWrite(handle, (int)use_custom);

  FileClose(handle);
  return true;
 }

//+------------------------------------------------------------------+
bool MaxMinScaler::Load(string prefix_name)
 {
  this.file_name_out = prefix_name + this.prefix_file;

  ResetLastError();
  int handle = FileOpen(file_name_out, FILE_READ | FILE_CSV | FILE_COMMON, "\n");
  if(handle == INVALID_HANDLE)
   {
    LogFatalError(StringFormat("Invalid handle Err= %d - Filename= %s", GetLastError(), this.file_name_out), FUNCION_ACTUAL);
    return false;
   }

  this.min_vals = string_to_vector(FileReadString(handle));
  this.max_vals = string_to_vector(FileReadString(handle));
  this.count_cols = (ulong)StringToInteger(FileReadString(handle));
  this.start_col = (ulong)StringToInteger(FileReadString(handle));
  this.excluyed_cols = (ulong)StringToInteger(FileReadString(handle));
  this.use_custom = (bool)StringToInteger(FileReadString(handle));

  FileClose(handle);
  return true;
 }
//+------------------------------------------------------------------+
vector MaxMinScaler::fit_transform(const vector &X)
 {
  if(loaded_scaler)
   {
    LogWarning("Este es un escalador cargado | no es necesario ajustarlo a los nuevos datos, llame a otra instancia de una clase", FUNCION_ACTUAL);
    return transform(X);
   }

//---
  if(use_custom)
   {
    if(!CheckSizeCustom(X))
      return X;
   }
  else
   {
    if(!CheckSizeExcluded(X))
      return X;
   }

//---
  vector X_to_scale = ExtractVectorToScale(X);

//---
  double max = X_to_scale.Max();
  double min = X_to_scale.Min();
  if(max - min < 1e-10)
    max = min + 1.0;

//---
  for(ulong i = 0; i < X_to_scale.Size(); i++)
    X_to_scale[i] = (X_to_scale[i] - min) / (max - min);


//---
  return ReconstructVector(X, X_to_scale);
 }

//+------------------------------------------------------------------+
matrix MaxMinScaler::fit_transform(const matrix &X, bool save_data)
 {
  if(loaded_scaler)
   {
    LogWarning("Este es un escalador cargado | no es necesario ajustarlo a los nuevos datos, llame a otra instancia de una clase", FUNCION_ACTUAL);
    return transform(X);
   }

  LogInfo(StringFormat("Numero de columnas de entrada: %I64u", X.Cols()), FUNCION_ACTUAL);

//---
  if(use_custom)
   {
    if(!CheckSizeCustom(X))
      return X;
   }
  else
   {
    if(!CheckSizeExcluded(X))
      return X;
   }

//---
  matrix X_to_scale = ExtractMatrixToScale(X);

  LogInfo(StringFormat("Columnas a escalar: %I64u", X_to_scale.Cols()), FUNCION_ACTUAL);

//---
  vector min_vals_cts(X_to_scale.Cols());
  vector max_vals_cts(X_to_scale.Cols());


  for(ulong i = 0; i < X_to_scale.Cols(); i++)
   {
    min_vals_cts[i] = X_to_scale.Col(i).Min();
    max_vals_cts[i] = X_to_scale.Col(i).Max();

    if(fabs(max_vals_cts[i] - min_vals_cts[i]) < 1e-10)
     {
      LogWarning(StringFormat("Columna %I64u tiene rango muy peque<00>o (%.2e), usando rango 1.0", i, max_vals_cts[i] - min_vals_cts[i]), FUNCION_ACTUAL);
      max_vals_cts[i] = min_vals_cts[i] + 1.0;
     }
   }

//---
  matrix X_scaled(X_to_scale.Rows(), X_to_scale.Cols());
  for(ulong row = 0; row < X_to_scale.Rows(); row++)
    for(ulong col = 0; col < X_to_scale.Cols(); col++)
      X_scaled[row][col] = (X_to_scale[row][col] - min_vals_cts[col]) / (max_vals_cts[col] - min_vals_cts[col]);


  if(save_data)
   {
    this.min_vals = min_vals_cts;
    this.max_vals = max_vals_cts;
   }

//---
  return ReconstructMatrix(X, X_scaled);
 }

//+------------------------------------------------------------------+
matrix MaxMinScaler::transform(const matrix &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Escalador no entrenado. Llame primero a fit_transform o load()", FUNCION_ACTUAL);
    return X;
   }
//---
  if(use_custom)
   {
    if(!CheckSizeCustom(X))
      return X;
   }
  else
   {
    if(!CheckSizeExcluded(X))
      return X;
   }

