[Eeglablist] Memory and MATLAB "pack" command

[arno]

Dear Tommaso,

using the "pack" command will not help much. "pack" simply defragment 
Matlab memory but does not increase it. Usually, not being able to read 
data has two causes: your recording sampling rate is high and the files 
are "very" big or the loading routines are not optimized enough and use 
too much memory. Here are some workaround to read your files.
1) close all existing programs before running Matlab or buy more memory 
(Matlab supports up to 16 Gb of Ram under 64-bit opterons).

2) read portion of continuous or epoch files, resample them, then 
concatenate them using menu "Edit > Merge datasets". We will update the 
merge command to be able to merge files "on disk" and delete the 
existing datasets in memory. This way, it is sure that you will have 
enough memory to merge any number of datasets (although you might not 
have enough memory to reload the merged file!).

3) read a few channels at a time, then resample your data, and when you 
are done fuse the files. For instance after reading the first channels 
into dataset 1 and the rest into dataset 2, to fuse them on the command 
line using type:

EEG = ALLEEG(2);
EEG.data(end+1: end+ALLEEG(1).nchan,:) = ALLEEG(1).data;
if ~isempty(EEG.chanlocs)
    EEG.chanlocs(end+1: end+ALLEEG(1).nchan) = ALLEEG(1).chanlocs;
end;
EEG = eeg_checkset(EEG);
[ALLEEG EEG CURRENTSET] = eeg_store(ALLEEG, EEG);
eeglab redraw;

4) optimize the reading routines to minimize memory usage (and send them 
back to us please!).

Arno

Tommaso Fadini wrote:
> Hello to all eeglablist members,
> I write to you about the problem which some of you have already  
> posted: not enough memory to load files.
>
> I would like to ask if someone knows if it is possible to use the  
> "pack" command (or some other one while running EEGLAB) with EEGLAB  
> 6.01b to write the loaded file to disk under linux (http:// 
> www.mathworks.com/support/tech-notes/1100/1106.html).
>
> Using Matlab, the swap space under linux is managed automatically or  
> there is a way to handle it?
>