Omit bad channels when gathering data used for decoding

docs/source/v1.5.md.inc

@@ -51,4 +51,6 @@ All users are encouraged to update.
 - Fixed bug where head position files were not written with a proper suffix and extension (#761 by @larsoner)
 - Fixed bug where default values for `decoding_csp_times` and `decoding_csp_freqs` were not set dynamically based on the config parameters (#779 by @larsoner)
 - Fixed bug where the MNE logger verbosity was not respected inside parallel jobs (#813 by @larsoner)
-- A number of processing steps erroneously **always** operated on un-cleaned epochs (`sensor/decoding_full_epochs`, `sensor/decoding_time_by_time`, `sensor/decoding_csp`); or operated on un-cleaned epochs (without PTP rejection) if no ICA or SSP was requested (`sensor/ime_frequency`, `sensor/make_cov`) The bug in `sensor/make_cov` could propagate to the source level, as the covariance matrix is used for inverse modeling. (#796 by @hoechenberger)
+- A number of processing steps erroneously **always** operated on un-cleaned epochs (`sensor/decoding_full_epochs`, `sensor/decoding_time_by_time`, `sensor/decoding_csp`); or operated on un-cleaned epochs (without PTP rejection) if no ICA or SSP was requested (`sensor/ime_frequency`, `sensor/make_cov`) The bug in `sensor/make_cov` could propagate to the source level, as the covariance matrix is used for inverse modeling. (#796 by @hoechenberger)
+- Bad channels may have been submitted to MVPA (full epochs decoding, time-by-time decoding, CSP-based decoding) when not using Maxwell filtering
+  (i.e., usually only EEG data was affected). This has now been fixed and data from bad channels is omitted from decoding. (#817 by @hoechenberger)

mne_bids_pipeline/steps/sensor/_02_decoding_full_epochs.py

@@ -118,7 +118,7 @@ def run_epochs_decoding(
     n_cond1 = len(epochs[epochs_conds[0]])
     n_cond2 = len(epochs[epochs_conds[1]])
 
-    X = epochs.get_data()
+    X = epochs.get_data(picks="data")  # omit bad channels
     y = np.r_[np.ones(n_cond1), np.zeros(n_cond2)]
 
     classification_pipeline = make_pipeline(

mne_bids_pipeline/steps/sensor/_03_decoding_time_by_time.py

@@ -129,7 +129,7 @@ def run_time_decoding(
     if cfg.decoding_time_generalization and decim > 1:
         epochs.decimate(decim, verbose="error")
 
-    X = epochs.get_data()
+    X = epochs.get_data(picks="data")  # omit bad channels
     y = np.r_[np.ones(n_cond1), np.zeros(n_cond2)]
     # ProgressBar does not work on dask, so only enable it if not using dask
     verbose = get_parallel_backend_name(exec_params=exec_params) != "dask"

mne_bids_pipeline/steps/sensor/_05_decoding_csp.py

@@ -254,7 +254,7 @@ def _fmt_contrast(cond1, cond2, fmin, fmax, freq_range_name, tmin=None, tmax=Non
             epochs=epochs, contrast=contrast, fmin=fmin, fmax=fmax, cfg=cfg
         )
         # Get the data for all time points
-        X = epochs_filt.get_data()
+        X = epochs_filt.get_data(picks="data")  # omit bad channels
 
         # We apply PCA before running CSP:
         # - much faster CSP processing
@@ -327,7 +327,7 @@ def _fmt_contrast(cond1, cond2, fmin, fmax, freq_range_name, tmin=None, tmax=Non
         if tmax is not None:  # avoid warnings about outside the interval
             tmax = min(tmax, epochs_filt.times[-1])
         epochs_filt.crop(tmin, tmax)
-        X = epochs_filt.get_data()
+        X = epochs_filt.get_data(picks="data")  # omit bad channels
         X_pca = pca.transform(X)
         del X