Portfolio rl strategy (#758)

Author: maurerle

assume/reinforcement_learning/learning_utils.py

@@ -336,3 +336,32 @@ def transfer_weights(
             )
 
     return new_state_copy
+
+
+def encode_time_features(start: datetime) -> list:
+    """
+    Encode time features for a given datetime object.
+    This function extracts various time-related features from the datetime object
+    and encodes them using sine and cosine transformations to capture periodicity.
+
+    Args:
+        start (datetime): The datetime object to encode.
+
+    Returns:
+        list: A list containing the encoded time features.
+    """
+
+    start_hour = start.hour / 24.0
+    start_hour_cos = np.cos(2 * np.pi * start_hour)
+    start_hour_sin = np.sin(2 * np.pi * start_hour)
+
+    start_month = start.month / 12.0
+    start_month_cos = np.cos(2 * np.pi * start_month)
+    start_month_sin = np.sin(2 * np.pi * start_month)
+
+    return (
+        start_hour_cos,
+        start_hour_sin,
+        start_month_cos,
+        start_month_sin,
+    )

assume/strategies/__init__.py

@@ -120,6 +120,10 @@
         RenewableEnergyLearningSingleBidStrategy
     )
 
+    from assume.strategies.portfolio_learning_strategies import (
+        PortfolioLearningStrategy,
+    )
 
+    bidding_strategies["portfolio_learning"] = PortfolioLearningStrategy
 except ImportError:
     pass

assume/strategies/learning_strategies.py

@@ -349,8 +349,6 @@ class EnergyLearningStrategy(TorchLearningStrategy, MinMaxStrategy):
         Number of time steps for which the agent forecasts market conditions. Defaults to 12.
     max_bid_price : float
         Maximum allowable bid price. Defaults to 100.
-    max_demand : float
-        Maximum demand capacity of the unit. Defaults to 10e3.
     device : str
         Device for computation, such as "cpu" or "cuda". Defaults to "cpu".
     float_type : str
@@ -363,8 +361,6 @@ class EnergyLearningStrategy(TorchLearningStrategy, MinMaxStrategy):
         Class of the neural network architecture used for the actor network. Defaults to MLPActor.
     actor : torch.nn.Module
         Actor network for determining actions.
-    order_types : list[str]
-        Types of market orders supported by the strategy. Defaults to ["SB"].
     action_noise : NormalActionNoise
         Noise model added to actions during learning to encourage exploration. Defaults to None.
     collect_initial_experience_mode : bool
@@ -390,9 +386,6 @@ def __init__(self, *args, **kwargs):
             **kwargs,
         )
 
-        # define allowed order types
-        self.order_types = kwargs.get("order_types", ["SB"])
-
     def calculate_bids(
         self,
         unit: SupportsMinMax,
@@ -835,8 +828,6 @@ class StorageEnergyLearningStrategy(TorchLearningStrategy, MinMaxChargeStrategy)
         Number of time steps for forecasting market conditions. Defaults to 24.
     max_bid_price : float
         Maximum allowable bid price. Defaults to 100.
-    max_demand : float
-        Maximum demand capacity of the storage. Defaults to 10e3.
     device : str
         Device used for computation ("cpu" or "cuda"). Defaults to "cpu".
     float_type : str
@@ -849,8 +840,6 @@ class StorageEnergyLearningStrategy(TorchLearningStrategy, MinMaxChargeStrategy)
         Class of the neural network for the actor network. Defaults to MLPActor.
     actor : torch.nn.Module
         The neural network used to predict actions.
-    order_types : list[str]
-        Types of market orders used by the strategy. Defaults to ["SB"].
     action_noise : NormalActionNoise
         Noise model added to actions during learning for exploration. Defaults to None.
     collect_initial_experience_mode : bool
@@ -876,9 +865,6 @@ def __init__(self, *args, **kwargs):
             **kwargs,
         )
 
-        # define allowed order types
-        self.order_types = kwargs.get("order_types", ["SB"])
-
     def get_individual_observations(
         self, unit: SupportsMinMaxCharge, start: datetime, end: datetime
     ):
@@ -1146,8 +1132,6 @@ class RenewableEnergyLearningSingleBidStrategy(EnergyLearningSingleBidStrategy):
         Class of the neural network for the actor network. Defaults to MLPActor.
     actor : torch.nn.Module
         The neural network used to predict actions.
-    order_types : list[str]
-        Types of market orders used by the strategy. Defaults to ["SB"].
     action_noise : NormalActionNoise
         Noise model added to actions during learning for exploration. Defaults to None.
     collect_initial_experience_mode : bool
@@ -1173,9 +1157,6 @@ def __init__(self, *args, **kwargs):
             **kwargs,
         )
 
-        # define allowed order types
-        self.order_types = kwargs.get("order_types", ["SB"])
-
     def get_individual_observations(
         self, unit: SupportsMinMaxCharge, start: datetime, end: datetime
     ):