test_irradiance.test_perez_components fails

[KonstantinTr]

Here is the error message I keep receiving

E       AssertionError: Series are different
E       
E       Series values are different (25.0 %)
E       [left]:  [0.0, 31.4635076992, 0.0, 45.4597896363]
E       [right]: [0.0, 31.46046871, nan, 45.45539877]

[cwhanse]

I don't know why the test is producing different results. But I think the fix is to remove the shortcut that was taken in test_irradiance. At the head of the file are calls to solar_position.get_solarposition and .get_clearsky that provide ephemeris and irradiance values for the unit tests; it is possible that a change to one of these functions is causing the test failure. In my view, its not the best design for a unit test to implicitly rely on other functions.

What would be better is to explicitly code the solar position and irradiance values in the test file. Over time these values may diverge from the functions we use to create them, but we'll have unit tests that are isolated to one function.

Other opinions?

[wholmgren]

I agree with Cliff regarding the bad test design, though I'm less sure that it's really the cause of this particular test failure.

It could be that the failure is related to a difference between your environment and the test environments. Can you run pandas.show_versions() and paste the results here?

[KonstantinTr]

INSTALLED VERSIONS

commit: None
python: 3.6.3.final.0
python-bits: 64
OS: Windows
OS-release: 10
machine: AMD64
processor: Intel64 Family 6 Model 94 Stepping 3, GenuineIntel
byteorder: little
LC_ALL: None
LANG: None
LOCALE: None.None

pandas: 0.20.3
pytest: 3.2.1
pip: None
setuptools: 36.5.0.post20170921
Cython: 0.26.1
numpy: 1.13.3
scipy: 0.19.1
xarray: None
IPython: 6.1.0
sphinx: 1.6.3
patsy: 0.4.1
dateutil: 2.6.1
pytz: 2017.2
blosc: None
bottleneck: 1.2.1
tables: 3.4.2
numexpr: 2.6.2
feather: None
matplotlib: 2.1.0
openpyxl: 2.4.8
xlrd: 1.1.0
xlwt: 1.3.0
xlsxwriter: 1.0.2
lxml: 4.1.0
bs4: 4.6.0
html5lib: 0.9999999
sqlalchemy: 1.1.13
pymysql: None
psycopg2: None
jinja2: 2.9.6
s3fs: None
pandas_gbq: None
pandas_datareader: None

[wholmgren]

It may be due to bottleneck's nan sum rules: nan + nan = 0. Try removing bottleneck and rerunning the test. Also note that the new pandas 0.21 disables bottleneck for sum operations.

[KonstantinTr]

I've removed the 'bottleneck' lib and now test passes.
@wholmgren Should I close the issue or should we redesign the test with Cliff's suggestions?

[wholmgren]

Seems to me that we've found and solved the problem, so I suggest closing the issue if you're satisfied. There are a number of other tests that should be redesigned in the same way, so maybe we should make a separate issue for cataloging them.

[cwhanse]

I agree - it merits a separate issue.