Repository navigation
travis failures as scipy circmean no longer equal to astropy's #5198
Description
Recent travis builds are failing on test_circstats.test_circmean_against_scipy. This may be because scipy has changed how its version works (I haven't investigated, but there was a recent scipy release).
@mirca - could you check what is happening? Ideally, I think the test should be rewritten so that it works for all versions. I would also suggest not to use random input data.
Activity
mirca commentedon Jul 27, 2016
@mhvk Sure! I'm gonna look into it this afternoon
bsipocz commentedon Jul 27, 2016
This is an upstream scipy bug, and they are working on it. scipy/scipy#6420
I would simply suggest to limit the scipy version to <0.18 for now, it's easy for CI, but maybe also for the code, threat 0.18 scipy the same as a missing scipy.
mhvk commentedon Jul 27, 2016
My suggestion would be to rewrite the test so that it uses input that is known to work for all versions of scipy.
bsipocz commentedon Jul 27, 2016
@mhvk - But then it would still be buggy from the user side. Maybe emit a warning if they're using scipy 0.18?
mhvk commentedon Jul 27, 2016
@bsipocz - why would it be buggy? astropy's version of the statistics made by @mirca does not have these problems.
bsipocz commentedon Jul 27, 2016
Oh, ignore me then. I blindly assumed that scipy is used, but now I see that we are only testing against it, sorry about it. @mirca for the win.