fixed eval func

cove9988 · Jan 12, 2025 · 2709024 · 2709024
1 parent 0c3d2ce
commit 2709024
Show file tree

Hide file tree

Showing 3 changed files with 51 additions and 50 deletions.
diff --git a/src/eval.py b/src/eval.py
@@ -4,39 +4,50 @@
 import pandas as pd
 import time
 import datetime
-from stable_baselines3 import PPO
-from src.ppo_model import ForexTradingEnv, load_data
 import logging
-logging.basicConfig(level=logging.CRITICAL, format='%(asctime)s - %(levelname)s - %(message)s')
-logger = logging.getLogger(__name__)
+from stable_baselines3 import PPO
+from src.ppo_model import ForexTradingEnv
+from src.util.read_config import EnvConfig
+from src.util.logger_config import setup_logging
 
-features = ['open', 'high', 'low', 'close', 'minute', 'hour', 'day', 'macd', 'boll_ub', 'boll_lb', 'rsi_30', 'dx_30', 'close_30_sma', 'close_60_sma']
-model_file = '/home/paulg/github/tradesformer/data/model/AUDUSD/weekly/AUDUSD_2024_120.zip'
-# csv_file = "/home/paulg/github/tradesformer/data/split/EURUSD/weekly/EURUSD_2024_103.csv"
-data_directory = "/home/paulg/github/tradesformer/data/split/AUDUSD/weekly"
-csv_files = glob.glob(os.path.join(data_directory, "*.csv"))
-run_time = 10
-_run = 1
-for file in csv_files :
-    if _run > run_time: break
-    # Read the CSV file
-    env = ForexTradingEnv(file,features)
-    model = PPO.load(model_file, env=env)
-# %%
-    observation, info = env.reset()
-    done = False
-    total_buy = 0
-    total_sell = 0
-    totoal_rewards = 0
-    mode = 'graph' #'graph', 'human'
-    while not done:
-        action, _states = model.predict(observation)
-        observation, reward, terminated, truncated, info = env.step(action)
-        done = terminated or truncated
-    env.render(mode = mode)
-    print(f'------rewards:{totoal_rewards}-----buy:{total_buy}--sell:{total_sell}------')
-    _run += 1
+logger = logging.getLogger(__name__)
 
+def eval(data_directory, env_config_file, model_file, asset, run_time = 10, mode = 'graph', sequence_length=24):
+    csv_files = glob.glob(os.path.join(data_directory, "*.csv"))
+    cf = EnvConfig(env_config_file)
+    features = cf.env_parameters("observation_list")
+    print(features)
 
+    _run = 1
+    for file in csv_files :
+        if _run > run_time: break
+        # Read the CSV file
+        env = ForexTradingEnv(file,cf,asset,features,sequence_length)
+        model = PPO.load(model_file, env=env)
+    # %%
+        observation, info = env.reset()
+        done = False
+        total_buy = 0
+        total_sell = 0
+        total_rewards = 0
+        step = 0
+        while not done:
+            action, _states = model.predict(observation)
+            observation, reward, terminated, truncated, info = env.step(action)
+            done = terminated or truncated
+            print(f'step:{step} rwd:{reward} action:{action} ')
+            step += 1
+            total_rewards += reward
+            if action == 1: total_buy += 1
+            if action == 2: total_buy += 1
+        env.render(mode = mode)
+        print(f'------rewards:{total_rewards}-----buy:{total_buy}--sell:{total_sell}------')
+        _run += 1
 
-# %%
+if __name__ == "__main__":
+    asset = "AUDUSD"   
+    env_config_file = '/home/paulg/github/tradesformer/src/configure.json'
+    model_file = '/home/paulg/github/tradesformer/data/model/AUDUSD/weekly/AUDUSD_2024_120.zip'
+    data_directory = "/home/paulg/github/tradesformer/data/split/AUDUSD/weekly"
+    setup_logging(asset=asset, console_level=logging.ERROR, file_level=logging.INFO)
+    eval(data_directory, env_config_file, model_file, asset, run_time= 5, mode = 'graph', sequence_length=24)
diff --git a/src/ppo_model.py b/src/ppo_model.py
@@ -160,6 +160,7 @@ def __init__(self, file, cf, asset, features, sequence_length = 24, logger_show
         self.balance = self.balance_initial
         self.positions = []
         self.ticket_id = 0
+        self.total_rewards = 0
         self.action_aggregator =ActionAggregator()
         # self.reward_calculator = RewardCalculator(
         #     self.data, cf, self.shaping_reward, self.stop_loss, self.profit_taken, self.backward_window
@@ -182,6 +183,7 @@ def reset(self, seed=None, options=None):
         self.balance = self.balance_initial
         self.positions = []
         self.ticket_id = 0
+        self.total_rewards = 0
         self.action_aggregator =ActionAggregator()
 
         # self.current_step = np.random.randint(self.sequence_length, self.max_steps)
@@ -282,10 +284,11 @@ def step(self, action):
                 position_reward, closed = self._calculate_reward(position)
                 if not closed: open_positon += 1
                 reward += position_reward
-
+                self.total_rewards += position_reward
         # Execute action
         _action, stability_reward = self.action_aggregator.add_action(action) 
         reward += stability_reward
+        self.total_rewards += stability_reward
         # logger.info(f"Step:{self.current_step}: action: {action}, real: {_action} stability reward:{stability_reward} ")
         if _action in (1, 2) and open_positon < self.max_current_holding :
             self.ticket_id += 1
@@ -319,7 +322,7 @@ def step(self, action):
             logger.info(f"Step:{self.current_step} Tkt:{position['Ticket']} {position['Type']} Rwd:{position['pips']} SL:{position['SL']} PT:{position['PT']}")
         else:
             reward -= 1 # no open any position, encourage open position
-
+            self.total_rewards -= 1
         # Move to the next time step
         self.current_step += 1
 
@@ -337,7 +340,7 @@ def step(self, action):
                 if position["Type"] == "Buy":
                     buy +=1
 
-            logger.info(f'--- Position:{len(self.positions)}/Buy:{buy} Balance: {self.balance} step {self.current_step }')
+            logger.info(f'--- Position:{len(self.positions)}/Buy:{buy} TtlRwds: {self.total_rewards} Balance: {self.balance} step {self.current_step }')
         # Additional info
         info = {}
         truncated = False
@@ -367,18 +370,3 @@ def render(self, mode='human', title=None, **kwargs):
             p.plot()
 
 
-def eval(model_file,env):
-    # Evaluate the agent
-    model = PPO.load(model_file, env=env)
-    observation, info = env.reset()
-    done = False
-
-    while not done:
-        action, _states = model.predict(observation)
-        observation, reward, terminated, truncated, info = env.step(action)
-        done = terminated
-        env.render()
-
-    # Save the model
-    logger.info("Model saved to 'ppo_forex_transformer'")
-
diff --git a/src/util/logger_config.py b/src/util/logger_config.py
@@ -24,4 +24,6 @@ def setup_logging(asset, console_level=logging.INFO, file_level=logging.DEBUG):
     logger.addHandler(file_handler)
 
     # Optional: Avoid duplicate logs in libraries by disabling propagation
-    logging.getLogger("matplotlib").setLevel(logging.WARNING)  # Example: Suppress matplotlib logs
+    logging.getLogger("matplotlib").setLevel(logging.ERROR)  # Suppress matplotlib logs
+    logging.getLogger("torch").setLevel(logging.ERROR)  
+    logging.getLogger("mplfinance").setLevel(logging.ERROR)