Inconsistent behaviour of tstart and tseg after using lc.truncate

[andresgur]

Description of the Bug

If a lightcurve with a gti is truncated, tstart and tseg behave inconsistently before and after the truncation. Prior to truncation, both are defined based on the times and dt e.g. tstart is defined in the usual way time[0] - dt/2.

After truncating tstart is defined by the first GTI starting time and tseg (incorrectly) based on the gti first and last times. I would have reopened #751 but I don't have privileges.

In truncate method of base.py, I don't think this is correct:

new_ts.tstart = new_ts.gti[0, 0]
new_ts.tseg = new_ts.gti[-1, 1] - new_ts.gti[0, 0]

I think the new tstart and tseg should be defined based on the updated times, unless I'm missing something. If we agree on that I could try to fix it myself.

Steps/Code to Replicate the Bug

from stingray import Lightcurve
import numpy as np

times = np.arange(50)

mean_flux = 100.0
std_flux = 2.0 
flux = np.random.normal(loc=mean_flux, scale=std_flux, size=len(times))
flux_err = np.ones_like(flux) * std_flux


lc = Lightcurve(times, flux, err=flux_err, err_dist="None", dt=1.0, skip_checks=True)

# No GTI
lc = Lightcurve(times, flux, err=flux_err, err_dist="None", dt=1.0, skip_checks=True)
print("No GTI\n=====")
print("Tstart before truncating", lc.tstart)
print("Tseg before truncating", lc.tseg)
newlc = lc.truncate(10, method="time")
print("Tstart after truncating", newlc.tstart)
print("Tstart after truncating", newlc.tseg)

# GTI
gti = [[25, 30], [40, 45]]

lc = Lightcurve(times, flux, err=flux_err, gti=gti, err_dist="None", dt=1.0, skip_checks=True)
print("GTI\n=====")
print("Tstart before truncating", lc.tstart)
print("Tseg tefore truncating", lc.tseg)
newlc = lc.truncate(10, method="time")
print("Tstart after truncating", newlc.tstart)
print("Tseg after truncating", newlc.tseg)

Expected Results

tstart should be defined consistently throughout, as lc.time[0] - lc.dt/2 I would say.
Same thing for tseg, which should be defined based on lc.time and lc.dt and not on the gti.

Actual Results

No GTI

Tstart before truncating -0.5
Tseg before truncating 50.0
Tstart after truncating 9.5
Tstart after truncating 40.0

GTI

Tstart before truncating -0.5
Tseg tefore truncating 50.0
Tstart after truncating 25.0
Tseg after truncating 20.0

[ankitkhushwaha]

this issue might require some more thoughts, @matteobachetti your guidance would be very helpful here.

[matteobachetti]

@andresgur in all our analysis, what counts is the good light curve bins, those inside GTIs. AveragedPowerspectrum etc will always ignore data outside GTIs. Therefore, tstart and tseg are better defined using the data intervals that actually matter. I hope this helps

[andresgur]

Hi @matteobachetti I can see that you obviously want to use the GTI in any subsequent analysis but from an outsider perspective, I was expecting tstart and tseg to return properties intrinsic to the lightcurve, not the GTI. I guess both can work as long as the code is clear on what each attributes means and it is almost more of a "personal" decision which one you return (which I'm not here to decide obvs). Or I guess if the powerspectrum objects are using these attributes directly, then I you need to stick to the GTI. In any case the definitions need to be consistent throughout, which needs fixing.

One thing though,

new_ts.tseg = new_ts.gti[-1, 1] - new_ts.gti[0, 0]

I can easily see this definition being problematic as the GTI will not necessarily be continuous, do you agree with this?

[matteobachetti]

@andresgur I thought a little about this, and I don't know how to solve this without breaking the API. tseg is used sparsely and always assumed to be from a single GTI, I agree that it loses meaning when GTIs are interrupted in the middle, I guess we should think to remove it completely in future versions of the code.