I've recently seen the @jakevdp's supersmoother code and it looks like a nice option for trying to smooth the Kepler data - in particular the periodic functionality that it has. I should implement this soon and seen how it effects flares, or whether it is able to leave them relatively untouched.

I have created a patch that allows the background variations in the light curve to be modelled by a number of sinusoids as well as a polynomial variation. If you specify how many sinusoids to fit then the code will perform a Lomb-Scargle periodogram of the light curve and find the frequencies of the peaks. It will sort the peaks in descending amplitude and use the number of sinusoids you specified from this descending list. The amplitude and phase of each of these will then be analytically marginalised over (by splitting each sinusoid into a sine and cosine part with separate amplitudes, e.g. Bretthort, and marginalising over the amplitudes in the same way as is done for the polynomial coefficients. This will hopefully help allow the code to be used when high frequency variations are present, and remove some of the effects of the sliding background window.

Some other minor things in the patch:

• In the Bayes factor functions I've now removed the bits of code that tried to deal with edge effects (e.g. the model sliding onto and off of the data). This was just confusing code and was generally redundant as you would generally just ignore the edges of the data anyway.
• In the cython codes now just refer to the number of background models and not the background polynomial order. This makes things more general and slightly less confusing.

Hello,

when I try to import the package I get the following error, I'm running it on a Mac:

import bayesflare
Traceback (most recent call last):
File "", line 1, in
File "/Users/aaron/anaconda/lib/python2.7/site-packages/bayesflare/init.py", line 36, in
from .models import Model, Flare, Transit, Expdecay, Impulse, Gaussian, Step, ModelCurve
File "/Users/aaron/anaconda/lib/python2.7/site-packages/bayesflare/models/init.py", line 5, in
from .model import Model, Flare, Transit, Expdecay, Impulse, Gaussian, Step, ModelCurve
File "/Users/aaron/anaconda/lib/python2.7/site-packages/bayesflare/models/model.py", line 4, in
from ..stats.general import logminus
File "/Users/aaron/anaconda/lib/python2.7/site-packages/bayesflare/stats/init.py", line 11, in
from .general import *
ImportError: No module named general

If we continue maintaining bayesflare we should probably switch to using lightkurve for loading and storing Kepler light curves (it also works with TESS light curves too).

When calculating the Lomb-Scargle periodogram there appears to be a fast method available in astroML/gatspy (see also here). It may be worth implementing this as the current periodogram calculation does take a fair bit of time.

In the models branch there have been enhancements so that there is now a generic Model class. The individual signal models (e.g. the Flare model) now inherit the characteristics of the Model class.

In this update the methods within the classes that define the model function now only take in dictionaries of parameter values, with the dictionaries having to contain parameters required for the given model.

A new model class for a Step function is also included.

This has been tested and the plot_lightcurve script works correctly, but other tests should be performed (i.e. do the analysis scripts, and threshold/efficiency scripts still work). It should then be merged with the master branch and a new release be created.

The Lightcurve class, which is used to handle observational data within the package is currently suitable only for working with Kepler data.

An improved approach should include means of working with data from solar observatories such as GOES and RHESSI, with the option to work with other data sources, including neutron observations.

I propose to rewrite the Lightcurve class, while preserving backwards-compatibility, and utilising pydata/pandas, further, I propose incorporating `transientlunatic/lightcurve' as a Git submodule to do this, to allow the code to be reused outside the BayesFlare project.

An important consideration will be maintaining backwards-compatibility, and to do this a number of methods may need to be added purely for compatibility.

I would love to see this tested using the analytic light curve flare model we developed last year!