ml_engine/patterns.py

# ml_engine/patterns.py
# (V8.8 - الحل النهائي: إعادة فحوصات السلامة (Column Checks) لجميع المؤشرات)

import pandas as pd
import numpy as np
import joblib
import asyncio
import io

# (يجب التأكد من أن pandas-ta مثبت في بيئة Hugging Face)
try:
    import pandas_ta as ta
except ImportError:
    print("❌❌ [PatternEngineV8] مكتبة pandas_ta غير موجودة! هذا المحرك سيفشل.")
    ta = None

class ChartPatternAnalyzer:
    
    def __init__(self, r2_service=None, 
                 model_key="lgbm_pattern_model_combined.pkl", 
                 scaler_key="scaler_combined.pkl",
                 window_size=60):
        """
        تهيئة المحرك بتحميل النماذج من R2.
        """
        self.window_size = window_size
        self.model = None
        self.scaler = None
        self.class_names = ["Bearish Pattern", "Neutral / No Pattern", "Bullish Pattern"]
        self.r2_service = r2_service
        self.model_key = model_key
        self.scaler_key = scaler_key
        
        # (الوصفة V8.7 - 35 عموداً رقمياً)
        self.feature_names = [
            'open', 'high', 'low', 'close', 'volume', 
            'RSI_14', 'MACD_12_26_9', 'MACDh_12_26_9', 'MACDs_12_26_9', 'SMA_20', 
            'EMA_20', 'BBL_5_2.0_2.0', 'BBM_5_2.0_2.0', 'BBU_5_2.0_2.0', 'BBB_5_2.0_2.0', 
            'BBP_5_2.0_2.0', 'STOCHk_14_3_3', 'STOCHd_14_3_3', 'STOCHh_14_3_3', 
            'ADX_14', 'ADXR_14_2', 'DMP_14', 'DMN_14', 'VWAP_D', 'MIDPOINT_14', 
            'TEMA_20', 'OBV', 'AD', 'ATRr_14', 'DPO_20', 'KVO_34_55_13', 
            'KVOs_34_55_13', 'CMO_14', 'ROC_10', 'WILLR_14'
        ]

        if not self.r2_service:
            print("⚠️ [PatternEngineV8] R2Service غير متوفر. يجب التحميل يدوياً.")

    async def initialize(self):
        """
        يجب استدعاؤها من app.py أو data_manager لتحميل النماذج.
        """
        if self.model and self.scaler:
            return True 
            
        if not self.r2_service:
            print("❌ [PatternEngineV8] لا يمكن التهيئة بدون R2 Service.")
            return False
            
        try:
            print(f"   > [PatternEngineV8] تحميل {self.model_key} من R2...")
            model_obj = self.r2_service.s3_client.get_object(Bucket=self.r2_service.BUCKET_NAME, Key=self.model_key)
            model_bytes = io.BytesIO(model_obj['Body'].read())
            self.model = joblib.load(model_bytes)
            
            print(f"   > [PatternEngineV8] تحميل {self.scaler_key} من R2...")
            scaler_obj = self.r2_service.s3_client.get_object(Bucket=self.r2_service.BUCKET_NAME, Key=self.scaler_key)
            scaler_bytes = io.BytesIO(scaler_obj['Body'].read())
            self.scaler = joblib.load(scaler_bytes)

            print("✅ [PatternEngineV8] تم تحميل النموذج (58%) والمقياس بنجاح.")
            
            if hasattr(self.scaler, 'feature_names_in_'):
                 print(f"   > يتوقع المقياس {len(self.scaler.feature_names_in_)} خاصية.")
                 if len(self.scaler.feature_names_in_) == len(self.feature_names):
                     print("   > ✅ (V8.8) عدد الخصائص (35) متطابق مع المقياس.")
                 else:
                     print(f"   > ⚠️ (V8.8) تحذير: عدم تطابق الخصائص! الكود يتوقع {len(self.feature_names)}, المقياس يتوقع {len(self.scaler.feature_names_in_)}")
            
            return True
            
        except Exception as e:
            print(f"❌❌ [PatternEngineV8] فشل فادح في تحميل النماذج من R2: {e}")
            self.model = None
            self.scaler = None
            return False

    # 🔴 --- START OF CHANGE (V8.8) --- 🔴
    # (إعادة فحوصات السلامة (Column Checks) لجميع المؤشرات)
    def _extract_features(self, df_ranged: pd.DataFrame, df_indexed: pd.DataFrame) -> pd.DataFrame:
        """
        (الوصفة V8.8 - إرجاع 35 عموداً + فحوصات سلامة كاملة)
        """
        if not ta:
            raise ImportError("مكتبة pandas-ta غير مثبتة.")

