Split daily validation posting on missing periods and week (#435)

* make sure data is sorted before slicing

* sort data in pre-post too

* split daily validation post split by uneven frequency and week

* update whatsnew

* flake8

* fix test to make sure frame assertion happens

Author: alorenzo175

docs/source/whatsnew/1.0.0rc1.rst

@@ -19,12 +19,16 @@ Enhancements
 ~~~~~~~~~~~~
 * Reference net load forecasts provided using week-ahead persistence. (:issue:`55`) (:pull:`392`)
 * Datamodel now supports ``'net_load'`` as an allowed variable. (:issue:`55`) (:pull:`392`)
+* Posting of daily validation now splits requests to avoid missing periods and
+  limit each request to one week of data (:issue:`424`) (:pull:`435`)
 
 
 Bug fixes
 ~~~~~~~~~
 * Fix incorrect ordering of months and weekdays in metrics plots.
   (:issue:`428`) (:pull:`430`)
+* Ensure data is sorted from reference data sources before slicing and
+  posting to the API (:pull:`435`)
 
 
 Contributors

solarforecastarbiter/io/reference_observations/common.py

@@ -283,7 +283,8 @@ def update_site_observations(api, fetch_func, site, observations,
         start = get_last_site_timestamp(api, site_observations, end)
     logger.debug('Fetching data for %s from %s to %s', site.name, start, end)
     obs_df = fetch_func(api, site, start, end)
-    data_in_range = obs_df[start:end]
+    # must be sorted for proper inexact start:end slicing
+    data_in_range = obs_df.sort_index()[start:end]
     if data_in_range.empty:
         return
     for obs in site_observations:
@@ -295,6 +296,9 @@ def _prepare_data_to_post(data, variable, observation, start, end):
     to prepare for posting"""
     data = data[[variable]]
     data = data.rename(columns={variable: 'value'})
+    # ensure data is sorted before slicing and for optimal order in the
+    # database
+    data = data.sort_index()
     # remove all future values, some files have forward filled nightly data
     data = data[start:min(end, _utcnow())]
     # we assume any reference data is given at the proper intervals

solarforecastarbiter/io/reference_observations/tests/test_common.py

@@ -492,6 +492,24 @@ def test_update_site_observations_no_data(
     mock_api.assert_not_called()
 
 
+def test_update_site_observations_out_of_order(
+        mock_api, site_objects_param, mocker,
+        observation_objects_param, fake_ghi_data):
+    start = pd.Timestamp('20190101T1200Z')
+    end = pd.Timestamp('20190101T1230Z')
+    fetch = mocker.MagicMock()
+    fetch.return_value = fake_ghi_data.sample(frac=1)
+    common.update_site_observations(
+        mock_api, fetch, site_objects[1], observation_objects_param,
+        start, end)
+    args, _ = mock_api.post_observation_values.call_args
+    assert args[0] == ''
+    pd.testing.assert_frame_equal(
+        args[1], fake_ghi_data.rename(
+            columns={'ghi': 'value'})[start:end].resample(
+                args[1].index.freq).first())
+
+
 @pytest.fixture()
 def template_fx(mock_api, mocker):
     mock_api.create_forecast = mocker.MagicMock(side_effect=lambda x: x)

solarforecastarbiter/validation/tasks.py

@@ -6,7 +6,7 @@
 
 from solarforecastarbiter import pvmodel
 from solarforecastarbiter.io.api import APISession
-from solarforecastarbiter.validation import validator
+from solarforecastarbiter.validation import validator, quality_mapping
 
 
 logger = logging.getLogger(__name__)
@@ -391,7 +391,7 @@ def _daily_validation(session, observation, start, end, base_url):
     logger.info('Validating data for %s from %s to %s',
                 observation.name, start, end)
     observation_values = session.get_observation_values(
-        observation.observation_id, start, end)
+        observation.observation_id, start, end).sort_index()
     value_series = observation_values['value'].astype(float)
     if len(value_series.dropna()) < 10:
         raise IndexError(
@@ -410,9 +410,28 @@ def _daily_validation(session, observation, start, end, base_url):
 
