FIRST WORKING CASE

.gitignore

@@ -33,3 +33,5 @@ crypto/brian/models/best/*
 crypto/brian/models/last/*
 crypto/brian/live_chart.html
 crypto/gogo2/models/*
+crypto/gogo2/trading_bot.log
+*.log

crypto/gogo2/main.py

@@ -21,6 +21,7 @@ import torch.cuda.amp as amp  # Add this import at the top
 from sklearn.preprocessing import MinMaxScaler
 import copy
 import argparse
+import traceback
 
 # Configure logging
 logging.basicConfig(
@@ -428,30 +429,26 @@ class TradingEnvironment:
             return False
     
     def step(self, action):
-        """Take an action in the environment"""
+        """Take an action in the environment and return the next state, reward, and done flag"""
-        # Store previous balance for reward calculation
+        # Store current price before taking action
-        prev_balance = self.balance
-        
-        # Update current price
-        if self.current_step < len(self.data) - 1:
-            self.current_step += 1
         self.current_price = self.data[self.current_step]['close']
-        else:
-            # End of data
-            return self.get_state(), 0, True
         
-        # Check for stop loss or take profit
+        # Process action (0: HOLD, 1: BUY/LONG, 2: SELL/SHORT, 3: CLOSE)
-        self._check_sl_tp()
-        
-        # Calculate reward based on action
         reward = self.calculate_reward(action)
         
-        # Check if we've reached the end of the data
+        # Check for stop loss / take profit hits
+        self.check_sl_tp()
+        
+        # Move to next step
+        self.current_step += 1
         done = self.current_step >= len(self.data) - 1
         
-        return self.get_state(), reward, done
+        # Get new state
+        next_state = self.get_state()
         
-    def _check_sl_tp(self):
+        return next_state, reward, done
+    
+    def check_sl_tp(self):
         """Check if stop loss or take profit has been hit"""
         if self.position == 'flat':
             return
@@ -686,9 +683,6 @@ class TradingEnvironment:
         if hasattr(self, 'predicted_prices') and len(self.predicted_prices) > 0:
             # Normalize predictions relative to current price
             pred_norm = np.array(self.predicted_prices[:3]) / latest_price - 1.0
-            # Pad if needed
-            if len(pred_norm) < 3:
-                pred_norm = np.pad(pred_norm, (0, 3 - len(pred_norm)), 'constant')
             state_components.append(pred_norm)
         else:
             # Add zeros if no predictions
@@ -1065,6 +1059,97 @@ class TradingEnvironment:
         
         return analysis
     
+    def initialize_price_predictor(self, device="cpu"):
+        """Initialize the price prediction model"""
+        self.price_predictor = PricePredictionModel(input_size=30, hidden_size=128, output_size=5)
+        self.price_predictor.to(device)
+        self.price_predictor_optimizer = optim.Adam(self.price_predictor.parameters(), lr=1e-3)
+        self.predicted_prices = np.array([])
+    
+    def train_price_predictor(self):
+        """Train the price prediction model on recent data"""
+        if len(self.features['price']) < 35:
+            return 0.0
+        
+        # Get price history
+        price_history = self.features['price']
+        
+        # Train the model
+        loss = self.price_predictor.train_on_new_data(
+            price_history, 
+            self.price_predictor_optimizer,
+            epochs=5
+        )
+        
+        return loss
+    
+    def update_price_predictions(self):
+        """Update price predictions"""
+        if len(self.features['price']) < 30:
+            self.predicted_prices = np.array([])
+            return
+        
+        # Get price history
+        price_history = self.features['price']
+        
+        # Get predictions
+        self.predicted_prices = self.price_predictor.predict_next_candles(price_history, num_candles=5)
+
+    def identify_optimal_trades(self):
+        """Identify optimal entry and exit points based on local extrema"""
+        if len(self.features['price']) < 20:
+            return
+        
+        # Find local bottoms and tops
+        bottoms, tops = find_local_extrema(self.features['price'], window=5)
+        
+        # Store optimal trade points
+        self.optimal_bottoms = bottoms  # Buy points
+        self.optimal_tops = tops  # Sell points
+        
+        # Create optimal trade signals
+        self.optimal_signals = np.zeros(len(self.features['price']))
+        for i in bottoms:
+            if 0 <= i < len(self.optimal_signals):  # Ensure index is valid
+                self.optimal_signals[i] = 1  # Buy signal
+        for i in tops:
+            if 0 <= i < len(self.optimal_signals):  # Ensure index is valid
+                self.optimal_signals[i] = -1  # Sell signal
+        
+        logger.info(f"Identified {len(bottoms)} optimal buy points and {len(tops)} optimal sell points")
+
+    def calculate_position_size(self):
+        """Calculate position size based on current balance and risk parameters"""
+        # Use a fixed percentage of balance for each trade
+        risk_percent = 5.0  # Risk 5% of balance per trade
+        
+        # Calculate position size with leverage
+        position_size = self.balance * (risk_percent / 100) * MAX_LEVERAGE
+        
+        # Apply a safety factor to avoid liquidation
+        safety_factor = 0.8
+        position_size *= safety_factor
+        
+        # Ensure minimum position size
+        min_position = 10.0  # Minimum position size in USD
+        position_size = max(position_size, min(min_position, self.balance * 0.5))
+        
+        # Ensure position size doesn't exceed balance * leverage
+        max_position = self.balance * MAX_LEVERAGE
+        position_size = min(position_size, max_position)
+        
+        return position_size
+
+    def calculate_fees(self, position_size):
+        """Calculate trading fees for a given position size"""
+        # Typical fee rate for crypto exchanges (0.1%)
+        fee_rate = 0.001
+        
+        # Calculate fee
+        fee = position_size * fee_rate
+        
+        return fee
+
 # Ensure GPU usage if available
 def get_device():
     """Get the best available device (CUDA GPU or CPU)"""
@@ -1177,33 +1262,28 @@ class Agent:
             return random.randrange(self.action_size)
     
