Include fewer variables in div plot grid in PyStanSamplingProcessor?

[taliaweiss]

I like that the PyStanSamplingProcessor creates divergence plots by default. However, if there are more than a few interestParams (for future saving/study), the plots and labels in the figure become very small.

Since this figure is primarily intended to alert the user of whether there are divergences (and of the fact that they can track where in parameter space divs occur), perhaps PyStanSamplingProcessor could make one divergence plot with only two variables – by default, the first two in interestParams. If the plot shows that there are divergences but doesn't show any localization of the divs for those two parameters, the user can then make additional divergence plots after-the-fact.

What do you think of this proposed change?

[guiguem]

This is an interesting proposal!

• When you say too small, you mean not readable even when zooming in?
  If so we can always increase the window size proportionally to the number of parameters
• I like this idea of defining the number of the first interest parameters to implement.
  We could say 5-6 but give the option to change this number by the user in the configuration file too.
  What do you think?
• Having this as a standalone processor (using common functions with the PyStan processor) makes a lot of sense to me; definitely a processor to have for diagnostics.

[taliaweiss]

Good question – it is readable when zooming in. The issue is more that so much zooming is required that it becomes not-so-user-friendly, in my opinion. Also, many of the tick labels get covered by the subplots, which does not occur when ~3 parameters are included.

Plotting the first 5 interest parameters by default but allowing this to be changed by the user in the configuration dict/file sounds very good to me.

Regarding having a standalone processor – what are you envisioning this processor would do that is different from the Histo2dDivergence processor?

divergence_2d_histo_Q_mass_survival_prob_Bfield_sigma_S_B_1kev_KEmin_KE_sample_Nfit_signal_Nfit_bkgd_lp_prob.pdf