[Mne_analysis] Applying ICA with No Components Selected out but Signal Changes Using Default Parameters

[Rockhill, Alexander P.]

Hi Alex and Mainak,

    The n_components argument is given None which yields 57 components and there are 64 channels with 7 bad channels which are not included so no I don't think it's because of the dimensionality reduction. Maybe it's some whitening.

To see something similar to what I'm looking at as far as scaling you can use the script below but I haven't been able to replicate the changes after ICA with sample data. I filedropped you both test epochs to the emails you responded to the thread with that does show that.

from time import time
import matplotlib.pyplot as plt
import mne
from mne.preprocessing import ICA
from mne.datasets import sample

'''data_path = sample.data_path()
raw_fname = data_path + '/MEG/sample/sample_audvis-raw.fif'

raw = mne.io.Raw(raw_fname, preload=True)
events = mne.find_events(raw)
epochs = mne.Epochs(raw, events, preload=True)

'''
epochs = mne.read_epochs('test-epo.fif', preload=True)

epochs = epochs.pick_types(meg=False, eeg=True)

fig, (ax0, ax1) = plt.subplots(1,2)
epochs.average().plot(axes=ax0, show=False)

ica = ICA(method='fastica', random_state=0)
t0 = time()
ica.fit(epochs)
fit_time = time() - t0
epochs = ica.apply(epochs, exclude=ica.exclude)
epochs.average().plot(axes=ax1, show=False)
ica.plot_sources(epochs)

Thanks,

Alex

Translational NeuroEngineering Laboratory
Division of Neurotherapeutics, Department of Psychiatry
Massachusetts General Hospital, Martinos Center
149 13th St Charlestown #2301, Boston, MA 02129
________________________________
From: mne_analysis-bounces at nmr.mgh.harvard.edu <mne_analysis-bounces at nmr.mgh.harvard.edu> on behalf of Alexandre Gramfort <alexandre.gramfort at inria.fr>
Sent: Wednesday, July 3, 2019 2:30 AM
To: Discussion and support forum for the users of MNE Software
Subject: Re: [Mne_analysis] Applying ICA with No Components Selected out but Signal Changes Using Default Parameters


        External Email - Use Caution

hi,

can you check that the number of components you fit is equal to the number of channels?
If it's less you have a dimensionality reduction step.

Alex


On Tue, Jul 2, 2019 at 11:28 PM Mainak Jas <mainakjas at gmail.com<mailto:mainakjas at gmail.com>> wrote:

        External Email - Use Caution

Hi Alex,

Could you provide us a full script on the MNE sample data that we can run?

Mainak

On Tue, Jul 2, 2019 at 5:14 PM Rockhill, Alexander P. <AROCKHILL at mgh.harvard.edu<mailto:AROCKHILL at mgh.harvard.edu>> wrote:
Also, of note the ica scale is off by quite a lot in the plot_sources plot, it is way too zoomed in.

Alex

Translational NeuroEngineering Laboratory
Division of Neurotherapeutics, Department of Psychiatry
Massachusetts General Hospital, Martinos Center
149 13th St Charlestown #2301, Boston, MA 02129
________________________________
From: mne_analysis-bounces at nmr.mgh.harvard.edu<mailto:mne_analysis-bounces at nmr.mgh.harvard.edu> <mne_analysis-bounces at nmr.mgh.harvard.edu<mailto:mne_analysis-bounces at nmr.mgh.harvard.edu>> on behalf of Rockhill, Alexander P. <AROCKHILL at mgh.harvard.edu<mailto:AROCKHILL at mgh.harvard.edu>>
Sent: Tuesday, July 2, 2019 3:25 PM
To: mne_analysis at nmr.mgh.harvard.edu<mailto:mne_analysis at nmr.mgh.harvard.edu>
Subject: [Mne_analysis] Applying ICA with No Components Selected out but Signal Changes Using Default Parameters

Hi,

    In an analysis, I am running:

ica = ICA(method='fastica', n_components=n_components,  # n_components=None
          random_state=seed)
ica.fit(inst2)
...
inst2 = ica.apply(inst2, exclude=ica.exclude)

    and when I skip all intermediate steps and just fit the ICA and apply it with an empty list for ica.exclude the signal still changes, quite a bit. I thought if no components were selected out and all the max PCA components were used the signal would be unchanged or basically unchanged. Is this a bug or something with my implementation?

Thanks,

Alex

Translational NeuroEngineering Laboratory
Division of Neurotherapeutics, Department of Psychiatry
Massachusetts General Hospital, Martinos Center
149 13th St Charlestown #2301, Boston, MA 02129
_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu<mailto:Mne_analysis at nmr.mgh.harvard.edu>
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
_______________________________________________
Mne_analysis mailing list
Mne_analysis at nmr.mgh.harvard.edu<mailto:Mne_analysis at nmr.mgh.harvard.edu>
https://mail.nmr.mgh.harvard.edu/mailman/listinfo/mne_analysis
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190703/10d16706/attachment-0001.html