Erroneous sorting in DiagnosticToolbox methods

[adam-a-a]

Sorting does not work as expected for methods such as display_constraints_with_extreme_jacobians and display_variables with_extreme_jacobians.

Here is a snapshot of output from display_constraints_with_extreme_jacobians:

[lbianchi-lbl]

@adam-a-a would you be able to provide an example (code snippet) to reproduce this behavior?

[adam-a-a]

@lbianchi-lbl After attempting to solve a model, m, which has extreme jacobian elements, and assuming you created the following instance of the DiagnosticToolbox ( dt = DiagnosticToolbox(m)), you could then check the following:

# This would report numerical issues and hint at checking extreme jacobian elements
dt.report_numerical_issues()
# Then check the extreme jac elements:
dt.display_constraints_with_extreme_jacobians()

[andrewlee94]

@adam-a-a What we need is an example of a model that exhibits the behaviour. Right now we have no idea why the ordering is not working nor any way to test that we have fixed it.

[dallan-keylogic]

I have ascertained that the Diagnostics Toolbox is currently sorting variables and constraints as intended: from most extreme to least extreme, i.e., in order of descending abs(log(norm)). We can discuss about what would be a more intuitive sorting order, like whether we should have separate lists of rows with large and small norms, but it's not a bug.

I would like to hear what @adam-a-a and @andrewlee94 think, if they have time.

[andrewlee94]

@adam-a-a @dallan-keylogic I think the first question needs to be "what a user would expect to see", with the caveat that many users do not really know how to interpret these results. From the numerical standpoint, i.e. from most extreme to least extreme, i.e., in order of descending abs(log(norm)) is probably the most useful, but it does result in some unintuitive behaviours like the example above (which on review is actually correct).