'pip install seaborn' doesn't auto-install dependencies about seaborn HOT 9 CLOSED

Yeah this came up in #5. Do you have a pointer for how to get that to work?

from seaborn.

I'm not a packaging expert, but some searching lead me to believe that adding some install_requires=... lines to your setup.py might solve the issue.

See this for reference:
http://stackoverflow.com/questions/6947988/when-to-use-pip-requirements-file-versus-install-requires-in-setup-py

from seaborn.

Yeah, my understanding was that install_requirements triggers a failure when you try to pip install without everything listed, but not that it installs the dependencies for you? I have to look into it more closely, though.

from seaborn.

install_requires would normally work, but since the requirements here (e.g. statsmodels) don't specify their own requirements with this flag, requirements are installed in the wrong order. Will see if there's a workaround.

from seaborn.

It looks like there used to be some other commands like setup_requires, but the following page says that install_requires will download and build the packages.

http://peak.telecommunity.com/DevCenter/setuptools#declaring-dependencies

from seaborn.

Oh geeze, I didn't even think about how complex it would be trying to auto-install all of numpy / scipy / statsmodels / matplotlib etc.... that is definitely a serious undertaking.

What I was imagining was something a bit less ambitions: using install_requires only for packages that aren't part of the standard scientific python stack (which is usually installed already via Anaconda etc.)

Specifically, husl and moss are two that should (?) be relatively easy to get working with install_requires, and it I think it would solve this issue for the vast majority of people without requiring you to package the whole stack.

Perhaps scikit-learn could also be added to those two.

from seaborn.

My understanding is that scikit-learn is among the most difficult packages to auto-install. But there's only a single function (in moss) that depends on it, so I was actually probably going to move that function elsewhere to remove the scikit-learn dependency in seaborn.

I think husl is a single file, so it might be reasonable to just package that with seaborn. I don't think the author updates it very often.

from seaborn.

Ahh, well I take back what I said about scikit-learn; I've never tried to install that one myself.

Either auto-installing moss and husl, or removing dependencies and doing what you suggest both seem like workable solutions.

from seaborn.

OK I have to admit that was considerably easier than I thought it would be. To test it, I ran (from the source directory):

conda create -n seaborn_dep_test anaconda
pip install .
make test

Which ran with no errors. I think this is good to go, thanks for hounding me!

from seaborn.