fix T training memory usage (due for more improvement)

core/orchestrator.py

@@ -836,22 +836,10 @@ class TradingOrchestrator:
             try:
                 from NN.models.dqn_agent import DQNAgent
 
-                # Determine actual state size from BaseDataInput
-                try:
-                    base_data = self.data_provider.build_base_data_input(self.symbol)
-                    if base_data:
-                        actual_state_size = len(base_data.get_feature_vector())
-                        logger.info(f"Detected actual state size: {actual_state_size}")
-                    else:
-                        actual_state_size = 7850  # Fallback based on error message
-                        logger.warning(
-                            f"Could not determine state size, using fallback: {actual_state_size}"
-                        )
-                except Exception as e:
-                    actual_state_size = 7850  # Fallback based on error message
-                    logger.warning(
-                        f"Error determining state size: {e}, using fallback: {actual_state_size}"
-                    )
+                # Use known state size instead of building data (which triggers massive API calls)
+                # The state size is determined by BaseDataInput structure and doesn't change
+                actual_state_size = 7850  # Known size from BaseDataInput.get_feature_vector()
+                logger.info(f"Using known state size: {actual_state_size}")
 
                 action_size = self.config.rl.get("action_space", 3)
                 self.rl_agent = DQNAgent(