Use mean when calculating actor loss

32cef6a4 · Nicola Gatto · 5d9ba4fa · 32cef6a4
Commit 32cef6a4 authored Jun 25, 2019 by Nicola Gatto
Hide whitespace changes
Inline Side-by-side

Showing with 1 addition and 1 deletion

src/main/resources/templates/gluon/reinforcement/agent/Agent.ftl ...n/resources/templates/gluon/reinforcement/agent/Agent.ftl +1 -1

No files found.
--- a/src/main/resources/templates/gluon/reinforcement/agent/Agent.ftl
+++ b/src/main/resources/templates/gluon/reinforcement/agent/Agent.ftl
@@ -548,7 +548,7 @@ class DdpgAgent(Agent):
                        actor_qvalues = tmp_critic(states, self._actor(states))
                        # For maximizing qvalues we have to multiply with -1
                        # as we use a minimizer
-                        actor_loss = -1 * actor_qvalues
+                        actor_loss = -1 * actor_qvalues.mean()
                    actor_loss.backward()
                    trainer_actor.step(self._minibatch_size)