dash and training wip

core/realtime_rl_cob_trader.py

@@ -731,7 +731,8 @@ class RealtimeRLCOBTrader:
             with self.training_lock:
                 # Check if we have enough data for training
                 predictions = list(self.prediction_history[symbol])
-                if len(predictions) < 10:
+                # Train with fewer samples to kickstart learning
+                if len(predictions) < 6:
                     return
                 
                 # Calculate rewards for recent predictions
@@ -739,11 +740,11 @@ class RealtimeRLCOBTrader:
                 
                 # Filter predictions with calculated rewards
                 training_predictions = [p for p in predictions if p.reward is not None]
-                if len(training_predictions) < 5:
+                if len(training_predictions) < 3:
                     return
                 
                 # Prepare training batch
-                batch_size = min(32, len(training_predictions))
+                batch_size = min(16, len(training_predictions))
                 batch_predictions = training_predictions[-batch_size:]
                 
                 # Train model