Bug in validate method

[zachdj]

After #76 landed, I get the following stack trace when evaluating models:

Traceback (most recent call last):
  File "/home/zach/anaconda3/envs/apollo/lib/python3.7/runpy.py", line 193, in _run_module_as_main
    "__main__", mod_spec)
  File "/home/zach/anaconda3/envs/apollo/lib/python3.7/runpy.py", line 85, in _run_code
    exec(code, run_globals)
  File "/home/zach/Develop/apollo/apollo/bin/evaluate.py", line 93, in <module>
    main()
  File "/home/zach/Develop/apollo/apollo/bin/evaluate.py", line 78, in main
    multioutput=multioutput)
  File "/home/zach/Develop/apollo/apollo/models/base.py", line 180, in validate
    axis=1, join='inner')
  File "/home/zach/anaconda3/envs/apollo/lib/python3.7/site-packages/pandas/core/reshape/concat.py", line 228, in concat
    copy=copy, sort=sort)
  File "/home/zach/anaconda3/envs/apollo/lib/python3.7/site-packages/pandas/core/reshape/concat.py", line 381, in __init__
    self.new_axes = self._get_new_axes()
  File "/home/zach/anaconda3/envs/apollo/lib/python3.7/site-packages/pandas/core/reshape/concat.py", line 448, in _get_new_axes
    new_axes[i] = self._get_comb_axis(i)
  File "/home/zach/anaconda3/envs/apollo/lib/python3.7/site-packages/pandas/core/reshape/concat.py", line 469, in _get_comb_axis
    sort=self.sort)
  File "/home/zach/anaconda3/envs/apollo/lib/python3.7/site-packages/pandas/core/indexes/api.py", line 70, in _get_objs_combined_axis
    return _get_combined_index(obs_idxes, intersect=intersect, sort=sort)
  File "/home/zach/anaconda3/envs/apollo/lib/python3.7/site-packages/pandas/core/indexes/api.py", line 115, in _get_combined_index
    index = index.intersection(other)
  File "/home/zach/anaconda3/envs/apollo/lib/python3.7/site-packages/pandas/core/indexes/datetimes.py", line 645, in intersection
    tz=result.tz, freq=None)
AttributeError: 'Index' object has no attribute 'tz'

Looks like the offending line is

matched = pd.concat([predictions, ground_truth], axis=1, join='inner')

ground_truth is loaded with the apollo.datasets.ga_power.open_sqlite method, which uses pandas Timestamp parsing instead of our new function in apollo.timestamps.
The timestamps in the index of DataArrays returned by open_sqlite don't have timezone info.

[zachdj]

I ran into some trouble trying to fix this.
The apollo.datasets.ga_power.open_sqlite function makes a call to pandas read_sql_query function:

df = pd.read_sql_query(
        sql=query,
        con=connection,
        params=params,
        index_col='reftime',
        parse_dates=['reftime'])

This will parse the 'reftime' column as a date and create a pandas Dataframe with a DatetimeIndex (underlying type is datetime64[ns]). When the Dataframe is converted to an xarray Dataset, everything is hunky-dory, but the Dataset's index is not timezone-aware.

We can make the index of the pandas Dataframe timezone-aware by replacing the above call to read_sql_query with the following:

df = pd.read_sql_query(
        sql=query,
        con=connection,
        params=params,
        index_col='reftime',
        parse_dates={
            'reftime': {
                'utc': True,
                'unit': 's'
            }
        })

This creates a Dataframe with a DatetimeIndex which is timezone-aware (underlying type isdatetime64[ns, UTC]). But when the Dataframe is converted to xarray, the index gets messed up because xarray doesn't handle the timezone-aware DatetimeIndex very well. Check out darothen's comment on xarray issue #1490:

But for some weird reason, numpy doesn't recognize its own datetime dtypes when they have timezone information.

So it looks like there may be some wrinkles making the Datasets returned from apollo.datasets.ga_power timezone-aware, which was my original plan for fixing this issue.