I don't care if you are a data scientist or a machine learning engineer:
Test your darn code!
Not every piece of code is easy to test, but that's not justification to test none of it.
I'm not speaking about writing your tests first or any other dogmatic way of doing things. I'm just talking about a having a test suit that you can run after making changes to ensure everything is still working.
If your codebase doesn't have automatic tests:
• It's harder to debug and find problems in it
• It's harder to change and improve it
• It's harder to understand
• It has a lower quality
• It's likely more complex than it should be
It's funny to see the AI field going through the same pains and discoveries the software industry already experienced.
We've been there. We know what works and what doesn't.
Start writing automatic tests.