//+------------------------------------------------------------------+
//|                                                    OnnxModel.mqh |
//|                                  Copyright 2025, Niquel Mendoza. |
//|                          https://www.mql5.com/es/users/nique_372 |
//+------------------------------------------------------------------+
#property copyright "Copyright 2025, Niquel Mendoza."
#property link      "https://www.mql5.com/es/users/nique_372"
#property strict

#ifndef EASYSB_MODELS_ONNXDEF_MQH
#define EASYSB_MODELS_ONNXDEF_MQH

//+------------------------------------------------------------------+
//| Include                                                          |
//+------------------------------------------------------------------+
#include <TSN\\MQLArticles\\Utils\\Basic.mqh>


//+------------------------------------------------------------------+
//| Defines                                                          |
//+------------------------------------------------------------------+
#define SIMPLE_MODEL_ONNX_CLASES 2

//+------------------------------------------------------------------+
//| Clase para clasificacion                                         |
//+------------------------------------------------------------------+
class COnnxSimpleModel : public CLoggerBase
 {
private:
  long                m_onnx_handle;

  struct Map
   {
    long                key[];
    float               value[];
   };


public:
                     COnnxSimpleModel(void) : m_onnx_handle(INVALID_HANDLE) {}
                    ~COnnxSimpleModel(void) {}

  //---
  bool               Init(const uchar& data[], int input_num);
  bool               Init(const string& file_name, bool comon, int input_num);
  bool               Predict(vector &features, vector& probas, ulong& best_class_index, long &out_best_class, long onnx_exe_flags);
 };

//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
bool COnnxSimpleModel::Init(const uchar &data[], int input_num)
 {
//---
  ResetLastError();
  m_onnx_handle = OnnxCreateFromBuffer(data, 0);
  if(m_onnx_handle == INVALID_HANDLE)
   {
    LogError(StringFormat("Fallo al crear el modelo onnx por buffer, ultimo error = %d", GetLastError()), FUNCION_ACTUAL);
    return false;
   }

//---
  ulong input_shape[] = {1, input_num};
  ResetLastError();
  if(!OnnxSetInputShape(m_onnx_handle, 0, input_shape))
   {
    LogError(StringFormat("Fallo al setear el input shape el modelo onnx por buffer [f=%d], ultimo error = %d", input_num, GetLastError()), FUNCION_ACTUAL);
    return false;
   }

//---
  ulong ouput_shape_label[] = {1};  // 0=N (valor dincamiso de muestras) | 1 = solo un valor por muestra (como un matrix de 1 col)
  OnnxSetOutputShape(m_onnx_handle, 0, ouput_shape_label);
  ulong ouput_shape_map[] = {1, 2}; //0 = N (netron no lo espeifica por lo isot) | 2 = key y float (aunque aqui tambine es complejo)
  OnnxSetOutputShape(m_onnx_handle, 1, ouput_shape_map);

//---
  return true;
 }

//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
bool COnnxSimpleModel::Init(const string &file_name, bool comon, int input_num)
 {
//---
  ResetLastError();
  m_onnx_handle = OnnxCreate(file_name, (comon ? ONNX_COMMON_FOLDER : 0));
  if(m_onnx_handle == INVALID_HANDLE)
   {
    LogError(StringFormat("Fallo al crear el modelo onnx por file = %s, ultimo error = %d", file_name, GetLastError()), FUNCION_ACTUAL);
    return false;
   }

//---
  ulong input_shape[] = {1, input_num};
  ResetLastError();
  if(!OnnxSetInputShape(m_onnx_handle, 0, input_shape))
   {
    LogError(StringFormat("Fallo al setear el input shape el modelo onnx por buffer [f=%d], ultimo error = %d", input_num, GetLastError()), FUNCION_ACTUAL);
    return false;
   }

//---
  ulong ouput_shape_label[] = {1};  // 0=N (valor dincamiso de muestras) | 1 = solo un valor por muestra (como un matrix de 1 col)
  OnnxSetOutputShape(m_onnx_handle, 0, ouput_shape_label);
  ulong ouput_shape_map[] = {1, 2}; //0 = N (netron no lo espeifica por lo isot) | 2 = key y float (aunque aqui tambine es complejo)
  OnnxSetOutputShape(m_onnx_handle, 1, ouput_shape_map);

//---
  return true;
 }



//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
bool COnnxSimpleModel::Predict(vector &features, vector& probas, ulong& best_class_index, long &out_best_class, long onnx_exe_flags)
 {
  static vectorf x_float;
  x_float.Assign(features);

//--- Label
  long out_put_data[];
  ArrayResize(out_put_data, 1);

