Create oracle_engine.py

ml_engine/oracle_engine.py

@@ -0,0 +1,283 @@
+import os
+import numpy as np
+import pandas as pd
+import pandas_ta as ta
+import lightgbm as lgb
+from scipy.signal import find_peaks
+import warnings
+from typing import Dict, Any, List, Optional
+
+# --- [ 0. إعدادات ] ---
+warnings.filterwarnings('ignore', category=FutureWarning)
+
+PIPELINE_SETTINGS = {
+    'SWING_PROMINENCE_PCT': 0.02, # 2%
+}
+
+# (بناءً على تقريرك [cite: 5-8]، الدقة فوق 75% ممتازة)
+DECISION_CONFIDENCE_THRESHOLD = 0.75 
+N_STRATEGY_MODELS = 11
+
+STRATEGY_MAP = {
+    0: 'WAIT',
+    1: 'SWING_LONG',
+    2: 'SCALP_LONG',
+    3: 'SWING_SHORT',
+    4: 'SCALP_SHORT'
+}
+
+# (الأطر الزمنية التي تم تدريب النماذج عليها)
+TIMEBYTES_TO_PROCESS = ['15m', '1h', '4h']
+
+
+class OracleEngine:
+    def __init__(self, model_dir: str = "ml_models/Unified_Models_V1"):
+        self.model_dir = model_dir
+        self.strategy_boosters: List[lgb.Booster] = []
+        self.quantile_boosters: Dict[str, lgb.Booster] = {}
+        self.feature_names: List[str] = []
+        self.initialized = False
+        print("🧠 [OracleEngine V2] تم الإنشاء (Multi-Timeframe). جاهز للتهيئة.")
+
+    async def initialize(self):
+        """
+        تحميل جميع النماذج الـ 15 (11 استراتيجية + 4 أهداف) إلى الذاكرة.
+        """
+        if self.initialized:
+            return True
+            
+        print(f"🧠 [OracleEngine V2] جاري تحميل 15 نموذجاً من {self.model_dir}...")
+        try:
+            # 1. تحميل نماذج "لجنة القرار" (Strategy Ensemble)
+            for i in range(N_STRATEGY_MODELS):
+                model_file = os.path.join(self.model_dir, f"lgbm_strategy_fold_{i}.txt")
+                if not os.path.exists(model_file):
+                    print(f"❌ [Oracle Error] ملف نموذج مفقود: {model_file}")
+                    return False
+                booster = lgb.Booster(model_file=model_file)
+                self.strategy_boosters.append(booster)
+            
+            print(f"   ✅ تم تحميل {len(self.strategy_boosters)} نماذج استراتيجية.")
+
+            # 2. تحميل نماذج "لجنة الأهداف" (Quantile Models)
+            quantile_names = ['tp_p20', 'tp_p50', 'tp_p80', 'sl_p80']
+            for name in quantile_names:
+                model_file = os.path.join(self.model_dir, f"lgbm_{name}.txt")
+                if not os.path.exists(model_file):
+                    print(f"❌ [Oracle Error] ملف نموذج مفقود: {model_file}")
+                    return False
+                booster = lgb.Booster(model_file=model_file)
+                self.quantile_boosters[name] = booster
+            
+            print(f"   ✅ تم تحميل {len(self.quantile_boosters)} نماذج أهداف.")
+
+            # 3. حفظ قائمة الميزات
+            self.feature_names = self.strategy_boosters[0].feature_name()
+            self.initialized = True
+            
+            print(f"✅ [OracleEngine V2] جاهز. (Threshold: {DECISION_CONFIDENCE_THRESHOLD*100}%)")
+            print(f"   -> سيعمل على الأطر: {TIMEBYTES_TO_PROCESS}")
+            return True
+
+        except Exception as e:
+            print(f"❌ [OracleEngine V2] فشل فادح أثناء التهيئة: {e}")
+            self.initialized = False
+            return False
+
+    # --- [ دوال هندسة الميزات (مطابقة 100% للتدريب) ] ---
+    
+    def _calculate_base_ta(self, df: pd.DataFrame) -> pd.DataFrame:
+        df.ta.rsi(length=14, append=True)
+        df.ta.adx(length=14, append=True)
+        df.ta.macd(fast=12, slow=26, signal=9, append=True)
+        df.ta.bbands(length=20, std=2, append=True)
+        df.ta.atr(length=14, append=True)
+        for length in [9, 21, 50, 100, 200]:
+            df[f'EMA_{length}'] = ta.ema(df['close'], length=length)
+        return df
+
+    def _calculate_market_structure(self, df: pd.DataFrame, prominence_pct: float) -> pd.DataFrame:
+        prominence_value = df['close'].mean() * prominence_pct
+        high_peaks_idx, _ = find_peaks(df['high'], prominence=prominence_value)
