Vertex-level descriptive indices appear to be broken

[sgoodreau]

To see the issue:

1. Load faux.mesa.high (or seemingly any dataset).
2. Go to Network Descriptves > More > Vertex-level descriptive indices

The calculations remain blank, no matter which vertex is chosen.

This is true in both the online version and the downloaded package version.
Graph-level descriptive indivces seem to work just fine.

statnetWeb_0.5.8

[sgoodreau]

Update: the values do end up appearing after all. But for me it took something close to 10 minutes before they did so, even on a small network like faux.mesa.high. So it would seem to be something about the underlying algorithms changing rather than a simple bug.

[mbojan]

@sgoodreau , I noticed that if you load a small network first, such as flobusiness and then faux.mesa.high things get calculated more quickly. After that, if you keep on switching datasets things get calculated almost instantly.

It seems to me that the app does some things upon startup or upon first data load that take extra time - of all things, it pre-loads all the default datasets, loads packages and so on. Once these are done, it works relatively smoothly. I'm not a Shiny expert though and these apps are a nightmare to debug...

[martinamorris]

Here's what I'm seeing at the simplest reprex level:

run_sw(),
load faux.mesa,
Network Descriptives / More

There are 4 options here. 2 of them seem to be working -- mixingmatrix and graph level indices -- but the other 2 don't -- the null hypothesis tests (the "Run" button doesn't respond) and the vertex level indices are empty.

If I go back and activate any of the other tabs from the Network Descriptives -- Attributes, degree, geodesic -- then come back, everything works on the 2 problem options.

@mbojan does that provide a clue?

[mbojan]

I see all the descriptives, but, similarly to #95 (comment) there is a delay, which for some reason can be speed-up by choosing faux.mesa, then say flobusiness, and going back to faux.mesa. I see all, including vertex-level measures, and can do the null-model tests.