more elaborate RL training

core/extrema_trainer.py

@@ -331,8 +331,39 @@ class ExtremaTrainer:
             
             # Get all available price data for better extrema detection
             all_candles = list(self.context_data[symbol].candles)
-            prices = [candle['close'] for candle in all_candles]
-            timestamps = [candle['timestamp'] for candle in all_candles]
+            prices = []
+            timestamps = []
+
+            for i, candle in enumerate(all_candles):
+                # Handle different candle formats
+                if isinstance(candle, dict):
+                    if 'close' in candle:
+                        prices.append(candle['close'])
+                    else:
+                        # Fallback to other price fields
+                        price = candle.get('price') or candle.get('high') or candle.get('low') or candle.get('open') or 0
+                        prices.append(price)
+
+                    # Handle timestamp with fallbacks
+                    if 'timestamp' in candle:
+                        timestamps.append(candle['timestamp'])
+                    elif 'time' in candle:
+                        timestamps.append(candle['time'])
+                    else:
+                        # Generate timestamp based on index if none available
+                        timestamps.append(datetime.now() - timedelta(minutes=len(all_candles) - i))
+                else:
+                    # Handle non-dict candle formats (e.g., tuples, lists)
+                    if hasattr(candle, '__getitem__'):
+                        prices.append(float(candle[3]))  # Assume OHLC format: [O, H, L, C]
+                        timestamps.append(datetime.now() - timedelta(minutes=len(all_candles) - i))
+                    else:
+                        # Skip invalid candle data
+                        continue
+
+            # Ensure we have enough data
+            if len(prices) < self.window_size * 3:
+                return detected
             
             # Use a more sophisticated extrema detection algorithm
             window = self.window_size