To clarify further, double precision is most necessary when running iterative calculations, incuding ICA decomposition, in which the 32-bit floating-point round-off errors would increase (at least) geometrically... The raw data, however are < 32-bit resolution, so saving it in double precision is not necessary.<div>
<br></div><div>Scott Makeig<br><br><div class="gmail_quote">On Sat, Apr 16, 2011 at 9:08 AM, Arnaud Delorme <span dir="ltr"><<a href="mailto:arno@ucsd.edu">arno@ucsd.edu</a>></span> wrote:<br><blockquote class="gmail_quote" style="margin:0 0 0 .8ex;border-left:1px #ccc solid;padding-left:1ex;">
Hi Niko,<br>
<br>
the reason EEGLAB saves in single precision is because no manufacturer can record precision of more than 32-bit (single precision). The actual resolution of most systems is 18 meaningful bits. The BDF format for instance only saves the first 24 bits (the EDF format only saves the first 16 bits so it might truncate some information).<br>
<br>
Then you might ask, why bother with double precision at all (double precision is 64-bit). The reason is that data processing algorithms (ICA for instance) do complex calculation on the data and that sometimes the result might not be accurate if you use 32 bits. This was especially true at the start of Matlab 7.x when single-precision routine had just been introduced and when single-precision routines were using some not-so-well-chosen approximations. For instance, when running ICA, we still systematically convert the data to double precision (unless there is not enough memory) even if you have selected single precision in your options.<br>
<br>
So the basic answer is: we sometimes need double precision for some complex calculation but it is not necessary to save as double precision since the first 18 bits of the data contain all of the relevant EEG information.<br>
<br>
Hope this helps,<br>
<br>
Arno<br>
<div><div></div><div class="h5"><br>
On Apr 12, 2011, at 5:07 PM, Niko Busch wrote:<br>
<br>
> Hi everyone,<br>
><br>
> I recently had trouble using signal processing functions because my EEG data are in single precision, whereas these functions expect double precision. On my machine, RAM is not really an issue, so I unchecked the EEGLAB option to use single precision to save RAM.<br>
><br>
> After digging through a number of EEGLAB functions, which were used for pre-processing the data, I found that most functions do read the eeg_options file and use single/double precision accordingly. However, pop_saveset seems to force the data to single precision when the data are stored to the hard drive no matter which options are set in the eeg_options file:<br>
><br>
> if ~strcmpi(class(EEG.data), 'memmapdata') & ~strcmpi(class(EEG.data), 'single')<br>
> tmpdata = single(reshape(EEG.data, EEG.nbchan, EEG.pnts*EEG.trials));<br>
><br>
> Of course I could delete these lines or use double(EEG.data) whenever I need double precision. However, I would like to understand why pop_saveset forces single precision and why it does not use the eeg_options.<br>
><br>
> Any ideas, anyone?<br>
><br>
> Best<br>
> Niko<br>
</div></div>> _______________________________________________<br>
> Eeglablist page: <a href="http://sccn.ucsd.edu/eeglab/eeglabmail.html" target="_blank">http://sccn.ucsd.edu/eeglab/eeglabmail.html</a><br>
> To unsubscribe, send an empty email to <a href="mailto:eeglablist-unsubscribe@sccn.ucsd.edu">eeglablist-unsubscribe@sccn.ucsd.edu</a><br>
> For digest mode, send an email with the subject "set digest mime" to <a href="mailto:eeglablist-request@sccn.ucsd.edu">eeglablist-request@sccn.ucsd.edu</a><br>
<br>
<br>
_______________________________________________<br>
Eeglablist page: <a href="http://sccn.ucsd.edu/eeglab/eeglabmail.html" target="_blank">http://sccn.ucsd.edu/eeglab/eeglabmail.html</a><br>
To unsubscribe, send an empty email to <a href="mailto:eeglablist-unsubscribe@sccn.ucsd.edu">eeglablist-unsubscribe@sccn.ucsd.edu</a><br>
For digest mode, send an email with the subject "set digest mime" to <a href="mailto:eeglablist-request@sccn.ucsd.edu">eeglablist-request@sccn.ucsd.edu</a><br>
</blockquote></div><br><br clear="all"><br>-- <br>Scott Makeig, Research Scientist and Director, Swartz Center for Computational Neuroscience, Institute for Neural Computation & Adj. Prof. of Neurosciences, University of California San Diego, La Jolla CA 92093-0559, <a href="http://sccn.ucsd.edu/~scott" target="_blank">http://sccn.ucsd.edu/~scott</a><br>
</div>