apply ICA method

[MegaSear]

Description of the problem

_apply_raw contains "picks = pick_types( raw.info, meg=False, include=self.ch_names, exclude="bads", ref_meg=False)"
But ICA.fit contains "picks = _picks_to_idx(inst.info, picks, allow_empty=False, with_ref_meg=self.allow_ref_meg)"
Interesting, I can train ICA on raw with bads channels
But using the same raw in the apply method I get an error that there are less channels due to bads

Steps to reproduce

raw.info['bads'] = ['Fpz']
n_components = 16
ica = ICA(n_components=n_components)
ICA.fit(raw)
ICA.apply(raw, exclude=[])

Link to data

No response

Expected results

The apply method is not expected to remove bad channels

Actual results

"operands could not be broadcast together with shapes (54,2455179) (63,1) (54,2455179)".

Additional information

Platform Windows-10-10.0.19045-SP0
Python 3.10.8 (tags/v3.10.8:aaaf517, Oct 11 2022, 16:50:30) [MSC v.1933 64 bit (AMD64)]
Executable C:\Users\searg\AppData\Local\Programs\Python\Python310\python.exe
CPU Intel(R) Core(TM) i5-9300H CPU @ 2.40GHz (8 cores)
Memory 15.8 GiB

Core

• mne 1.9.0 (latest release)
• numpy 1.26.4 (OpenBLAS 0.3.23.dev with 8 threads)
• scipy 1.15.2
• matplotlib 3.10.1 (backend=qtagg)

Numerical (optional)

• sklearn 1.6.1
• numba 0.59.1
• cupy 12.3.0
• pandas 2.2.3
• h5io 0.2.4
• h5py 3.13.0

• unavailable nibabel, nilearn, dipy, openmeeg

[larsoner]

Any chance you could make a fully reproducible example that could be run using the sample dataset?

[MegaSear]

import mne 
from mne.preprocessing import ICA

from mne.datasets import eegbci
from mne.io import read_raw_edf
from pathlib import Path
import os 

mne_data_dir = Path.home() / 'mne_data'
mne_data_dir.mkdir(parents=True, exist_ok=True)

os.environ['MNE_DATA'] = str(mne_data_dir)
mne.set_config('MNE_DATA', os.environ['MNE_DATA'], set_env=True)

subject = 1
runs = [6]
edf_paths = eegbci.load_data(subject, runs)
raw = read_raw_edf(edf_paths[0], preload=True)

print(raw.info)

raw.info["bads"] = [raw.ch_names[0]] # the first channel was chosen for nothing
raw.info["bads"]

example_channels = raw.ch_names

ica = ICA(n_components=20) # the 20 components was chosen for nothing

ica.fit(inst=raw.copy(), picks=example_channels) # pick ALL channels

print(ica.pca_components_.shape)
print(ica.mixing_matrix_.shape)
print(ica.get_sources(raw.copy()).get_data().shape)

ica.apply(inst=raw.copy(), exclude=[]) 

[cbrnr]

I just hit this issue as well, in a script that has worked previously, so at some point some change seems to have broken it. If I have the time I can try to debug (and find the corresponding commit/PR).