Open archenroot opened 5 years ago
I have also run into this error, but have found no success in debugging it. It is caused by Inf or NaN making it into the model's network, though I am unsure how, as the observation space, action space, and reward space all actively replace nan and abs(inf) with 0. Any ideas?
hm, need to do more tracing...
Hi,
We just release a guide in the documentation to tackle this type of issue. Feel free to open an issue on stable-baselines repo if you find something wrong coming from the library.
@archenroot has anyone debugged this anymore using VecCheckNan
from stable-baselines?
While the optimize.py continue running, I observed one exception, but the process continue...