Workflow recommendation for end to end learning

avisingh599 / reward-learning-rl

[RSS 2019] End-to-End Robotic Reinforcement Learning without Reward Engineering

https://sites.google.com/view/reward-learning-rl/

Other

367 stars 68 forks source link

Workflow recommendation for end to end learning #17

Closed seivazi closed 4 years ago

seivazi commented 4 years ago

Hi, Could you share with us some insights into how you approach the end to end learning development workflow?

Specifically, how do you deal with long iteration times for changes (few hours)? Is there some deployment infrastructure (other than Ray, Docker, and git) in your group? For example, do you try out one by one or do you submit a batch of changes and somehow a software reports it for you later on.

Cheers, Shahram

avisingh599 commented 4 years ago

We usually submit a batch of changes and then re-run everything. The long running times are indeed problematic, which is why the public version of the code is not updated very frequently.