search

dwkim78 / upsilon

Automated Classification of Periodic Variable Stars Using Machine Learning
MIT License
42 stars 11 forks source link

Period Problem #7

Closed azim7091 closed 4 years ago

azim7091 commented 4 years ago

lc_out00026.txt

Hi I have a light curve of an obvious variable with really obvious period, but the code cannot find the correct period? where does this problem come from and how can improve code ability to guess a better period?

I have attached the light curve, but let me know if you cannot find it (sorry, I am new to github).

dwkim78 commented 4 years ago

Hi,

You first need to understand Lomb-Scargle algorithm which UPSILoN used to find period. Your lightcurve's periodogram looks like:

Screen Shot 2020-05-25 at 9 49 02 AM

, which is not ideal for finding period.

Cheers,

azim7091 commented 4 years ago

Thank you so much for your responses. I will check the algorithm. Another problem is about amplitude estimation, it is even worse. The code gives "amplitude', 141.59747680075938" which is wrong.

Is it also related to Lomb-Scargle algorithm?

Best Regards Azim

kirxkirx commented 4 years ago

If I may add... The input lightcurve is short: two consecutive cycles of a delta Scuti type variable. This is probably not what the code is expecting. But in principle, the lightcurve is sufficient to correctly determine the period and variability type. The trick is to restrict the trial period range to a sensible range (something like 0.1-0.01d) after visually inspecting the lightcurve and approximately guessing the period. In this particular case, the most important is to set the short-period limit of 0.01d in order to cut off the false peaks at high frequencies (where the period search algorithm is trying to fit a sine wave with a period equal to the distance between the constitutive data points -- this is not what we want to do physically and the amplitude of such wave will not be properly constrained, of course).

azim7091 commented 4 years ago

Thanks a lot for the explanation. I have checked this suggestion for the variable, and I found that it can identify the variable (DSCT) in a better way for fewer data points and gives a better estimation for period (2 times the real one). But for very few data points (.01 day) the result is not acceptable.

I have attached the results. 13points.txt 33points.txt 45points.txt 95points.txt