Update ml_engine/oracle_engine.py

ml_engine/oracle_engine.py

@@ -1,10 +1,9 @@
 # ==============================================================================
-# 🧠 ml_engine/oracle_engine.py (V4.5 - LightGBM Golden Threshold Edition)
+# 🧠 ml_engine/oracle_engine.py (V4.6 - Fix List/DataFrame Bug)
 # ==============================================================================
 # GEM-Architect Approved
-# - Uses the trained LightGBM model.
-# - Implements the "Golden Threshold" strategy (0.5% Predicted Return).
-# - Integrates CNN probabilities + Market Context properly.
+# - Fixes AttributeError: 'list' object has no attribute 'empty'
+# - Auto-converts raw list OHLCV to DataFrame.
 # ==============================================================================
 
 import os
@@ -23,14 +22,11 @@ class OracleEngine:
         self.model = None
         self.initialized = False
         
-        # 🏆 THE GOLDEN CONFIGURATION (From Testing)
-        # Threshold 0.005 (0.5% return) gave 77% Win Rate.
+        # 🏆 THE GOLDEN CONFIGURATION
         self.CONFIDENCE_THRESHOLD = 0.005 
         
-        # Context Features (Must match training)
+        # Context Features
         self.ctx_features = ["ret_var_30", "ret_skew_30", "ret_kurt_30"]
-        # CNN Features inputs
-        self.cnn_cols = ["cnn_prob_neutral", "cnn_prob_loss", "cnn_prob_win"]
 
     async def initialize(self):
         """Load LightGBM Model"""
@@ -51,12 +47,33 @@ class OracleEngine:
             print(f"❌ [Oracle] Init Error: {e}")
             return False
 
+    def _prepare_dataframe(self, data_input):
+        """Helper to safely convert List or DF to DataFrame"""
+        try:
+            if data_input is None: return None
+            
+            # If it's already a DataFrame
+            if isinstance(data_input, pd.DataFrame):
+                return data_input if not data_input.empty else None
+            
+            # If it's a List (CCXT format: [ts, o, h, l, c, v])
+            if isinstance(data_input, list):
+                if len(data_input) == 0: return None
+                df = pd.DataFrame(data_input, columns=['timestamp', 'open', 'high', 'low', 'close', 'volume'])
+                return df
+                
+            return None
+        except Exception:
+            return None
+
     def _calc_context_features(self, df):
-        """Calculate statistical context features from OHLCV"""
+        """Calculate statistical context features from OHLCV DataFrame"""
         try:
             if len(df) < 30: return np.zeros(3)
             
+            # Ensure close is float
             close = df['close'].values.astype(float)
+            
             prev_close = np.roll(close, 1); prev_close[0] = close[0]
             log_ret = np.log(close / np.maximum(prev_close, 1e-9))
             
@@ -75,28 +92,27 @@ class OracleEngine:
     async def predict(self, symbol_data: dict) -> dict:
         """
         Decision Core.
-        Input: symbol_data containing 'ohlcv' and 'titan_probs'.
+        Input: symbol_data containing 'ohlcv' (List or DF) and 'pattern_probs'.
         """
         if not self.initialized:
-            return {'action': 'WAIT', 'reason': 'Oracle Not Init', 'confidence': 0.0}
+            return {'action': 'WAIT', 'reason': 'Oracle Not Init', 'oracle_score': 0.0}
 
         try:
             # 1. Get Inputs
-            # Titan Probs: [Neutral, Loss, Win]
-            titan_probs = symbol_data.get('titan_probs') 
+            # Support both key names for compatibility
+            titan_probs = symbol_data.get('pattern_probs') or symbol_data.get('titan_probs')
             if not titan_probs or len(titan_probs) != 3:
-                return {'action': 'WAIT', 'reason': 'No Titan Input', 'confidence': 0.0}
+                return {'action': 'WAIT', 'reason': 'No Pattern Input', 'oracle_score': 0.0}
             
