diff --git a/autoagora_agents/algorithm.py b/autoagora_agents/algorithm.py
index 6b59f57..35ad39f 100644
--- a/autoagora_agents/algorithm.py
+++ b/autoagora_agents/algorithm.py
@@ -13,6 +13,18 @@
 
 
 class Algorithm(ABC):
+    """Base class for algorithms.
+
+    Concretions must implement :meth:`__call__`.
+
+    Attributes:
+        niterations (int): Number of times the algorithm has been called.
+        nupdates (int): Number of times the algorithm has been updated.
+        group (str): The group to which the algorithm belongs.
+        i (int): The index of the algorithm.
+        name (str): The group and index of the algorithm.
+    """
+
     def __init__(self, *, group: str, i: int) -> None:
         self.niterations = 0
         self.nupdates = 0
@@ -21,9 +33,11 @@ def __init__(self, *, group: str, i: int) -> None:
         self.name = f"{group}_{i}"
 
     def reset(self) -> None:
+        """Reset the algorithm's state."""
         self.niterations = 0
 
-    def update(self):
+    def update(self) -> None:
+        """Update the algorithm's parameters."""
         self.nupdates += 1
 
     @abstractmethod
@@ -35,6 +49,17 @@ def __call__(
         reward: float,
         done: bool,
     ) -> np.ndarray:
+        """Run the algorithm forward.
+
+        Keyword Arguments:
+            observation (np.ndarray): The observation seen by the agent.
+            action (np.ndarray): The previous action taken by the agent.
+            reward (float): The reward of the agent.
+            done (bool): If True, the agent is no longer in the game.
+
+        Returns:
+            np.ndarray: The next action taken by the agent.
+        """
         pass
 
     @staticmethod
@@ -153,7 +178,15 @@ def __call__(
         self.niterations += 1
         return act
 
-    def logprob(self, actions):
+    def logprob(self, actions: torch.Tensor) -> torch.Tensor:
+        """Compute the log probability of the action given the distribution.
+
+        Arguments:
+            actions (torch.Tensor): The actions for which to compute the log probability
+
+        Returns:
+            torch.Tensor: The log probability of the actions.
+        """
         return self.actiondist.logprob(actions)