Suggestions + improvements

[pavolgaj]

I start to make a big code change. I mainly want to use the emcee package for Monte Carlo fitting (instead of quite old pymc v2). If someone has some suggestions on what to change in the OCFit package and/or GUI, please, write here. I also have some;)
Hopefully, the completely new version will be published this year...

Planned changes

• MC fitting using emcee - changes in fitting function, InfoMCMC class and GUI (Fit.Params window - 9f53a82) - done in b5e4f7c
• calculation of absolute params. in GUI (New functions in GUI #4) - maybe add a new window to set required params (82155fc)
• remember values set in FitParams in GUI and set them after reopening this window (New functions in GUI #4) - 0136380 (for old MCMC)
• adding some new models: Quad+Apsidal (added in dc833b3), maybe ETV model of Borkovits (models for ETVs #5)
• epoch calculation for elliptic orbits (Secondary minima for eccentric orbit #1) - in OCFit (2b6d743) and GUI (81823d1)
• store data in JSON except for the pickle file - done in aa4e97a
• change fitting from GA to scipy DE (https://docs.scipy.org/doc/scipy/reference/generated/scipy.optimize.differential_evolution.html) - add DE as new function (6478182 and 60730b2 - GUI)
• calculation of additional params (e.g. dP, dM) from quadratic model - dc833b3
• add buttons for running InfoMCMC (or maybe also InfoGA) to GUI - fdf1557
• change FitLinear/FitQuad in GUI - fitting methods (standard, robust, MC) - new Window FitParams (537bb2e)
• GUI: update ephemeris after button press (4229b95)
• update docs on GitHub/Wiki page
• add models: LiTE3+Apsidal and LiTE3Quad+Apsidal (or LiTE3+ApsidalQuad) - 5f51946

[kolarjakubmuni]

Hi. I would suggest the possibility of fitting more than one phenomenon in the O-C diagram, e.g. parabolic change with apsidal motion, etc. For my field of study (quadruple eclipsing systems) we have usually two O-C diagrams, for pair A and B. We could fit them separately for knowing what is happening with the pair. However, these two pairs are gravitationally bound and influence each other. So their O-C diagrams can show us their mutual motion. Maybe it should be very useful to fit both O-C diagrams together. Of course both pairs have different ephemeris values and it would be quite large change in the software.

From the paper Zasche et al. 2019, there is an example of the O-C diagrams for both pairs.

[pavolgaj]

A combination of parabolic trend and apsidal motion is easy to implement.

The quadrupole systems are fascinating but fitting the O-Cs from both pairs together would be really complicated and I don't know if it would have some benefits. It would be necessary to load and store O-Cs separately which requires changing a big part of the code (initialization, fitting, plotting etc.). And only parameter that should be the same for O-Cs of both pairs is their mutual orbital period (label as P3 in OCFit). I don't plan to implement this possibility, now - maybe if I'll study such systems, my opinion changes;)

[pavolgaj]

I just changed the FitMCMC function to use the emcee package (commit b5e4f7c in a dev branch). I'm open to any of your comments or possible issues with it.
The original function (used pymc) was renamed to FitMCMC_old and the old InfoMC class moved to file info_mc_old.py - emcee and pymc generate a bit different output files.

I compared pymc and emcee fitting and they give very similar results. Maybe, the emcee gives more realistic posterior distributions. And from my point of view, fitting using an emcee is faster.

[kolarjakubmuni]

Maybe the fitting primary and secondary minima, which are vertically shifted simultaneously. Something like this.

(From my preliminary results of one system.)

[pavolgaj]

@kolarjakubmuni Interesting... The difference in O-Cs for primary and secondary for nearly 6 hours is something strange. Could you write some specifications for this system? Isn't it eccentric?

[pavolgaj]

Change default file format from pickle to JSON (aa4e97a). The format can be changed manually with parameter format in the Save function - it has 2 possible values 'json' or 'pickle'. Detection of the file format before loading it is done automatically. So, loading files saved in the older version of OCFit (pickle format) is possible.

[pavolgaj]

Added new function FitDE for fitting using Differential Evolution from scipy (6478182). It's an alternative method to Genetic Algorithms in the FitGA function. DE should be a bit faster compared to GA because of internal converge control - not all iterations need to be run.

[pavolgaj]

@kolarjakubmuni:

Hi. I would suggest the possibility of fitting more than one phenomenon in the O-C diagram, e.g. parabolic change with apsidal motion, etc.

I just implement a new model called ApsidalQuad. I don't have any good data to test it but I don't expect any problems.

[kolarjakubmuni]

@kolarjakubmuni Interesting... The difference in O-Cs for primary and secondary for nearly 6 hours is something strange. Could you write some specifications for this system? Isn't it eccentric?

Yes, the difference is quite large. We don't know much about the system, it should be quadruple doubly eclipsing. So there is probably a combination with more effects and the system is probably eccentric. However, the secondary minima are very shallow and we need to confirm these differences (this is from the TESS data).

[pavolgaj]

Commit 2b6d743 solves the problem with eccentric orbits (#1). New function DeltaEpoch calculate phase (difference in epoch) for secondary minima from eccentricity and argument of pericenter. All classes have a new input parameter dE which is used to calculate epochs and types of minima.

[pavolgaj]

@kjkoeller I start to modify GUI. I just added some new buttons - so I had to make the main window a bit larger. Could you check if it looks good also on bigger screens (with set fixed=False - line 2908). If not, please, change screenheight on line 2916.

[pavolgaj]

Added function to calculate or estimate phase (difference in epoch) of secondary minima in GUI (81823d1). Estimation is done from the difference between O-Cs of primary and secondary minima.

[pavolgaj]

Added new window InfoMCMC in GUI. It can load saved sampling of MCMC or GA/DE fitting from a file and generate some figures useful for analysing sampling. Files generated by pymc are not supported! Only files from emcee could be used.

[pavolgaj]

Implemented new fitting methods (Robust Regression and MCMC) for FitLinear and FitQuad in GUI (537bb2e). Added new window FitParams for setting necessary parameters.

[pavolgaj]

Merged to master branch. Uploaded to PyPI. Created a new release.

[pavolgaj]

Added models LiTE3Apsidal and LiTE3ApsidalQuad (5f51946).