Closed PierreExeter closed 4 years ago
Hello,
This means that the environment is not vectorized, it is not possible to specify the log directory and to monitor the training. Why did you make a special case for DQN and DDPG?
Good point. There was some issue at some point with DQN/DDPG and vectorized environment (the performances where different,i I did that as a safe option) so I need to check if it is still the case, it may be an issue with copying arrays.
Ok I will run a short experiment with and without vectorized environment and let you know if I see any difference.
FYI,
Vectorizing DQN or DDPG does not seem to affect the training, the average return or the training time significantly, see results attached.
Thanks, then I may change that =)
Btw, what did you use to generate that nice pdf?
You're welcome! I used Latex and Matplotlib for the figures.
Btw I'm not sure whether this will affect the results but I received this warning during training:
UserWarning: Training and eval env are not of the same type<Monitor<TimeLimit<PendulumEnv
You're welcome! I used Latex and Matplotlib for the figures.
I meant: do you have a script somewhere or you wrote the latex file manually?
Btw I'm not sure whether this will affect the results but I received this warning during training:
This is a normal warning when uing eval env, it is hard to check that the training env is the same as the eval env, so instead of throwing an error, we warn the user.
I meant: do you have a script somewhere or you wrote the latex file manually?
It's a template I use for writing report. It's pretty basic but here it is if that can be of any use.
Hello,
Thanks a lot for this amazing code.
I noticed that the environment is instantiated differently when using either DQN or DDPG. Specifically at line 249 of train.py, the env is created with:
in the case of DQN and DDPG whereas it is created with the make_env helper:
for all the other algorithms.
This means that the environment is not vectorized, it is not possible to specify the log directory and to monitor the training. Why did you make a special case for DQN and DDPG?
Thanks