Linke turbidity factor fixes

[mikofski]

fixes #263 and fixes #262

• shift outputmin to zero, since (1) zero based indexing, and (2) documentation states that indices of LT data are 0.0833[arcdeg] apart
• shift from edge to center
• raise IndexError for out of bound lat/lon
• add test_linke_turbidity_corners() to test corners are not out of bounds, PASSES, but older tests fail since increment is different.

[wholmgren]

Thanks for digging into this!

I'm a bit skeptical that the new log messages are going to be useful to anyone after this is merged. I've been slowly removing all of the silly logging that I added before I was using pdb.

You can get rid of that todo comment, as far as I'm concerned. I think users have a responsibility to follow our conventions.

The tests that throw IndexErrors should use pytest as shown here: http://doc.pytest.org/en/latest/assert.html#assertions-about-expected-exceptions

[mikofski]

@wholmgren do you want me to revise the 3 failing tests with new expected values?

pvlib/test/test_clearsky.py::test_lookup_linke_turbidity FAILED
pvlib/test/test_clearsky.py::test_lookup_linke_turbidity_months FAILED
pvlib/test/test_clearsky.py::test_lookup_linke_turbidity_nointerp_months FAILED

[mikofski]

Reverse engineering the current expected LT values to make sure the new calculation is consistent:

test_lookup_linke_turbidity()

lat, lon = 32.2, -111
# using _linearly_scale
lat = (lat - 90.) * (-2159. / 180.) + 1.  # 694.2788888888888
lon = (lon - (-180.)) * (4319. / 360.) + 1.  # 828.8083333333333
# now use new lookup method, shifting the indices to their corrected inputs at (32.125, -110.875)
clearsky.lookup_linke_turbidity(
    times, 90. - np.around(lat) / 12. - (1. / 24.), np.around(lon) / 12. - 180. + (1. / 24.)
)
# 2014-06-24 00:00:00-07:00    3.101266
# 2014-06-24 12:00:00-07:00    3.101266
# 2014-06-25 00:00:00-07:00    3.114430
# Freq: 12H, dtype: float64

yields the same result as the expected answer

[wholmgren]

Can you look into doing the same thing for the rest of the tests? I'm guessing that it will work for some but not others. Maybe can adjust the precision of the tests too.

[mikofski]

Can you tell me which tests aren't passing? All test_clearsky.py pass for me.

[wholmgren]

There is a failing test in test_location and a bunch in test_modelchain

https://travis-ci.org/pvlib/pvlib-python/jobs/176473332

Other than copy/pasting a bunch of new values, the easiest way to fix the modelchain tests might be to manually supply the irradiance data in the mc.run_model calls. Is that asking too much?

[mikofski]

There is an error in test_forecast.test_process_data[NDFD] that I don't think has anything to do with Linke turbidity factors, but I might be wrong?

==================================== ERRORS ====================================
__________________ ERROR at setup of test_process_data[NDFD] ___________________
request = <SubRequest 'model' for <Function 'test_process_data[NDFD]'>>
    @requires_siphon
    @pytest.fixture(scope='module', params=_modelclasses)
    def model(request):
        amodel = request.param()
        amodel.raw_data = \
>           amodel.get_data(_latitude, _longitude, _start, _end)
pvlib/test/test_forecast.py:51: 
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
pvlib/forecast.py:252: in get_data
    self.data = self._netcdf2pandas(self.netcdf_data, self.query_variables)
pvlib/forecast.py:333: in _netcdf2pandas
    self.set_time(netcdf_data.variables[time_var])
pvlib/forecast.py:357: in set_time
    times = num2date(time[:].squeeze(), time.units)
netCDF4/_netCDF4.pyx:3695: in netCDF4._netCDF4.Variable.__getitem__ (netCDF4/_netCDF4.c:37923)
    ???
_ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ _ 
>   ???
E   IndexError
netCDF4/_netCDF4.pyx:4374: IndexError

[wholmgren]

Awesome, thanks! Two minor things left to do:

1. add @requires_scipy to your new corner cases test. That's the reason for the failure on the travis py27-min build.
2. add some notes to the 0.4.2 whatsnew file. Be sure to mention that typical modeling results could change by ~1%, depending on location, if they depend on the turbidity table.

The test_forecast error is unrelated (sorry).

[wholmgren]

Thanks @mikofski!