Fixed incorrect actor_loss shape for SAC continuous, addresses issue #…

…379 (#383)
vwxyzjn · Oct 4, 2023 · 0fceeef · 0fceeef · vercel · Oct 4, 2023
1 parent 7e24ae2
commit 0fceeef
Showing 1 changed file with 1 addition and 1 deletion.
diff --git a/cleanrl/sac_continuous_action.py b/cleanrl/sac_continuous_action.py
@@ -266,7 +266,7 @@ def get_action(self, x):
                     pi, log_pi, _ = actor.get_action(data.observations)
                     qf1_pi = qf1(data.observations, pi)
                     qf2_pi = qf2(data.observations, pi)
-                    min_qf_pi = torch.min(qf1_pi, qf2_pi).view(-1)
+                    min_qf_pi = torch.min(qf1_pi, qf2_pi)
                     actor_loss = ((alpha * log_pi) - min_qf_pi).mean()
 
                     actor_optimizer.zero_grad()