On Wed, 26 Feb 2003, Rene Brun wrote: > Hi Gong, > > It is difficult to come with a diagnostic without having your histogram. > Could you send me a small file with your histogram + a script > with your fit funnction? > > Rene Brun > > Datao Gong wrote: > > > > Hi,Rooters, > > I fit my ntuple with likelihood method. I found if i tightly cut, > > the fit result is much different and fit quality worse (ch2/ndf far > > away from 1). Compare to min_fit, i found it is from the empty bins > > in histogram. Likelihood method seems do not take into account > > empty bin default. (I read Burn's email in roottalk02). In my case, > > I fit my histogram with option "EMVL0", the ch2/ndf is 237/189, meanwhile, > > mn_fit give me 190/191. I know there are two empty bins in my histogram. > > If i set "W" option, the chi2/ndf is up to 3067/191. Is there something > > wrong? If you have an empty bin its Error is undefined (the error is usually sqrt(n)) or zero so that chi^2 would divide by zero, so these are left out. If you explicitly set error to, say 1 or so, then the fit behaves differently. Dipl.-Phys. Henning Kiel Lehrstuhl fuer Experimentelle Physik IV Universitaet Dortmund Otto-Hahn-Str.4 44227 Dortmund Germany --------------------------------------- Phone: +49 231 755 3542 Fax : +49 231 755 3688 --------------------------------------- God is real, unless declared integer. "She sells sea shells by the sea shore. The shells she sells are sea shells I'm sure."
This archive was generated by hypermail 2b29 : Thu Jan 01 2004 - 17:50:09 MET