improve default reward, fix bugs in environment

freqtrade/freqai/RL/Base5ActionRLEnv.py

@@ -140,30 +140,32 @@ class Base5ActionRLEnv(gym.Env):
             if action == Actions.Neutral.value:
                 self._position = Positions.Neutral
                 trade_type = "neutral"
+                self._last_trade_tick = None
             elif action == Actions.Long_enter.value:
                 self._position = Positions.Long
                 trade_type = "long"
+                self._last_trade_tick = self._current_tick
             elif action == Actions.Short_enter.value:
                 self._position = Positions.Short
                 trade_type = "short"
+                self._last_trade_tick = self._current_tick
             elif action == Actions.Long_exit.value:
                 self._position = Positions.Neutral
                 trade_type = "neutral"
+                self._last_trade_tick = None
             elif action == Actions.Short_exit.value:
                 self._position = Positions.Neutral
                 trade_type = "neutral"
+                self._last_trade_tick = None
             else:
                 print("case not defined")
 
-            # Update last trade tick
-            self._last_trade_tick = self._current_tick
-
             if trade_type is not None:
                 self.trade_history.append(
                     {'price': self.current_price(), 'index': self._current_tick,
                      'type': trade_type})
 
-        if self._total_profit < 0.2:
+        if self._total_profit < 0.5:
             self._done = True
 
         self._position_history.append(self._position)
@@ -221,8 +223,7 @@ class Base5ActionRLEnv(gym.Env):
     def is_tradesignal(self, action: int):
         # trade signal
         """
-        not trade signal is :
+        Determine if the signal is a trade signal
-        Determine if the signal is non sensical
         e.g.: agent wants a Actions.Long_exit while it is in a Positions.short
         """
         return not ((action == Actions.Neutral.value and self._position == Positions.Neutral) or
@@ -237,6 +238,24 @@ class Base5ActionRLEnv(gym.Env):
                     (action == Actions.Long_exit.value and self._position == Positions.Short) or
                     (action == Actions.Long_exit.value and self._position == Positions.Neutral))
 
+    def _is_valid(self, action: int):
+        # trade signal
+        """
+        Determine if the signal is valid.
+        e.g.: agent wants a Actions.Long_exit while it is in a Positions.short
+        """
+        # Agent should only try to exit if it is in position
+        if action in (Actions.Short_exit.value, Actions.Long_exit.value):
+            if self._position not in (Positions.Short, Positions.Long):
+                return False
+
+        # Agent should only try to enter if it is not in position
+        if action in (Actions.Short_enter.value, Actions.Long_enter.value):
+            if self._position != Positions.Neutral:
+                return False
+
+        return True
+
     def _is_trade(self, action: Actions):
         return ((action == Actions.Long_enter.value and self._position == Positions.Neutral) or
                 (action == Actions.Short_enter.value and self._position == Positions.Neutral))
@@ -278,13 +297,8 @@ class Base5ActionRLEnv(gym.Env):
         if self._is_trade(action) or self._done:
             pnl = self.get_unrealized_profit()
 
-            if self._position == Positions.Long:
+            if self._position in (Positions.Long, Positions.Short):
-                self._total_profit = self._total_profit + self._total_profit * pnl
+                self._total_profit *= (1 + pnl)
-                self._profits.append((self._current_tick, self._total_profit))
-                self.close_trade_profit.append(pnl)
-
-            if self._position == Positions.Short:
-                self._total_profit = self._total_profit + self._total_profit * pnl
                 self._profits.append((self._current_tick, self._total_profit))
                 self.close_trade_profit.append(pnl)
 

freqtrade/freqai/RL/BaseReinforcementLearningModel.py

@@ -19,7 +19,6 @@ from typing import Callable
 from datetime import datetime, timezone
 from stable_baselines3.common.utils import set_random_seed
 import gym
-from pathlib import Path
 logger = logging.getLogger(__name__)
 
 torch.multiprocessing.set_sharing_strategy('file_system')
@@ -112,8 +111,6 @@ class BaseReinforcementLearningModel(IFreqaiModel):
         test_df = data_dictionary["test_features"]
         eval_freq = self.freqai_info["rl_config"]["eval_cycles"] * len(test_df)
 
