SPE fit algorithm: account for fit convergence to set valid status

[tibaldo]

While looking at he new implementation of the gain calibration via SPE fitting using run 3936, I found that for two pixels I got poor results with the fitted function not matching the data


Yet, the gain value is not flagged in the results file, that is, it has 'is_valid = True'.
By looking at the code it seems that presently the only condition to set 'is_valid = True' is that the fit algorithm returns some values. I think the fit convergence should be also taken into account to avoid this kind of problem.
I explored this is my branch tibaldo:spefit_convergence_issues and it seems rather straighforward. In this exemple of a possible implementation I checked three outputs from Minuit: fit convergence status, validity of parameters and that the parameters have not reached their limits.
What do you think @guillaumegrolleron @jlenain? If that seems like a good addition I can make any modifications needed and then make a pull request from my branch.
PS: I did not dig into the origin of the problem for the two pixels. For pixel 658 there is a failure in the preliminary Gaussian fits to set the starting value for the parameters, for pixel 660 the charge distribution looks unusual, but I'm unsure if it is just a plot range feature.

[jlenain]

Hi @tibaldo , sorry for the delay and thanks for this!
Indeed, it would be very useful to add some criteria to flag fit results as valid. The ones you implemented in https://github.com/tibaldo/nectarchain/tree/spefit_convergence_issues seem fully valid, so yes, please go ahead with a pull request, that would be much appreciated!

[tibaldo]

Excellent, thank you. I'll make sure this is covered in the unit tests and make apull request probably tomorrow.

[jlenain]

So, discussing with @guillaumegrolleron , it seems he is already implementing these changes while working on #89 - where he is working on forcing the spawn implementation on Linux as well (which would then also be future-proof). So maybe better to coordinate with @guillaumegrolleron for such a pull request. Many thanks!

[tibaldo]

@guillaumegrolleron let me know how you want to proceeds, anything works for me