Open vlad17 opened 6 years ago
Which test exposes this issue? At a cursory glance it looks like the env and the agents are properly seeded, but I can dig deeper.
No test in particular, this is just what happens when you run the same setting (Eg. main_ddpg.py) twice. don't worry about this for now.
The same run on the same seed, even on tensorflow CPU, is giving different learning curves. There might be some non-determinism somewhere, maybe in the environment-seeding.