-        # environments
-        if not self.train_env:
         self.train_env = MyRLEnv(df=train_df, prices=prices_train, window_size=self.CONV_WIDTH,
                                  reward_kwargs=self.reward_params, config=self.config)
         self.eval_env = Monitor(MyRLEnv(df=test_df, prices=prices_test,
@@ -122,17 +119,6 @@ class BaseReinforcementLearningModel(IFreqaiModel):
         self.eval_callback = EvalCallback(self.eval_env, deterministic=True,
                                           render=False, eval_freq=eval_freq,
                                           best_model_save_path=str(dk.data_path))
-        else:
-            self.train_env.reset()
-            self.eval_env.reset()
-            self.train_env.reset_env(train_df, prices_train, self.CONV_WIDTH, self.reward_params)
-            self.eval_env.reset_env(test_df, prices_test, self.CONV_WIDTH, self.reward_params)
-            # self.eval_callback.eval_env = self.eval_env
-            # self.eval_callback.best_model_save_path = str(dk.data_path)
-            # self.eval_callback._init_callback()
-            self.eval_callback.__init__(self.eval_env, deterministic=True,
-                                        render=False, eval_freq=eval_freq,
-                                        best_model_save_path=str(dk.data_path))
 
     @abstractmethod
     def fit_rl(self, data_dictionary: Dict[str, Any], dk: FreqaiDataKitchen):
@@ -284,30 +270,43 @@ class MyRLEnv(Base5ActionRLEnv):
 
     def calculate_reward(self, action):
 
-        if self._last_trade_tick is None:
+        # first, penalize if the action is not valid
-            return 0.
+        if not self._is_valid(action):
+            return -15
 
         pnl = self.get_unrealized_profit()
-        max_trade_duration = self.rl_config.get('max_trade_duration_candles', 100)
+        rew = np.sign(pnl) * (pnl + 1)
+        factor = 100
+
+        # reward agent for entering trades
+        if action in (Actions.Long_enter.value, Actions.Short_enter.value):
+            return 25
+        # discourage agent from not entering trades
+        if action == Actions.Neutral.value and self._position == Positions.Neutral:
+            return -15
+
+        max_trade_duration = self.rl_config.get('max_trade_duration_candles', 300)
         trade_duration = self._current_tick - self._last_trade_tick
 
-        factor = 1
         if trade_duration <= max_trade_duration:
             factor *= 1.5
         elif trade_duration > max_trade_duration:
             factor *= 0.5
 
+        # discourage sitting in position
+        if self._position in (Positions.Short, Positions.Long):
+            return -50 * trade_duration / max_trade_duration
+
         # close long
         if action == Actions.Long_exit.value and self._position == Positions.Long:
-            if self.close_trade_profit and self.close_trade_profit[-1] > self.profit_aim * self.rr:
+            if pnl > self.profit_aim * self.rr:
                 factor *= self.rl_config['model_reward_parameters'].get('win_reward_factor', 2)
-            return float(pnl * factor)
+            return float(rew * factor)
 
         # close short
         if action == Actions.Short_exit.value and self._position == Positions.Short:
-            factor = 1
+            if pnl > self.profit_aim * self.rr:
-            if self.close_trade_profit and self.close_trade_profit[-1] > self.profit_aim * self.rr:
                 factor *= self.rl_config['model_reward_parameters'].get('win_reward_factor', 2)
-            return float(pnl * factor)
+            return float(rew * factor)
 
         return 0.

