covmat regularization test fails

[wilsonmr]

I think there are two issues here. One is that the random matrix I picked to be part of the test has a zero eigenvalue which was not very sensible. The second is that the procedure should probably still be able to handle this anyway

whilst I'm fixing this I can also add np.assert_allclose as suggested in #749

see #651 and #749 for tests failing on some machines

cc @tgiani @scarlehoff @Zaharid

[tgiani]

@wilsonmr the thing I don t understand is that it looks to me that the procedure is actually able to handle the fact of having a zero eigenvalue..not sure why this is machine dependent

[wilsonmr]

well I think, but need to check, that it can handle a zero eigenvalue provided that it's like ~ +0 so that 1/+0 ~ +inf

but seeing as we get numeric instability around zero eigenvalue they can be +/- 0 and if the machine you're on comes out with -0 then the algorithm makes some change to the matrix which is measurable