epochs with no data?

✉️ Mailing List Archive (read-only)
1

External Email - Use Caution

Hello again!

Thank you for your responses on my last question, they were very helpful. I
am running into another problem now, however. I am epoching resting state
data into arbitrary 2 second windows. I am not automatically rejecting
epochs by annotation nor by a rejection parameter. I see that the epoch
object's drop_log is recording nearly the whole first half of the epochs as
"NO_DATA" epochs, however the raw file contains data for those time
indices. I am unsure what could be going on to result in this loss of data,
however it seems especially problematic due to the amount of epochs
dropped. Thank you in advance, again!

Best wishes,
Dillan
-------------- next part --------------
An HTML attachment was scrubbed...
URL: http://mail.nmr.mgh.harvard.edu/pipermail/mne_analysis/attachments/20190104/a73d8aa2/attachment.html

2

External Email - Use Caution

Hi Dillan,

This can happen when the epochs selected include (theoretical) samples beyond the (actual) data range. You should not have many of those epochs. Can you roughly tell how many no_data labels you found? I?d need to refresh my knowledge of the epochs reading code a bit to see if there may be other reasons for this. Let?s see what the others say in the meantime.

Best,
Denis

3

External Email - Use Caution

Hi Denis,

I limited the tmin and tmax of the epoch to 0 and 2, respectively, since I
don't any kind of stimulus onset in these epochs and don't want them to
overlap. This is partly what puzzles me, as I would think this would
prevent any inclusion of not-real data. My resting state data can be cut
into 93 two-second time windows, and this the number of events I am feeding
into the epoch object. It is labeling 41 of these 93 as 'NO_DATA'.

Thank you very much!

Dillan

4

External Email - Use Caution

Hi Dillan,

What is then the distance in seconds between any two events that you passed to the epochs constructor?

Denis

5

External Email - Use Caution

Hi Denis,

The events are 2 seconds apart from each other.

Thanks!

Dillan

6

External Email - Use Caution

Hum. That sounds puzzling. Can you reproduce this somehow with the MNE sample data? That could help to see if there is something wrong in the code or if something is special about your data.

Best,
Denis

7

External Email - Use Caution

Hi Denis,

Apologies for the delay in getting back to you. I've tried reproducing this
with a MNE sample dataset, and did not get the same effect. It would appear
then, that there might be something special about the data. Unfortunately
this dataset is not one that I collected. It was acquired from the Child
Mind Institute and as such, the extent of my knowledge about it goes only
so far as the public information provided on it!

I tried this on a different subject's data from the same CMI dataset. I got
the same results where the epoch.plot() is dropping over half of the
epochs, however this time this the drop_log is empty (!?!?!) and at this
point i am thoroughly confused. Here is the code where I am constructing
this Epoch object, so that you may see that there shouldn't be any
automatic rejection taking place:

mne.Epochs(eyes_closed, events=epoch_array, tmin=0, tmax=2,
event_id={'twoSec':7}, picks=scalpData, reject_by_annotation=False)

Any other guesses? Thank you for your help!

Best wishes,
Dillan

8

External Email - Use Caution

Hi Dillan-

How are you reading in the data? From the EGI formatted file? Can you provide more of the code?

9

External Email - Use Caution

Hi all,

I've written a previous email explaining in more detail the process leading
up to this epoching issue, but it seems that the email has been hung up
waiting for moderator approval. I'll start from the beginning:

Here is where I load up the raw data, which was imported from the CMI
database in a .raw format:

raw=mne.io.read_raw_egi (raw_file_path, montage=
mne.channels.read_montage(kind= 'GSM-HydrocCel-129'), preload=True,
verbose= True)

Output:

Reading EGI header from [raw_file_path] ....

Reading events..

Assembling measurement info...

Synthesizing trigger channel 'STI 014' ...

Excluding events {101,102,103.....[etc] } (the events listed as being
excluded here have the same labels as the nonsense channels talked about
below-- not sure what to make of that)

Reading 0 .... 168248 = 0.000 ... 336.496 secs ....

