[Eeglablist] ICA variability

Scott Makeig smakeig at gmail.com
Fri Apr 9 09:37:18 PDT 2021

Claudia -

ICA decomposition (e.g., runica using infomax, or AMICA) uses a random seed
to begin learning the decomposition from the data. Thus, ICA decomposition
run twice on the same data will return *slightly* different results. If the
results are *too* different, then you should investigate whether the
decompositions (one or both) were failures -- from not having enough data,
or having more channels than independent signals (for example, when one or
more channels are interpolated from the others).

Scott Makeig

On Fri, Apr 9, 2021 at 11:44 AM Krogmeier, Claudia M via eeglablist <
eeglablist at sccn.ucsd.edu> wrote:

> Hello,
> Is it normal for ICA decomposition to behave differently each run, despite
> no changes in settings?  I also notice more variability in the values I am
> outputting after preprocessing with datasets in which a channel must be
> interpolated.
> I discovered that my values were not always the same despite no changes to
> preprocessing or analysis when I was testing an automated preprocessing
> script. I realized that not only were my values from the automated script
> different from the values I get by manually clicking through the same
> processing functions myself - but these values are different every time
> they are preprocessed - manually or with an automated script.
> I have tried clearing eeglab between saving different datasets, in case
> files are not correctly being overwritten.
> I would not expect the values to be so different, but perhaps this is how
> ICA works. Thank you for any thoughts.
> Claudia
> _______________________________________________
> 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

Scott Makeig, Research Scientist and Director, Swartz Center for
Computational Neuroscience, Institute for Neural Computation, University of
California San Diego, La Jolla CA 92093-0559, http://sccn.ucsd.edu/~scott

More information about the eeglablist mailing list