Examples with deprecated fits module of flavio

[bhishmapitamah]

Usage of deprecated module (as mentioned in documentation):

import flavio.statistics.fits

Note that the FastFit class of the flavio.statistics.fits module was deprecated in favour of flavio.statistics.likelihood.FastLikelihood as of flavio v1.6.

Recent installations will have following error:

ImportError                               Traceback (most recent call last)
 in ()
      3 import flavio
      4 import flavio.plots
----> 5 import flavio.statistics.fits
      6 import matplotlib.pyplot as plt
      7 from collections import OrderedDict

ImportError: No module named 'flavio.statistics.fits'

Following ipython examples use the fits module:

FastFit_WilsonCoefficients_C7p.ipynb
FastFit_WilsonCoefficients_CS_CSp.ipynb
FrequentistFit_VubVcb_Profiler.ipynb

They should be replaced with the relevant flavio.statistics.likelihood methods.

[DavidMStraub]

True, the examples are all outdated.

[bhishmapitamah]

Can you point out what changes should be make to get these examples working in flavio v2.0.0, as the signature has changed for both FastLikelihood constructor and method log_likelihood.

[bhishmapitamah]

In the example FastFit_WilsonCoefficients_C7p.ipynb i changed fastfit_obs to

def fastfit_obs(name, obslist):
    return flavio.statistics.likelihood.FastLikelihood(
                name = name,
                observables = obslist
            )

and changed the likelihood functions like global_fastfit.log_likelihood to

par = flavio.default_parameters
par_dict = par.get_central_all()

def ll(x):
    wc_NP.set_initial({ 'C7p_bs': x[0] + 1j * x[1]}, scale=4.8)
    return global_fastfit.log_likelihood(par_dict, wc_NP)

@DavidMStraub can you please tell if these changes are sufficient?

This might not be the best place to ask but I also wanted to know how to add multiple instances of an observable with same bins from different experiments for likelihood as I didn't see any examples with this case.

[DavidMStraub]

Hi,

yes, looks good. The instantiation for wc_NP is missing, but I guess you have it somewhere. Personally, I would nowadays directly use wilson:

from wilson import Wilson
...
def ll(x):
    wc_NP = Wilson({"C7p_bs": ...}, scale=4.8, eft="WET", basis="flavio")
    ...

[bhishmapitamah]

@DavidMStraub can you please tell how to handle the case of adding multiple instances for some observable with same bins. I see the case with multiple measurements for an observable in you code for paper bkstartmumu, but how to handle the case when even the bins are same for an observable.

[DavidMStraub]

For every bin, all existing measurements will be taken into account automatically. Thus one has to be careful not to double-count. You should also consider using smelli which was developed precisely with the goal to abstract away these subtleties.

[bhishmapitamah]

I was trying to understand what how flavio takes multiple measurements into account.

For the example given in notebook FastFit_WilsonCoefficients_C7p the number of observables given by global_fastfit.full_measurement_likelihood.get_number_observations() is 9. However global_fastfit.exp_covariance._central_cov[0].shape gives (8, 8) shape.

Does it combine the distributions while calculating χ² in FastLikelihood?

[DavidMStraub]

As per the doc string, the number of observations is "defined as individual measurements of observables". Apparently, here there is one of the 8 observables that is measured by two experiments.

[bhishmapitamah]

So it combines the the distributions where it finds more than one measurement for an observable?

[DavidMStraub]

Sure. But it's in the user's responsability to be careful that these measurements are actually independent. For instance, for BR(Bs->mumu), flavio contains a global average as well as the individual measurements, so there is double counting if one doesn't exclude one.

[sambit-1236]

In the example FastFit_WilsonCoefficients_C7p.ipynb i changed fastfit_obs to

def fastfit_obs(name, obslist):
    return flavio.statistics.likelihood.FastLikelihood(
                name = name,
                observables = obslist
            )

and changed the likelihood functions like global_fastfit.log_likelihood to

par = flavio.default_parameters
par_dict = par.get_central_all()

def ll(x):
    wc_NP.set_initial({ 'C7p_bs': x[0] + 1j * x[1]}, scale=4.8)
    return global_fastfit.log_likelihood(par_dict, wc_NP)

@DavidMStraub can you please tell if these changes are sufficient?

This might not be the best place to ask but I also wanted to know how to add multiple instances of an observable with same bins from different experiments for likelihood as I didn't see any examples with this case.

Can you please share the updated

Usage of deprecated module (as mentioned in documentation):

import flavio.statistics.fits

Note that the FastFit class of the flavio.statistics.fits module was deprecated in favour of flavio.statistics.likelihood.FastLikelihood as of flavio v1.6.

Recent installations will have following error:

ImportError                               Traceback (most recent call last)
 in ()
      3 import flavio
      4 import flavio.plots
----> 5 import flavio.statistics.fits
      6 import matplotlib.pyplot as plt
      7 from collections import OrderedDict

ImportError: No module named 'flavio.statistics.fits'

Following ipython examples use the fits module:

FastFit_WilsonCoefficients_C7p.ipynb FastFit_WilsonCoefficients_CS_CSp.ipynb FrequentistFit_VubVcb_Profiler.ipynb

They should be replaced with the relevant flavio.statistics.likelihood methods.

Can you please share the updated .ipython for these 3 examples? I am new to this package and facing lots of error while trying to modify the outdated examples?