Upon importing the raw file there are quite a few channels that seem to be
not 'real' scalp or external electrodes included in the raw object. One of
them, 'STI 014,' is useful for reading the eyes closed/eyes open events but
there are about 50 other channels that contain no data and are randomly
labeled.

raw.info['ch_names'][129:] yields a list of these nonsensical channels: [
'101' , '102' , '3' , '11' ,........]

I split up and re-concatenate the raw dataset into its respective eyes-open
and eyes-closed sections, and proceed from there with 2 sets of raw data.

After going through several standard preprocessing steps (filtering,
re-referencing, ICA) without a hitch, I run into my issue with epoching the
data into 2 seconds.

The sample frequency is 500 Hz, so I construct an array of trigger events
by sampling every 1000th index and appending it to the array with a made-up
trigger code of 7.

print(epoch_array) yields [ [ 0, 0, 7], [1000, 0, 7], [2000, 0, 7], [ 3000,
0, 7].....]

I use this array to construct the epoch object:

twoSec_epoch = mne.Epochs(eyes_closed_raw, events=epoch_array, tmin=0,
tmax=0, event_id={'twoSec':7}, reject=None, flat=None,
reject_by_annotation=False)

Output:

93 matching events found

Applying baseline correction (mode: mean)

Not setting metadata

0 projection items activated

And of the 93 epochs in twoSec_epoch, the first 28 of them are being
dropped seemingly without reason.

twoSec_epoch.get_data()

Output:

Loading data for 93 events and 1001 original time points ...

28 bad epochs dropped

This happens whether or not the empty channels are dropped from the raw
object before creating the epoch object.

twoSec_epoch.drop_log[:28] yields [ ['NO_DATA'], ['NO_DATA'],
['NO_DATA'].....]

But I know that the 'real' electrodes (with the random fake channels
dropped) do contain data for these epochs:

raw_data_mat= eyes_closed_raw.get_data()

first_28_epochs= raw_data_mat[:128, 0:28*1000]

print(sum(sum(np.isnan(first_28_epochs))))

print(sum(sum(first_28_epochs==0)))

yields 0 and 0, ie there is data for each of the 128 channels.

I am unsure whether the issue is with the epoch_array I'm making or with
something else about this particular dataset. Any help/guesses would be
greatly appreciated!

Thank you and best wishes,

Dillan Cellier

10

External Email - Use Caution

hi Dillan,

yes you cannot post emails which are too big.

I would recommend you look at the stim channel in raw.plot
evenutally use raw.plot(events=events) to see if the events make
sense for your epochs.

I don't have access to your data so I can only give you hints
to visually debug

HTH
Alex

11

External Email - Use Caution

Hi Dillan and the rest,

I myself experienced an egi event exclusion problem and believe to have found a fix. I think (but am not completely sure) this might be the same issue as you are experiencing.

My data is recorded with 89 different events, and MNE will consistently exclude one or multiple events from the data, which of course affects further processing. This is happening since version 15. I noted this once in an issue on GitHub but this got burried under other issues.

If this fix works for you, I will try to open an issue again and maybe it can get fixed in the next version of MNE.

What worked is to change line 238 of egimff.py from

if event.sum() <= 1 and event_codes[ii]: to

if event.sum() <= 0 and event_codes[ii]:

if MNE is installed via anaconda you can probably find this file here:

C:\Users\ "your account" \Anaconda3\envs\mne\Lib\site-packages\mne\io\egi?

Greetings,

Lizette Heine

ps. First time posting on the mailinglist, sorry if this is not the usual way of doing things.

12

External Email - Use Caution

hi,

can you share a dataset to replicate this and we'll see how to
make this fix available for everyone.

A

13

External Email - Use Caution

Hi Lizette,

I actually ended up solving my issue in another way; the problem was due to
the mismatch between the _first_samps of my custom-constructed raw eyes
open and eyes closed datasets and the time indices I was feeding in to
create my 2 second epochs. I didn't notice a discrepancy between where MNE
was reading the beginning of my files and the event time indices in
epoch_array because when I would .get_data() and manually extract data
using those indices, it was not a problem. Only in conversion to an epoch
object did it begin skipping to the first in _first_samps.

Thank you for this follow-up, though!

Best wishes,
Dillan