Hacker News Reader

Comment by mckoss | Hacker News Reader

Why is the IPython and scientific package installation always such a nightmare?

I am very impressed with Anaconda's Free edition. Works well in Linux and has more packages [2] than Enthought's free edition. Anaconda is the commercial offering from Travis Oliphant, the creator of Numpy.

[1]: https://store.continuum.io/cshop/anaconda/

[2]: Packages: http://docs.continuum.io/anaconda/pkgs.html

Because some people like to use operating systems that don't have software installation and maintenance figured out.

    apt-get install ipython

Done.

That doesn't solve the problem. Here's another version:

Why is installing a specific version of a package you fancy (e.g the latest or some previous) a nightmare?

And there "apt-get" doesn't help you much.

    pip install mypackage==1.2.3

I believe the general formula is apt-get for python and python-pip, then use pip for everything past that point.

And in general, virtualenvs are the right way to go for anything other than throwaway scripts. I've got a global numpy/scipy/requests install so that I can just do `ipython -pylab` and start getting work done. But real project code happens in a virtualenv.

On a mac anaconda works pretty well. I sent a link to a marketer with minimal python experience and she was up and running in under an hour. (Admittedly, she's rather more adept than a normal marketer...)

First of all, you can use APT to install older versions. Second, you can always make up a case where the packages in the repos don't do 100% what you desire (maybe you want some different build options, or dependency versions, etc). The packages in the repos are designed to cover most cases, and to work together with the other packages on your system. So I have to ask: are you SURE you need something the packages don't give you? If so, then you have agreed to take on the extra work to get what you want, and it's understood that it may/may not be trivial. This applies to any software.

Even so, APT does help you quite a bit with build dependencies, so whatever non-standard thing you've decided to do, you can still do it with (hopefully) not too much work. Feel free to ask me about any specific thing you're having trouble with.

If you have trouble with it, you might prefer to try Enthought Canopy. I've used it a bit, and install is a breeze. I've not figured out how to install outside packages yet, but the actual base install is very simple and includes most of the scipy packages.

Just to note also: I'm completely unaffiliated with Enthought except that I have a registered account with them for occasional feedback.

Anaconda is another option. Comes with almost everything I need and you can use pip to install the few things it doesn't come with

My goto recommendation for a fresh installation: http://fonnesbeck.github.io/ScipySuperpack/

For the same reasons any set of non trivial softwares is: flexibility and # options means many things can go wrong.

Solutions like Enthought Canopy, superpack or Anaconda exist to solve that exact problem (disclaimer: I work for Enthought).

Anaconda and Python(x,y) on windows help out a ton. Similarly for Homebrew on OSX. On Linux, if you have a package manager you're probably good to go.

Because you are not using Anaconda...

Making some overgeneralizations here, software written by scientists aren't going to be as good as software written by software engineers. Their usual purpose (and their training/background) is to publish papers and writing code is just a means to an end.

Reply on news.ycombinator.com