//---
  matrix X_to_scale =  solo_escalar_lo_previsto ? ExtractMatrixToScale(X) : X;

  if(X_to_scale.Cols() != this.min_vals.Size())
   {
    LogError(StringFormat("Columnas a escalar %I64u != columnas entrenadas %u", X_to_scale.Cols(), this.min_vals.Size()), FUNCION_ACTUAL);
    return X;
   }

//---
  matrix X_scaled(X_to_scale.Rows(), X_to_scale.Cols());
  for(ulong row = 0; row < X_to_scale.Rows(); row++)
    for(ulong col = 0; col < X_to_scale.Cols(); col++)
      X_scaled[row][col] = (X_to_scale[row][col] - this.min_vals[col]) / (this.max_vals[col] - this.min_vals[col]);

//---
  if(reconstruir)
    return ReconstructMatrix(X, X_scaled);
  else
    return X_scaled;
 }

//+------------------------------------------------------------------+
vector MaxMinScaler::transform(const vector &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Escalador no entrenado. Llame primero a fit_transform o load()", FUNCION_ACTUAL);
    return X;
   }

//---
  vector X_to_scale = solo_escalar_lo_previsto ? ExtractVectorToScale(X) : X;

  if(X_to_scale.Size() != this.min_vals.Size())
   {
    LogError(StringFormat("Elementos a escalar %I64u != elementos entrenados %u", X_to_scale.Size(), this.min_vals.Size()), FUNCION_ACTUAL);
    return X;
   }

//---
  vector X_scaled(X_to_scale.Size());
  for(ulong i = 0; i < X_to_scale.Size(); i++)
    X_scaled[i] = (X_to_scale[i] - this.min_vals[i]) / (this.max_vals[i] - this.min_vals[i]);


//---
  if(reconstruir)
    return ReconstructVector(X, X_scaled);
  else
    return X_scaled;

 }

//+------------------------------------------------------------------+
matrix MaxMinScaler::inverse_transform(const matrix &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Escalador no entrenado. Llame primero a fit_transform", FUNCION_ACTUAL);
    return X_scaled;
   }
//---
  matrix X_to_unscale;
  if(solo_escalar_lo_previsto)
    X_to_unscale = ExtractMatrixToScale(X_scaled);
  else
    X_to_unscale = X_scaled;

//---
  if(X_to_unscale.Cols() != this.min_vals.Size())
   {
    LogError(StringFormat("Columnas escaladas %I64u != columnas entrenadas %u", X_to_unscale.Cols(), this.min_vals.Size()), FUNCION_ACTUAL);
    return X_scaled;
   }

//---
  matrix X_unscaled(X_to_unscale.Rows(), X_to_unscale.Cols());
  for(ulong row = 0; row < X_to_unscale.Rows(); row++)
   {
    for(ulong col = 0; col < X_to_unscale.Cols(); col++)
     {
      X_unscaled[row][col] = X_to_unscale[row][col] * (this.max_vals[col] - this.min_vals[col]) + this.min_vals[col];

     }
   }

//---
  if(reconstruir)
    return ReconstructMatrix(X_scaled, X_unscaled);
  else
    return X_unscaled;
 }

//+------------------------------------------------------------------+
vector MaxMinScaler::inverse_transform(const vector &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Escalador no entrenado. Llame primero a fit_transform", FUNCION_ACTUAL);
    return X_scaled;
   }

//---
  vector X_to_unscale;

  if(solo_escalar_lo_previsto)
    X_to_unscale = ExtractVectorToScale(X_scaled);
  else
    X_to_unscale = X_scaled;


  if(X_to_unscale.Size() != this.min_vals.Size())
   {
    LogError(StringFormat("Elementos escalados %I64u != elementos entrenados %u", X_to_unscale.Size(), this.min_vals.Size()), FUNCION_ACTUAL);
    return X_scaled;
   }

//---
  vector X_unscaled(X_to_unscale.Size());
  for(ulong i = 0; i < X_to_unscale.Size(); i++)
    X_unscaled[i] = X_to_unscale[i] * (this.max_vals[i] - this.min_vals[i]) + this.min_vals[i];

//---
  if(reconstruir)
    return ReconstructVector(X_scaled, X_unscaled);
  else
    return X_unscaled;
 }

//+------------------------------------------------------------------+
//|                    Robust Scaler                                 |
//+------------------------------------------------------------------+
class RobustScaler : public ScalerBase
 {
protected:
  vector             medians, iqrs;

  bool               Save() override;
  bool               Load(string prefix_name) override;

public:
                     RobustScaler() : ScalerBase() { this.prefix_file = "_median_iqr.csv"; }

  vector             fit_transform(const vector &X) override; //Para vectores no se guarda data
  matrix             fit_transform(const matrix &X, bool save_data) override;
  matrix             transform(const matrix &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
  vector             transform(const vector &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
  matrix             inverse_transform(const matrix &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
  vector             inverse_transform(const vector &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false) override;
 };

