NM solver: 0 results found for twig

[kathmcp]

Hi, since I have updated to 2.3.53, when I run the Nelder-Mead solver, it keeps coming up with the error:

Sun, 05 Sep 2021 05:55 SOLVER WARNING received error while setting values: 0 results found for twig: 'None', {'context': 'constraint', 'uniqueid': 'lbvYkJlglLqDoUHhjpVdApMtjumdVB', 'check_visible': False, 'check_default': True, 'check_advanced': False, 'check_single': False}. lnprobability=-inf

It does complete the solver, but looking at the results, the values for requiv and incl are the same as the initial inputs of 1.05 and 90, so it doesn't appear to have done anything to them. I was wondering if this is related to the above error?

I don't know if it is related to the new release, but I didn't get this error before. However, I have now added another passband (so now have B, V and R) and a mesh dataset as well, so am not sure if this could be the problem.

I also get this same error when I try to plot the residuals

As the Emcee solver takes days to run, it would be good to know if there is something I should change before setting it off again.
Thanks!

[kecnry]

That warning is definitely concerning. Can you show me the output of print(b.get_value('fit_parameters', expand=True))?

Was this a bundle that you saved in an old version and then imported after updating phoebe? If so, there is a chance some constraint isn't migrating correctly for contact binaries - please send me the bundle file (either here or by email) and any information about how you created it (which version(s) of phoebe, etc) and I'll try to track down what went wrong and get you back up and running.

As for the final error during plotting, you adopted the solution which changes the face values of the parameters, but you still need to run a forward-model with run_compute before you'll have a synthetic model available for plotting.

[kathmcp]

I mean the warnings in Nelder-Mead when it said 0 results found for twig. This is not appearing anymore since I removed the mesh dataset.

Yes, I can continue on now, hopefully with my final run! I will add in a plot from the mesh dataset from a previous file, as it will look good in my report :)

[kecnry]

I've tried to recreate the bug you were seeing from scratch, but something is clearly different in your system that is necessary to reproduce the bug. And it seems the notebooks you sent me don't raise the error as they are now.

So... when you get a chance (no rush - it can be after you finish your thesis), can you please save the bundle object (using b_cb.save('export.bundle') instead of the whole notebook), then try to create a separate short notebook/script that loads that bundle (b_cb = phoebe.load('export.bundle')) and reproduces the warnings/errors for Nelder-Mead? It would be great if that short notebook could show it working without the mesh, then add the mesh and show that it starts failing. If you can send me those two files (the bundle object and the short notebook showing just the bug itself in a few lines), then it'll be much easier to reproduce and track down the underlying issue.

Thanks and good luck wrapping up your thesis!

[kathmcp]

Hi Kyle, sorry for the delay in replying. I managed to hand my thesis in with half a day to spare :) I have taken a couple of weeks to recover and organise things, but I will start looking into this now. By the way, I gave you a special acknowledgement in my thesis, I literally couldn't have completed it without your help!

[kathmcp]

Hi again, I think this is what you need, but please let me know if there's anything else I can do.
NMIssueMeshNoMesh.zip

[kecnry]

Thanks, I'll take a look and try to track that down when I get a chance. Congratulations on completing your thesis!