Toy studies and inference

While a single fit is useful, it does not say a lot about the uncertainty of the result and whether the fit is biased in any way or not.

Many statistical methods, such as obtaining sWeights, (Feldman and Cousins) Confidence Interval, setting limits and more are all covered it the hepstats library, which work directly with zfit parts.

For other toy studies, models offer a sampler function that can be used for repeated sampling.

Playing with toys: Multiple samplings

The method create_sampler() returns a sampler that can be used like a Data object (e.g. for building a ZfitLoss). The sample generated depends on the original pdf at this point, e.g. parameters have the value they have when create_sampler() is called, respectively the values that are given explicitly as params to it.

Reusing the model, obs and parameters from Building a model, this is typically how toys look like:

import os
os.environ["ZFIT_DISABLE_TF_WARNINGS"] = "1"

import zfit
from zfit import z
import numpy as np

obs = zfit.Space('x', -10, 10)

mu1 = zfit.Parameter("mu1", 1)
sigma1 = zfit.Parameter("sigma1", 1.3)
gauss1 = zfit.pdf.Gauss(obs=obs, mu=mu1, sigma=sigma1)

mu2 = zfit.Parameter("mu2", 3.)
sigma2 = zfit.Parameter("sigma2", 2.)

# using the previous gaussians and obs to create a model

gauss3 = zfit.pdf.Gauss(obs=obs, mu=mu2, sigma=sigma2)
model = zfit.pdf.SumPDF([gauss1, gauss3], fracs=0.4)

sampler = model.create_sampler(n=1000)
nll = zfit.loss.UnbinnedNLL(model=model, data=sampler)

minimizer = zfit.minimize.Minuit()

results = []
nruns = 5
for run_number in range(nruns):
    sampler.resample()  # now the resampling gets executed


    # initialize the parameters randomly
    mu1.set_value(np.random.normal())
    sigma1.set_value(abs(np.random.normal()) + 0.5)

    result = minimizer.minimize(nll)
    results.append(result)

    # safe the result, collect the values, calculate errors...

If some parameters are constrained to values observed from external measurements, usually Gaussian constraints, then sampling of the observed values might be needed to obtain an unbiased sample from the model. Example:

TODO: the sample below is not correct and needs updating…

Make live