gryf/coach
1
0
Fork 0
mirror of https://github.com/gryf/coach.git synced 2025-12-17 11:10:20 +01:00
Code Issues Projects Releases Wiki Activity

preventing the evaluation agent from getting stuck in bad policies by updating from the global network during episodes

Browse Source
This commit is contained in:
Itai Caspi
2017-10-25 10:28:45 +03:00
parent d3c6860421
commit 39cf78074c
2 changed files with 6 additions and 0 deletions
Download Patch File Download Diff File Expand all files Collapse all files

5
agents/agent.py
View File

@@ -457,6 +457,11 @@ class Agent:
while not episode_ended:
episode_ended = self.act(phase=RunPhase.TEST)
if keep_networks_synced \
and self.total_steps_counter % self.tp.agent.update_evaluation_agent_network_after_every_num_steps:
for network in self.networks:
network.sync()
if self.tp.visualization.dump_gifs and self.total_reward_in_current_episode > max_reward_achieved:
max_reward_achieved = self.total_reward_in_current_episode
frame_skipping = int(5/self.tp.env.frame_skip)