Hi- I’m fitting a histogram with empty bins using the “LL” log-likelihood option, e.g.
TF1 *f1 = new TF1("f1","[0]+[1]*x");
hist->Fit(f1,"LL");
(I’m using “LL” because I have non-integer bin contents.) It’s my understanding that, for likelihood fits, Minuit includes empty bins in the fit. However, if after fitting I do
f1->GetNumberFitPoints()
it returns the number of non-empty bins rather than the total number of bins in the range. Can I be sure that Minuit takes into account the empty bins?
The empty bins are always ignored unless they have a non-null error.
Note that this has nothing to do with Minuit, but with TH1::Fit
that selects the information to be used by the minimisation function
called by minuit.
[quote=“brun”]The empty bins are always ignored unless they have a non-null error.
Note that this has nothing to do with Minuit, but with TH1::Fit
that selects the information to be used by the minimisation function
called by minuit.
[/quote]
I thought the main reason to use likelihood is in the case of low statistics, including potentially empty bins. I had found this ancient thread on roottalk that discussed how the error isn’t the right thing to check for in non-error-based fitting routines:
As indicated in my posting, empty bins are always ignored unless
the bin error is non-null.
When specifying the option “w”, all errors get an equal weight.
So, if you specify the option “LW” all bins will be taken into account,
even if in this case the bin error is not used in the log-likelihood fit.