Refactoring

Every codebase ages over time. Each addition of new lines generates bugs, and fixing them becomes increasingly time-consuming and costly. The difference is that some work gets discarded after 15 years, while others after just one year.

In the world of software engineering, old, decaying code is something every team eventually encounters. The world of ML is no different. Development Jupyter notebooks are full of old analyses that, over time, no one knows how to run or what they were even for. Python scripts refer to old packages that no one maintains anymore and require dependencies that conflict with yours. Docker containers attempt to download library versions that are no longer available on the internet.

If you recognize yourself in the previous paragraph, I can help. Sometimes an upgrade is enough. Other times, it requires switching to a different library. And in some cases, it's time to cut out the old code like a tumor. My favorite activity when writing code is deleting it. Of course, while maintaining the same functionality.