wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-06-29 20:17:17 +08:00
Code Issues Packages Projects Releases Wiki Activity

fix incorrect critic loss in TD3 (#10775)

Browse Source 
Co-authored-by: Manny Vindiola <manuel.m.vindiola.civ@mail.mil>
This commit is contained in:
mvindiola1
2020-09-20 23:01:51 -04:00
committed by GitHub
parent 3e94c690c7
commit 2b893d1bb5
3 changed files with 0 additions and 3 deletions
Download Patch File Download Diff File Expand all files Collapse all files
rllib/agents/ddpg/ddpg_tf_policy.py
-1
View File
@@ -175,7 +175,6 @@ def ddpg_actor_critic_loss(policy, model, _, train_batch):
if twin_q:
td_error = q_t_selected - q_t_selected_target
twin_td_error = twin_q_t_selected - q_t_selected_target
td_error = td_error + twin_td_error
if use_huber:
errors = huber_loss(td_error, huber_threshold) + \
huber_loss(twin_td_error, huber_threshold)
rllib/agents/ddpg/ddpg_torch_policy.py
-1
View File
@@ -124,7 +124,6 @@ def ddpg_actor_critic_loss(policy, model, _, train_batch):
if twin_q:
td_error = q_t_selected - q_t_selected_target
twin_td_error = twin_q_t_selected - q_t_selected_target
td_error = td_error + twin_td_error
if use_huber:
errors = huber_loss(td_error, huber_threshold) \
+ huber_loss(twin_td_error, huber_threshold)
rllib/agents/ddpg/tests/test_ddpg.py
-1
View File
@@ -495,7 +495,6 @@ class TestDDPG(unittest.TestCase):
td_error = q_t_selected - q_t_selected_target
twin_td_error = twin_q_t_selected - q_t_selected_target
td_error = td_error + twin_td_error
errors = huber_loss(td_error, huber_threshold) + \
huber_loss(twin_td_error, huber_threshold)