[Eeglablist] margaux.izac at u-bordeaux.fr sent you Rank issue via WeTransfer

Makoto Miyakoshi mmiyakoshi at ucsd.edu
Wed Nov 15 08:07:49 PST 2023


Hi Margaux,

Please re-upload your datasets to the temporary server. I missed it.

It's surprising that you obtained the 0.99 correlation across channels.
My initial guess is that you did something awfully wrong when using a
band-pass filter. Using a very narrow band-pass can cause rank deficiency.
Please check your filter parameters again.
Another possibility is electrode gel bridging as Torralba pointed out. But
generally it is less common.

Makoto

On Wed, Nov 15, 2023 at 8:09 AM Margaux Izac via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:

> Hello,
>
> Would someone please know :
> - Why does the window before ICA warns us about a rank issue although our
> rank is 32 ?
> - What could impact the data's rank when some files have a rank of 29,
> other than two electrodes recording the same signal ?
>
> (see detailed version of the issue further down)
>
> Best regards,
> Margaux IZAC
>
>
> Download link:
>
> https://urldefense.com/v3/__https://wetransfer.com/downloads/291a3b24a144df88f10fd25ecc0668f720231107180138/4239b16861b899f6e895d76196fb60aa20231107180153/8867e6?trk=TRN_TDL_01&utm_campaign=TRN_TDL_01&utm_medium=email&utm_source=sendgrid__;!!Mih3wA!B7mwW6QeCyDxtoO8oV9Oepb3GJkeMEx_K6TZ3yMAhM4Gxh70HI-nylDXZNvluCUmku4jkMAjFbon-L6YA38AAp5k$
>
>
>
> Message:
> Hello,
>
> We have recently used an eegosport 32 channels headset (ANT Neuro) for an
> experiment and
> are now analysing the data with eeglab. However we're struggling with what
> seems to be a
> rank issue.
> For the experiment we used OpenVibe and created a channel list of 34
> electrodes where CPz
> and AFz were at the very end so that OpenVibe could identify them as
> Reference and Ground.
> When importing our data from a GDF file on eeglab we therefore ended up
> with 34 electrodes
> in the Channel locations window.
> After confirmation from ANT Neuro, we manually rejected CPz and AFz to
> then follow the
> "Append Channel", "Look up locks", "Set reference" procedure from the
> section 5b-Retaining
> the reference channel chapter of the eeglab documentation. After these
> steps, we ended up
> with 32 electrodes and CPz set as a Reference in the blue eeglab window.
> We also tried to only reject AFz and directly set CPz as Reference but
> this procedure was
> slightly different as we ended up with 33 electrodes and CPz set as a
> Reference in the
> blue eeglab window.
>
> In both cases we tried launching ICA after only applying a bandpass filter
> and resampling
> the data to 250Hz and got a message (see ICA rank 0 message.png) saying
> the rank of the
> data matrix was lower than the number of input data channels. Hyeonseok et
> al., 2023 paper
> was useful to better understand this issue and suggested we check the rank
> by using the
> function rank(EEG.data).
>
> To our suprise, our rank was 30 (for Test_1.gdf).
> As two channels that have identical data induce a data rank of n-1 we
> checked our EEG.data
> matrix but values seemed correct. We therefore decided to take a look at a
> matrix
> correlation with corrcoef(EEG.data') and saw that sometimes, the values
> were close to 0,99
> (see Covariance_Test_1.mat).
>
> We tried different GDF files and some had a rank of 32 (see Test_2.gdf),
> others a rank of
> 29 (see Test_7.gdf). The amount of correlation values close to 0.99 varied
> from 0 to 4 in
> all our tested files.
> We tried launching an ICA everytime but got the same message whatever the
> rank.
>
> Therefore our questions are the following :
> - What could impact rank data ?
> - Why does the ICA error message still pop up when our rank is 32 ?
>
> I will attach to this message our code (ICA_rank_issue.m), three GDF files
> (Test_1.gdf;
> Test_2.gdf & Test_7.gdf) and their corresponding correlation matrix
> (Covariance_Test_1.mat, Covariance_Test_2.mat & Covariance_Test_7.mat).
>
> Thanking you in advance for your help.
>
> Best regards,
> Camille JEUNET-KELWAY, Emeline PIERRIEAU and Margaux IZAC
>
>
>
> 1 item
>
> Rank issue eeglab.zip - 33.4 MB
>
>
> [Get more out of WeTransfer, get Pro](
> https://urldefense.com/v3/__https://wetransfer.com/pro?trk=TRN_TDL_01&utm_campaign=TRN_TDL_01&utm_medium=email&utm_source=sendgrid__;!!Mih3wA!B7mwW6QeCyDxtoO8oV9Oepb3GJkeMEx_K6TZ3yMAhM4Gxh70HI-nylDXZNvluCUmku4jkMAjFbon-L6YA-Nnkf_s$
> )
>
>
>
> About WeTransfer:
> https://urldefense.com/v3/__https://wetransfer.com/about__;!!Mih3wA!B7mwW6QeCyDxtoO8oV9Oepb3GJkeMEx_K6TZ3yMAhM4Gxh70HI-nylDXZNvluCUmku4jkMAjFbon-L6YA_mPZqN2$
> Help:
> https://urldefense.com/v3/__https://wetransfer.zendesk.com/hc/en-us__;!!Mih3wA!B7mwW6QeCyDxtoO8oV9Oepb3GJkeMEx_K6TZ3yMAhM4Gxh70HI-nylDXZNvluCUmku4jkMAjFbon-L6YA1BEQ9Tc$
> Legal:
> https://urldefense.com/v3/__https://wetransfer.com/legal/terms__;!!Mih3wA!B7mwW6QeCyDxtoO8oV9Oepb3GJkeMEx_K6TZ3yMAhM4Gxh70HI-nylDXZNvluCUmku4jkMAjFbon-L6YA3lCemd2$
>
>
> To make sure our emails arrive, please add noreply at wetransfer.com to
> [your contacts](
> https://urldefense.com/v3/__https://wetransfer.zendesk.com/hc/en-us/articles/204909429__;!!Mih3wA!B7mwW6QeCyDxtoO8oV9Oepb3GJkeMEx_K6TZ3yMAhM4Gxh70HI-nylDXZNvluCUmku4jkMAjFbon-L6YAxaHWm92$
> ).
>
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu
> _______________________________________________
> Eeglablist page: http://sccn.ucsd.edu/eeglab/eeglabmail.html
> To unsubscribe, send an empty email to
> eeglablist-unsubscribe at sccn.ucsd.edu
> For digest mode, send an email with the subject "set digest mime" to
> eeglablist-request at sccn.ucsd.edu
>


More information about the eeglablist mailing list