freqtrade/freqai/prediction_models/ReinforcementLearner.py

@@ -6,6 +6,10 @@ from freqtrade.freqai.data_kitchen import FreqaiDataKitchen
 from freqtrade.freqai.RL.Base5ActionRLEnv import Actions, Base5ActionRLEnv, Positions
 from freqtrade.freqai.RL.BaseReinforcementLearningModel import BaseReinforcementLearningModel
 from pathlib import Path
+from pandas import DataFrame
+from stable_baselines3.common.callbacks import EvalCallback
+from stable_baselines3.common.monitor import Monitor
+import numpy as np
 
 logger = logging.getLogger(__name__)
 
@@ -49,6 +53,25 @@ class ReinforcementLearner(BaseReinforcementLearningModel):
 
         return model
 
+    def set_train_and_eval_environments(self, data_dictionary: Dict[str, DataFrame],
+                                        prices_train: DataFrame, prices_test: DataFrame,
+                                        dk: FreqaiDataKitchen):
+        """
+        User can override this if they are using a custom MyRLEnv
+        """
+        train_df = data_dictionary["train_features"]
+        test_df = data_dictionary["test_features"]
+        eval_freq = self.freqai_info["rl_config"]["eval_cycles"] * len(test_df)
+
+        self.train_env = MyRLEnv(df=train_df, prices=prices_train, window_size=self.CONV_WIDTH,
+                                 reward_kwargs=self.reward_params, config=self.config)
+        self.eval_env = Monitor(MyRLEnv(df=test_df, prices=prices_test,
+                                window_size=self.CONV_WIDTH,
+                                reward_kwargs=self.reward_params, config=self.config))
+        self.eval_callback = EvalCallback(self.eval_env, deterministic=True,
+                                          render=False, eval_freq=eval_freq,
+                                          best_model_save_path=str(dk.data_path))
+
 
 class MyRLEnv(Base5ActionRLEnv):
     """
@@ -58,30 +81,43 @@ class MyRLEnv(Base5ActionRLEnv):
 
     def calculate_reward(self, action):
 
-        if self._last_trade_tick is None:
+        # first, penalize if the action is not valid
-            return 0.
+        if not self._is_valid(action):
+            return -15
 
         pnl = self.get_unrealized_profit()
-        max_trade_duration = self.rl_config.get('max_trade_duration_candles', 100)
+        rew = np.sign(pnl) * (pnl + 1)
+        factor = 100
+
+        # reward agent for entering trades
+        if action in (Actions.Long_enter.value, Actions.Short_enter.value):
+            return 25
+        # discourage agent from not entering trades
+        if action == Actions.Neutral.value and self._position == Positions.Neutral:
+            return -15
+
+        max_trade_duration = self.rl_config.get('max_trade_duration_candles', 300)
         trade_duration = self._current_tick - self._last_trade_tick
 
-        factor = 1
         if trade_duration <= max_trade_duration:
             factor *= 1.5
         elif trade_duration > max_trade_duration:
             factor *= 0.5
 
+        # discourage sitting in position
+        if self._position in (Positions.Short, Positions.Long):
+            return -50 * trade_duration / max_trade_duration
+
         # close long
         if action == Actions.Long_exit.value and self._position == Positions.Long:
-            if self.close_trade_profit and self.close_trade_profit[-1] > self.profit_aim * self.rr:
+            if pnl > self.profit_aim * self.rr:
                 factor *= self.rl_config['model_reward_parameters'].get('win_reward_factor', 2)
-            return float(pnl * factor)
+            return float(rew * factor)
 
         # close short
         if action == Actions.Short_exit.value and self._position == Positions.Short:
-            factor = 1
+            if pnl > self.profit_aim * self.rr:
-            if self.close_trade_profit and self.close_trade_profit[-1] > self.profit_aim * self.rr:
                 factor *= self.rl_config['model_reward_parameters'].get('win_reward_factor', 2)
-            return float(pnl * factor)
+            return float(rew * factor)
 
         return 0.