[Eeglablist] Amica & SIFT

[Makoto Miyakoshi]

Dear Marjan,

> Actually, I think the reason behind my problem is PCA which keeps
components having more variance and I hope there is a solution.

I did not understand what you mean here.

> According to the SIFT toolbox, I have an issue to balance between
consistency and
residual whitening.

No one can pass the whiteness test though. See my past 6 publications. I
could not report successful achievement of complete whiteness in any of
them. You can still publish the result, and if you care, report it in the
limitation section. The residual being not completely white is different
from, for example, your result is biased toward false positives. I thought
it is benign.
https://urldefense.com/v3/__https://github.com/sccn/groupSIFT/blob/master/README.md__;!!Mih3wA!CN1uWLXSVFZgJ9yrvo76esnnceM1OThFkX_tAnmrnidunxCEgtqBXd5PBxak__hwet3z5M2zJ-LtwJgzs22YD8AvBMQ$ 

> I also have no idea which window length and optimal
order are suitable to strike a balance among those(consistency and residual
whitening).

If your data is from an event-related brain potential study and epoched, I
suggest -1 to 2 s for epoching and sliding window length 1 s so that you
can analyze down to 1 Hz.
If your stimulus onset asynchrony (SOA) is shorter than 2 s, use shorter
epoch windows accordingly, such as -0.5 to 1.5 s if SOA is 2 s.
You don't want to use too aggressively short sliding window. It'll make the
parameter-to-datapoint ratio more challenging. But I sometimes saw that
when I used 0.5 s sliding window, the whiteness tests showed good results.
I did not bother to investigate it because I already gave it up.

> Although I studied SIFT instruction, I have two problems, when
I increase the length of window, consistency become good (PC>80%), but
residual whitening become zero everytime and when I decrease winlenght
(PC<80٪) and residual become good ?

Oh I see, so you saw something similar.
Generally speaking, using a shorter time window means having improved local
stationarity. Those parameter changes you saw are related to it.
Tim's official SIFT manual recommends very idealistic values for the
quality check. I don't think many researcher's datasets can pass that
criteria. See my papers for the reported parameters. I think you'll feel
better.

Makoto

On Wed, Jul 20, 2022 at 10:27 AM m za via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:

> Dear Admin
>
> I am writing this email because I still have a problem with Amica .
> Actually, I think the reason behind my problem is PCA which keeps
> components having more variance and I hope there is a solution. According
> to the SIFT toolbox, I have an issue to balance between consistency and
> residual whitening. I also have no idea which window length and optimal
> order are suitable to strike a balance among those(consistency and residual
> whitening). Although I studied SIFT instruction, I have two problems, when
> I increase the lenght of window, consistency become good(PC>80%), but
> residual whitening become zero everytime and when I decrease winlenght
> (PC<80٪) and residual become good ?
>
> Best regards
> Marjan
>
>
> https://urldefense.com/v3/__https://drive.google.com/file/d/1LN92alwktqfIXtvmBJJP-Q37p39Ddy7X/view?usp=drivesdk__;!!Mih3wA!D8xBnpdJ7OAhiaSHPmtoMR4fgTmV6cGkNMHckLqauGq4XuTz0VbIesV3qGp3p7hO9ApPzEfDNMQyaCkRCuQPXa6b96o$
>
>
> https://urldefense.com/v3/__https://drive.google.com/file/d/1LT6ve0RUJSclr8jBT4vCeIkRwiGYUnA0/view?usp=drivesdk__;!!Mih3wA!D8xBnpdJ7OAhiaSHPmtoMR4fgTmV6cGkNMHckLqauGq4XuTz0VbIesV3qGp3p7hO9ApPzEfDNMQyaCkRCuQPi7C-Ex0$
> _______________________________________________
> 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