Visualize the modifier structure of a model

[alexander-held]

When working with a complex model, it may be useful to be able to visualize its structure. That could mean many things, but concretely it can be the modifiers used per channel, sample, and parameter. That can show where and how parameters act. There are three dimension to this (channel, sample, parameter), so this could be shown as multiple 2d tables.

Slightly related: instead of showing modifier types, it could also be interesting to show the effects of templates to identify which parameters have the biggest effects. Associated pyhf issue: scikit-hep/pyhf#1652.

[alexander-held]

As suggested by @kratsg, the opacity of each field could scale with the size of the impact of each modifier. That allows to still distinguish contributions with large impacts by color, while those with small impact blend more into the background.

[alexander-held]

#333 provides an example script, and should be refactored into the existing API. Using the 10.17182/hepdata.89408.v3 workspace as an example, the visualization currently looks like this:

[alexander-held]

#333 is going to support two different ways of structuring this visualization, with grids split by channel (default) or by sample (as in the visualization in the previous comment). With an update to a more colorblind-friendly colormap (and inclusion of the missing modifier types), the figures now look like this:

split by channel:

split by sample:

This visualization is focused on workspace structure, so it will not include an attempt at visualizing differences in impact across modifiers.