     quality_flags.name = 'quality_flag'
     observation_values.update(quality_flags)
-    session.post_observation_values(observation.observation_id,
-                                    observation_values,
-                                    params='donotvalidate')
+    return _group_continuous_week_post(
+        session, observation, observation_values)
+
+
+def _group_continuous_week_post(session, observation, observation_values):
+    # observation_values expected to be sorted
+    # observation values already have uneven frequency checked
+    gid = quality_mapping.check_if_series_flagged(
+        observation_values['quality_flag'], 'UNEVEN FREQUENCY').cumsum()
+    # make series of week + year integers to further
+    # split data to post at most one week at a time
+    # ~10,000 pts of 1min data
+    week_int = (gid.index.week + gid.index.year).values
+    # combine the continuous groups with groups of weeks
+    # gid is unique for each group since week_int and cumsum
+    # increase monotonically and are positive
+    gid += week_int
+    observation_values['gid'] = gid
+    for _, group in observation_values.groupby('gid'):
+        session.post_observation_values(observation.observation_id,
+                                        group[['value', 'quality_flag']],
+                                        params='donotvalidate')
 
 
 def daily_single_observation_validation(access_token, observation_id, start,

solarforecastarbiter/validation/tests/test_validation_tasks.py

@@ -632,8 +632,9 @@ def test_daily_observation_validation_ghi(mocker, make_observation,
         DESCRIPTION_MASK_MAPPING['UNEVEN FREQUENCY'] |
         LATEST_VERSION_FLAG
     ]
-    assert post_mock.called_once
-    assert_frame_equal(post_mock.call_args[0][1], out)
+    assert post_mock.called
+    posted_df = pd.concat([cal[0][1] for cal in post_mock.call_args_list])
+    assert_frame_equal(posted_df, out)
 
 
 def test_daily_observation_validation_ghi_zeros(mocker, make_observation,
@@ -677,8 +678,9 @@ def test_daily_observation_validation_ghi_zeros(mocker, make_observation,
         base | DESCRIPTION_MASK_MAPPING['NIGHTTIME'] |
         DESCRIPTION_MASK_MAPPING['UNEVEN FREQUENCY']
     ]
-    assert post_mock.called_once
-    assert_frame_equal(post_mock.call_args[0][1], out)
+    assert post_mock.called
+    posted_df = pd.concat([cal[0][1] for cal in post_mock.call_args_list])
+    assert_frame_equal(posted_df, out)
 
 
 def test_validate_daily_dc_power(mocker, make_observation, daily_index):
@@ -767,8 +769,9 @@ def test_daily_observation_validation_dc_power(mocker, make_observation,
         DESCRIPTION_MASK_MAPPING['NIGHTTIME'] |
         LATEST_VERSION_FLAG
     ]
-    assert post_mock.called_once
-    assert_frame_equal(post_mock.call_args[0][1], out)
+    assert post_mock.called
+    posted_df = pd.concat([cal[0][1] for cal in post_mock.call_args_list])
+    assert_frame_equal(posted_df, out)
 
 
 def test_validate_daily_ac_power(mocker, make_observation, daily_index):
@@ -863,8 +866,9 @@ def test_daily_observation_validation_ac_power(mocker, make_observation,
         DESCRIPTION_MASK_MAPPING['NIGHTTIME'] |
         LATEST_VERSION_FLAG
     ]
-    assert post_mock.called_once
-    assert_frame_equal(post_mock.call_args[0][1], out)
+    assert post_mock.called
+    posted_df = pd.concat([cal[0][1] for cal in post_mock.call_args_list])
+    assert_frame_equal(posted_df, out)
 
