wassname/ray
1
0
Fork 0
mirror of https://github.com/wassname/ray.git synced 2026-07-21 12:50:45 +08:00
Code Issues Packages Projects Releases Wiki Activity

[rllib] fix for rollout of lstm policies (#3643)

Browse Source 
* fix for lstm policies

* added call to local evaluator

* Update python/ray/rllib/rollout.py

Co-Authored-By: eugenevinitsky <eugenevinitsky@users.noreply.github.com>

* Update rollout.py

* Update rollout.py
This commit is contained in:
Eugene Vinitsky
2019-01-13 15:54:23 -08:00
committed by Eric Liang
parent 00e9f8d870
commit a5d1f03515
1 changed files with 16 additions and 1 deletions
Download Patch File Download Diff File Expand all files Collapse all files
python/ray/rllib/rollout.py
+16 -1
View File
@@ -104,6 +104,17 @@ def rollout(agent, env_name, num_steps, out=None, no_render=True):
env = agent.local_evaluator.env
else:
env = gym.make(env_name)
if hasattr(agent, "local_evaluator"):
state_init = agent.local_evaluator.policy_map[
"default"].get_initial_state()
else:
state_init = []
if state_init:
use_lstm = True
else:
use_lstm = False
if out is not None:
rollouts = []
steps = 0
@@ -114,7 +125,11 @@ def rollout(agent, env_name, num_steps, out=None, no_render=True):
done = False
reward_total = 0.0
while not done and steps < (num_steps or steps + 1):
action = agent.compute_action(state)
if use_lstm:
action, state_init, logits = agent.compute_action(
state, state=state_init)
else:
action = agent.compute_action(state)
next_state, reward, done, _ = env.step(action)
reward_total += reward
if not no_render: