search

madminer-tool / madminer

Machine learning–based inference toolkit for particle physics
https://arxiv.org/abs/1907.10621
MIT License
86 stars 45 forks source link

Phase-space-dependendant nuisance parameters #64

Open johannbrehmer opened 6 years ago

johannbrehmer commented 6 years ago

Let the user supply a function (or a string that can be saved in the HDF5 and parsed efficiently) that depends on x and the value of a nuisance parameter nu and returns a event-wise rescaling.

johannbrehmer commented 6 years ago

The simplest thing that we could implement immediately is a Taylor expansion:

dσ(x|θ,ν) = (1 + aᵢ(x) ν + bᵢ(x) ν² + cᵢ(x) ν³ + ...) dσ(x|θ),

with more brackets for more than one nuisance parameter. The functions aᵢ(x) etc. could for now be explicitly specified by the user, with the option to add a morphing setup later. The subscript i refers to the process generating an event (e.g. differentiating signal and different backgrounds). This is of course much less general than having the full (x,z) dependence, but also much less painful to set up.