Hi Rene,
RB>I agree with your theory. In the practical case of TProfile
RB>with low statistics in one bin and very small weights, do you have a
RB>better algorithm to propose? I mean an algorithm really working,
RB>and for example, not subject to rounding problems?
RB>Let me know. I will be happy to include it.
I thought a little bit about this problem and arrived at the following
conclusion / suggestion:
Is it possible to use a user-supplied error calculation function? In addition
of course to the other options (error-of-the-mean and spread option).
It is always the user's own responsibility to calculate their errors
correctly, as well as judging the outcome of a fit they make. What should be
the job of ROOT (IMHO), is to provide the tools to fit data points; even if
the fit is bad and the chi^2 is large, this is telling something.
A user cannot expect ROOT to take care for any ill-posed fit, it is necessary
to give a bad result in that case. There is more to fitting distributions than
just to run a program: when fits do not work well, the users may be advised to
think about the model they are fitting and not to complain to ROOT developers
about their (physics or) statistics problem.
just another $0.02 from me -- I am curious for other opinions.
Best Regards,
Martin
This archive was generated by hypermail 2b29 : Sat Jan 04 2003 - 23:51:00 MET