[Eeglablist] Re: component map

[Scott Makeig]

chriszl at brandeis.edu wrote:
> Hi, eeg team, I'm trying out EEGLAB on our new 64 channel data, I have a 
> suggestion that you include units in each and every scalp map, because some 
> of them mean dB, some of them mean milli volt, while others don't have any 
> units. 
> If I understand it right, after ICA, 64 channel data should give independent  
> 64 components, each of which is a time series, what does it mean to plot a 
> scalp map for each component? the color on the component map should mean 
> weight of each channel, right? so it doesn't have any unit. 
>  
> best 
> Chris 

Chris -

ICA divides the scaled data into the product of two matrices 
(maps*activations = data). Somehow we did not adopt the convention
to normalize either, which would throw the units into the other matrix 
(though  infomax more or less normalizes the activations). Else, we 
could adopt the PCA formalism,  maps*scalefactors*activations = data, 
where both matrices are normalized and the diagonal scalefactors matrix 
holds the units... (Your vote?)

Maps in themselves are unscaled because they vary widely in uV accounted 
for (if activations are normalized, leaving the maps to hold the 
scaling), so the color scale would have to be different for every map 
(i.e., different color bar by every map). We use the convention that 
green=0. For the 'jet' colormap we use, this means that the color limits 
(caxis) are symmetrical, but are (usually) individually scaled for the 
map so bright color contrasts are visible.

Scott Makeig