scikit-hep / pyhf

pure-Python HistFactory implementation with tensors and autodiff

https://pyhf.readthedocs.io/

Apache License 2.0

274 stars 81 forks source link

visualization of workspace as a calendar heat map #1097

Open kratsg opened 3 years ago

kratsg commented 3 years ago

Description

As title says. Something like https://github.com/rougier/scientific-visualization-book would be great to show to visualize the workspace in some way.

matthewfeickert commented 3 years ago

:fire:

matthewfeickert commented 3 years ago

For reference, the calendar heatmap code has its own repo: https://github.com/rougier/calendar-heatmap

alexander-held commented 3 years ago

One possible visualization, using the example in the repository linked above:

year -> channel
month -> sample
day of the week -> modifier
color -> normalization effect of modifier compared to nominal

Other permutations of channel/sample/modifier work too. This is a useful plot to quickly spot possible implementation issues with modifiers. With some scalar quantity defined to measure shape (not as straightforward...), could do the same for shape as well.

lukasheinrich commented 3 years ago

yeah I've tried doing some in my thesis

cranmer commented 3 years ago

I found this image from 2012 that I made to graphically summarize a model:

Rows are observables or parameters.
Columns are PDF components (eg a constraint term or the top level PDF for a particular category / channel (so bins of histogram are combined into one column)).
red dot indicate that the PDF depends on the parameter or observable (it's just a boolean)

This kind of visualization is useful for understanding the correlations in the model, which NPs are shared etc.