FAIL: Test median. with numpy-1.10

[jlec]

======================================================================
FAIL: Test median.
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/usr/lib64/python3.4/site-packages/nose/case.py", line 198, in runTest
    self.test(*self.arg)
  File "/var/tmp/portage/dev-python/bottleneck-1.0.0/work/Bottleneck-1.0.0-python3_4/lib/bottleneck/tests/reduce_test.py", line 95, in unit_maker
    assert_array_equal(actual, desired, err_msg)
  File "/usr/lib64/python3.4/site-packages/numpy/testing/utils.py", line 782, in assert_array_equal
    verbose=verbose, header='Arrays are not equal')
  File "/usr/lib64/python3.4/site-packages/numpy/testing/utils.py", line 708, in assert_array_compare
    raise AssertionError(msg)
AssertionError: 
Arrays are not equal

func median | input a0 (float64) | shape (0,) | axis -1

Input array:
[]

(mismatch 100.0%)
 x: array(nan)
 y: array('Crashed', 
      dtype='<U7')

----------------------------------------------------------------------
Ran 79 tests in 129.634s

FAILED (failures=1)
Running unit tests for bottleneck
NumPy version 1.10.1
NumPy relaxed strides checking option: True
NumPy is installed in /usr/lib64/python3.4/site-packages/numpy
Python version 3.4.3 (default, Aug 21 2015, 10:51:23) [GCC 4.9.3]
nose version 1.3.7

I haven't tested it with numpy-1.9 again but I really suspect numpy here.

[jlec]

Same problem with numpy-1.9.3

[kwgoodman]

I installed numpy 1.10.1 and see the same unit test failure. I don't see the unit test failure with numpy 1.9.1. Can you double-check that you are seeing the unit test failure with numpy 1.9.1?

Yeah, you're right. We'll have to ask numpy about this:

In [1]: a = np.zeros((0,))
In [2]: bn.median(a)
Out[2]: nan
In [3]: np.median(a)
<snip>
IndexError: index -1 is out of bounds for axis 0 with size 0

At the very least, the numpy error message is wrong.

[kwgoodman]

Bug has already been caught in numpy: numpy/numpy#6462

Looks like it will be fixed in numpy 1.10.2.

[jlec]

Still see it with numpy-1.9.1

[kwgoodman]

Can you post the output of the bottleneck unit tests with numpy 1.9.1 like you did with numpy 1.10.1?

[jlec]

======================================================================
ERROR: Test rankdata.
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/usr/lib64/python3.4/site-packages/nose/case.py", line 198, in runTest
    self.test(*self.arg)
  File "/var/tmp/portage/dev-python/bottleneck-1.0.0/work/Bottleneck-1.0.0-python3_4/lib/bottleneck/tests/list_input_test.py", line 38, in unit_maker
    desired = func0(*argsi)
  File "/var/tmp/portage/dev-python/bottleneck-1.0.0/work/Bottleneck-1.0.0-python3_4/lib/bottleneck/slow/nonreduce_axis.py", line 14, in rankdata
    from scipy.stats import rankdata as imported_rankdata
  File "/usr/lib64/python3.4/site-packages/scipy/stats/__init__.py", line 321, in <module>
    from .stats import *
  File "/usr/lib64/python3.4/site-packages/scipy/stats/stats.py", line 180, in <module>
    import scipy.special as special
  File "/usr/lib64/python3.4/site-packages/scipy/special/__init__.py", line 601, in <module>
    from ._ufuncs import *
  File "__init__.pxd", line 155, in init scipy.special._ufuncs (scipy/special/_ufuncs.c:28225)
ValueError: numpy.dtype has the wrong size, try recompiling

======================================================================
FAIL: Test rankdata.
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/usr/lib64/python3.4/site-packages/nose/case.py", line 198, in runTest
    self.test(*self.arg)
  File "/var/tmp/portage/dev-python/bottleneck-1.0.0/work/Bottleneck-1.0.0-python3_4/lib/bottleneck/tests/reduce_test.py", line 95, in unit_maker
    assert_array_equal(actual, desired, err_msg)
  File "/usr/lib64/python3.4/site-packages/numpy/testing/utils.py", line 739, in assert_array_equal
    verbose=verbose, header='Arrays are not equal')
  File "/usr/lib64/python3.4/site-packages/numpy/testing/utils.py", line 665, in assert_array_compare
    raise AssertionError(msg)
AssertionError:
Arrays are not equal

func rankdata | input a0 (float64) | shape (0,) | axis -1

Input array:
[]

(mismatch 100.0%)
 x: array([], dtype=float64)
 y: array('Crashed',
      dtype='<U7')

----------------------------------------------------------------------
Ran 79 tests in 91.001s

FAILED (errors=1, failures=1)
Running unit tests for bottleneck
NumPy version 1.9.1
NumPy is installed in /usr/lib64/python3.4/site-packages/numpy
Python version 3.4.3 (default, Aug 21 2015, 10:51:23) [GCC 4.9.3]
nose version 1.3.7

[jlec]

Looking at the error message it seems it is a scipy problem when not recompiled after downgrade.

[kwgoodman]

Oh, OK, that is probably it. Also it is for a different function, rankdata, instead of median.

Thanks for reporting. I hadn't tried numpy 1.10.1 until you reported the problem.

[jlec]

Bug has already been caught in numpy: numpy/numpy#6462

Backporting this commit makes this bug go away.

[kwgoodman]

Good check. OK, we'll sit tight until np 1.10.2 is released.

[kwgoodman]

No unit test failures with np 1.10.4. I'm closing.