Post-processing: normalisation of b-tag weights
We need to adapt the postprocessing script (merge_parquet.py
) to do the normalisation of b-tagging weights.
As discussed with Spandan, the normalisation should be done on an inclusive sample of a given process, per category used in an analysis, but without taking b-tagging cuts into account.
This is because the bTagShapeSF should not change overall yields or differential distributions, except the b-tagging ones.
For this, the processors running b-tag SF should save the nominal b-tagging SF as well, in addition to the event weight which is the product of all partial weights.
Tagging @mikumar for this.