If the eyetracker loses an eye (or during a blink), we get a series of nans in our time series. Currently these are not detected by any function. We could add functionality to detect these.
This does not prevent blink detection (left) because there is usually a steep period on each side, but complete dropouts (right) are unflagged:
There is also a related bug where comparing np.nan to anything will always return false, so any window with a nan in it cannot be rejected. That is probably more urgent
If the eyetracker loses an eye (or during a blink), we get a series of nans in our time series. Currently these are not detected by any function. We could add functionality to detect these.
This does not prevent blink detection (left) because there is usually a steep period on each side, but complete dropouts (right) are unflagged: