212 lines
7.5 KiB
MQL5
212 lines
7.5 KiB
MQL5
|
//+------------------------------------------------------------------+
|
||
|
|
//| GateSystemAutoLearning.mqh - Master Include for Auto-Learning |
|
||
|
|
//| Single include that activates full learning pipeline: |
|
||
|
|
//| - Shadow logging |
|
||
|
|
//| - Auto-tuning |
|
||
|
|
//| - ONNX model loading/reloading |
|
||
|
|
//| - Trigger-based retraining |
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
#ifndef GATESYSTEMAUTOLEARNING_MQH
|
||
|
|
#define GATESYSTEMAUTOLEARNING_MQH
|
||
|
|
|
||
|
|
// Core components
|
||
|
|
#include "CGateBase.mqh"
|
||
|
|
#include "CShadowLogger.mqh"
|
||
|
|
#include "CMLPolishGateONNX.mqh"
|
||
|
|
#include "EnhancedEfficientGateSystem.mqh"
|
||
|
|
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
//| Auto-Learning Configuration |
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
struct SAutoLearningConfig
|
||
|
|
{
|
||
|
|
bool enabled; // Master switch
|
||
|
|
bool shadow_mode; // Shadow logging only?
|
||
|
|
bool auto_tune_gates; // Enable auto-tuning?
|
||
|
|
bool load_onnx_models; // Load ONNX dynamic thresholds?
|
||
|
|
bool auto_reload_models; // Auto-reload when new model available?
|
||
|
|
int retrain_interval_hours; // Hours between retraining
|
||
|
|
string model_directory; // Where ONNX models are stored
|
||
|
|
|
||
|
|
void Init()
|
||
|
|
{
|
||
|
|
enabled = true;
|
||
|
|
shadow_mode = false;
|
||
|
|
auto_tune_gates = true;
|
||
|
|
load_onnx_models = true;
|
||
|
|
auto_reload_models = true;
|
||
|
|
retrain_interval_hours = 24;
|
||
|
|
model_directory = "DualEA/models/";
|
||
|
|
}
|
||
|
|
};
|
||
|
|
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
//| Auto-Learning Manager |
|
||
|
|
//| Monitors for model updates and triggers retraining |
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
class CAutoLearningManager
|
||
|
|
{
|
||
|
|
private:
|
||
|
|
SAutoLearningConfig m_config;
|
||
|
|
CEfficientGateManagerEnhanced* m_gate_manager;
|
||
|
|
|
||
|
|
// Model monitoring
|
||
|
|
datetime m_last_model_check;
|
||
|
|
string m_model_files[3]; // G5, G6, G7 models
|
||
|
|
datetime m_model_load_times[3];
|
||
|
|
|
||
|
|
// Trigger file paths
|
||
|
|
string m_trigger_file;
|
||
|
|
string m_reload_signal_file;
|
||
|
|
|
||
|
|
public:
|
||
|
|
CAutoLearningManager()
|
||
|
|
{
|
||
|
|
m_gate_manager = NULL;
|
||
|
|
m_last_model_check = 0;
|
||
|
|
m_trigger_file = "DualEA/retrain.trigger";
|
||
|
|
m_reload_signal_file = "DualEA/models/model.reload";
|
||
|
|
|
||
|
|
ArrayInitialize(m_model_load_times, 0);
|
||
|
|
m_model_files[0] = "DualEA/models/gate5_optimizer.onnx";
|
||
|
|
m_model_files[1] = "DualEA/models/gate6_optimizer.onnx";
|
||
|
|
m_model_files[2] = "DualEA/models/gate7_optimizer.onnx";
|
||
|
|
}
|
||
|
|
|
||
|
|
void Initialize(CEfficientGateManagerEnhanced* gm, SAutoLearningConfig &config)
|
||
|
|
{
|
||
|
|
m_gate_manager = gm;
|
||
|
|
m_config = config;
|
||
|
|
|
||
|
|
Print("[AutoLearning] Manager initialized");
|
||
|
|
Print("[AutoLearning] Shadow mode: " + (config.shadow_mode ? "YES" : "NO"));
|
||
|
|
Print("[AutoLearning] ONNX models: " + (config.load_onnx_models ? "YES" : "NO"));
|
||
|
|
Print("[AutoLearning] Auto-tuning: " + (config.auto_tune_gates ? "YES" : "NO"));
|
||
|
|
|
||
|
|
// Load initial models
|
||
|
|
if(config.load_onnx_models)
|
||
|
|
{
|
||
|
|
ReloadModels();
|
||
|
|
}
|
||
|
|
|
||
|
|
// Clear any stale reload signal
|
||
|
|
if(FileIsExist(m_reload_signal_file, FILE_COMMON))
|
||
|
|
{
|
||
|
|
FileDelete(m_reload_signal_file, FILE_COMMON);
