Mikio L. Braun

ML and AI expert

Back to archive

Thread

5 tweets

1
Totally with you on the research side, and against the swe-ization of DS. MLOps right now sometimes seems like DevOps applying what they learned without deeper understanding of how DS work really looks like.
2
@paul_rietschka I also think that notebooks can be great if you know what you‘re doing (prob. true for almost any tool), but there are also limits to what you can do and some things you have to do manually (eg refactorings).
3
@paul_rietschka I personally think that MATLAB was really great as a tool. You had the console and autoreloading of files. Really great to work on something and grow some structured codebase on the side. As a language it was quite limited, though.
4
@paul_rietschka I try to structure my notebooks, not copy and paste too much, but in the end, every cell is a bit of the whole thing and I need to mentally track which ones to re-evaluate based on changes. It‘s a bit tedious and could be easier with better tooling support.
5
@paul_rietschka Btw, the one thing the „SWE approach to DS“ misses to understand is that it is not enough to „clean up“ research code to put it to production. After you‘ve done that, you often need to go back and work on the next iteration. IMHO that is the big unsolved challenge.
1