LCMV Beamformer Regularization Parameter Does Not Affect Results

Support & Discussions
1

MNE version: 1.11.0

operating system: Ubuntu 18.04

Dear beamformer experts,

I am experimenting with parameters in mne.beamformer.make_lcmv. Some parameter changes affect the source results as expected. However, I am confused about the regularization parameter (reg).

Specifically, changing reg (e.g., 0, 0.05, 0.1) does not seem to affect the source results at all, with all other parameters and input data are kept the same.

My question: Why might changing the regularization parameter have no apparent effect on the LCMV solution? Perhaps other parameters need to be set to have reg working, under what conditions would the reg parameter meaningfully influence the results?

I have included screenshots showing the results across different reg values and my code snippet calling the mne.beamformer.make_lcmvfunction for reference. Thank you very much.

Best,

Zoe

image-4
image-41920×547 108 KB
filters = make_lcmv(evokeds_inv.info,
fwd,
data_cov,
reg=0,
noise_cov=noise_cov,
pick_ori=“max-power”,
weight_norm=“unit-noise-gain”,
rank=‘info’,
reduce_rank = True,
depth = 0.8
)

2

my first guess would be a bug in your analysis code somewhere. Can you see what it inside the filters dictionary and confirm they change when you specify different values for reg? Also, how is the data_cov computed exactly? Did you apply any regularization there?