failing test_moments after centered moving window introduced

I'm getting sporadic failures in
FAIL: test_rolling_std (pandas.stats.tests.test_moments.TestMoments)
FAIL: test_rolling_sum (pandas.stats.tests.test_moments.TestMoments)
FAIL: test_rolling_kurt (pandas.stats.tests.test_moments.TestMoments)
FAIL: test_rolling_max (pandas.stats.tests.test_moments.TestMoments)
and related

git bisected to: 915d261

I've upgraded to statsmodels '0.5.0.dev-c9062e4', did a clean/develop
but am still getting failures.

any ideas? @changhiskhan

[changhiskhan]

Do you have the failing output handy? I'll take a look, thanks.

On Tue, Dec 11, 2012 at 10:16 AM, y-p notifications@github.com wrote:

I'm getting sporadic failures in
FAIL: test_rolling_std (pandas.stats.tests.test_moments.TestMoments)
FAIL: test_rolling_sum (pandas.stats.tests.test_moments.TestMoments)
FAIL: test_rolling_kurt (pandas.stats.tests.test_moments.TestMoments)
FAIL: test_rolling_max (pandas.stats.tests.test_moments.TestMoments)
and related

git bisected to: 915d261915d261d298c54e887

I've upgraded to statsmodels '0.5.0.dev-c9062e4', did a clean/develop
but am still getting failures.

any ideas? @changhiskhan https://github.com/changhiskhan

—
Reply to this email directly or view it on GitHubhttps://github.com//issues/2490.

======================================================================
FAIL: test_rolling_count (pandas.stats.tests.test_moments.TestMoments)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 46, in test_rolling_count
    fill_value=0)
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 320, in _check_moment_func
    fill_value=fill_value)
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 380, in _check_ndarray
    assert_almost_equal(result[1], expected[10])
  File "/home/user1/src/pandas/pandas/util/testing.py", line 124, in assert_almost_equal
    1, a / b, decimal=5, err_msg=err_msg(a, b), verbose=False)
  File "/usr/local/lib/python2.7/dist-packages/numpy/testing/utils.py", line 468, in assert_almost_equal
    raise AssertionError(msg)
AssertionError: 
Arrays are not almost equal to 5 decimals expected 13.00000 but got 1.00000

======================================================================
FAIL: test_rolling_mean (pandas.stats.tests.test_moments.TestMoments)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 49, in test_rolling_mean
    self._check_moment_func(mom.rolling_mean, np.mean)
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 320, in _check_moment_func
    fill_value=fill_value)
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 387, in _check_ndarray
    self.assert_(np.isnan(result[14]))
AssertionError: False is not true

======================================================================
FAIL: test_rolling_min (pandas.stats.tests.test_moments.TestMoments)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 173, in test_rolling_min
    self._check_moment_func(mom.rolling_min, np.min)
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 320, in _check_moment_func
    fill_value=fill_value)
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 387, in _check_ndarray
    self.assert_(np.isnan(result[14]))
AssertionError: False is not true

======================================================================
FAIL: test_rolling_std (pandas.stats.tests.test_moments.TestMoments)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 235, in test_rolling_std
    lambda x: np.std(x, ddof=1))
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 320, in _check_moment_func
    fill_value=fill_value)
  File "/home/user1/src/pandas/pandas/stats/tests/test_moments.py", line 387, in _check_ndarray
    self.assert_(np.isnan(result[14]))
AssertionError: False is not true

----------------------------------------------------------------------

[wesm]

I have no idea and I can't reproduce (it doesn't seem to have anything to do with the scikits.timeseries testing dependency introduced). I'm going to cut the beta and let you guys debug

can either of you post output of updated ci/print_versions.py from here

[wesm]

INSTALLED VERSIONS
------------------
Python: 2.7.2.final.0
Cython: 0.17.2
Numpy: 1.6.1
Scipy: 0.10.0
statsmodels: 0.4.3
scikits.timeseries: 0.91.3
dateutil: 2.1
pytz: 2012d
PyTables: 2.3.1
matplotlib: 1.2.0
openpyxl: 1.6.1
xlrd: 0.7.1
xlwt: 0.7.2
sqlalchemy: 0.7.1


have stats

[changhiskhan]

INSTALLED VERSIONS

Python: 2.7.3.final.0
Cython: 0.18-pre
Numpy: 1.6.1
Scipy: 0.10.1
statsmodels: 0.5.0.dev-7720064
scikits.timeseries: 0.91.3
dateutil: 1.5
pytz: 2011n
PyTables: 2.3.1
matplotlib: 1.1.0
openpyxl: 1.6.1
xlrd: 0.7.6
xlwt: 0.7.3
sqlalchemy: 0.7.6

have stats

yeah, nothing there.

INSTALLED VERSIONS
------------------
Python: 2.7.3.candidate.2
Cython: 0.17.2
Numpy: 1.6.2
Scipy: 0.10.1
statsmodels: 0.5.0.dev-c9062e4
scikits.timeseries: 0.91.3
dateutil: 1.5
pytz: 2012d
PyTables: 2.4.0
matplotlib: 1.1.1rc2
openpyxl: 1.5.8
xlrd: 0.6.1
xlwt: 0.7.4
sqlalchemy: 0.7.8

tried in a virtualenv, and also a fresh clone - still failing.
no failures on travis either. hrmm.

