Make meteonorm remote_data tests robust to data changes

The meteonorm iotools tests pinned specific values, the full column
list, and exact column counts returned by the live API. Meteonorm
regularly updates their weather dataset and adds parameters to the
'all' set, neither of which is a breaking API change, but both cause
these tests to fail spuriously (e.g. new
'surface_tilt'/'surface_azimuth'
columns broke the exact-list assertion).

Assert only what pvlib controls, not what meteonorm's dataset controls:

- realtime: check the known columns are a subset of the response and
  that the column count is at least as large, instead of exact equality.
- training/tmy: drop the golden-value assert_frame_equal comparisons
  (and their fixtures); check columns, numeric dtype, and that at most
  20% of values are NaN (the guarantee meteonorm provides).

Row counts and index checks are kept since they are determined by the
requested time range, not the dataset.

Author: maschwanden

docs/sphinx/source/whatsnew/v0.15.3.rst

@@ -47,4 +47,5 @@ Contributors
 ~~~~~~~~~~~~
 * Eesh Saxena (:ghuser:`eeshsaxena`)
 * Karl Hill (:ghuser:`karlhillx`)
+* Mathias Aschwanden (:ghuser:`maschwanden`)
 * Yonry Zhu (:ghuser:`yonryzhu`)

tests/iotools/test_meteonorm.py

@@ -61,30 +61,6 @@ def expected_meteonorm_index():
     return expected_meteonorm_index
 
 
-@pytest.fixture
-def expected_meteonorm_data():
-    # The first 12 rows of data
-    columns = ['ghi', 'global_horizontal_irradiance_with_shading']
-    expected = [
-        [0.0, 0.0],
-        [0.0, 0.0],
-        [0.0, 0.0],
-        [0.0, 0.0],
-        [0.0, 0.0],
-        [0.0, 0.0],
-        [0.0, 0.0],
-        [3.75, 3.74],
-        [57.25, 57.20],
-        [149.0, 148.96],
-        [242.25, 242.24],
-        [228.0, 227.98],
-    ]
-    index = pd.date_range('2025-01-01 00:30', periods=12, freq='1h', tz='UTC')
-    index.freq = None
-    expected = pd.DataFrame(expected, index=index, columns=columns)
-    return expected
-
-
 @pytest.fixture
 def expected_columns_all():
     columns = [
@@ -114,8 +90,7 @@ def expected_columns_all():
 @pytest.mark.remote_data
 @pytest.mark.flaky(reruns=RERUNS, reruns_delay=RERUNS_DELAY)
 def test_get_meteonorm_training(
-        demo_api_key, demo_url, expected_meta, expected_meteonorm_index,
-        expected_meteonorm_data):
+        demo_api_key, demo_url, expected_meta, expected_meteonorm_index):
     data, meta = pvlib.iotools.get_meteonorm_observation_training(
         latitude=50, longitude=10,
         start='2025-01-01', end='2026-01-01',
@@ -128,10 +103,12 @@ def test_get_meteonorm_training(
     for key in ['version', 'commit']:
         assert key in meta  # value changes, so only check presence
     pd.testing.assert_index_equal(data.index, expected_meteonorm_index)
-    # meteonorm API only guarantees similar, not identical, results between
-    # calls.  so we allow a small amount of variation with atol.
-    pd.testing.assert_frame_equal(data.iloc[:12], expected_meteonorm_data,
-                                  check_exact=False, atol=1)
+    # don't pin values: meteonorm may update the dataset without it being a
+    # breaking change.  check parsing instead.
+    assert list(data.columns) == \
+        ['ghi', 'global_horizontal_irradiance_with_shading']
+    assert data.dtypes.map(pd.api.types.is_numeric_dtype).all()
+    assert (data.isna().mean() <= 0.2).all()  # meteonorm guarantees <=20% NaN
 
 
 @pytest.mark.remote_data
@@ -156,8 +133,11 @@ def test_get_meteonorm_realtime(demo_api_key, demo_url, expected_columns_all):
     assert meta['surface_tilt'] == 20
     assert meta['surface_azimuth'] == 10
 
-    assert list(data.columns) == expected_columns_all
-    assert data.shape == (241, 19)
+    # meteonorm may add parameters to 'all' at any time, so only check that
+    # the columns we know about are present, not that the set matches exactly.
+    assert set(expected_columns_all).issubset(data.columns)
+    assert data.shape[0] == 241  # row count is determined by the time range
+    assert data.shape[1] >= len(expected_columns_all)
     # can't test the specific index as it varies due to the
     # use of pd.Timestamp.now
     assert type(data.index) is pd.core.indexes.interval.IntervalIndex
@@ -259,38 +239,10 @@ def expected_meteonorm_tmy_meta():
     return meta
 
 
-@pytest.fixture
-def expected_meteonorm_tmy_data():
-    # The first 12 rows of data
-    columns = ['diffuse_horizontal_irradiance']
-    expected = [
-        [0.],
-        [0.],
-        [0.],
-        [0.],
-        [0.],
-        [0.],
-        [0.],
-        [0.],
-        [9.07],
-        [8.44],
-        [86.64],
-        [110.44],
-    ]
-    index = pd.date_range(
-        '2030-01-01', periods=12, freq='1h', tz=3600)
-    index.freq = None
-    interval_index = pd.IntervalIndex.from_arrays(
-        index, index + pd.Timedelta(hours=1), closed='left')
-    expected = pd.DataFrame(expected, index=interval_index, columns=columns)
-    return expected
-
-
 @pytest.mark.remote_data
 @pytest.mark.flaky(reruns=RERUNS, reruns_delay=RERUNS_DELAY)
 def test_get_meteonorm_tmy(
-        demo_api_key, demo_url, expected_meteonorm_tmy_meta,
-        expected_meteonorm_tmy_data):
+        demo_api_key, demo_url, expected_meteonorm_tmy_meta):
     data, meta = pvlib.iotools.get_meteonorm_tmy(
         latitude=50, longitude=10,
         api_key=demo_api_key,
@@ -312,10 +264,11 @@ def test_get_meteonorm_tmy(
     assert meta.items() >= expected_meteonorm_tmy_meta.items()
     for key in ['version', 'commit']:
         assert key in meta  # value changes, so only check presence
-    # meteonorm API only guarantees similar, not identical, results between
-    # calls.  so we allow a small amount of variation with atol.
-    pd.testing.assert_frame_equal(data.iloc[:12], expected_meteonorm_tmy_data,
-                                  check_exact=False, atol=1)
+    # don't pin values: meteonorm may update the dataset without it being a
+    # breaking change.  check parsing instead.
+    assert list(data.columns) == ['diffuse_horizontal_irradiance']
+    assert data.dtypes.map(pd.api.types.is_numeric_dtype).all()
+    assert (data.isna().mean() <= 0.2).all()  # meteonorm guarantees <=20% NaN
 
 
 @fail_on_pvlib_version('0.16.0')