Fix powerspectrum from lightcurve - wrong size, float casting

[matteobachetti]

Resolves two recently reported issues:

• avg_*s_from_event methods not accepting complex fluxes and
• avg_*s_from_event methods throwing errors of "no segment shorter than gti"

The first one comes from our forcing a float casting on input fluxes (to avoid integers - that can lead to casting errors in Numba routines, and quadruple precision, which is overkill), that kills complex numbers.

The second one comes from our re-defining dt to accommodate an exact number of bins inside a segment. This works well with actual events, but does damage when you have a light curve that was sampled with a given dt.

[codecov]

Codecov Report

Merging #760 (9267e68) into main (5fac201) will increase coverage by 0.62%.
The diff coverage is 100.00%.

❗ Current head 9267e68 differs from pull request most recent head 962543a. Consider uploading reports for the commit 962543a to get more accurate results

@@            Coverage Diff             @@
##             main     #760      +/-   ##
==========================================
+ Coverage   96.62%   97.25%   +0.62%     
==========================================
  Files          41       41              
  Lines        7388     7396       +8     
==========================================
+ Hits         7139     7193      +54     
+ Misses        249      203      -46     

Files	Coverage Δ
stingray/fourier.py	99.76% <100.00%> (+<0.01%)	⬆️

... and 6 files with indirect coverage changes

📣 We’re building smart automated test selection to slash your CI/CD build times. Learn more

[mgullik]

If we have this line, do you think it's better to set segment_size = None as the default value?

[matteobachetti]

Yeah, it would indeed make sense. However, this would be a breaking change and would require to make a number of modifications to the rest of the code as well.

[mgullik]

I understand, it makes sense

[mgullik]

Hi @matteobachetti,

great upgrade to the code!
If I understood correctly you define
dt = np.median(np.diff(times[:100]))
in the case dt is not given. I was wondering what happens if there are less than 100 bins in the time array? And also if the function checks whether or not the light curve is evenly sampled?

There are also a couple of comments attached to the file

[matteobachetti]

times[:100] is safe if the array is shorter than 100, it will just give the full array. These functions assume that data are evenly sampled. Making the check every time would add significant computation in large datasets (checks of this kind were one of the reasons Stingray was so slow in the past 😅 )