remove_nans doesn't work

Problem description

lightkurve.lightcurve.FoldedLightCurve.remove_nans doesn’t seem to remove nans at all.

Example

import numpy as np
from lightkurve import KeplerLightCurveFile

KIC = 4570949
lc = KeplerLightCurveFile.from_archive(str(KIC), quarter=3, verbose=False).PDCSAP_FLUX
print(len(lc.flux[np.isnan(lc.flux)]))
lc.remove_nans()
print(len(lc.flux[np.isnan(lc.flux)]))

The result prints as 6 6 which suggests to me that there are as many nans before as after applying remove_nans.

Expected behavior

I expected the flux array to no longer contain nans.

Environment:

• platform (e.g. Linux, OSX, Windows): OSx
• lightkurve version (e.g. 1.0b6): 1.0b15
• installation method (e.g. pip, conda, source): pip

[barentsen]

This behavior by design. Methods like remove_nans, remove_outliers, bin, flatten, normalize, etc all create and return a new LightCurve object rather than modifying the existing one. This enables method chaining, e.g. you can do lc.remove_outliers().bin().plot(). It also makes it easier to preserve the original LightCurve object.

This is similar to the design of pandas where e.g. df1 = df.sort_values(by='foo') does not modify df and instead creates a brand new dataframe df1. However pandas does provide an inplace parameter to change this behavior, e.g. df.sort_values(by='foo', inplace=True) does modify df directly.

I am open to a discussion about whether or not the “method chaining” design we are using here is the right approach for Lightkurve. My default would be to keep it because we’re already invested in it, but we can still change it before the true 1.0 release if there are very compelling reasons!

[gully]

We could also consider adding an inplace operator. I am inclined to keep the current behavior for the purpose of method chaining.