ubuntu 12.04 python3.2 test_cummethods_bool: AssertionError: (very low values) expected 1.00000 but got 0.00000

[yarikoptic]

that is 0.13.1

======================================================================
FAIL: test_cummethods_bool (pandas.tests.test_series.TestSeries)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/tmp/buildd/pandas-0.13.1/debian/tmp/usr/lib/python3/dist-packages/pandas/tests/test_series.py", line 5202, in test_cummethods_bool
    assert_series_equal(res, expecteds[method])
  File "/tmp/buildd/pandas-0.13.1/debian/tmp/usr/lib/python3/dist-packages/pandas/util/testing.py", line 458, in assert_series_equal
    assert_almost_equal(left.values, right.values, check_less_precise)
  File "testing.pyx", line 58, in pandas._testing.assert_almost_equal (pandas/src/testing.c:2561)
  File "testing.pyx", line 93, in pandas._testing.assert_almost_equal (pandas/src/testing.c:1803)
  File "testing.pyx", line 135, in pandas._testing.assert_almost_equal (pandas/src/testing.c:2342)
AssertionError: (very low values) expected 1.00000 but got 0.00000

INSTALLED VERSIONS
------------------
commit: None
python: 3.2.3.final.0
python-bits: 32
OS: Linux
OS-release: 3.2.0-4-amd64
machine: x86_64
processor: x86_64
byteorder: little
LC_ALL: C
LANG: C

pandas: 0.13.1
Cython: None
numpy: 1.6.1
scipy: 0.9.0
statsmodels: None
IPython: None
sphinx: 1.1.3
patsy: None
scikits.timeseries: None
dateutil: 2.0
pytz: 2012c
bottleneck: None
tables: None
numexpr: None
matplotlib: None
openpyxl: None
xlrd: None
xlwt: None
xlsxwriter: None
sqlalchemy: None
lxml: None
bs4: 4.0.2
html5lib: None
bq: None
apiclient: None

[jreback]

how do you even install numpy 1.6.1 with 3.2....fails with http://comments.gmane.org/gmane.comp.python.numeric.general/43801

and can't reprod with 1.6.2

[yarikoptic]

isn't Debian magic indeed? ;) I install it as usual with the same old 'apt-get install python3-numpy' ;-) here it is in precise
http://packages.ubuntu.com/precise/python3-numpy
and as you can see carries a set of patches on top of stock 1.6.1 to actually make it work: http://changelogs.ubuntu.com/changelogs/pool/main/p/python-numpy/python-numpy_1.6.1-6ubuntu1/changelog

if you were to follow http://neuro.debian.net/blog/2012/2012-04-14_ndtools.html then you could easily sudo nd_login nd+ubuntu precise amd64 $PWD to get into a clean cow chroot to then apt-get build-dep python-pandas to get all needed build-depends for pandas and reproduce it to full extent ;-)

[yarikoptic]

heh -- thought to make a step forward and just provide precise chroot on that sparc box, but then recalled that canonical doesn't bother with sparc... heh

[jreback]

i have a debian box...but easiest if I can just do it in a virtual env....let me see

[jreback]

@yarikoptic still appearing in master?

[jreback]

@yarikoptic going to close

pls report any new issues you see

[yarikoptic]

finally got to rebuild 0.14.0+git17-g3849d5d-1 and the issue is there on 12.04 ubuntu... will upload without a build for 12.04 for now

======================================================================
FAIL: test_cummethods_bool (pandas.tests.test_series.TestSeries)
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/tmp/buildd/pandas-0.14.0+git17-g3849d5d/debian/tmp/usr/lib/python3/dist-packages/pandas/tests/test_series.py", line 5436, in test_cummethods_bool
    assert_series_equal(res, expecteds[method])
  File "/tmp/buildd/pandas-0.14.0+git17-g3849d5d/debian/tmp/usr/lib/python3/dist-packages/pandas/util/testing.py", line 532, in assert_series_equal
    assert_almost_equal(left.values, right.values, check_less_precise)
  File "testing.pyx", line 58, in pandas._testing.assert_almost_equal (pandas/src/testing.c:2550)
  File "testing.pyx", line 93, in pandas._testing.assert_almost_equal (pandas/src/testing.c:1873)
  File "testing.pyx", line 131, in pandas._testing.assert_almost_equal (pandas/src/testing.c:2319)
AssertionError: (very low values) expected 1.00000 but got 0.00000, with decimal 5

[jreback]

@yarikoptic if you could report any error on current master would be gr8.

[yarikoptic]

Traveling atm, will try to not forget to do in the evening when at the hotel again

On July 2, 2014 8:58:10 AM EDT, jreback [email protected] wrote:

@yarikoptic if you could report any error on current master would be
gr8.

---

Reply to this email directly or view it on GitHub:
#6270 (comment)

[yarikoptic]

still the case with 0.14.0+git345-g8cd3dd6-1

[jreback]

can you show me: res, expecteds[methods], method, e.g. this is a loop of 4 tests, need to know which/where is failing

also tell me if something like this fixes:

the expected is somethign like:

np.cumsum(s.values)

does np.cumsum(s.values,dtype='int64') fix?

[yarikoptic]

need to run, but got to the failure point:

(Pdb) up
> /tmp/buildd/pandas-0.14.0+git345-g8cd3dd6/debian/tmp/usr/lib/python3/dist-packages/pandas/tests/test_series.py(5572)test_cummethods_bool()
-> assert_series_equal(res, expecteds[method])
(Pdb) print res
0    False
1    False
2      NaN
3     -inf
dtype: object
(Pdb) print method
'cummax'
(Pdb) print expecteds[method]
0    False
1     True
2      NaN
3     True
dtype: object
(Pdb) print np.cumsum(s.values)
array([1, 2, 3, 4, 5, 6, 7])
(Pdb) print np.cumsum(s.values,dtype='int64')
array([1, 2, 3, 4, 5, 6, 7])

[jreback]

I cannot for somereason build 3.2 with 1.6.1 (and its on 32-bit). I think their might be a bug in numpy in np.maximum.accumulate, but can't repro. best to just skip it (on your side).

[yarikoptic]

could you be so kind @jreback to skip on your side? (since that is where it belongs?)

this environment is a stock precise -- you could get it on any debian/ubuntu system using debootstrap, let me know if you need precise commands

[jreback]

ok, if you'd try with #7675 this should skip in this instance

[jreback]

@yarikoptic lmk if you can try with #7675 to see if that skips the wonky test