Another test failure on 0.5.0 (due to sporadic numerical error?)

[jakirkham]

Looks either like a numerical error or too tight of a constraint. This likely results in sporadic failures due to some sort of round off error. There may also be a question of numerical stability here. Not sure. Have not looked close enough to say either way. Here is the error. Using NumPy 1.10.4, Python 2.7.11, gcc 4.8.5.

python /zopt/conda2/conda-bld/work/cmt-0.5.0/code/cmt/python/tests/mcgsm_test.py
....F.......
======================================================================
FAIL: test_gradient (__main__.Tests)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/zopt/conda2/conda-bld/work/cmt-0.5.0/code/cmt/python/tests/mcgsm_test.py", line 270, in test_gradient
    self.assertLess(err, 1e-8)
AssertionError: 3.513801324170163e-08 not less than 1e-08

----------------------------------------------------------------------
Ran 12 tests in 1.064s

FAILED (failures=1)

[FlorianFranzen]

Is this fixed in the latest release v0.6.0?

[jakirkham]

I have seen sporadic failures due to this test in the latest version as well.