MNE IO RAW Decimate

[umair-hassan]

Dear MNE Maintainers,

I am trying to decimate the raw data (loaded using raw.io), I do not care about the antialiasing and filters. I just need the decimated date (downsampled ratio basically). I am not sure if I can create epoch easily for the whole data and then downsample it, therefore I think inside resample (there could be an option to switch off the filters) or there could be a generic decimate method for raw io object as well.

Happy to help implementing it though.

Best, Umair

[welcome[bot]]

Hello! 👋 Thanks for opening your first issue here! ❤️ We will try to get back to you soon. 🚴🏽‍♂️

[agramfort]

@umair-hassan look at the decim parameter in mne.Epochs class

close the issue if it fixes your issue

[umair-hassan]

I understand that there exist a decim param in Epochs, but I do not want to Epoch the data at the continous raw stage when I am planning to decimate. Basically just loading the data using mne.io.raw method and just directly apply decimation. I also understand that it will impact my events, but I do not care about that at this stage of processing.

[cbrnr]

Why do you want to decimate raw (continuous) data without an anti-alias filter? Do you take care of that yourself? You can resample with raw.resample(), which does apply a suitable anti-aliasing filter though.

[umair-hassan]

I need it for a concurrent EEG fMRI artefact correction step. I cannot just apply antialiasing in the presence of huge gradient artefacts. The data was sampled at exceptionally high Fs therefore downsampling it to a reasonable Fs is also necessary. I will take care of anti-aliasing myself.

I think generally speaking for all brain stimulation, neuromodulation, TMS-EEG analysis too, this would be quite desirable.

[cbrnr]

You can always grab the continuous data with raw.get_data() and then decimate the NumPy array yourself. Would that work?

[umair-hassan]

Yes in principle.

[cbrnr]

We might consider adding a new parameter antialias=True to Raw.resample(), but I'd rather not unless it would be helpful for a larger group of users. So if you can live with the proposed workaround LMK, otherwise we'll wait for other devs to chime in.

[umair-hassan]

Thanks!

[larsoner]

People do ask for this from time to time. Because this keeps coming up (and since we have a non-antialiased .decimate for epochs+evoked), I'm okay with adding one that just takes every Nth sample as long as we state clearly that bad things can happen when you do this.

[cbrnr]

OK! Then instead of adding a parameter to resample(), it is better to add a new decimate() method as you suggested.

[umair-hassan]

Ping, just checking for an update here.

[jasmainak]

see related issue: https://github.com/mne-tools/mne-python/issues/10447