change is_action_continuous to _is_action_continuous

Summary: Sometimes we use is_action_continuous sometimes we use _is_action_continuous as an attribute of a policy learner. This diff makes it consistent by changing is_action_continuous to _is_action_continuous

Reviewed By: rodrigodesalvobraz

Differential Revision: D65853193

fbshipit-source-id: 847c82ddb31ae46a60da2698b02e232049abf91e

pearl/pearl_agent.py

@@ -111,8 +111,8 @@ def __init__(
         )
 
         # set here so replay_buffer and policy_learner are in sync
-        self.replay_buffer.is_action_continuous = (
-            self.policy_learner.is_action_continuous
+        self.replay_buffer._is_action_continuous = (
+            self.policy_learner._is_action_continuous
         )
         self.replay_buffer.device_for_batches = self.device
 
@@ -192,7 +192,7 @@ def observe(
             terminated=action_result.terminated,
             max_number_actions=(
                 self.policy_learner.action_representation_module.max_number_actions
-                if not self.policy_learner.is_action_continuous
+                if not self.policy_learner._is_action_continuous
                 else None
             ),  # max number of actions for discrete action space
             cost=action_result.cost,

pearl/policy_learners/policy_learner.py

@@ -108,7 +108,7 @@ def __init__(
         self._batch_size = batch_size
         self._training_steps = 0
         self.on_policy = on_policy
-        self.is_action_continuous = is_action_continuous
+        self._is_action_continuous = is_action_continuous
         self.distribution_enabled: bool = is_distribution_enabled()
         self.requires_tensors = requires_tensors
 

pearl/policy_learners/sequential_decision_making/actor_critic_base.py

@@ -115,7 +115,7 @@ def __init__(
 
         self._action_dim: int = (
             self.action_representation_module.representation_dim
-            if self.is_action_continuous
+            if self._is_action_continuous
             else self.action_representation_module.max_number_actions
         )
 
@@ -239,7 +239,7 @@ def act(
         # Step 1: compute exploit_action
         # (action computed by actor network; and without any exploration)
         with torch.no_grad():
-            if self.is_action_continuous:
+            if self._is_action_continuous:
                 exploit_action = self._actor.sample_action(subjective_state)
                 action_probabilities = None
             else:

pearl/policy_learners/sequential_decision_making/implicit_q_learning.py

@@ -225,7 +225,7 @@ def _actor_loss(self, batch: TransitionBatch) -> torch.Tensor:
             actor_loss = (advantage * loss).mean()
 
         else:
-            if self.is_action_continuous:
+            if self._is_action_continuous:
                 log_action_probabilities = self._actor.get_log_probability(
                     batch.state, batch.action
                 ).view(-1)

pearl/replay_buffers/sequential_decision_making/bootstrap_replay_buffer.py

@@ -94,7 +94,7 @@ def sample(self, batch_size: int) -> TransitionWithBootstrapMaskBatch:
             # pyre-fixme[6]: For 1st argument expected `List[Transition]` but got
             #  `List[Union[Transition, TransitionBatch]]`.
             transitions=samples,
-            is_action_continuous=self.is_action_continuous,
+            is_action_continuous=self._is_action_continuous,
         )
         # pyre-fixme[16]: Item `Transition` of `Union[Transition, TransitionBatch]`
         #  has no attribute `bootstrap_mask`.