Most programmers never do or use machine learning, and in fact most people who call themselves "data scientists" don't know the first thing about it. Knowing a package with some canned algorithms in it is not one tenth of the battle. If you can write a logistic regression or SVM from scratch (not that it has to be as good as the existing product; in practice, you'd probably to use an existing QP solver, for example) we'll talk.

I also don't think Scala is part of "the full stack". Worth knowing? Sure. Powerful? Yeah, it's quite a large step up from Java. You'll want to know it to use Spark, and it's a great "gateway drug" for Haskell. But I don't think it's on the list of things that a person needs to know to call himself a "full stack web developer".

Anyway, someone needs to write a terrible language that isn't so bad that no one uses it. Take all the worst features of C++ and Java, bolt on some enterprise crap, make your shitty ORM a first-class citizen, and then sell the shit out of it. (It can't be unusable like Brainfuck, but it can be worse than Java and still sellable to the enterprise.) It'll be a nine-figure effort, but I'll spot the first $200. Here's the one rule: this shitty language has to be named FullStack. Then we can be done with this shitty buzzword because "Full Stack Developer" will mean something else. Bonus points if it's attached to a "hip" but broken database system called Big Data.