Merge WeightedUnrealisedProfit

.gitignore

@@ -8,3 +8,4 @@ data/postgres/*
 data/log/*
 data/reports/*
 *.pkl
+venv/*

lib/RLTrader.py

@@ -14,7 +14,7 @@
 from stable_baselines import PPO2
 
 from lib.env.TradingEnv import TradingEnv
-from lib.env.reward import BaseRewardStrategy, IncrementalProfit
+from lib.env.reward import BaseRewardStrategy, IncrementalProfit, WeightedUnrealisedProfit
 from lib.data.providers.dates import ProviderDateFormat
 from lib.data.providers import BaseDataProvider,  StaticDataProvider, ExchangeDataProvider
 from lib.util.logger import init_logger

lib/env/TradingEnv.py

@@ -7,7 +7,7 @@
 from typing import List, Dict
 
 from lib.env.render import TradingChart
-from lib.env.reward import BaseRewardStrategy, IncrementalProfit
+from lib.env.reward import BaseRewardStrategy, IncrementalProfit, WeightedUnrealisedProfit
 from lib.env.trade import BaseTradeStrategy, SimulatedTradeStrategy
 from lib.data.providers import BaseDataProvider
 from lib.data.features.transform import max_min_normalize, mean_normalize, log_and_difference, difference
@@ -109,6 +109,7 @@ def _take_action(self, action: int):
         elif asset_sold:
             self.asset_held -= asset_sold
             self.balance += sale_revenue
+            self.reward_strategy.reset_reward()
 
             self.trades.append({'step': self.current_step, 'amount': asset_sold,
                                 'total': sale_revenue, 'type': 'sell'})
@@ -191,6 +192,8 @@ def reset(self):
         self.asset_held = 0
         self.current_step = 0
 
+        self.reward_strategy.reset_reward()
+
         self.account_history = pd.DataFrame([{
             'balance': self.balance,
             'asset_bought': 0,

lib/env/reward/BaseRewardStrategy.py

@@ -9,6 +9,10 @@ class BaseRewardStrategy(object, metaclass=ABCMeta):
     def __init__(self):
         pass
 
+    @abstractmethod
+    def reset_reward(self):
+        raise NotImplementedError()
+
     @abstractmethod
     def get_reward(self,
                    current_step: int,

lib/env/reward/IncrementalProfit.py

@@ -12,12 +12,15 @@ class IncrementalProfit(BaseRewardStrategy):
     def __init__(self):
         pass
 
+    def reset_reward(self):
+        pass
+
     def get_reward(self,
                    current_step: int,
                    current_price: Callable[[str], float],
                    observations: pd.DataFrame,
                    account_history: pd.DataFrame,
-                   net_worths: List[float]):
+                   net_worths: List[float]) -> float:
         reward = 0
 
         curr_balance = account_history['balance'].values[-1]

lib/env/reward/WeightedUnrealisedProfit.py

@@ -0,0 +1,41 @@
+from collections import deque
+
+import pandas as pd
+import numpy as np
+from typing import List, Callable
+
+from lib.env.reward.BaseRewardStrategy import BaseRewardStrategy
+
+
+class WeightedUnrealisedProfit(BaseRewardStrategy):
+    def __init__(self, **kwargs):
+        self.decay_rate = kwargs.get('decay_rate', 1e-2)
+        self.decay_denominator = np.exp(-1 * self.decay_rate)
+
+        self.reset_reward()
+
+    def reset_reward(self):
+        self.rewards = deque(np.zeros(1, dtype=float))
+        self.sum = 0.0
+
+    def calc_reward(self, reward):
+        self.sum = self.sum - self.decay_denominator * self.rewards.popleft()
+        self.sum = self.sum * self.decay_denominator
+        self.sum = self.sum + reward
+
+        self.rewards.append(reward)
+
+        return self.sum / self.decay_denominator
+
+    def get_reward(self,
+                   current_step: int,
+                   current_price: Callable[[str], float],
+                   observations: pd.DataFrame,
+                   account_history: pd.DataFrame,
+                   net_worths: List[float]) -> float:
+        if account_history['asset_sold'].values[-1] > 0:
+            reward = self.calc_reward(account_history['sale_revenue'].values[-1])
+        else:
+            reward = self.calc_reward(account_history['asset_held'].values[-1] * current_price)
+
+        return reward

lib/env/reward/__init__.py

@@ -1,2 +1,3 @@
 from lib.env.reward.IncrementalProfit import IncrementalProfit
+from lib.env.reward.WeightedUnrealisedProfit import WeightedUnrealisedProfit
 from lib.env.reward.BaseRewardStrategy import BaseRewardStrategy