Garnet regression errors

[agitter]

When testing #29, I noticed that our Garnet integration tests fail with newer versions of SciPy and NumPy. There is a divide by 0 error during the linear regression. We should either explore why that happens or catch the error more gracefully.

[sgosline]

Do you have some code/data that reproduces this error?

[agitter]

Our Garnet test case can reproduce the error on recent SciPy and NumPy versions. This runs Garnet on the FOXA1 ChIP data in the a549 example directory and confirms that the output files match what we expect. The error comes from motif_regression.py, so we should be able to test directly on that call using the input files events_to_genes_with_motifs.pkl from tests/integration_test_standard/ and Tgfb_exp.txt from the example directory.

The Travis build logs in #29 show some SciPy and NumPy versions that cause the error.

• SciPy 1.0.0 and NumPy 1.14.0 failed.
• SciPy 0.19.1 and NumPy 1.14.0 failed.
• SciPy 0.19.1 and NumPy 1.13.3 worked.

[AmandaKedaigle]

I believe I ran into this problem before - I'll see if I can figure out what I did back then to get around it. (it might just have been changing/updating numpy). But definitely agree we should catch the error

[sgosline]

Just checking on this too - I'm just getting a warning, not an error. But I haven't run all the tests. We should maybe just update the docs - we don't support Python 3.x at all either, because of the print statements.