+        low_peaks_idx, _ = find_peaks(-df['low'], prominence=prominence_value)
+        df['last_SH_price'] = df.iloc[high_peaks_idx]['high'].reindex(df.index).ffill().bfill()
+        df['last_SL_price'] = df.iloc[low_peaks_idx]['low'].reindex(df.index).ffill().bfill()
+        df['BOS_Long'] = np.where(df['close'] > df['last_SH_price'].shift(1), 1, 0)
+        df['BOS_Short'] = np.where(df['low'] < df['last_SL_price'].shift(1), 1, 0)
+        return df
+
+    def _calculate_fibonacci_matrix(self, df: pd.DataFrame) -> pd.DataFrame:
+        wave_range = df['last_SH_price'] - df['last_SL_price']
+        df['fibo_0.382'] = df['last_SH_price'] - (wave_range * 0.382)
+        df['fibo_0.500'] = df['last_SH_price'] - (wave_range * 0.500)
+        df['fibo_0.618'] = df['last_SL_price'] + (wave_range * 0.618)
+        df['fibo_ext_1.618'] = df['last_SH_price'] + (wave_range * 1.618)
+        df['dist_to_0.618_pct'] = (df['close'] - df['fibo_0.618']) / (df['close'] + 1e-9)
+        df['dist_to_1.618_pct'] = (df['fibo_ext_1.618'] - df['close']) / (df['close'] + 1e-9)
+        df['is_in_golden_pocket'] = np.where(
+            (df['close'] < df['fibo_0.500']) & (df['close'] > df['fibo_0.618']), 1, 0
+        )
+        df.replace([np.inf, -np.inf], np.nan, inplace=True)
+        return df
+
+    def _calculate_alpha_strategies(self, df: pd.DataFrame) -> pd.DataFrame:
+        df['volume_zscore'] = (df['volume'] - df['volume'].rolling(50).mean()) / (df['volume'].rolling(50).std() + 1e-9)
+        df['dist_from_EMA200_pct'] = (df['close'] - df['EMA_200']) / (df['EMA_200'] + 1e-9)
+        
+        # (إصلاح Bollinger Bands كما في كود التجميع)
+        bbu_col = next((col for col in df.columns if 'BBU_20_2.0' in str(col)), None)
+        bbl_col = next((col for col in df.columns if 'BBL_20_2.0' in str(col)), None)
+        bbm_col = next((col for col in df.columns if 'BBM_20_2.0' in str(col)), None)
+
+        if all([bbu_col, bbl_col, bbm_col]):
+            df['BBW_pct'] = (df[bbu_col] - df[bbl_col]) / (df[bbm_col] + 1e-9)
+            df['is_squeeze'] = np.where(df['BBW_pct'] < df['BBW_pct'].rolling(100).min(), 1, 0)
+        else:
+            df['BBW_pct'] = np.nan
+            df['is_squeeze'] = 0
+
+        df['is_trending'] = np.where(df['ADX_14'] > 25, 1, 0)
+        df['ATR_pct'] = (df['ATRr_14'] / df['close']) * 100
+        return df
+
+    def _create_feature_vector(self, ohlcv_tf_data: List) -> Optional[pd.DataFrame]:
+        """
+        تشغيل خط أنابيب الميزات الكامل على بيانات إطار زمني واحد.
+        """
+        if ohlcv_tf_data is None or len(ohlcv_tf_data) < 200:
+             return None
+        
+        df = pd.DataFrame(ohlcv_tf_data, columns=['timestamp', 'open', 'high', 'low', 'close', 'volume'])
+        df = df.astype(float)
+        df['datetime'] = pd.to_datetime(df['timestamp'], unit='ms')
+        df = df.set_index('datetime')
+
+        # تشغيل خط الأنابيب
+        df = self._calculate_base_ta(df)
+        df = self._calculate_market_structure(df, PIPELINE_SETTINGS['SWING_PROMINENCE_PCT'])
+        df = self._calculate_fibonacci_matrix(df)
+        df = self._calculate_alpha_strategies(df)
+        
+        # ملء أي قيم NaN متبقية (مهم جداً للتنبؤ)
+        df = df.ffill().bfill()
+        
+        # أخذ آخر صف فقط
+        latest_features = df.iloc[-1:]
+        
+        # التأكد من وجود جميع الميزات بالترتيب الصحيح
+        try:
+            feature_vector = latest_features[self.feature_names]
+            # التأكد من عدم وجود NaN نهائياً
+            if feature_vector.isnull().values.any():
+                print("⚠️ [Oracle Warning] Feature vector contains NaN after fill.")
+                return None
+            return feature_vector
+        except Exception as e:
+            print(f"❌ [Oracle Error] عدم تطابق الميزات: {e}")
+            return None
+
+
+    async def predict(self, symbol_data: Dict[str, Any]) -> Dict[str, Any]:
+        """
+        الدالة الرئيسية: تحليل إشارة مرشحة وإرجاع قرار كامل.