        # (1. البدء بآخر صف من البيانات الأساسية)
        df_features = df_ranged.iloc[-1:].copy()
        
        # (2. بيانات مفهرسة لـ VWAP)
        h_idx = df_indexed['high']
        l_idx = df_indexed['low']
        c_idx = df_indexed['close']
        v_idx = df_indexed['volume']
        
        # (3. بيانات غير مفهرسة (السريعة) لباقي المؤشرات)
        c = df_ranged['close']
        h = df_ranged['high']
        l = df_ranged['low']
        v = df_ranged['volume']
        
        try:
            # --- حساب الـ 30 مؤشر (مع فحوصات السلامة) ---
            
            # (المؤشرات التي تُرجع سلسلة Series - آمنة نسبياً)
            df_features['RSI_14'] = ta.rsi(c, length=14).iloc[-1]
            df_features['SMA_20'] = ta.sma(c, length=20).iloc[-1]
            df_features['EMA_20'] = ta.ema(c, length=20).iloc[-1]
            df_features['MIDPOINT_14'] = ta.midpoint(c, length=14).iloc[-1]
            df_features['TEMA_20'] = ta.tema(c, length=20).iloc[-1]
            df_features['OBV'] = ta.obv(c, v).iloc[-1]
            df_features['AD'] = ta.ad(h, l, c, v).iloc[-1]
            df_features['ATRr_14'] = ta.atr(h, l, c, percent=True, length=14).iloc[-1]
            df_features['DPO_20'] = ta.dpo(c, length=20).iloc[-1]
            df_features['CMO_14'] = ta.cmo(c, length=14).iloc[-1]
            df_features['ROC_10'] = ta.roc(c, length=10).iloc[-1]
            df_features['WILLR_14'] = ta.willr(h, l, c, length=14).iloc[-1]

            # (الاستثناء: VWAP يستخدم بيانات مفهرسة)
            vwap_series = ta.vwap(h_idx, l_idx, c_idx, v_idx)
            if vwap_series is not None: 
                df_features['VWAP_D'] = vwap_series.iloc[-1]

            # --- (المؤشرات التي تُرجع DataFrame - تحتاج فحص سلامة) ---
            
            macd_data = ta.macd(c, fast=12, slow=26, signal=9)
            if macd_data is not None and not macd_data.empty and 'MACD_12_26_9' in macd_data.columns:
                df_features['MACD_12_26_9'] = macd_data['MACD_12_26_9'].iloc[-1]
                df_features['MACDh_12_26_9'] = macd_data['MACDh_12_26_9'].iloc[-1]
                df_features['MACDs_12_26_9'] = macd_data['MACDs_12_26_9'].iloc[-1]

            bb_data = ta.bbands(c, length=5, std=2.0)
            if bb_data is not None and not bb_data.empty and 'BBL_5_2.0' in bb_data.columns:
                df_features['BBL_5_2.0_2.0'] = bb_data['BBL_5_2.0'].iloc[-1] 
                df_features['BBM_5_2.0_2.0'] = bb_data['BBM_5_2.0'].iloc[-1]
                df_features['BBU_5_2.0_2.0'] = bb_data['BBU_5_2.0'].iloc[-1]
                df_features['BBB_5_2.0_2.0'] = bb_data['BBB_5_2.0'].iloc[-1]
                df_features['BBP_5_2.0_2.0'] = bb_data['BBP_5_2.0'].iloc[-1]

            stoch_data = ta.stoch(h, l, c, k=14, d=3, smooth_k=3)
            if stoch_data is not None and not stoch_data.empty and 'STOCHk_14_3_3' in stoch_data.columns:
                df_features['STOCHk_14_3_3'] = stoch_data['STOCHk_14_3_3'].iloc[-1]
                df_features['STOCHd_14_3_3'] = stoch_data['STOCHd_14_3_3'].iloc[-1]
                df_features['STOCHh_14_3_3'] = stoch_data['STOCHh_14_3_3'].iloc[-1]

            adx_data = ta.adx(h, l, c, length=14, adxr=2)
            if adx_data is not None and not adx_data.empty and 'ADX_14' in adx_data.columns:
                df_features['ADX_14'] = adx_data['ADX_14'].iloc[-1]
                df_features['ADXR_14_2'] = adx_data['ADXR_14_2'].iloc[-1]
                df_features['DMP_14'] = adx_data['DMP_14'].iloc[-1]
                df_features['DMN_14'] = adx_data['DMN_14'].iloc[-1]

