Travis builds are slow

[averagehat]

Options as I see them:

We could support both pip- and conda- style installations for bio_pieces, but use conda for the travis builds (you could also have another travis build script that did a traditional pip install which you run every once in a while). User's at WRAIR can continue to use the pip-style install. This would, however, make installing pathdiscov more painful. so:

1. Move group_references into a separate project, or into pathdiscov itself (like originally)
2. Separate requirements.txt files/Optional build structure? (aka, don't install make_pca by default). Not sure how this would be done.
3. Change pathdiscov to use conda.

Note: scikit-bio does not work on windows (not that we have any windows users)

[necrolyte2]

Another option is to just to put in the documentation for installation a piece about scikit-bio as well as make_pca.

Anywhere that scikit-bio is imported would then have a try,except block that exits with an error if it is not installed and explains that it has to be installed manually.

[averagehat]

Agreed. I'll put that in place.

[necrolyte2]

Example to use mini-conda: https://gist.github.com/dan-blanchard/7045057

[necrolyte2]

I'll probably work on this today if #82 works out fine

matplolib and pandas also slow down the build quite a bit so would like to make it all faster so we can get results back quicker

[averagehat]

We excluded scikit-bio from the requirements file so this is no longer tested and no longer slow

[averagehat]

reopening as we are going to convert the build to miniconda

[necrolyte2]

Ya, if you look at RTD builds you will see them failing as well due to slow builds. I think it was mostly numpy, but ya