Comment by data-ottawa

That one has a bit more friction than pandas because the return schema requirement -- pandas let's you get away with this bad practice.

It also does batches when you declare scalar outputs, but you can't control the batch size, which usually isn't an issue, but I've run into situations where it is.