Remove explict forward calls (#753)

* Remove explict forward calls

* Changelog and commit checks.

* Reverted test forward removal for super call.

Co-authored-by: Anssi <kaneran21@hotmail.com>

docs/misc/changelog.rst

@@ -26,6 +26,7 @@ Bug Fixes:
 - Fixed a bug in ``HumanOutputFormat``. Distinct keys truncated to the same prefix would overwrite each others value,
   resulting in only one being output. This now raises an error (this should only affect a small fraction of use cases
   with very long keys.)
+- Routing all the nn.Module calls through implicit rather than explict forward as per pytorch guidelines (@manuel-delverme)
 
 Deprecations:
 ^^^^^^^^^^^^^
@@ -917,4 +918,4 @@ And all the contributors:
 @benblack769 @bstee615 @c-rizz @skandermoalla @MihaiAnca13 @davidblom603 @ayeright @cyprienc
 @wkirgsn @AechPro @CUN-bjy @batu @IljaAvadiev @timokau @kachayev @cleversonahum
 @eleurent @ac-93 @cove9988 @theDebugger811 @hsuehch @Demetrio92 @thomasgubler @IperGiove @ScheiklP
-@simoninithomas @armandpl
+@simoninithomas @armandpl @manuel-delverme

stable_baselines3/common/on_policy_algorithm.py

@@ -166,7 +166,7 @@ class OnPolicyAlgorithm(BaseAlgorithm):
             with th.no_grad():
                 # Convert to pytorch tensor or to TensorDict
                 obs_tensor = obs_as_tensor(self._last_obs, self.device)
-                actions, values, log_probs = self.policy.forward(obs_tensor)
+                actions, values, log_probs = self.policy(obs_tensor)
             actions = actions.cpu().numpy()
 
             # Rescale and perform action

stable_baselines3/dqn/policies.py

@@ -66,7 +66,7 @@ class QNetwork(BasePolicy):
         return self.q_net(self.extract_features(obs))
 
     def _predict(self, observation: th.Tensor, deterministic: bool = True) -> th.Tensor:
-        q_values = self.forward(observation)
+        q_values = self(observation)
         # Greedy action
         action = q_values.argmax(dim=1).reshape(-1)
         return action

stable_baselines3/sac/policies.py

@@ -182,7 +182,7 @@ class Actor(BasePolicy):
         return self.action_dist.log_prob_from_params(mean_actions, log_std, **kwargs)
 
     def _predict(self, observation: th.Tensor, deterministic: bool = False) -> th.Tensor:
-        return self.forward(observation, deterministic)
+        return self(observation, deterministic)
 
 
 class SACPolicy(BasePolicy):

stable_baselines3/sac/sac.py

@@ -253,7 +253,7 @@ class SAC(OffPolicyAlgorithm):
             # Compute actor loss
             # Alternative: actor_loss = th.mean(log_prob - qf1_pi)
             # Mean over all critic networks
-            q_values_pi = th.cat(self.critic.forward(replay_data.observations, actions_pi), dim=1)
+            q_values_pi = th.cat(self.critic(replay_data.observations, actions_pi), dim=1)
             min_qf_pi, _ = th.min(q_values_pi, dim=1, keepdim=True)
             actor_loss = (ent_coef * log_prob - min_qf_pi).mean()
             actor_losses.append(actor_loss.item())

stable_baselines3/td3/policies.py

@@ -80,7 +80,7 @@ class Actor(BasePolicy):
     def _predict(self, observation: th.Tensor, deterministic: bool = False) -> th.Tensor:
         # Note: the deterministic deterministic parameter is ignored in the case of TD3.
         #   Predictions are always deterministic.
-        return self.forward(observation)
+        return self(observation)
 
 
 class TD3Policy(BasePolicy):