I would teach a data scientist how to calibrate a classifier. I would teach a data scientist how to use version control. I would teach a data scientist how to do continuous integration. I would teach a data scientist how to package software for maven or pypi. I would teach a data scientist how to play Poker.

I would not teach a data scientist Kubernetes.

Maybe not essential, but say I (a data scientist) was thinking about learning Kubernetes what would be the reasons you'd give for avoiding it?

Because your time and attention are limited. You can learn all aspects of software engineering, but to what purpose? As a data scientist, you are not spending time understanding the data, spending time with business to understand the problem and come up with the right objective functions to optimize for, you are probably not being a data scientist.

I would go even wider than that and say that Kubernetes may be an ‘appropriate technology’ for Google but when it is exported to startups it is harmful, like OKRs. IBM had a technology for managing distributed applications for mainframes in the 1990s called “Parallel Sysplex” that was simple in comparison.