[FIX] Reading an EDF with preload=False and mixed frequency (mne-tools#13069)

Co-authored-by: pre-commit-ci[bot] <66853113+pre-commit-ci[bot]@users.noreply.github.com>
Co-authored-by: Eric Larson <[email protected]>

Author: 3 people

doc/changes/devel/13069.bugfix.rst

@@ -0,0 +1 @@
+Fix bug cause by unnecessary assertion when loading mixed frequency EDFs without preloading :func:`mne.io.read_raw_edf` by `Simon Kern`_.

mne/conftest.py

@@ -186,6 +186,7 @@ def pytest_configure(config: pytest.Config):
     ignore:.*builtin type swigvarlink has no.*:DeprecationWarning
     # eeglabio
     ignore:numpy\.core\.records is deprecated.*:DeprecationWarning
+    ignore:Starting field name with a underscore.*:
     # joblib
     ignore:process .* is multi-threaded, use of fork/exec.*:DeprecationWarning
     """  # noqa: E501

mne/io/edf/edf.py

@@ -436,21 +436,24 @@ def _read_segment_file(data, idx, fi, start, stop, raw_extras, filenames, cals,
                 ones[orig_idx, smp_read : smp_read + len(one_i)] = one_i
                 n_smp_read[orig_idx] += len(one_i)
 
+        # resample channels with lower sample frequency
         # skip if no data was requested, ie. only annotations were read
-        if sum(n_smp_read) > 0:
+        if any(n_smp_read) > 0:
             # expected number of samples, equals maximum sfreq
             smp_exp = data.shape[-1]
-            assert max(n_smp_read) == smp_exp
 
             # resample data after loading all chunks to prevent edge artifacts
             resampled = False
+
             for i, smp_read in enumerate(n_smp_read):
                 # nothing read, nothing to resample
                 if smp_read == 0:
                     continue
                 # upsample if n_samples is lower than from highest sfreq
                 if smp_read != smp_exp:
-                    assert (ones[i, smp_read:] == 0).all()  # sanity check
+                    # sanity check that we read exactly how much we expected
+                    assert (ones[i, smp_read:] == 0).all()
+
                     ones[i, :] = resample(
                         ones[i, :smp_read].astype(np.float64),
                         smp_exp,

mne/io/edf/tests/test_edf.py

@@ -259,6 +259,24 @@ def test_edf_different_sfreqs(stim_channel):
     assert_allclose(times1, times2)
 
 
+@testing.requires_testing_data
+@pytest.mark.parametrize("stim_channel", (None, False, "auto"))
+def test_edf_different_sfreqs_nopreload(stim_channel):
+    """Test loading smaller sfreq channels without preloading."""
+    # load without preloading, then load a channel that has smaller sfreq
+    # as other channels, produced an error, see mne-python/issues/12897
+
+    for i in range(1, 13):
+        raw = read_raw_edf(input_fname=edf_reduced, verbose="error", preload=False)
+
+        # this should work for channels of all sfreq, even if larger sfreqs
+        # are present in the file
+        x1 = raw.get_data(picks=[f"A{i}"], return_times=False)
+        # load next ch, this is sometimes with a higher sometimes a lower sfreq
+        x2 = raw.get_data([f"A{i + 1}"], return_times=False)
+        assert x1.shape == x2.shape
+
+
 def test_edf_data_broken(tmp_path):
     """Test edf files."""
     raw = _test_raw_reader(