Tuesday, August 01, 2017

Anaconda considered harmful

I was inspired to start using Python based on this website. For getting going with Python, the authors recommend using the Anaconda installation, which automatically installs a large number of the numeric and scientific packages available for the language.

This probably is good advice for an economist who simply wants to use the language to play around with different models. But if you intend become a serious Python developer, this is not the right way to get going.

The problem with Anaconda is that it makes things too easy for you: you come to simply assume that all this "stuff" that Anaconda includes will be available for anyone who wants to use your work. No, if you want to develop Python code that will be used by others, is much better to start with a bare-bones Python installation, carefully consider what third party packages you really need to depend upon, and, in seeing what you need to do to install them for yourself, learn what your users will need to do to install those packages for themselves.

1 comment:

  1. Or add "Install Anaconda" to your documentation. Depends on what you think is a good use of people's time.

    You WANT to use numeric and other libraries. You want them to come in plug and play form. Just as in C++ I wanted to use the STL, rather than roll my own libraries, or use a patchwork of multiple bespoke packages from different sources.


Distraction Deterrents in Small Contexts

"distracted from distraction by distraction" - T.S. Eliot I've been reading a little on how Facebook and other social netwo...