Thread

Me and @DJCordhose have been discussing how to go from notebooks to production and back, and in an ideal world we agreed that you'd start copying parts that are stable to a library that you'd also be using in production and re-import those. Here's a 🧶 why I was wrong

2/5 Spending serious time in notebooks I realize that this isn't just some interactive way to work with data, but actually a version of "REPL-based development" practical.li/clojure-stagin… It's messy because you need to jump back and forth to re-run parts of codes but it works.

3/5 You could move parts into python modules and use the autoreload feature in Jupyter. And this technically works (apart from minor issues), but I'm finding it breaks the agility of what I'm doing.

4/5 Suddenly you need to think about whether what you're doing breaks other analyses you have elsewhere, or production code. You'd need to switch to test driven development to make sure you don't, and it breaks the flow.