//+------------------------------------------------------------------+
bool RobustScaler::Save()
 {
  FileDelete(this.file_name_out);
  ResetLastError();

  int handle = FileOpen(this.file_name_out, FILE_WRITE | FILE_CSV | FILE_COMMON);
  if(handle == INVALID_HANDLE)
   {
    LogFatalError(StringFormat("Invalid handle Err= %d - Filename= %s", GetLastError(), this.file_name_out), FUNCION_ACTUAL);
    return false;
   }

  FileWrite(handle, vector_to_string(this.medians));
  FileWrite(handle, vector_to_string(this.iqrs));
  FileWrite(handle, count_cols);
  FileWrite(handle, start_col);
  FileWrite(handle, excluyed_cols);
  FileWrite(handle, (int)use_custom);

  FileClose(handle);
  return true;
 }

//+------------------------------------------------------------------+
bool RobustScaler::Load(string prefix_name)
 {
  this.file_name_out = prefix_name + this.prefix_file;
  ResetLastError();

  int handle = FileOpen(file_name_out, FILE_READ | FILE_CSV | FILE_COMMON, "\n");
  if(handle == INVALID_HANDLE)
   {
    LogFatalError(StringFormat("Invalid handle Err= %d - Filename= %s", GetLastError(), this.file_name_out), FUNCION_ACTUAL);
    return false;
   }

  this.medians = string_to_vector(FileReadString(handle));
  this.iqrs = string_to_vector(FileReadString(handle));
  this.count_cols = (ulong)StringToInteger(FileReadString(handle));
  this.start_col = (ulong)StringToInteger(FileReadString(handle));
  this.excluyed_cols = (ulong)StringToInteger(FileReadString(handle));
  this.use_custom = (bool)StringToInteger(FileReadString(handle));

  FileClose(handle);
  return true;
 }

//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
vector RobustScaler::fit_transform(const vector &X)
 {
  if(loaded_scaler)
   {
    LogWarning("Este es un escalador cargado | no es necesario ajustarlo a los nuevos datos, llame a otra instancia de una clase", FUNCION_ACTUAL);
    return transform(X);
   }

//---
  if(use_custom)
   {
    if(!CheckSizeCustom(X))
      return X;
   }
  else
   {
    if(!CheckSizeExcluded(X))
      return X;
   }

//---
  vector X_to_scale = ExtractVectorToScale(X);

//---
  double medians_cts = X_to_scale.Median();
  double q75 = X_to_scale.Percentile(75);
  double q25 = X_to_scale.Percentile(25);
  double iqrs_cts = q75 - q25;

  if(fabs(iqrs_cts) < 1e-10)
    iqrs_cts = 1.00;


//---
  for(ulong i = 0; i < X_to_scale.Size(); i++)
    X_to_scale[i] = (X_to_scale[i] - medians_cts) / iqrs_cts;

//---
  return ReconstructVector(X, X_to_scale);
 }

//+------------------------------------------------------------------+
matrix RobustScaler::fit_transform(const matrix &X, bool save_data)
 {
  if(loaded_scaler)
   {
    LogWarning("Este es un escalador cargado | no es necesario ajustarlo a los nuevos datos, llame a otra instancia de una clase", FUNCION_ACTUAL);
    return transform(X);
   }

  LogInfo(StringFormat("Numero de columnas de entrada: %I64u", X.Cols()), FUNCION_ACTUAL);

//---
  if(use_custom)
   {
    if(!CheckSizeCustom(X))
      return X;
   }
  else
   {
    if(!CheckSizeExcluded(X))
      return X;
   }

//---
  matrix X_to_scale = ExtractMatrixToScale(X);
  LogInfo(StringFormat("Columnas a escalar: %I64u", X_to_scale.Cols()), FUNCION_ACTUAL);

//---
  vector medians_cts(X_to_scale.Cols());
  vector iqrs_cts(X_to_scale.Cols());

//---
  for(ulong i = 0; i < X_to_scale.Cols(); i++)
   {
    vector col = X_to_scale.Col(i);
    medians_cts[i] = col.Median();
    double q75 = col.Percentile(75);
    double q25 = col.Percentile(25);
    iqrs_cts[i] = q75 - q25;
    
    if(fabs(iqrs_cts[i]) < 1e-10)
     {
      LogWarning(StringFormat("Columna %I64u tiene IQR muy peque<00>o (%.2e), usando 1.0", i, iqrs_cts[i]), FUNCION_ACTUAL);
      iqrs_cts[i] = 1.0;
     }
   }

//---
  matrix X_scaled(X_to_scale.Rows(), X_to_scale.Cols());
  for(ulong row = 0; row < X_to_scale.Rows(); row++)
    for(ulong col = 0; col < X_to_scale.Cols(); col++)
      X_scaled[row][col] = (X_to_scale[row][col] - medians_cts[col]) / iqrs_cts[col];

  if(save_data)
   {
    this.medians = medians_cts;
    this.iqrs = iqrs_cts;
   }