[changhiskhan]

When you say "sporadic", how often is it occurring? Are you running the multi-processing test suite or regular?

test_fast or nosetests pandas.stats.tests.test_moment both produce failures.
it fails almost always, but which specific tests fail changes between runs, ocassionaly
it passes without failure.

here's a break on failed comparison in from test_Rolling_median->check_moment_func->check_ndarray:

> /home/user1/src/pandas/pandas/stats/tests/test_moments.py(387)_check_ndarray()
-> self.assert_(np.isnan(result[14]))
(Pdb) l
382                     self.assert_(np.isnan(result[-9:]).all())
383                 else:
384                     self.assert_((result[-9:] == 0).all())
385                 if has_min_periods:
386                     self.assert_(np.isnan(expected[23]))
387  ->                 self.assert_(np.isnan(result[14]))
388                     self.assert_(np.isnan(expected[-5]))
389                     self.assert_(np.isnan(result[-14]))
390     
391         def _check_structures(self, func, static_comp,
392                               has_min_periods=True, has_time_rule=True,
(Pdb) result
array([        nan,         nan,         nan,         nan,  0.37457507,
        0.30156993,  0.22856479,  0.30628624,  0.22856479,         nan,
               nan,         nan,         nan,  0.37457507,  0.30156993,
        0.22856479,  0.30628624,  0.22856479,         nan,         nan,
               nan,         nan,  0.37457507,  0.30156993,  0.22856479,
        0.30628624,  0.22856479,         nan,         nan,         nan,
               nan,  0.37457507,  0.30156993,  0.22856479,  0.30628624,
        0.22856479,         nan,         nan,         nan,         nan,
               nan,         nan,         nan,         nan,         nan,
               nan,         nan,         nan,         nan,         nan])
(Pdb) expected
array([        nan,         nan,         nan,         nan,         nan,
               nan,         nan,         nan,         nan,         nan,
               nan,         nan,         nan,         nan,         nan,
               nan,         nan,         nan,         nan,         nan,
               nan,         nan,         nan,         nan,  0.06942868,
        0.05451945,  0.06942868,  0.14899674,  0.22856479,  0.30156993,
        0.14899674,  0.30156993,  0.30156993,  0.30156993,  0.14899674,
        0.30156993,  0.30156993,  0.37929138,  0.37929138,  0.37929138,
        0.37457507,  0.30156993,  0.22856479,  0.30628624,  0.22856479,
               nan,         nan,         nan,         nan,         nan])

[changhiskhan]

Try replacing _center_window in moments.py with this:

def _center_window(rs, window, axis):
    offset = int((window - 1) / 2.)
    if isinstance(rs, (Series, DataFrame, Panel)):
        rs = rs.shift(-offset, axis=axis)
    else:
        rs_indexer = [slice(None)] * rs.ndim
        rs_indexer[axis] = slice(None, -offset)

        lead_indexer = [slice(None)] * rs.ndim
        lead_indexer[axis] = slice(offset, None)

        na_indexer = [slice(None)] * rs.ndim
        na_indexer[axis] = slice(-offset, None)

        rs[tuple(rs_indexer)] = rs[tuple(lead_indexer)]
        rs[tuple(na_indexer)] = np.nan
    return rs

no effect.

[changhiskhan]

The result looks completely mangled. center=True should just be a shift operation. If you look at moments.py, the only difference is that _center_window is called after computing the results.

Can you pdb into _center_window and see what's going on?

rs at entry to center_window in test_rolling_apply

[        nan         nan         nan         nan         nan         nan
         nan         nan         nan         nan         nan         nan
         nan         nan         nan         nan         nan         nan
         nan         nan         nan         nan         nan         nan
  0.21432634  0.25726326  0.26078034  0.31504732  0.24231212  0.24028765
  0.28524033  0.25578755  0.22841107  0.18792591  0.19638916  0.25554294
  0.32495223  0.14671714  0.12936415  0.15640616  0.219897    0.26083415
  0.2166916   0.19919935  0.22116124         nan         nan         nan
         nan         nan]
n

rs returned from center_window in test_rolling_apply

[        nan         nan         nan         nan  0.219897    0.26083415
  0.2166916   0.19919935  0.22116124         nan         nan         nan
         nan  0.219897    0.26083415  0.2166916   0.19919935  0.22116124
         nan         nan         nan         nan  0.219897    0.26083415
  0.2166916   0.19919935  0.22116124         nan         nan         nan
         nan  0.219897    0.26083415  0.2166916   0.19919935  0.22116124
         nan         nan         nan         nan         nan         nan
         nan         nan         nan         nan         nan         nan
         nan         nan]

is this the proper behaviour for shift?

mkdf(10,1,r_idx_type='dt')
Out[10]: 
C0         C_l0_g0
R0                
2000-01-03    R0C0
2000-01-04    R1C0
2000-01-05    R2C0
2000-01-06    R3C0
2000-01-07    R4C0
2000-01-10    R5C0
2000-01-11    R6C0
2000-01-12    R7C0
2000-01-13    R8C0
2000-01-14    R9C0

a=mkdf(10,1,r_idx_type='dt')
a.shift()
Out[12]: 
C0         C_l0_g0
R0                
2000-01-03     NaN
2000-01-04    R0C0
2000-01-05    R1C0
2000-01-06    R2C0
2000-01-07    R3C0
2000-01-10    R4C0
2000-01-11    R5C0
2000-01-12    R6C0
2000-01-13    R7C0
2000-01-14    R8C0

[changhiskhan]

yes, but _center_window implements shift for ndarrays separately.

take a look at the following:

rs[tuple(rs_indexer)]
rs[tuple(na_indexer)]
rs[tuple(lag_indexer)]

And also what each *_indexer is