better sticks

train_rl_with_realtime.py

@@ -107,6 +107,12 @@ class RLTrainingIntegrator:
         self.trade_count = 0
         self.win_count = 0
         
+        # Add session-wide PnL tracking
+        self.session_pnl = 0.0
+        self.session_trades = 0
+        self.session_wins = 0
+        self.session_balance = 100.0  # Start with $100 balance
+        
         # Track current position state
         self.in_position = False
         self.entry_price = None
@@ -151,7 +157,28 @@ class RLTrainingIntegrator:
                     # Calculate profit if we have entry data
                     pnl = None
                     if self.entry_price is not None:
-                        pnl = (price - self.entry_price) / self.entry_price
+                        # Calculate percentage change
+                        pnl_pct = (price - self.entry_price) / self.entry_price
+                        
+                        # Cap extreme PnL values to more realistic levels (-90% to +100%)
+                        pnl_pct = max(min(pnl_pct, 1.0), -0.9)
+                        
+                        # Apply to current balance
+                        trade_amount = self.session_balance * 0.1  # Use 10% of balance per trade
+                        trade_profit = trade_amount * pnl_pct
+                        self.session_balance += trade_profit
+                        
+                        # Ensure session balance doesn't go below $1
+                        self.session_balance = max(self.session_balance, 1.0)
+                        
+                        # For normalized display in charts and logs
+                        pnl = pnl_pct
+                        
+                        # Update session-wide PnL
+                        self.session_pnl += pnl
+                        self.session_trades += 1
+                        if pnl > 0:
+                            self.session_wins += 1
                     
                     # Log the complete trade on the chart
                     if self.chart:
@@ -162,10 +189,12 @@ class RLTrainingIntegrator:
                         # Record the trade with PnL
                         if hasattr(self.chart, 'add_trade'):
                             self.chart.add_trade(
-                                action=action_str,
                                 price=price,
                                 timestamp=timestamp,
-                                pnl=pnl
+                                pnl=pnl,
+                                amount=0.1,
+                                action=action_str,
+                                type=action_str  # Add explicit type
                             )
                     
                     # Update trade counts
@@ -185,6 +214,21 @@ class RLTrainingIntegrator:
                     'reward': reward,
                     'timestamp': timestamp.isoformat()
                 })
+            else:
+                # Hold action
+                action_str = "HOLD"
+                timestamp = datetime.now()
+            
+            # Update chart trading info
+            if self.chart and hasattr(self.chart, 'update_trading_info'):
+                # Determine current position size (0.1 if in position, 0 if not)
+                position_size = 0.1 if self.in_position else 0.0
+                self.chart.update_trading_info(
+                    signal=action_str,
+                    position=position_size,
+                    balance=self.session_balance,
+                    pnl=self.session_pnl
+                )
             
             # Track reward for all actions (including hold)
             self.reward_history.append(reward)
@@ -205,6 +249,23 @@ class RLTrainingIntegrator:
             logger.info(f"  Win rate: {info['win_rate']:.4f}")
             logger.info(f"  Trades: {info['trades']}")
             
+            # Log session-wide PnL
+            session_win_rate = self.session_wins / self.session_trades if self.session_trades > 0 else 0
+            logger.info(f"  Session Balance: ${self.session_balance:.2f}")
+            logger.info(f"  Session Total PnL: {self.session_pnl:.4f}")
+            logger.info(f"  Session Win Rate: {session_win_rate:.4f}")
+            logger.info(f"  Session Trades: {self.session_trades}")
+            
+            # Update chart trading info with final episode information
+            if self.chart and hasattr(self.chart, 'update_trading_info'):
+                # Reset position since we're between episodes
+                self.chart.update_trading_info(
+                    signal="HOLD",
+                    position=0.0,
+                    balance=self.session_balance,
+                    pnl=self.session_pnl
+                )
+            
             # Reset position state for new episode
             self.in_position = False
             self.entry_price = None
@@ -441,8 +502,8 @@ async def start_realtime_chart(symbol="BTC/USDT", port=8050):
     
     try:
         logger.info(f"Initializing RealTimeChart for {symbol}")
-        # Create the chart
+        # Create the chart with sample data enabled and no-ticks warnings disabled
-        chart = RealTimeChart(symbol)
+        chart = RealTimeChart(symbol, use_sample_data=True, log_no_ticks_warning=False)
         
         # Start the WebSocket connection in a separate thread
         # The _start_websocket_thread method already handles this correctly
@@ -466,10 +527,18 @@ async def start_realtime_chart(symbol="BTC/USDT", port=8050):
         logger.error(traceback.format_exc())
         raise
 
-def run_training_thread(chart, num_episodes=5000, max_steps=2000):
+def run_training_thread(chart):
-    """Run the RL training in a separate thread"""
+    """Start the RL training in a separate thread"""
     integrator = RLTrainingIntegrator(chart)
-    thread = Thread(target=lambda: integrator.start_training(num_episodes, max_steps))
+    
+    def training_thread_func():
+        try:
+            # Use a small number of episodes to test termination handling
+            integrator.start_training(num_episodes=2, max_steps=500)
+        except Exception as e:
+            logger.error(f"Error in training thread: {str(e)}")
+    
+    thread = threading.Thread(target=training_thread_func)
     thread.daemon = True
     thread.start()
     logger.info("Started RL training thread")
@@ -490,8 +559,17 @@ def test_signals(chart):
     # Add a test SELL signal
     chart.add_nn_signal("SELL", datetime.now(), 0.85)
     
-    # Add a test trade
+    # Add a test trade if the method exists
-    chart.add_trade("BUY", 50000.0, datetime.now(), 0.05)
+    if hasattr(chart, 'add_trade'):
+        chart.add_trade(
+            price=83000.0, 
+            timestamp=datetime.now(), 
+            pnl=0.05, 
+            action="BUY",
+            type="BUY"  # Add explicit type
+        )
+    else:
+        logger.warning("RealTimeChart has no add_trade method - skipping test trade")
 
 async def main():
     """Main function that coordinates the realtime chart and RL training"""
@@ -520,6 +598,18 @@ async def main():
     except Exception as e:
         logger.error(f"Unexpected error: {str(e)}")
     finally:
+        # Log final PnL summary
+        if hasattr(integrator, 'session_pnl'):
+            session_win_rate = integrator.session_wins / integrator.session_trades if integrator.session_trades > 0 else 0
+            logger.info("=" * 50)
+            logger.info("FINAL SESSION SUMMARY")
+            logger.info("=" * 50)
+            logger.info(f"Final Session Balance: ${integrator.session_balance:.2f}")
+            logger.info(f"Total Session PnL: {integrator.session_pnl:.4f}")
+            logger.info(f"Total Session Win Rate: {session_win_rate:.4f} ({integrator.session_wins}/{integrator.session_trades})")
+            logger.info(f"Total Session Trades: {integrator.session_trades}")
+            logger.info("=" * 50)
+        
         # Clean up
         if realtime_websocket_task:
             realtime_websocket_task.cancel()