【機械学習】AIを取り入れた新しいトレード方法

matrixf GetLiveData(uint start, uint total)
 {
   matrixf return_matrix(total, 3);
   
   
    OPEN.CopyRates(Symbol(), PERIOD_CURRENT,COPY_RATES_OPEN, start, total);
    HIGH.CopyRates(Symbol(), PERIOD_CURRENT,COPY_RATES_HIGH, start, total);
    LOW.CopyRates(Symbol(), PERIOD_CURRENT,COPY_RATES_LOW, start, total);
        
    return_matrix.Col(OPEN, 0);
    return_matrix.Col(HIGH, 1);
    return_matrix.Col(LOW, 2);
          
      
   return return_matrix;
 }

switch(NORM)
   {
    case  NORM_MEAN_NORM:
      
       //--- saving the mean
       
       norm_params.Assign(norm_x.mean_norm_scaler.mean);
       WriteCsv(normparams_folder+csv_name_+".mean_norm_scaler.mean.csv",norm_params,x_vars);
       
       //--- saving the min
       
       norm_params.Assign(norm_x.mean_norm_scaler.min);
       WriteCsv(normparams_folder+csv_name_+".mean_norm_scaler.min.csv",norm_params,x_vars);
       
       //--- saving the max
       
       norm_params.Assign(norm_x.mean_norm_scaler.max);
       WriteCsv(normparams_folder+csv_name_+".mean_norm_scaler.max.csv",norm_params,x_vars);
           
      break;
      
    // ... (他のケースは省略)
   }

# 多層パーセプトロンニューラルネットワーク(MLP)の構築
class NeuralNetworkClass():
    def __init__(self, csv_name, target_column, batch_size=32):
        # データの読み込みと前処理
        self.data = pd.read_csv(dataset_path+"\\"+csv_name)
        if self.data.empty:
            print(f"No such dataset or Empty dataset csv = {csv_name}")
            quit()
        
        print(self.data.head())
        
        self.target_column = target_column
        X = self.data.drop(columns=self.target_column).to_numpy()
        Y = self.data[self.target_column].to_numpy()
        
        self.train_x, self.test_x, self.train_y, self.test_y = train_test_split(X, Y, test_size=0.3, random_state=42)
        
        print(f"train x shape {self.train_x.shape}\ntest x shape {self.test_x.shape}")
        
        self.input_size = self.train_x.shape[-1]
        self.output_size = 1
        self.batch_size = batch_size
        self.model = None
        
        self.plots_directory = "Plots"
        self.models_directory = "Models"

    def BuildNeuralNetwork(self, activation_function='relu', neurons = 10):
        self.model = keras.Sequential([
            keras.layers.Input(shape=(self.input_size,)),
            keras.layers.Dense(units=neurons, activation=activation_function, activity_regularizer=l2(0.01), kernel_initializer="he_uniform"),
            keras.layers.Dense(units=self.output_size, activation='linear', activity_regularizer=l2(0.01), kernel_initializer="he_uniform")  
        ])
        self.model.summary()

    # ... (train_network と test_network メソッドは省略)

def saveONNXModel(self, folder="ONNX Models"):
    
    path = data_path + "\\MQL5\\Files\\" + folder 
    
    if not os.path.exists(path): # create this path if it doesn't exist
        os.makedirs(path)
    
    onnx_model_name = f"MLP.REG.{self.target_column}.{self.data.shape[0]}.onnx"
    path +=  "\\" + onnx_model_name
    
    
    loaded_keras_model = load_model(f"Models\\MLP.REG.{self.target_column}.{self.data.shape[0]}.h5") 
    
    onnx_model, _ = tf2onnx.convert.from_keras(loaded_keras_model, output_path=path)

    onnx.save(onnx_model, path )
    
    print(f'Saved model to {path}')

#resource "\\Files\\ONNX Models\\MLP.REG.CLOSE.10000.onnx" as uchar RNNModel[]

long mlp_onnxhandle;

#include <MALE5\preprocessing.mqh>
CPreprocessing<vectorf, matrixf> *norm_x;

int inputs[], outputs[];

vectorf OPEN,
       HIGH, 
       LOW;
       
//+------------------------------------------------------------------+
//| Expert initialization function                                   |
//+------------------------------------------------------------------+
int OnInit()
  {
//---
   
  if (!LoadNormParams()) //Load the normalization parameters we saved once
    {
      Print("Normalization parameters csv files couldn't be found \nEnsure you are collecting data and Normalizing them using [ONNX get data.ex5] Script \nTrain the Python model again if necessary");
      return INIT_FAILED;
    }
   
//--- ONNX SETTINGS
 
  mlp_onnxhandle = OnnxCreateFromBuffer(RNNModel, MQLInfoInteger(MQL_DEBUG) ? ONNX_DEBUG_LOGS : ONNX_DEFAULT); //creating onnx handle buffer | rUN DEGUG MODE during debug mode
  
  if (mlp_onnxhandle == INVALID_HANDLE)
    {
       Print("OnnxCreateFromBuffer Error = ",GetLastError());
       return INIT_FAILED;
    }

//--- since not all sizes defined in the input tensor we must set them explicitly
//--- first index - batch size, second index - series size, third index - number of series (only Close)
   
   OnnxTypeInfo type_info; //Getting onnx information for Reference In case you forgot what the loaded ONNX is all about

   long input_count=OnnxGetInputCount(mlp_onnxhandle);
   Print("model has ",input_count," input(s)");
   for(long i=0; i<input_count; i++)
     {
      string input_name=OnnxGetInputName(mlp_onnxhandle,i);
      Print(i," input name is ",input_name);
      if(OnnxGetInputTypeInfo(mlp_onnxhandle,i,type_info))
        {
          PrintTypeInfo(i,"input",type_info);
          ArrayCopy(inputs, type_info.tensor.dimensions);
        }
     }

   long output_count=OnnxGetOutputCount(mlp_onnxhandle);
   Print("model has ",output_count," output(s)");
   for(long i=0; i<output_count; i++)
     {
      string output_name=OnnxGetOutputName(mlp_onnxhandle,i);
      Print(i," output name is ",output_name);
      if(OnnxGetOutputTypeInfo(mlp_onnxhandle,i,type_info))
       {
         PrintTypeInfo(i,"output",type_info);
         ArrayCopy(outputs, type_info.tensor.dimensions);
       }
     }
   
//---

   if (MQLInfoInteger(MQL_DEBUG))
    {
      Print("Inputs & Outputs");
      ArrayPrint(inputs);
      ArrayPrint(outputs);
    }
   
   const long input_shape[] = {batch_size, 3};
   
   if (!OnnxSetInputShape(mlp_onnxhandle, 0, input_shape)) //Giving the Onnx handle the input shape
     {
       printf("Failed to set the input shape Err=%d",GetLastError());
       return INIT_FAILED;
     }
   
   const long output_shape[] = {batch_size, 1};
   
   if (!OnnxSetOutputShape(mlp_onnxhandle, 0, output_shape)) //giving the onnx handle the output shape
     {
       printf("Failed to set the input shape Err=%d",GetLastError());
       return INIT_FAILED;
     }
   
//---

   return(INIT_SUCCEEDED);
  }

void OnTick()
  {
   matrixf input_data = GetLiveData(0,1);
   vectorf output_data(1);
   
   
   if (!OnnxRun(mlp_onnxhandle, ONNX_NO_CONVERSION, input_data, output_data))
     {
       Print("Failed to Get the Predictions Err=",GetLastError());
       ExpertRemove();
       return;
     }
   
   Comment("inputs_data\n",input_data,"\npredictions\n",output_data);
}