Issue with epoch end (sec) when splitting continuous recording using eeg_regepochs

[Jazelle759]

Description

I am not too sure if this is an issue or I misunderstood how to use eeg_regepochs.

After applying average reference to my continuous data, the epoch start (sec) is 0 and epoch end (sec) is 432.660.
I wanted to split the data into 10 seconds epochs so I tried using eeg_regepochs to split it.
I was hoping it will be divided into 43 epochs with each epoch start to be 0 and end to be 10.
It did get divided into 43 epochs with start of 0 but the epoch end was 0.996 instead of anywhere near 10 secs which seemed odd.

Screenshot links
before_epoch_screenshot
after_epoch_screenshot

I've uploaded the dataset I tried using eeg_regepochs on. Steps has been applied until average reference step using the code below before saving.
Example Dataset

#### Steps to Reproduce

1. Downsample data to 250
   EEG = pop_resample( EEG, 250);

2. Bandpass filter using IIR Butterworth filter in ERPLAB (1-30)
   EEG = pop_basicfilter( EEG, 1:128 , 'Boundary', 'boundary', 'Cutoff', [ 1 30], 'Design', 'butter', 'Filter', 'bandpass', 'Order', 2 );

3. Add channel locations
   EEG = pop_chanedit(EEG, 'load',{'\GSN-HydroCel-129.sfp' 'filetype' 'autodetect'},'changefield',{132 'datachan' 0},'setref',{'1:130' 'Cz'});

4. Apply average reference
   EEG = pop_reref( EEG, [],'refloc',struct('labels',{'Cz'},'Y',{0},'X',{5.4492e-16},'Z',{8.8992},'sph_theta',{0},'sph_phi',{90},'sph_radius',{8.8992},'theta',{0},'radius',{0},'type',{''},'ref',{''},'urchan',{132},'datachan',{0}));

5. Create epochs of 10 seconds each
   EEG = eeg_regepochs(EEG, 'recurrence', 10.0, 'rmbase', NaN);

#### Expected behavior: Epoch end (sec) to be 10
#### Actual behavior: Epoch end (sec) 0.996
#### Versions

OS version	Windows Server 2016 Standard
Matlab version	9.5.0.1033004/R2018b
EEGLAB version	v2020.0

[arnodelorme]

Yes, it is not a bug. It is because your samples have a specific length. In your case 1/250 = 0.004 s. So the latency of your first sample is 0, the second one 0.004, etc... However, even with 2 sample, the data length is 0.008 not 0.004. So sample at latency 0.996 is from time range 0.996 to 1.000

[ronjafa]

Hello

I had the same problem and through trial and error came to the solution. Namely, I think that you also have to define ‚limits'. This code worked for me to segment my data into 4 second epochs non overlapping.

EEG= eeg_regepochs (EEG, 'recurrence', 4, 'limits', [0 4])

[gdemirezen]

Yes, calling the function with defining limits explicitly is a work-around.

I see the problem is related to the usage of nargin in eeg_regepochs.m.
For example, the call eeg_regepochs(EEG, 'recurrence', 2) results in a local nargin value of 3.
Lines 106-108:
if nargin < 3
g.limits = [0 g.recurrence];
end
Since nargin=3, g.limits is not set as the default value of [0 recurrence_interval] as written in the function description but remains at [0 1], which was the default value input to finputcheck() at line 95.
Then at line 130, eplength is calculated as 1 since g.limits = [0 1] though it needed to be 2 in this case.

A control structure around lines 106-108 as in lines 87-94 can be used as a solution if ordered values without keys are to be input as well.

Note: I am using eeglab2021.1 and MATLAB 2020a.