Hey everyone, we would like to start offering “office hours” for RLlib, which will basically be 30min video chats with @sven1977 to debug/unblock issues together.
If you are interested, please fill out this form here (~2min) to let us know what times and days-of-week would work best for you. We’ll get back to you with a calendly link then to schedule the actual day+time.
Hey everyone, thanks for signing up. Will send out the emails tomorrow morning! Sorry for the delay, just busy with release testing these days. @Lars_Simon_Zehnder@rusu24edward
Hey @thgehr , sorry, but we are no longer offering these office hours. Please ask your questions here on this forum or file an issue on github, instead.
Fyi: The RLlib team has ramped up its community support efforts and we are spending 1.5 full engineer days per week on each a) the forum here and b) github issues. We are carefully evaluating whether this cadence will help us catch up on older posts/issues over the next few months and adjust as necessary.
Hey everyone , we would like to re-start office hours on a regular bi-weekly basis!
8am PT Tuesdays. These will be video chats with members of the RLlib team to debug/unblock your RLlib projects. If you are interested, please add your questionto this doc here. We are asking if you would please link to either a github issue or discuss link in your question. This will help to share the learnings with others. Upcoming office hours schedule:
The last Office Hours github issue, “Handling Conditional Callbacks from Ray Tune”, has been commented with link to youtube video of the live coding debugging.
The end result is a pattern in RLlib for how to train/compare different models, where same algorithm is trained using different hyperparameters. Create 3 different algorithm configs, then pass these config names into a Ray Tune grid search. Model metrics are calculated using Conditional Callbacks. Model metrics can be sanity-check visualized in TensorBoard.
Last Office Hours, we discussed 1) IMPALA with MixinReplayBuffer to answer Renos’ question. 2) PettingZoo deeper integration rather than just using existing RLlib wrapper. 3) Custom DDPO with PyTorch to answer Simon’s question. 4) Renos another question about multi heterogeneous agents. Here is full Office Hours Playlist on YouTube.
Next Tuesday, Artur is hosting. Please add your questionto this doc here . Please link to either a github issue or discuss link in your question. This will help to share the learnings with others.