TnPcal

This project is meant to be a versatile, flexible and user friendly way to measure efficiencies using the Tag and Probe method.

How does it work?

The logic of the method is that one would apply a requirement to a dataset and out of it, a certain number of samples would pass it. The fraction of successes is called, the efficiency. E.g. one takes the population of a country and requires it to be male, 50% of the population passes, the efficiency is 0.5.

However in HEP one needs to deal with datasets that are a mixture of different categories, for simplicity, we will call them signal and background. We are only interested on the signal; thus, we need a way to remove the background when calculating the signal efficiency. There are two ways to do this:

• sWeigths: Weights are calculated through the sPlot technique or other approach and the weighted dataset is assumed to be pure signal.o

• Fits: The distribution of interest (usually the mass of a particle) is fitted. Thus we can separate the signal yield and study it.

Fitting approach

In order to extract a good (low uncertainty) signal yield from the fit, such that we can know the efficiency of a cut on it, we have to make sure that the sample we start with is clean enough and unbiased. By this we mean:

• clean enough: We can see a nice, large peak with little background.

• unbiased: The properties of this clean sample are the same as the ones for the original one. I.e. the cleaning process does not change the efficiencies.

Thus a tagging requirement is needed. Once the tagging is done, we fit and extract the efficiencies.

Usage

Fitting approach

from tnpcal.fitter import Fitter as TnPFitter

obj = TnPFitter(data=df, model=mod, cfg=cfg)
obj.run()

The code will take:

• A pandas dataframe with the data
• The fitting model, implemented with zfit
• The configuration through the cfg dictionary

and run all the fits as well as calculate all the efficiencies. The configuration is as shown below:

observable:
    name    : mass
    min_val : 5000
    max_val : 6000
selection:
    tag   : 'purity > 0.8' #Quantity that drives signal purity, e.g 1 means only signal 0 means full mix
    probe : 'quantity'     #Quantity, whose efficiency needs to be measured