Detect invalid metadata when loading observations

tests/test_obs_rename_columns.py

+import h5py
+import pypfilt
+import scipy.stats
+
+
+class FixedNormalObs(pypfilt.obs.Univariate):
+    def distribution(self, ctx, snapshot):
+        return scipy.stats.norm(loc=0, scale=1)
+
+
+def test_obs_rename_columns(tmp_path):
+    """
+    Ensure that table metadata is correct when loading observations from files
+    whose columns names are not ``'time'`` and ``'value'``.
+    """
+    obs_unit = 'observations'
+    settings = {}
+    obs_model = FixedNormalObs(obs_unit, settings)
+
+    obs_file = tmp_path / 'test_obs.ssv'
+    time_scale = pypfilt.Datetime()
+    with open(obs_file, 'w') as f:
+        f.write('When What\n')
+        f.write('2024-09-18 4\n')
+        f.write('2024-09-19 5\n')
+        f.write('2024-09-20 6\n')
+
+    table = obs_model.from_file(
+        obs_file, time_scale, time_col='When', value_col='What'
+    )
+    obs_file.unlink()
+
+    # Verify that the observations table has correct metadata.
+    assert 'string_columns' in table.dtype.metadata
+    assert table.dtype.metadata['string_columns'] == []
+    assert 'time_columns' in table.dtype.metadata
+    assert table.dtype.metadata['time_columns'] == ['time']
+
+    # Verify that we can save the observations table to disk.
+    out_file = tmp_path / 'test_obs.hdf5'
+    with h5py.File(out_file, 'w') as f:
+        pypfilt.io.save_dataset(time_scale, f, 'test_obs', table)