chore: extra comment on term_or_trunc vs terminal

instadeepai · Oct 22, 2024 · b735334 · b735334
1 parent 63eb99f
commit b735334
Showing 1 changed file with 4 additions and 0 deletions.
diff --git a/mava/systems/q_learning/anakin/rec_qmix.py b/mava/systems/q_learning/anakin/rec_qmix.py
@@ -154,6 +154,10 @@ def replicate(x: Any) -> Any:
     init_hidden_state = replicate(init_hidden_state)
 
     init_acts = env.action_spec().generate_value()
+
+    # NOTE: Term_or_trunc refers to the the joint done, ie. when all agents are done or when the
+    # episode has terminated. We use this exclusively in QMIX. Terminal refers to individual agent
+    # dones. We keep this in this file for consistency with IQL.
     init_transition = Transition(
         obs=init_obs,  # (A, ...)
         action=init_acts,  # (A,)