Kubeflow Jupyter notebook images are large because of all the extra Python libs; can we shrink to improve pull times?

[inc0]

And that's an issue for minikube for example. There are all sorts of things we can do to shrink them.
Couple examples:

• make sure all compile artifacts are gone
• cherry-pick required things with multi-stage builds
• change base image to something lightweight

[willingc]

Images are large because there are a ton of dependencies and third party libraries beyond Jupyter notebook being added to the container image.

A minimal Jupyter notebook version is available from jupyter/docker-stacks repo.

Let's amend this issue title to more accurately reflect the actual issue (lots of installed libraries beyond Jupyter). Thanks!

cc/ @jlewi @inc0

[willingc]

Thanks @jlewi

[inc0]

@willingc true, issue is on our side rather than jupyter. We create 2 conda envs ~5gig each. I made an attempt to shrink images a bit here #579

[ankushagarwal]

Closing since #689 has been merged