> Dear Root Developers,
>
> I would like to make a 3D fit (e.g. plane) to a set of 3D
> points (x,y,z) with their errors. Both the points and the errors
> come from another calculation and I want to set directly the
> points with their errors, like in TGraphErrors, but then in 3D.
>
> So, something like TGraph2DErrors class could do the job.
> Would it be useful to add this to ROOT?
> I was surprised you didn't put it in yet (at least I didn't find it).
>
> Currently I am using TH2F, but each bin has one entry (with
> weight=z), and the assumed error on this in a Fit is
> certainly wrong (1.0 I guess), so the fit is wrong.
>
> So I am thinking of making a new class like
> class TH2FErrors : public TH2F {
> void Fill(Axis_t x, Axis_t y, Stat_t w, Stat_t e)
> }
>
> void TH2FErrors::Fill(Axis_t x, Axis_t y, Stat_t w, Stat_t e)
> {
> Int_t binx, biny, bin;
> TH2F::Fill(x,y,w);// ensure bookkeeping
>
> binx = fXaxis.FindBin(x);
> biny = fYaxis.FindBin(y);
> bin = biny*(fXaxis.GetNbins()+2) + binx;
> // overwrite bin contents + bin error
> SetBinContent(bin,w);
> SetBinError(bin,e);
> }
>
> But I'm not sure this is a good idea, and it might even be
> dangerous. Maybe you can help me here.
>
> Thanks!
>
> Martin.
Martin,
Your solution above cannot work.
If you have equidistant points, you can use the already existing class
TH3F.
This creates a 3-D histogram that you can fill with
TH1::Fill(x,y,z,w)
then use TH1::Fit to fit your 3-d distribution.
Use the example 13 in the tutorials "Fitting with a user-defined
function"
as a model.
In case you have non-equidistant points, look at the batch example
in $ROOTSYS/test/minexam. This illustrates how to make a fit
on n dimensions with non-equidistant points. This examples shows also
how to define your own fitting model.
Rene Brun