Rating changes in pyjibe after running all curves and then clicking individually on each curve

[PinkShnack]

Mentioned in Pyjibe here (AFM-analysis/PyJibe#43 (comment)), the rating display in pyjibe changes after clicking in this scenario:

• I click "Apply Model and Fit All". It works as expected.
• I click on individual curves (files) and the displayed rating value changes.

This seems like a bug. It only happens (afaik) with the KVM extension model.

Todo:

• make main.UiForceDistance.on_fit_all and main.UiForceDistance.on_curve_list do the same thing. At the moment, on_fit_all does not run self.tab_fit.fit_update_parameters(fdist), which I guess it should.
• find out why ratings are different -> likely due to initial params being set from previous fit, which makes fit (possibly) different each time.

[PinkShnack]

After some investigation I found there is a difference in these values (between clicking on the curve vs. clicking "Fit All"). Biggest rating change I found was changing between 4.5 and 5.6.

This happens when I have more than one curve loaded and is present regardless of curve order.
I cannot reproduce it when just one curve is loaded.

• ancillary max_indent. Differences quite small e.g. 2.662e-6 vs. 2.636e-6
• fit_parameters.params_initial:
  • time_ind: e.g. 0.11865 vs. 0.61742
  • eta: e.g. 4.49e-13 vs. 2.7827 (Large difference)
  • lmd: e.g. 0.059 vs. 0.308
• `fit_parameters.params_fitted (just some examples):
  • Eu: 1719 vs. 594
  • Eapp: 806 vs. 594
  • lmd: 0.0237 vs. 6.163
  • time_ind, eta same as above
  • contact_point: -7.31e-7 vs -7.056e-7
  • baseline: 3.08 vs 3.68

I would be worried especially about the variation in calculated Eu and Eapp.

[PinkShnack]

The initial fitting parameters are set from the default initial parameters defined by the model in get_parameter_defaults. To me this is logical and it is reproducible.

However, if the parameters are available in the GUI, then the GUI value is taken as the initial parameter. (as done in TabFit.fit_parameters in tab_fit.py line 308). This seems to be the source of the discrepancy between the above values. Sometimes the default value is used, sometimes the GUI value from the previous fitted curve. The line specifically that does this is in TabFit.fit_parameters in tab_fit.py line 333 (p.set).

QUESTION:

I would assume this is a bug, but I can also see why this might be a feature? As in, we could use initial params from the previous fit. I guess I need someone more familiar with AFM data analysis to say if it's a bug.

And it only happens for the KVM sometimes, and I guess only because ancillary features are sometimes taken from the GUI.

[paulmueller]

This is definitely a bug. All curves should fit with the same initial parameters and not with the parameters from the previous curve, which would introduce a bias. The fit parameters should be taken from the values put in by the user.

[PinkShnack]

This is definitely a bug. All curves should fit with the same initial parameters and not with the parameters from the previous curve, which would introduce a bias. The fit parameters should be taken from the values put in by the user.

The user being in this case Shada and her model's initial params?

EDIT: I get you, you mean if I type in a new viscosity value in the GUI, that should be taken for all fits, and not something else. I agree. However in my pyjibe it seems trying to set params manually just doesn't work when using the KVM model. I can change the Tip Radius, click Fit all, and the tip radius is back to a different value.

So it seems this is just a small part of a bigger issue with the extension model and how the GUI works with it.