yandexdataschool / Practical_RL

A course in reinforcement learning in the wild

Geek Repo:Geek Repo

Github PK Tool:Github PK Tool

Discourage Coursera students from using Coursera-hosted notebooks?

dniku opened this issue · comments

Jupyter containers hosted by Coursera cause a lot of trouble. Perhaps more than they are worth.

  • They are very limited in terms of lifetime and CPU. The docs say 90 minutes / 0.5-2 CPUs. That's definitely insufficient to train Breakout, for example.
  • Updating them is inconvenient. We don't have a way to overwrite any of the existing files, as documented here (and here is a much longer doc). The only way to update a file in a student's container is to have them rename/delete it and then relogin.

Perhaps Colab could be a viable default option, at least for assignments that require doing DL?

Already kind of done, but the current solution is temporary.

Mostly done. I have removed all notebook launchers from the course, except for the Numpy primer, in order to enable access to the notebooks that are still on Coursera's servers. I've also added there a deprecation notice and a link to the current version of the ML primer on Colab.