-
Notifications
You must be signed in to change notification settings - Fork 177
New issue
Have a question about this project? Sign up for a free GitHub account to open an issue and contact its maintainers and the community.
By clicking “Sign up for GitHub”, you agree to our terms of service and privacy statement. We’ll occasionally send you account related emails.
Already on GitHub? Sign in to your account
RecurrentPPO: 9x speedup - whole sequence batching #118
base: master
Are you sure you want to change the base?
Conversation
…g speed improvement
…pping those batches
@araffin have you been able to take a look at this yet? I am very curious what you think about it. |
no, not yet, still on my stack... and going on holidays soon, so, I'll probably take a look next week or in january. |
Cool. Let me know if you need any help running experiments/coding |
…ble-baselines3-contrib into Stable-Baselines-Team-master
merge with master
Hello,
|
My bad. Bug is fixed now! |
I had to set To reproduce:
On CartPole, I have another error:
Also, SDE seems not supported (that's ok, but need to be checked at runtime). Finally, I experienced some NaN issue from time to time when
|
Also an error when using CNN:
|
The error for CartPole seems to be still there... |
Yes, it has only been implemented for Box action spaces so that might be it. I have not much time to work on this anymore. So feel free to do it. |
Description
Moving from 2d batches to 3d batches of whole sequences leads to a 5-9 times speedup in terms of fps while keeping results similar. Proof.
Context
Types of changes
Its currently implemented as an additional feature but would probably be more optimal to replace the original.
Checklist:
make format
(required)make check-codestyle
andmake lint
(required)make pytest
andmake type
both pass. (required)Note: we are using a maximum length of 127 characters per line