TST: Erratic test failure in test_cossin_separate

[endolith]

Intermittent test failure during https://github.com/scipy/scipy/pull/4607/checks?check_run_id=1507492909

It seems like every time I push, something irrelevant fails somewhere. So I guess I should report each of these so the tests can be made more deterministic or find out what is causing the occasional failures?

Error message:

=================================== FAILURES ===================================
________________________ test_cossin_separate[float64] _________________________
scipy/linalg/tests/test_decomp_cossin.py:150: in test_cossin_separate
    assert_allclose(u1_2, u1, rtol=0., atol=10*np.finfo(dtype_).eps)
E   AssertionError: 
E   Not equal to tolerance rtol=0, atol=2.22045e-15
E   
E   Mismatched elements: 4514 / 6400 (70.5%)
E   Max absolute difference: 0.80714081
E   Max relative difference: 2.
E    x: array([[-0.012286, -0.063705, -0.046882, ...,  0.067121,  0.042775,
E           -0.038814],
E          [ 0.189033,  0.030199,  0.020188, ..., -0.041924,  0.135174,...
E    y: array([[ 0.012286,  0.063705, -0.046882, ...,  0.067121, -0.042775,
E            0.038814],
E          [-0.189033, -0.030199,  0.020188, ..., -0.041924, -0.135174,...
        X          = array([[-0.01224596,  0.11111008, -0.05583939, ...,  0.00863433,
        -0.02529069, -0.05999699],
       [-0.0022407...007,  0.07247924],
       [ 0.03405866, -0.05719033,  0.02516655, ...,  0.02967258,
         0.04382338,  0.01344216]])
        _          = 0
        dlw        = <fortran object>
        drv        = <fortran object>
        dtype_     = <class 'numpy.float64'>
        lwval      = 10780
        lwvals     = {'lwork': 10780}
        m          = 250
        p          = 80
        pfx        = 'or'
        q          = 170
        theta      = array([0.00541383, 0.00856157, 0.02748419, 0.02978311, 0.0718637 ,
       0.07503233, 0.07906807, 0.10008097, 0.115568...54, 0.99574407, 1.01896566, 1.02267398, 1.04836181,
       1.05885172, 1.09333057, 1.09419429, 1.13869658, 1.18214442])
        theta2     = array([0.00541383, 0.00856157, 0.02748419, 0.02978311, 0.0718637 ,
       0.07503233, 0.07906807, 0.10008097, 0.115568...54, 0.99574407, 1.01896566, 1.02267398, 1.04836181,
       1.05885172, 1.09333057, 1.09419429, 1.13869658, 1.18214442])
        u1         = array([[ 0.01228641,  0.06370488, -0.0468819 , ...,  0.06712145,
        -0.04277498,  0.03881356],
       [-0.1890334...168,  0.16848187],
       [-0.04684342,  0.05722923, -0.11347679, ..., -0.03189451,
        -0.1163678 , -0.03369047]])
        u1_2       = array([[-0.01228641, -0.06370488, -0.0468819 , ...,  0.06712145,
         0.04277498, -0.03881356],
       [ 0.1890334...168, -0.16848187],
       [ 0.04684342, -0.05722923, -0.11347679, ..., -0.03189451,
         0.1163678 ,  0.03369047]])
        u2         = array([[ 0.09401129,  0.09900626, -0.07349002, ...,  0.10217532,
        -0.0603292 ,  0.0933257 ],
       [-0.1354097...708, -0.01956046],
       [ 0.01315501, -0.09415009,  0.01847025, ..., -0.10221244,
         0.06842986,  0.55079481]])
        u2_2       = array([[-0.09401129, -0.09900626, -0.07349002, ...,  0.10217532,
        -0.0603292 ,  0.0933257 ],
       [ 0.1354097...708, -0.01956046],
       [-0.01315501,  0.09415009,  0.01847025, ..., -0.10221244,
         0.06842986,  0.55079481]])
        v1t        = array([[ 0.02071111, -0.12575399, -0.03392104, ...,  0.0561229 ,
         0.00561223,  0.10164257],
       [-0.0456795...206,  0.02532954],
       [-0.00360381, -0.02058857,  0.0545874 , ..., -0.0498076 ,
        -0.06530985,  0.11992337]])
        v1t_2      = array([[-0.02071111,  0.12575399,  0.03392104, ..., -0.0561229 ,
        -0.00561223, -0.10164257],
       [ 0.0456795...206,  0.02532954],
       [-0.00360381, -0.02058857,  0.0545874 , ..., -0.0498076 ,
        -0.06530985,  0.11992337]])
        v2t        = array([[-0.18492015, -0.12474194,  0.07637733, ...,  0.0724234 ,
         0.02643485,  0.16669119],
       [ 0.1832701...928,  0.17164494],
       [-0.12861266,  0.02723643,  0.05142215, ..., -0.15568402,
        -0.10115197, -0.03535843]])
        v2t_2      = array([[ 0.18492015,  0.12474194, -0.07637733, ..., -0.0724234 ,
        -0.02643485, -0.16669119],
       [-0.1832701...928, -0.17164494],
       [ 0.12861266, -0.02723643, -0.05142215, ...,  0.15568402,
         0.10115197,  0.03535843]])
=========================== short test summary info ============================
FAILED scipy/linalg/tests/test_decomp_cossin.py::test_cossin_separate[float64]
= 1 failed, 30316 passed, 2066 skipped, 10992 deselected, 77 xfailed, 5 xpassed in 622.38s (0:10:22) =

   Normal return from subroutine COBYLA

   NFVALS =   50   F = 2.485185E+01    MAXCV = 1.999965E-10
   X = 4.955358E+00   6.666553E-01

 NNLS quitting on iteration count.

Scipy/Numpy/Python version information:

Nightly CPython (3.9)

[ilayn]

Yeah @WarrenWeckesser already brought this in #11524 but then I forgot. I'll take a look

[ilayn]

It looks like we didn't set the random seed in the test and hence the intermittent failures. I'll send a PR now