Move scipy import in clearsky.py

[wholmgren]

As discussed in #13.

I briefly looked into adding a real test for this, but it wasn't immediately obvious how to do it. Instead, I ran the test suite on a virtualenv without scipy. Here's the output, as expected:

$ nosetests pvlib/test/test_clearsky.py -v
pvlib.test.test_clearsky.test_ineichen_required ... ERROR
pvlib.test.test_clearsky.test_ineichen_supply_linke ... ok
pvlib.test.test_clearsky.test_ineichen_solpos ... ok
pvlib.test.test_clearsky.test_ineichen_airmass ... ok
pvlib.test.test_clearsky.test_ineichen_keys ... ok
pvlib.test.test_clearsky.test_haurwitz ... ok
pvlib.test.test_clearsky.test_haurwitz_keys ... ok
pvlib.test.test_clearsky.test_disc_keys ... ok
pvlib.test.test_clearsky.test_disc_value ... ok

======================================================================
ERROR: pvlib.test.test_clearsky.test_ineichen_required
----------------------------------------------------------------------
Traceback (most recent call last):
  File "/Users/holmgren/.pyenv/versions/pvlib-3.4.3/lib/python3.4/site-packages/nose/case.py", line 198, in runTest
    self.test(*self.arg)
  File "/Users/holmgren/git_repos/pvlib-python/pvlib/test/test_clearsky.py", line 34, in test_ineichen_required
    clearsky.ineichen(times, tus)
  File "/Users/holmgren/git_repos/pvlib-python/pvlib/clearsky.py", line 134, in ineichen
    'supply your own turbidities.')
nose.proxy.ImportError: The Linke turbidity lookup table requires scipy. You can still use clearsky.ineichen if you supply your own turbidities.
-------------------- >> begin captured logging << --------------------
pvlib: DEBUG: creating Location object
pvlib: DEBUG: using PyEphem to calculate solar position
pvlib: DEBUG: tz_localize to US/Arizona and then tz_convert to UTC
pvlib: DEBUG: irradiance.extraradiation()
pvlib: DEBUG: Calculating ET rad using Spencer method
pvlib: DEBUG: using PyEphem to calculate solar position
pvlib: DEBUG: tz_localize to US/Arizona and then tz_convert to UTC
--------------------- >> end captured logging << ---------------------

----------------------------------------------------------------------
Ran 9 tests in 1.986s

FAILED (errors=1)

[wholmgren]

Resolved the merge conflict in the whats new file.

[wholmgren]

@bmu or @Calama-Consulting are we good to go on this one?

[bmu]

Ooops, maybe I was to hasty. When I run the tests without scipy I get for errors:

• pvlib.test.test_clearsky.test_ineichen_required ... ERROR
• pvlib.test.test_clearsky.test_disc_value ... ok
Failure: ImportError (The Linke turbidity lookup table requires scipy. You can still use clearsky.ineichen if you supply your own turbidities.) ... ERROR
• pvlib.test.test_location.test_location_print_pytz ... ok
Failure: ImportError (The Linke turbidity lookup table requires scipy. You can still use clearsky.ineichen if you supply your own turbidities.) ... ERROR
• pvlib.test.test_solarposition.test_calc_time ... ERROR

Not sure, where the errors are thrown in all cases (I don't see any use of ineichen or scipy in Location), I just copied the output of nosetests -v -w pvlib --with-coverage --cover-package=pvlib.

I think we should skip all tests that use scipy, if scipy is not installed.
Should I revert the merge?

[wholmgren]

Ok, that was very confusing but I tracked it down to test_irradiance.py and test_pvsystem.py failing on module import without scipy. They both use clearsky.ineichen to calculate the irradiance data used in later tests. I changed the calls to use a fixed turbidity.

The nosetests log now looks right to me.

I agree that we should skip the tests that require scipy if scipy is not installed, however, I would suggest that we make that a 0.2 milestone.

[wholmgren]

Ah, not sure what the best git approach is here, but I ended up making a new PR in #34 to resolve this.