//--- Mapa
  Map mapa_out[];
  ArrayResize(mapa_out, 1);
  ArrayResize(mapa_out[0].key, SIMPLE_MODEL_ONNX_CLASES);
  ArrayResize(mapa_out[0].value, SIMPLE_MODEL_ONNX_CLASES);

//---
  ResetLastError();
  if(!OnnxRun(m_onnx_handle, onnx_exe_flags, x_float, out_put_data, mapa_out))
   {
    LogFatalError(StringFormat("No se pudo ejecutar el modelo onnx, ultimo error = %d", GetLastError()), FUNCION_ACTUAL);
    out_best_class = -1;
    probas.Resize(0);
    best_class_index = -1;
    return false;
   }

//---
  probas.Assign(mapa_out[0].value);
// VectorPrint(v, 4);
  best_class_index = probas.ArgMax();
  out_best_class = mapa_out[0].key[best_class_index]; // Clave maxcimo valor
  return true;
 }


//+------------------------------------------------------------------+
//| Clase para modelos de regresion                                  |
//+------------------------------------------------------------------+
class COnnxModelRegresoin : public CLoggerBase
 {
private:
  long                m_onnx_handle;

public:
                     COnnxModelRegresoin(void) : m_onnx_handle(INVALID_HANDLE) {}
                    ~COnnxModelRegresoin(void) { if(m_onnx_handle != INVALID_HANDLE) OnnxRelease(m_onnx_handle); }

  //---
  bool               Init(const uchar& data[], int input_num);
  bool               Init(const string& file_name, bool comon, int input_num);
  bool               Predict(vector& entrada, long onnx_exe_flags, double& output);
 };

//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
bool COnnxModelRegresoin::Init(const uchar &data[], int input_num)
 {
//---
  ResetLastError();
  m_onnx_handle = OnnxCreateFromBuffer(data, 0);
  if(m_onnx_handle == INVALID_HANDLE)
   {
    LogError(StringFormat("Fallo al crear el modelo onnx por buffer, ultimo error = %d", GetLastError()), FUNCION_ACTUAL);
    return false;
   }

//---
  ulong input_shape[] = {1, input_num};
  ResetLastError();
  if(!OnnxSetInputShape(m_onnx_handle, 0, input_shape))
   {
    LogError(StringFormat("Fallo al setear el input shape el modelo onnx por buffer [f=%d], ultimo error = %d", input_num, GetLastError()), FUNCION_ACTUAL);
    return false;
   }

//---
  ulong ouput_shape_label[] = {1};  // 1=N (valor dincamiso de muestras) solo queremos 1 salida
  OnnxSetOutputShape(m_onnx_handle, 0, ouput_shape_label);

//---
  return true;
 }


//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
bool COnnxModelRegresoin::Init(const string &file_name,bool comon,int input_num)
 {
//---
  ResetLastError();
  m_onnx_handle = OnnxCreate(file_name, (comon ? ONNX_COMMON_FOLDER : 0));
  if(m_onnx_handle == INVALID_HANDLE)
   {
    LogError(StringFormat("Fallo al crear el modelo onnx por file = %s, ultimo error = %d", file_name, GetLastError()), FUNCION_ACTUAL);
    return false;
   }

//---
  ulong input_shape[] = {1, input_num};
  ResetLastError();
  if(!OnnxSetInputShape(m_onnx_handle, 0, input_shape))
   {
    LogError(StringFormat("Fallo al setear el input shape el modelo onnx por buffer [f=%d], ultimo error = %d", input_num, GetLastError()), FUNCION_ACTUAL);
    return false;
   }

//---
  ulong ouput_shape_label[] = {1};  // 1=N (valor dincamiso de muestras) solo queremos 1 salida
  OnnxSetOutputShape(m_onnx_handle, 0, ouput_shape_label);

//---
  return true;
 }

//+------------------------------------------------------------------+
//|                                                                  |
//+------------------------------------------------------------------+
bool COnnxModelRegresoin::Predict(vector& entrada, long onnx_exe_flags, double& output)
 {
//---
  static matrixf x_float;
  x_float.Assign(entrada);
// Print(x_float.Cols());
//Print(x_float.Rows());

//---
  static matrixf out;
  out.Resize(1, 1);

//---
  ResetLastError();
  if(!OnnxRun(m_onnx_handle, onnx_exe_flags, x_float, out))
   {
    LogError(StringFormat("Fallo al eejeuctar la prediccion del modelo onnx, ultimo error = %d", GetLastError()), FUNCION_ACTUAL);
    return false;
   }

//---
  output = out[0][0];
  return true;
 }

//+------------------------------------------------------------------+
#endif //EASYSB_MODELS_ONNXDEF_MQH
//+------------------------------------------------------------------+