We need gradual complexity, but I think it's very hard to design. It needs knowledge + didactic capability.
It's the numpy, scipy, jupyter tools, as well as Anaconda, that make the sale.
Too, it's not just the analysis, but the extract/transform/load (ETL) beforehand. Python's general purpose wealth of tools had me transforming the FCC antenna coverage .zip[1] to shapefile and parking it in a spatialite file for QGIS in a relative jiffy.
Doubtless I could do all that in Julia or R, but I get to hone my python-fu at the office doing cloud stuff by day--another argument in favor of the general-purpose tool.
[1] ftp://ftp.fcc.gov/pub/Bureaus/MB/Databases/fm_service_contour_data/FM_service_contour_current.zip
What python brought to the table, was that it was freely installed on most machines.
At the company I work at there’s a pypi mirror with quite recent packages and it works out of the box - for R there’s only an ancient version and some hacks to get it to work with juptyer notebooks.
As an aside, emacs has so much potential, I'm so glad the community continues to thrive. I use emacs exclusively for science data analysis and consistently impress my colleagues with how quickly things can be accomplished. If rough edges were polished in some of the python corners there could easily be an emacs renaissance in academia, a la nicolas rougier et al.
For just how popular python is in science now, the workflows are abysmal at best. 2 out of 3 of my colleagues won't touch it because Anaconda makes no sense, a command line work flow is too obtuse, and Matlab for all its disgusting nature is at least "natural" to work with as an environment in some sense. Which makes me sad. I have dreams that emacs makes that more approachable.