     def learn(self):
-        """Learn from experience replay with mixed precision"""
+        """Learn from a batch of experiences"""
         if len(self.memory) < BATCH_SIZE:
             return None
         
         try:
-            # Sample batch from memory
+            # Sample a batch of experiences
             experiences = self.memory.sample(BATCH_SIZE)
             
-            # Check if any experience has None values
+            # Convert experiences to tensors
-            for exp in experiences:
+            states = torch.FloatTensor([e.state for e in experiences]).to(self.device)
-                if exp.state is None or exp.next_state is None:
+            actions = torch.LongTensor([e.action for e in experiences]).to(self.device)
-                    return None
+            rewards = torch.FloatTensor([e.reward for e in experiences]).to(self.device)
-            
+            next_states = torch.FloatTensor([e.next_state for e in experiences]).to(self.device)
-            # Convert to tensors
+            dones = torch.FloatTensor([e.done for e in experiences]).to(self.device)
-            states = torch.FloatTensor([exp.state for exp in experiences]).to(self.device)
-            actions = torch.LongTensor([exp.action for exp in experiences]).unsqueeze(1).to(self.device)
-            rewards = torch.FloatTensor([exp.reward for exp in experiences]).to(self.device)
-            next_states = torch.FloatTensor([exp.next_state for exp in experiences]).to(self.device)
-            dones = torch.FloatTensor([exp.done for exp in experiences]).to(self.device)
             
             # Use mixed precision for forward/backward passes
             if self.device.type == "cuda":
                 with amp.autocast():
                     # Compute Q values
-                    current_q_values = self.policy_net(states).gather(1, actions)
+                    current_q_values = self.policy_net(states).gather(1, actions.unsqueeze(1))
                     
-                    # Compute next state values using target network
+                    # Compute next Q values with target network
                     with torch.no_grad():
                         next_q_values = self.target_net(next_states).max(1)[0]
                         target_q_values = rewards + (GAMMA * next_q_values * (1 - dones))
@@ -1214,21 +1294,25 @@ class Agent:
                     # Compute loss
                     loss = F.smooth_l1_loss(current_q_values, target_q_values)
                     
-                # Optimize with gradient scaling
+                # Backward pass with mixed precision
                 self.optimizer.zero_grad()
                 self.scaler.scale(loss).backward()
+                
+                # Gradient clipping to prevent exploding gradients
                 self.scaler.unscale_(self.optimizer)
-                torch.nn.utils.clip_grad_norm_(self.policy_net.parameters(), 1.0)
+                torch.nn.utils.clip_grad_norm_(self.policy_net.parameters(), max_norm=1.0)
+                
                 self.scaler.step(self.optimizer)
                 self.scaler.update()
             else:
-                # Standard precision training
+                # Standard precision for CPU
                 # Compute Q values
-                current_q_values = self.policy_net(states).gather(1, actions)
+                current_q_values = self.policy_net(states).gather(1, actions.unsqueeze(1))
                 
-                # Compute next state values using target network
+                # Compute next Q values with target network
                 with torch.no_grad():
                     next_q_values = self.target_net(next_states).max(1)[0]
+                    target_q_values = rewards + (GAMMA * next_q_values * (1 - dones))
                 
                 # Reshape target values to match current_q_values
                 target_q_values = target_q_values.unsqueeze(1)
@@ -1236,18 +1320,27 @@ class Agent:
                 # Compute loss
                 loss = F.smooth_l1_loss(current_q_values, target_q_values)
                 
-                # Optimize the model
+                # Backward pass
                 self.optimizer.zero_grad()
                 loss.backward()
-                torch.nn.utils.clip_grad_norm_(self.policy_net.parameters(), 1.0)
+                
+                # Gradient clipping to prevent exploding gradients
+                torch.nn.utils.clip_grad_norm_(self.policy_net.parameters(), max_norm=1.0)
+                
                 self.optimizer.step()
             
+            # Update steps done
+            self.steps_done += 1
+            
+            # Update target network
+            if self.steps_done % TARGET_UPDATE == 0:
+                self.target_net.load_state_dict(self.policy_net.state_dict())
+                
             return loss.item()
             
         except Exception as e:
             logger.error(f"Error during learning: {e}")
-            import traceback
+            logger.error(f"Traceback: {traceback.format_exc()}")
-            logger.error(traceback.format_exc())
             return None
     
     def update_target_network(self):

crypto/gogo2/training_stats.csv

@@ -0,0 +1 @@
+episode_rewards,episode_lengths,balances,win_rates,episode_pnls,cumulative_pnl,drawdowns,prediction_accuracy