Change bare asserts to numpy asserts.
Change bare asserts to numpy asserts
Coverage remained the same when pulling 7374fbb on kshedden:cor_nearest_maint into 48f7a21 on statsmodels:master.
assert_allclose allows better control over tolerance
i.e, atol=5e-3 and rtol=0 or irrelevant in this case
I think assert_almost_equal is now only interesting in the case when we have reference numbers with a printed precision given by the number of decimals.
I'm merging this to see what the results on Ubuntu are.
Trying out the test case on my computer, the tests would pass at atol=5e-5.
If we loose a lot of precision on some machines, then there might be some other problems in the code.