            kvo_data = ta.kvo(h, l, c, v, fast=34, slow=55, signal=13)
            if kvo_data is not None and not kvo_data.empty and 'KVO_34_55_13' in kvo_data.columns:
                df_features['KVO_34_55_13'] = kvo_data['KVO_34_55_13'].iloc[-1]
                df_features['KVOs_34_55_13'] = kvo_data['KVOs_34_55_13'].iloc[-1]
        
        except Exception as e:
            # (هذا الخطأ يجب ألا يظهر الآن إلا في حالات نادرة جداً)
            print(f"❌ [PatternEngineV8.8] خطأ أثناء حساب المؤشرات وظيفياً: {e}")
            pass
        # --- (نهاية حساب المؤشرات) ---

        # (ملء أي قيم مفقودة (NaN) بـ 0 قبل إرسالها للمقياس)
        df_features.fillna(0, inplace=True)
        
        # (التأكد من أننا نرسل فقط الـ 35 عموداً التي يتوقعها المقياس)
        final_features_df = pd.DataFrame(columns=self.feature_names)
        
        for col in self.feature_names:
            if col in df_features:
                final_features_df[col] = df_features[col].values
            else:
                final_features_df[col] = 0 

        return final_features_df
    # 🔴 --- END OF CHANGE (V8.8) --- 🔴

    async def detect_chart_patterns(self, ohlcv_data: dict) -> dict:
        """
        (الدالة الرئيسية - لا تغيير هنا عن V8.7)
        """
        best_match = {
            'pattern_detected': 'no_clear_pattern',
            'pattern_confidence': 0,
            'predicted_direction': 'neutral',
            'timeframe': None,
            'details': {}
        }
        
        if not self.model or not self.scaler:
            if not hasattr(self, '_init_warned'):
                print("⚠️ [PatternEngineV8] النموذج/المقياس غير محمل. يجب استدعاء .initialize() أولاً.")
                self._init_warned = True
            return best_match
            
        all_results = []

        for timeframe, candles in ohlcv_data.items():
            if len(candles) >= max(self.window_size, 200):
                try:
                    window_candles = candles[-200:]
                    
                    # (1. نسخة غير مفهرسة (RangeIndex 0,1,2...))
                    df_ranged = pd.DataFrame(window_candles, columns=['timestamp', 'open', 'high', 'low', 'close', 'volume'])
                    
                    # (2. نسخة مفهرسة (DatetimeIndex))
                    df_indexed = df_ranged.copy()
                    df_indexed['timestamp'] = pd.to_datetime(df_indexed['timestamp'], unit='ms')
                    df_indexed.set_index('timestamp', inplace=True)

                    # (3. استخراج الخصائص (V8.8))
                    features_df = self._extract_features(df_ranged, df_indexed)
                    
                    if features_df is None or features_df.empty:
                        continue 
                        
                    # (4. تطبيع الخصائص (Scaler))
                    features_scaled = self.scaler.transform(features_df)
                    
                    # (5. التنبؤ بالاحتماليات (Probabilities))
                    probabilities = self.model.predict_proba(features_scaled)[0]
                    
                    best_class_index = np.argmax(probabilities)
                    confidence = probabilities[best_class_index]
                    pattern_name = self.class_names[best_class_index]
                    
                    if pattern_name != "Neutral / No Pattern" and confidence > 0.5:
                        all_results.append({
                            'pattern': pattern_name,
                            'confidence': float(confidence),
                            'timeframe': timeframe
                        })
                        
                except Exception as e:
                    print(f"❌ [PatternEngineV8.8] فشل التنبؤ لـ {timeframe}: {e}")
                    
        # (6. اختيار أفضل نمط)
        if all_results:
            best_result = max(all_results, key=lambda x: x['confidence'])
            
            direction = 'neutral'
            if "Bullish" in best_result['pattern']: direction = 'up'
            elif "Bearish" in best_result['pattern']: direction = 'down'

            best_match['pattern_detected'] = best_result['pattern']
            best_match['pattern_confidence'] = best_result['confidence']
            best_match['timeframe'] = best_result['timeframe']
            best_match['predicted_direction'] = direction
            best_match['details'] = {'ml_confidence': best_result['confidence']}

        return best_match

print("✅ ML Module: Pattern Engine V8.8 (Robust DataFrame Checks) loaded")