-            # Market Context (From 15m data)
-            ohlcv_15m = symbol_data.get('ohlcv', {}).get('15m')
-            if ohlcv_15m is None or ohlcv_15m.empty:
-                return {'action': 'WAIT', 'reason': 'No Market Data', 'confidence': 0.0}
+            # Market Context (From 15m data) - ✅ FIXED HERE
+            raw_15m = symbol_data.get('ohlcv', {}).get('15m')
+            df_15m = self._prepare_dataframe(raw_15m)
             
-            # 2. Build Feature Vector
-            # Order: [cnn_p0, cnn_p1, cnn_p2, ctx_var, ctx_skew, ctx_kurt]
-            ctx_vals = self._calc_context_features(ohlcv_15m)
+            if df_15m is None:
+                return {'action': 'WAIT', 'reason': 'No Market Data', 'oracle_score': 0.0}
             
-            # Handle NaN/Inf
+            # 2. Build Feature Vector
+            ctx_vals = self._calc_context_features(df_15m)
             ctx_vals = np.nan_to_num(ctx_vals, nan=0.0)
             
             input_vector = np.concatenate([titan_probs, ctx_vals]).reshape(1, -1)
@@ -104,40 +120,31 @@ class OracleEngine:
             # 3. Predict (Net Expected Return)
             predicted_pnl = float(self.model.predict(input_vector)[0])
             
-            # 4. Decision Logic (The Golden Rule)
-            # Titan Win Prob (Raw Confidence)
+            # 4. Decision Logic
             cnn_win_prob = titan_probs[2] 
-            
-            # We combine Oracle PnL Prediction AND Titan Win Prob
-            # Oracle says "How much money?", Titan says "How likely?"
-            
             is_buy = False
             reason = ""
             
             if predicted_pnl > self.CONFIDENCE_THRESHOLD:
-                # Strong signal
                 is_buy = True
-                strength = "HIGH"
-                reason = f"Golden Setup (Exp. Ret: {predicted_pnl*100:.2f}%)"
+                reason = f"Golden Setup (Exp: {predicted_pnl*100:.2f}%)"
             elif predicted_pnl > (self.CONFIDENCE_THRESHOLD * 0.5) and cnn_win_prob > 0.7:
-                # Moderate return but very high certainty
                 is_buy = True
-                strength = "MODERATE"
-                reason = f"High Certainty (Win Prob: {cnn_win_prob:.2f})"
+                reason = f"High Certainty (Win: {cnn_win_prob:.2f})"
             else:
-                reason = f"Weak Signal (Exp: {predicted_pnl*100:.2f}% < {self.CONFIDENCE_THRESHOLD*100:.1f}%)"
+                reason = f"Weak (Exp: {predicted_pnl*100:.2f}%)"
 
             # 5. Build Result
             result = {
-                'confidence': float(cnn_win_prob), # For compatibility with old logic
-                'oracle_score': float(predicted_pnl), # The real juice
+                'confidence': float(cnn_win_prob), 
+                'oracle_score': float(predicted_pnl), 
                 'target_class': "TP2" if predicted_pnl > 0.01 else "TP1",
                 'action_type': 'BUY',
-                'analysis_summary': f"Oracle: {predicted_pnl*100:.2f}% Return | Titan: {cnn_win_prob:.2f} Win"
+                'analysis_summary': f"Oracle: {predicted_pnl*100:.2f}% | Pattern: {cnn_win_prob:.2f}"
             }
             
             if is_buy:
-                result['action'] = 'WATCH' # System will upgrade to BUY after Governance
+                result['action'] = 'WATCH' 
             else:
                 result['action'] = 'WAIT'
                 result['reason'] = reason
@@ -145,6 +152,6 @@ class OracleEngine:
             return result
 
         except Exception as e:
-            print(f"❌ [Oracle] Inference Error: {e}")
-            traceback.print_exc()
-            return {'action': 'WAIT', 'reason': 'Error', 'confidence': 0.0}
+            # print(f"❌ [Oracle] Inference Error: {e}") # Disabled to reduce log spam
+            # traceback.print_exc()
+            return {'action': 'WAIT', 'reason': 'Error', 'oracle_score': 0.0}