Page 1 of 1

Questions about multi-column input X in function crqa

Posted: Sat Dec 10, 2016 21:19
by fan.mi
Hi everyone,

I am currently doing research which applies RQA to EEG data. This toolbox is amazing and I really appreciate the genius sharing of Dr. Marwan for this software.

The problem I am dealing with involves multi-channel EEG, which means they have N*m matrix as an input, where N is number of channels and m is number of time points. Up to now, I compute RQA measures channel by channel - but it is really slow. I wonder how it works when I directly input the N*m matrix to the crqa function, because I found this in the file:

% The input vectors can be multi-column vectors, where
% each column will be used as a component of the
% phase-space vector. However, if the first column is
% monotonically increasing, it will be used as an
% time scale for plotting.

I am wondering about how this works in terms of computation. As I understand, this means we are going to use the entire column as a vector to represents the coordinates in the re-constructed space - however, how the embedded dimension works in this case? Or does it means all embedded dimensions are set to 1 (no embedded dimension)? Any ideas will be helpful. Thanks a lot!

Re: Questions about multi-column input X in function crqa

Posted: Wed Dec 21, 2016 11:00
by Norbert
Hello Fan.Mi,

thanks for your kind words.

Regarding your question: it is something different when you calculate the RQA for each of the N channels separately or when you combine all N channels to one phase space trajectory (of then dimension N), because then you get only one RQA result. It depends on your research question whether it makes sense.

When you use the crqa function from the toolbox for such a N-dimensional vector, then it is not necessary to embed. Embedding is only necessary when you have only one time series and need to reconstruct the phase space. If the N channels are considered to be the N components of the phase space vector, then you have it already and not reconstruction/embedding is necessary.

In the toolbox you could sill apply embedding to the N-dimensional vector, but usually it makes only sense in some special cases.

I hope this answers has helped.

Best wishes
Norbert