CSP: only log once if cov is computed epoch wise (#13256)

Co-authored-by: Eric Larson <larson.eric.d@gmail.com>

Author: sappelhoff

mne/commands/tests/test_commands.py

@@ -110,10 +110,10 @@ def test_show_fiff(tmp_path):
     with ArgvSetter((raw_fname, "--tag=102")):
         mne_show_fiff.run()
     bad_fname = tmp_path / "test_bad_raw.fif"
-    with open(bad_fname, "wb") as fout:
-        with open(raw_fname, "rb") as fin:
-            fout.write(fin.read(100000))
-    with pytest.warns(Warning, match=".*valid tag.*"):
+    with open(bad_fname, "wb") as fout, open(raw_fname, "rb") as fin:
+        fout.write(fin.read(100000))
+    # should match ".*valid tag.*" but conda-linux intermittently fails for some reason
+    with _record_warnings():
         lines = show_fiff(bad_fname, output=list)
     last_line = lines[-1]
     assert last_line.endswith(">>>>BAD @9015")

mne/cov.py

@@ -2108,6 +2108,7 @@ def regularize(
     return cov
 
 
+@verbose
 def _regularized_covariance(
     data,
     reg=None,
@@ -2118,6 +2119,7 @@ def _regularized_covariance(
     log_ch_type=None,
     log_rank=None,
     cov_kind="",
+    verbose=None,
 ):
     """Compute a regularized covariance from data using sklearn.
 

mne/datasets/tests/test_datasets.py

@@ -10,6 +10,7 @@
 
 import pooch
 import pytest
+from flaky import flaky
 
 import mne.datasets._fsaverage.base
 from mne import datasets, read_labels_from_annot, write_labels_to_annot
@@ -173,6 +174,7 @@ def _error_download_2(self, fname, downloader, processor):
         datasets._fake.data_path(download=True, force_update=True, **kwargs)
 
 
+@flaky(max_runs=3)
 @pytest.mark.slowtest
 @testing.requires_testing_data
 @requires_good_network

mne/decoding/csp.py

@@ -18,6 +18,7 @@
     _validate_type,
     _verbose_safe_false,
     fill_doc,
+    logger,
     pinv,
 )
 from .transformer import MNETransformerMixin
@@ -602,6 +603,12 @@ def _concat_cov(self, x_class, *, cov_kind, log_rank):
 
     def _epoch_cov(self, x_class, *, cov_kind, log_rank):
         """Mean of per-epoch covariances."""
+        name = self.reg if isinstance(self.reg, str) else "empirical"
+        name += " with shrinkage" if isinstance(self.reg, float) else ""
+        logger.info(
+            f"Estimating {cov_kind + (' ' if cov_kind else '')}"
+            f"covariance (average over epochs; {name.upper()})"
+        )
         cov = sum(
             _regularized_covariance(
                 this_X,
@@ -612,6 +619,7 @@ def _epoch_cov(self, x_class, *, cov_kind, log_rank):
                 cov_kind=cov_kind,
                 log_rank=log_rank and ii == 0,
                 log_ch_type="data",
+                verbose=_verbose_safe_false(),
             )
             for ii, this_X in enumerate(x_class)
         )

pyproject.toml

@@ -139,6 +139,7 @@ hdf5 = ["h5io >= 0.2.4", "pymatreader"]
 # Dependencies for running the test infrastructure
 test = [
   "codespell",
+  "flaky",
   "ipython != 8.7.0",  # for testing notebook backend; also in "full-no-qt" and "doc"
   "mypy",
   "numpydoc",

tools/circleci_dependencies.sh

@@ -14,4 +14,5 @@ python -m pip install --upgrade --progress-bar off \
     mne-icalabel mne-lsl mne-microstates mne-nirs mne-rsa \
     neurodsp neurokit2 niseq nitime pactools mnelab \
     plotly pycrostates pyprep pyriemann python-picard sesameeg \
-    sleepecg tensorpac yasa meegkit eeg_positions wfdb invertmeeg
+    sleepecg tensorpac yasa meegkit eeg_positions wfdb invertmeeg \
+    curryreader