Closed rluquer closed 5 months ago
Thanks for bringing this to our attention! Just to confirm, as far as you can tell from what outputs and plots were made, the fit looks like it converged well to a reasonable set of values and it is only this plot that is causing an issue? I think I've likely encountered this before as well - I'll see if I've already patched this in one of my in prep. pull requests.
I'd say so, but it is hard to judge from the plots because the transit model + systematics or the data - systematics do not get drawn in the figures produced when isplots_S5=1
Any news?
Sorry, I had a paper that needed to be resubmitted rapidly that is now taken care of. I'll try my best to resolve this before the end of the week.
I'd say so, but it is hard to judge from the plots because the transit model + systematics or the data - systematics do not get drawn in the figures produced when isplots_S5=1
Wait, do you mean that there are no model lines drawn on these figures? That's a different issue that what I thought was going on, but I think I know how to take care of that issue too. Can you confirm what reduced chi-squared value and scatter_mult value you're getting (should be very near 1.0 and >1.0, respectively); I ask since sometimes issues happen when scatter_mult goes negative which can happen with broad Normal priors.
But other issues have sometimes arisen for me where masked values don't play nicely with the GP code for reasons I don't fully understand
In Figs510X:
scatter_mult is 1.4+-0.01, no negative values (I've already avoided that with U-priors) reduced chi-squared ---
Thanks, that helps me narrow down the issue a lot!!
So the fit apparently is done correctly, I've tried dynasty emcee and lsq. All break the code when isplots>=3, but when isplots=1 and the fit finishes, then I have those figures lacking the model and/or data
the models are certainly missing the GP term for what I can see in Fig 5104
FYI, I was able to reproduce this bug with one of my own datasets today which is helping me narrow down the issue. Still working on the solution though
Great news, thank you @taylorbell57. Keep me posted :)
Caleb Canas may have solved the problem. He says it works for him.
Oh, this is great! Is there any solution that I can test in the meantime?
@rluquer, I just came up with a quick patch that works for me - you could try checking out the pull request above
FAQ check
Instrument
Light curve fitting (Stages 4-6)
What happened?
When using GPs (celerite, Matern32 kernel) on a S5 fit, there is a problem with the plotting that impedes the function
s5_meta = eureka.S5_lightcurve_fitting.s5_fit.fitlc(eventlabel)
to finish, thus not enabling S6 to run properly. The actual fit works and finishes (I've tried lsq and dynesty, both works), the problem seems to be with plotting whenisplots_S5 >= 3
.Error traceback output
What operating system are you using?
MacOS Monterey (12.6.5)
What version of Python are you running?
Python 3.9.7
What Python packages do you have installed?
Code of Conduct