Improve performance of solarposition.ephemeris

[cwhanse]

solarposition.ephemeris returns results in dataframe with a localized time index, rather than UTC time. This results in slow performance because the dataframe is assembled from Series and np.arrays that are calculated using UTC time, requiring a timezone conversion.

To Reproduce
Change line 693 from
DFOut = pd.DataFrame(index=time)

to

DFOut = pd.DataFrame(index=time_utc)

to observe the speedup: roughly 23 sec calculation time for 500k timesteps with index=time, and <5 seconds for index=time_utc

• pvlib.__version__: 0.5.2
• pandas.__version__: 0.20.3

Recommend that solarposition.ephemeris return a dataframe indexed in UTC time.

[wholmgren]

The returned index is the same as the input index, and is not necessarily localized. I think the bottleneck is the same as in #502 -- in each case the DataFrame index is not the same as the first Series index. I agree with the proposed change, but suggest that we also add

DFOut.index = time

following the construction of the DataFrame. So the whole thing would look like

    # make output DataFrame
    DFOut = pd.DataFrame(index=time_utc)
    DFOut['apparent_elevation'] = ApparentSunEl
    DFOut['elevation'] = SunEl
    DFOut['azimuth'] = SunAz
    DFOut['apparent_zenith'] = 90 - ApparentSunEl
    DFOut['zenith'] = 90 - SunEl
    DFOut['solar_time'] = SolarTime
    DFOut.index = time

Have not tested it, but I think it would work.