 
 @pytest.mark.parametrize('var', ['air_temperature', 'wind_speed', 'dni', 'dhi',
@@ -885,13 +889,14 @@ def test_daily_observation_validation_other(var, mocker, make_observation,
         return_value=data)
     post_mock = mocker.patch(
         'solarforecastarbiter.io.api.APISession.post_observation_values')
-    validate_mock = mocker.MagicMock()
+    validated = pd.Series(2, index=daily_index)
+    validate_mock = mocker.MagicMock(return_value=validated)
     mocker.patch.dict(
         'solarforecastarbiter.validation.tasks.IMMEDIATE_VALIDATION_FUNCS',
         {var: validate_mock})
     tasks.daily_single_observation_validation(
         '', obs.observation_id, data.index[0], data.index[-1])
-    assert post_mock.called_once
+    assert post_mock.called
     assert validate_mock.called
 
 
@@ -914,13 +919,14 @@ def test_daily_observation_validation_many(mocker, make_observation,
         return_value=data)
     post_mock = mocker.patch(
         'solarforecastarbiter.io.api.APISession.post_observation_values')
-    validate_mock = mocker.MagicMock()
+    validated = pd.Series(2, index=daily_index)
+    validate_mock = mocker.MagicMock(return_value=validated)
     mocker.patch.dict(
         'solarforecastarbiter.validation.tasks.IMMEDIATE_VALIDATION_FUNCS',
         {'dhi': validate_mock, 'dni': validate_mock})
     tasks.daily_observation_validation(
         '', data.index[0], data.index[-1])
-    assert post_mock.called_once
+    assert post_mock.called
     assert validate_mock.call_count == 2
 
 
@@ -967,3 +973,51 @@ def test_daily_observation_validation_not_enough(mocker, make_observation):
         '', data.index[0], data.index[-1])
     assert out is None
     assert log.called
+
+
+def test__group_continuous_week_post(mocker, make_observation):
+    split_dfs = [
+        pd.DataFrame([(0, LATEST_VERSION_FLAG)],
+                     columns=['value', 'quality_flag'],
+                     index=pd.date_range(
+                         start='2020-05-03T00:00',
+                         end='2020-05-03T23:59',
+                         tz='UTC',
+                         freq='1h')),
+        # new week split
+        pd.DataFrame([(0, LATEST_VERSION_FLAG)],
+                     columns=['value', 'quality_flag'],
+                     index=pd.date_range(
+                         start='2020-05-04T00:00',
+                         end='2020-05-04T11:59',
+                         tz='UTC',
+                         freq='1h')),
+        # missing 12
+        pd.DataFrame(
+            [(0, LATEST_VERSION_FLAG | DESCRIPTION_MASK_MAPPING['UNEVEN FREQUENCY'])] +  # NOQA
+            [(1, LATEST_VERSION_FLAG)] * 7,
+            columns=['value', 'quality_flag'],
+            index=pd.date_range(
+                start='2020-05-04T13:00',
+                end='2020-05-04T20:00',
+                tz='UTC',
+                freq='1h')),
+        # missing a week+
+        pd.DataFrame(
+            [(9, LATEST_VERSION_FLAG | DESCRIPTION_MASK_MAPPING['UNEVEN FREQUENCY'])] +  # NOQA
+            [(3, LATEST_VERSION_FLAG)] * 7,
+            columns=['value', 'quality_flag'],
+            index=pd.date_range(
+                start='2020-05-13T09:00',
+                end='2020-05-13T16:59',
+                tz='UTC',
+                freq='1h')),
+    ]
+    ov = pd.concat(split_dfs, axis=0)
+    obs = make_observation('ghi')
+    session = mocker.MagicMock()
+    tasks._group_continuous_week_post(session, obs, ov)
+    call_list = session.post_observation_values.call_args_list
+    assert len(call_list) == 4
+    for i, cal in enumerate(call_list):
+        assert_frame_equal(split_dfs[i], cal[0][1])