I know plenty of people who are comfortable in both. But I don’t know any “data science” people as good at sql as a good dba. But that’s ok.

I think a good analysis requires both sql and some flat file, if only for reproducibility. I’ve encountered many point in time analyses that can’t be recreated because they were just sql against db and there’s no way to reproduce the result months later. Are their methods accurate? Or auditable? It’s much harder because the source data is changed.

So I like talking about only-sqlers about the importantance of incorporating the data management of extracts and archives into the overall solution. And doing so in a way that doesn’t try to redesign the database trying to avoid the need for users to file away point in time snapshots of data.

I’ve also tried to do things in sql only to realize that I could have done it easier in pandas.

I think it’s usually a mix of both for me, but I try to at least get a decent “tidy” dataset out of sql and then do all the manipulation in pandas because a python pipeline is more portable than a sql pipeline, I think.

I wrote a fairly complex mango query a while back, and trying to understand it now is a nightmare. Compare that to SQL, which is one of the most readable languages out there in my opinion.