wassname/DeepRL
1
0
Fork 0
mirror of https://github.com/wassname/DeepRL.git synced 2026-09-10 11:40:58 +08:00
Code Issues Packages Projects Releases Wiki Activity

Fix a critical bug in DDPG

Browse Source
This commit is contained in:
Shangtong Zhang
2017-10-15 10:05:33 -06:00
parent 3ed6f2a9db
commit 1b6de16b4f
6 changed files with 103 additions and 221 deletions
Download Patch File Download Diff File Expand all files Collapse all files
agent/DDPG_agent.py
+5 -3
View File
@@ -89,11 +89,13 @@ class DDPGAgent:
critic_loss.backward()
self.critic_opt.step()
actor_loss = -critic.predict(states, actor.predict(states, False))
actor_loss = actor_loss.mean()
actions = actor.predict(states, False)
var_actions = Variable(actions.data, requires_grad=True)
q = critic.predict(states, var_actions)
q.backward(torch.ones(q.size()))
actor.zero_grad()
actor_loss.backward()
actions.backward(-var_actions.grad.data)
self.actor_opt.step()
self.soft_update(self.target_network, self.learning_network)