Open ntadej opened 4 years ago
I believe this is a very important ingredient needed to allow people to fully move to pyhf
. As @lukasheinrich mentions in #731, it's not so easy to calculate the post-fit uncertainty band and I've seen calculations starting to break especially with high parameter correlations. I think it would be useful to not only get the uncertainty of the total model prediction, but also per sample (e.g. needed for yield tables).
Since GitHub does not seem to link discussions back to issues: #1414 is related to this request.
The usual HEP workflow is to create a workspace with several channels and several systematics. In the final paper we need to show the post-fit distributions with appropriate errors.
The request is to have a simple interface to get the channel distribution with errors, ideally just two arrays (one with yields and one with errors), supporting:
Plotting the actual distributions is then certainly out of scope of pyhf.