Support light curves centered around zero

Problem description

For light curves input by the user, it is not uncommon for the time series flux to be normalized and mean-subtracted so that the relative flux is centered around zero.

This causes problems when lightkurve attempts to re-normalize (often unnecessarily) the flux by dividing the median which is very close or equal to zero.

The minimum thing to add is a warning of possible bad behavior when normalize is called on a light curve that contains negative values.

We should also remove unnecessary calls to lc.normalize(), such as the one in periodogram.from_lightcurve that has no practical effect. I will submit a pull request for this soon. Perhaps auto-renormalization could also not be the default where it’s an option, like in lightcurve.plot (change default to normalize=False).

Has zero-centered flux been previously considered for the lightkurve internal default?

Example

import numpy as np
import lightkurve as lk

time = np.arange(100)
amp = 0.1 #What we want to measure
freq = 0.08412
flux = amp*np.cos(2*np.pi*np.arange(100)*freq)

lc = lk.LightCurve(time,flux)

ls = lc.to_periodogram(normalization='amplitude')
print(ls.max_power)

Gives… 23627464 ppm

Expected behavior

Expect to measure close to the input amplitude of 1e5 ppm. Renormalization under the hood (division by small number) causes the variability to blow up.

Environment

• platform (e.g. Linux, OSX, Windows): OSX
• lightkurve version (e.g. 1.0b6): 1.0.1
• installation method (e.g. pip, conda, source): conda

[barentsen]

Thanks Keaton!! Please do keep thinking about this and feel free to open new issues with examples. I love that you are making us think about this!!

(I’m going to close this issue for clarity, because I think the original problem has been resolved!)

[keatonb]

That was basically my suggestion. Let’s table it for now and see if anyone else has need for this. You’re right that this can be done manually as is.