That's super cool!

In the past, I've resorted to using their docker images and then hacking pip-based packages in, but that was pretty brittle.

Their accelerated scikit tooling is really a godsent for some things such as clustering.