Fill missing values issue and workaround

[nachonacho2]

Hi,
In the analysis example notebook, 'fill missing values' incorrectly interpolates using other track data if present.
Instead of using locations = fill_missing(locations), my workaround is to call the function individually for each node, axis, and track (easier than writing a new interpolation function from scratch). For example, with my data (single node, 5 tracks):

fill_missing(locations[:, 0, 0, 0])
fill_missing(locations[:, 0, 1, 0])
fill_missing(locations[:, 0, 0, 1])
fill_missing(locations[:, 0, 1, 1])
fill_missing(locations[:, 0, 0, 2])
fill_missing(locations[:, 0, 1, 2])
fill_missing(locations[:, 0, 0, 3])
fill_missing(locations[:, 0, 1, 3])
fill_missing(locations[:, 0, 0, 4])
fill_missing(locations[:, 0, 1, 4])

On a related note, I'm wondering what was your reason for using cubic interpolation rather than linear?
Cheers

[talmo]

Hi @nachonacho2,

Happy new year and sorry about the delay!

Thanks for pointing this out. I think the behavior of the interpolation function might have changed with a recent scipy version (or we forgot to update the notebook) -- I'll make the changes to ensure that it works.

Regarding interpolation: this is a bit of an arbitrary decision (see relevant discussion in #6). I think here the reasoning was that linear interpolation would result in constant velocity, whereas cubic allows for a bit more "wiggliness" in the fit. That said, linear (or even nearest) is probably a safer general purpose interpolant for imputation. I'll change the default in the notebook.

Cheers,

Talmo

[talmo]

Analysis notebook updated and should take care of this issue, but feel free to re-open if you're still having trouble. Cheers!

[nachonacho2]

Happy New Year @talmo!
Thanks for the fix!
Cheers