+        (هذه الدالة تحل محل llm_service.get_trading_decision)
+        """
+        if not self.initialized:
+            return {'action': 'WAIT', 'reason': 'Oracle Engine not initialized'}
+
+        ohlcv_data = symbol_data.get('ohlcv')
+        current_price = symbol_data.get('current_price')
+        if not ohlcv_data or not current_price:
+            return {'action': 'WAIT', 'reason': 'Missing OHLCV or price data'}
+
+        try:
+            all_tf_decisions = []
+
+            # --- [ الخطوة 1: تحليل كل إطار زمني على حدة ] ---
+            for tf in TIMEBYTES_TO_PROCESS:
+                feature_vector = self._create_feature_vector(ohlcv_data.get(tf))
+                
+                if feature_vector is None:
+                    print(f"   -> {symbol_data['symbol']} @ {tf}: Skipped (Insufficient data)")
+                    continue
+                
+                # 1. تشغيل "لجنة القرار" (Strategy Ensemble)
+                all_probs = [
+                    booster.predict(feature_vector, num_iteration=booster.best_iteration)
+                    for booster in self.strategy_boosters
+                ]
+                ensemble_probs = np.mean(all_probs, axis=0)[0] # (نأخذ التنبؤ الأول)
+                
+                # 2. تحليل القرار والثقة
+                predicted_strategy_idx = np.argmax(ensemble_probs)
+                confidence = ensemble_probs[predicted_strategy_idx]
+                strategy_name = STRATEGY_MAP.get(predicted_strategy_idx, 'WAIT')
+
+                all_tf_decisions.append({
+                    'timeframe': tf,
+                    'strategy': strategy_name,
+                    'confidence': float(confidence),
+                    'feature_vector': feature_vector # (نحتفظ به لاستخدامه لاحقاً)
+                })
+
+            if not all_tf_decisions:
+                return {'action': 'IGNORE', 'reason': 'Feature calculation failed for all TFs'}
+
+            # --- [ الخطوة 2: اختيار القرار الأفضل (أعلى ثقة) ] ---
+            best_decision = max(all_tf_decisions, key=lambda x: x['confidence'])
+
+            strategy_name = best_decision['strategy']
+            confidence = best_decision['confidence']
+            best_tf = best_decision['timeframe']
+
+            # --- [ الخطوة 3: تطبيق فلتر الثقة (الأهم) ] ---
+            if confidence < DECISION_CONFIDENCE_THRESHOLD or strategy_name == 'WAIT':
+                return {
+                    'action': 'IGNORE',
+                    'reason': f"Best signal ({strategy_name} @ {best_tf}) confidence ({confidence:.2f}) is below threshold ({DECISION_CONFIDENCE_THRESHOLD})",
+                    'confidence': confidence,
+                    'strategy': strategy_name
+                }
+
+            # --- [ الخطوة 4: (نجحت الثقة) - تشغيل "لجنة الأهداف" ] ---
+            winning_feature_vector = best_decision['feature_vector']
+            preds_quantile = {}
+            for name, booster in self.quantile_boosters.items():
+                preds_quantile[name] = booster.predict(winning_feature_vector, num_iteration=booster.best_iteration)[0]
+
+            # --- [ الخطوة 5: تحديد الأهداف النهائية ] ---
+            tp_pct = preds_quantile['tp_p50'] # (الهدف الواقعي) [cite: 79]
+            sl_pct = preds_quantile['sl_p80'] # (وقف الخسارة الآمن) [cite: 80]
+            
+            if tp_pct <= 0 or sl_pct <= 0:
+                return {'action': 'IGNORE', 'reason': f'Quantile model predicted negative TP/SL ({tp_pct=}, {sl_pct=})'}
+
+            if "LONG" in strategy_name:
+                tp_price = current_price * (1 + tp_pct)
+                sl_price = current_price * (1 - sl_pct)
+                action_type = "BUY"
+            elif "SHORT" in strategy_name:
+                tp_price = current_price * (1 - tp_pct)
+                sl_price = current_price * (1 + sl_pct)
+                action_type = "SELL" # (إذا كان النظام يدعم البيع)
+            else:
+                return {'action': 'IGNORE', 'reason': 'Strategy not actionable'}
+
+            # --- [ الخطوة 6: إرجاع القرار الكامل ] ---
+            return {
+                'action': 'WATCH', # (للتوافق مع `app.py` القديم)
+                'confidence': confidence,
+                'analysis_summary': f"Oracle Consensus @ {best_tf}: {strategy_name} (Conf: {confidence:.2%})",
+                'strategy': strategy_name,
+                'action_type': action_type, # (إضافة: BUY أو SELL)
+                'tp_price': float(tp_price),
+                'sl_price': float(sl_price),
+                'quantile_tp_pct': float(tp_pct),
+                'quantile_sl_pct': float(sl_pct)
+            }
+
+        except Exception as e:
+            print(f"❌ [OracleEngine V2] فشل فادح أثناء التنبؤ: {e}")
+            import traceback
+            traceback.print_exc()
+            return {'action': 'WAIT', 'reason': f'Exception: {e}'}