I am using some Roofit to fit my data.
By default I was using an unbinned fit and everything seems ok except that a single fit takes about 5/6 minutes
to finish (and this for toys studies is not optimal for me…).
I tried with a binned fit. given a dataset
I created a
and fit with
I have 8 free parameters and the statistic is pretty large (45 bins. the less populated bin has ~2000 entries)
and so I expected to not see large difference between binned and unbinned fit result.
But I observe that the two fits converge on two different minima: the binned fit gets the “worst” minimum
( much larger chi2/ndf). If I set the initial parameters of the binned fit identical to the parameters obtained from unbinned fit it still moves into the other minimum. This is a bit puzzling.
is there something wrong in what I am doing?