Extreme min/max for gaze coordinates

[kalexandriabond]

Hi,

When I run the following minimal example, I get extreme values for the min and max gaze position (0.0 and 100000000.0, respectively) that are well outside the stated gaze coordinates (GAZE_COORDS 0.00 0.00 1919.00 1079.00). Do you know why this is? Thanks in advance for your help.

from pyedfread import edf

samples, events, messages = edf.pread('SUB001.EDF')

print('min gaze coordinates ', samples.gx_right.min(), samples.gy_right.min())
print('max gaze coordinates ', samples.gx_right.max(), samples.gy_right.max())

[gschwim]

Hi @KMBond ,

It could be bad samples from outside of a recorded trial. Try bounding the samples by the start and end times of a trial and see if its still like that. It could also be loss of lock on the eye (e.g. blink, etc).

[kalexandriabond]

Hmm, okay. Should this be the case using the sample data provided with the package (SUB001.EDF)?

These extreme values don't appear to be in the raw EDF or the ASC files. So I think it would be worth considering whether there's an issue with the parsing of the raw data to a pd.DataFrame.

[gschwim]

Hmm. I admit I never looked at the sample data. I haven't noticed a problem with my own data but it's worth checking out. Will see if I can get it going in the next few days.

[gschwim]

Hi. I took a look at one of our recent trial data sets. I can confirm that I am seeing similar to you. Outside of a trial I do get 0.0 for the minimum but not inside, possibly due to a lack of blinks but I'm speculating. The max is the same regardless. I don't see these numbers in the .asc.

Maybe the source code will reveal why?

[selimonat]

I think your issue might be due to blinks, as they distort the actual x and y gaze positions. When the eyelids are passing over the pupil the calibrated model might be interpreting this as upward or downward movement leading to such extreme values.

…

On Fri 27. Sep 2019 at 17:54, Greg Schwimer ***@***.***> wrote: Hi. I took a look at one of our recent trial data sets. I can confirm that I am seeing similar to you. Outside of a trial I do get 0.0 for the minimum but not inside, possibly due to a lack of blinks but I'm speculating. The max is the same regardless. I don't see these numbers in the .asc. Maybe the source code will reveal why? — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#16>, or mute the thread <https://github.com/notifications/unsubscribe-auth/AA7F73OPCKY4XJRTSQBC5ZLQLYUDTANCNFSM4I27Y4WA> .

[kalexandriabond]

If the extreme gaze data were due to blinks, then those extreme values should also be present in the ASC file (but they aren't). @selimonat?

[behinger]

Hi! some time ago, I took over maintenance and I finally looked into this. While there is no 100_000_000 in the ASC file, there are some parts where empty (?) samples exist:

These are precisely the location where we get that high gx_right values. I didnt find anything in pyedfread nor in the header files of the API that could explain this. I'll add a warning if such values are observed - not sure how else to proceed here

[flyingfalling]

Chiming in here, I used to filter out INT_MAX and INT_MIN (for their 16bit ints?) samples since it seems they were used as special markers in the EDF files (to indicate NAN, i.e. missing values).