//---
  return ReconstructMatrix(X, X_scaled);
 }

//+------------------------------------------------------------------+
matrix RobustScaler::transform(const matrix &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Escalador no entrenado. Llame primero a fit_transform o load()", FUNCION_ACTUAL);
    return X;
   }

//---
  if(use_custom)
   {
    if(!CheckSizeCustom(X))
      return X;
   }
  else
   {
    if(!CheckSizeExcluded(X))
      return X;
   }

//---
  matrix X_to_scale = solo_escalar_lo_previsto ? ExtractMatrixToScale(X) : X;

  if(X_to_scale.Cols() != this.medians.Size())
   {
    LogError(StringFormat("Columnas a escalar %I64u != columnas entrenadas %u", X_to_scale.Cols(), this.medians.Size()), FUNCION_ACTUAL);
    return X;
   }

//---
  matrix X_scaled(X_to_scale.Rows(), X_to_scale.Cols());
  for(ulong row = 0; row < X_to_scale.Rows(); row++)
    for(ulong col = 0; col < X_to_scale.Cols(); col++)
      X_scaled[row][col] = (X_to_scale[row][col] - this.medians[col]) / this.iqrs[col];

//---
  if(reconstruir)
    return ReconstructMatrix(X, X_scaled);
  else
    return X_scaled;
 }

//+------------------------------------------------------------------+
vector RobustScaler::transform(const vector &X, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Escalador no entrenado. Llame primero a fit_transform o load()", FUNCION_ACTUAL);
    return X;
   }

//---
  vector X_to_scale = solo_escalar_lo_previsto ? ExtractVectorToScale(X) : X;

  if(X_to_scale.Size() != this.medians.Size())
   {
    LogError(StringFormat("Elementos a escalar %I64u != elementos entrenados %u", X_to_scale.Size(), this.medians.Size()), FUNCION_ACTUAL);
    return X;
   }

//---
  vector X_scaled(X_to_scale.Size());
  for(ulong i = 0; i < X_to_scale.Size(); i++)
    X_scaled[i] = (X_to_scale[i] - this.medians[i]) / this.iqrs[i];


//---
  if(reconstruir)
    return ReconstructVector(X, X_scaled);
  else
    return X_scaled;
 }

//+------------------------------------------------------------------+
matrix RobustScaler::inverse_transform(const matrix &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Escalador no entrenado. Llame primero a fit_transform", FUNCION_ACTUAL);
    return X_scaled;
   }

  matrix X_to_unscale;

  if(solo_escalar_lo_previsto)
    X_to_unscale = ExtractMatrixToScale(X_scaled);
  else
    X_to_unscale = X_scaled;


  if(X_to_unscale.Cols() != this.medians.Size())
   {
    LogError(StringFormat("Columnas escaladas %I64u != columnas entrenadas %u", X_to_unscale.Cols(), this.medians.Size()), FUNCION_ACTUAL);
    return X_scaled;
   }

//---
  matrix X_unscaled(X_to_unscale.Rows(), X_to_unscale.Cols());
  for(ulong row = 0; row < X_to_unscale.Rows(); row++)
    for(ulong col = 0; col < X_to_unscale.Cols(); col++)
      X_unscaled[row][col] = X_to_unscale[row][col] * this.iqrs[col] + this.medians[col];

//---
  if(reconstruir)
    return ReconstructMatrix(X_scaled, X_unscaled);
  else
    return X_unscaled;

 }

//+------------------------------------------------------------------+
vector RobustScaler::inverse_transform(const vector &X_scaled, bool solo_escalar_lo_previsto = false, bool reconstruir = false)
 {
  if(!loaded_scaler)
   {
    LogError("Escalador no entrenado. Llame primero a fit_transform", FUNCION_ACTUAL);
    return X_scaled;
   }

  vector X_to_unscale;

  if(solo_escalar_lo_previsto)
    X_to_unscale = ExtractVectorToScale(X_scaled);
  else
    X_to_unscale = X_scaled;


  if(X_to_unscale.Size() != this.medians.Size())
   {
    LogError(StringFormat("Elementos escalados %I64u != elementos entrenados %u", X_to_unscale.Size(), this.medians.Size()), FUNCION_ACTUAL);
    return X_scaled;
   }

//---
  vector X_unscaled(X_to_unscale.Size());
  for(ulong i = 0; i < X_to_unscale.Size(); i++)
    X_unscaled[i] = X_to_unscale[i] * this.iqrs[i] + this.medians[i];


//---
  if(reconstruir)
    return ReconstructVector(X_scaled, X_unscaled);
  else
    return X_unscaled; 
 }
//+------------------------------------------------------------------+
#endif