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
//| Check and reload models if updated |
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
void CheckModelUpdates()
|
||
|
|
{
|
||
|
|
if(!m_config.enabled || !m_config.auto_reload_models)
|
||
|
|
return;
|
||
|
|
|
||
|
|
// Check every 5 minutes
|
||
|
|
if(TimeCurrent() - m_last_model_check < 300)
|
||
|
|
return;
|
||
|
|
|
||
|
|
m_last_model_check = TimeCurrent();
|
||
|
|
|
||
|
|
// Check for reload signal from Python
|
||
|
|
if(FileIsExist(m_reload_signal_file, FILE_COMMON))
|
||
|
|
{
|
||
|
|
Print("[AutoLearning] Model reload signal detected");
|
||
|
|
ReloadModels();
|
||
|
|
FileDelete(m_reload_signal_file, FILE_COMMON);
|
||
|
|
}
|
||
|
|
|
||
|
|
// Also check file modification times
|
||
|
|
for(int i=0; i<3; i++)
|
||
|
|
{
|
||
|
|
if(!FileIsExist(m_model_files[i], FILE_COMMON))
|
||
|
|
continue;
|
||
|
|
|
||
|
|
datetime file_time = (datetime)FileGetInteger(m_model_files[i],
|
||
|
|
FILE_MODIFY_DATE,
|
||
|
|
FILE_COMMON);
|
||
|
|
|
||
|
|
if(file_time > m_model_load_times[i])
|
||
|
|
{
|
||
|
|
Print(StringFormat("[AutoLearning] Model %d updated, reloading...", i));
|
||
|
|
ReloadModels();
|
||
|
|
break;
|
||
|
|
}
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
//| Reload all ONNX models |
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
void ReloadModels()
|
||
|
|
{
|
||
|
|
Print("[AutoLearning] Reloading ONNX models...");
|
||
|
|
|
||
|
|
// Note: In actual implementation, you would access individual gates
|
||
|
|
// and reload their ONNX models. For now, log the intent.
|
||
|
|
|
||
|
|
for(int i=0; i<3; i++)
|
||
|
|
{
|
||
|
|
if(FileIsExist(m_model_files[i], FILE_COMMON))
|
||
|
|
{
|
||
|
|
m_model_load_times[i] = (datetime)FileGetInteger(m_model_files[i],
|
||
|
|
FILE_MODIFY_DATE,
|
||
|
|
FILE_COMMON);
|
||
|
|
Print(StringFormat("[AutoLearning] Loaded model %d: %s", i, m_model_files[i]));
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
Print("[AutoLearning] Model reload complete");
|
||
|
|
}
|
||
|
|
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
//| Trigger retraining (called when sufficient data collected) |
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
void TriggerRetraining()
|
||
|
|
{
|
||
|
|
if(!m_config.enabled)
|
||
|
|
return;
|
||
|
|
|
||
|
|
// Create trigger file for Python optimizer
|
||
|
|
int handle = FileOpen(m_trigger_file,
|
||
|
|
FILE_WRITE|FILE_TXT|FILE_COMMON);
|
||
|
|
|
||
|
|
if(handle != INVALID_HANDLE)
|
||
|
|
{
|
||
|
|
FileWriteString(handle, "Retraining triggered at " +
|
||
|
|
TimeToString(TimeCurrent()) + "\n");
|
||
|
|
FileClose(handle);
|
||
|
|
|
||
|
|
Print("[AutoLearning] Retraining triggered");
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
//| Record trade outcome for learning |
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
void RecordOutcome(double pnl)
|
||
|
|
{
|
||
|
|
if(m_gate_manager != NULL)
|
||
|
|
{
|
||
|
|
m_gate_manager.RecordTradeOutcome(pnl);
|
||
|
|
}
|
||
|
|
}
|
||
|
|
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
//| Periodic update (call from OnTick or OnTimer) |
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
void OnTickUpdate()
|
||
|
|
{
|
||
|
|
CheckModelUpdates();
|
||
|
|
}
|
||
|
|
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
//| Getters |
|
||
|
|
//+------------------------------------------------------------------+
|
||
|
|
bool IsEnabled() const { return m_config.enabled; }
|
||
|
|
bool IsShadowMode() const { return m_config.shadow_mode; }
|
||
|
|
};
|
||
|
|
|
||
|
|
#endif // GATESYSTEMAUTOLEARNING_MQH
|