Best version mix for ray rllib v1 and rllib v2 including soft dependencies (e.g. tensorflow))

How severe does this issue affect your experience of using Ray?

  • Medium: It contributes to significant difficulty to complete my task, but I can work around it.

Dear all,
with the recent major API changes coming in ray 2.0.0, I wonder if the community can provide some “best practice” mixture of version compatibility regarding the major libraries.
I am thinking about the following

ray1 version 1.13.0 or 1.12.0 works well with

  • numpy version XX
  • tensorflow version XX
  • torch version XX

I am asking that because even within the soft-dependent libraries I observe ground-breaking changes, e.g. in tensorflow when suddenly I get the message of “missing module keras” (caused by a call within rllib code), or the removal of np.bool in numpy>=1.20.

In the last days I already came across pipdeptree, but this is rather post-mortem analysis and a good twin brother for pip check.

Thanks in advance!