RooDataSet: 2D weights?

[siragoni]

Dear All,
could you please help me out a bit with Roofit?
My problem is that I have a dataset formed of two variables, cos(theta) and phi.
I have to attach weights to this dataset to eventually perform a 2D fit.
If the event has the same weight for both observables, I can do:

        RooRealVar t("t", "Cos(theta) ", -0.6, 0.6) ;
        RooRealVar p("p", "phi angle ",   0, 2*3.14) ;


        RooRealVar* weight = new RooRealVar("weight", "weight", 0.0, 1000. );
        RooArgSet*  ArgSet = new RooArgSet("args");
        ArgSet->add(t);
        ArgSet->add(p);
        ArgSet->add(*weight);
        RooDataSet* dataAnglOld = new RooDataSet("dataAnglOld", "dataAnglOld", *ArgSet);

        ...

      t.setVal(ct);
      p.setVal(phiF);
      weight->setVal(w);
      // dataAngl->add(*ArgSet, w, 0);
      dataAnglOld->add(*ArgSet);

And when the dataset is complete, I create the new one like this:

    RooDataSet* dataAngl = new RooDataSet("dataAngl", "dataAngl", 
    dataAnglOld,*dataAnglOld->get(),0,weight->GetName());
    dataAngl->get(0)->Print("V");
    cout << dataAngl->weight() << endl;
    dataAngl->get(1)->Print("V");
    cout << dataAngl->weight() << endl;
    dataAngl->get(2)->Print("V");
    cout << dataAngl->weight() << endl;

With the last few couts I could tell if weights had been set properly, and they were.
However, it is not clear to me how to give a 2D weight there…

Would you be able to advise me?

Just a note, I would need to do it this way to do a 2D fit.

Thanks in advance,
Simone

[moneta]

Hi,

I am sorry I don’t understand the 2D weights. A weight is a scalar associated with each single event and not to the variables. The event then can have multi-dimensional observable variables, like cos(theta) and phi in your case

Lorenzo

[siragoni]

Buongiorno,
Thanks for the reply.
I understand what you mean.

The point here is that in this case, cos(theta) and phi have two different weights in a single event. This is typically the case when the weight is due to the acceptance, and it is not possible to compute the acceptance as a function of both observables, but only one at a time.

In fact, the example code I put is for this 2D “acceptance”.

Please, let me know if I have been able to clarify it.
Otherwise I can go more into detail.

Thanks again for the reply.
Simone

[moneta]

Hi,

Still I am not sure I have understood the problem well. If you compute one at a time, you neglect correlation between variables and it is then like to build a 1D model where you have a dataset on cos(theta) and one on phi and not a 2D data set.

Lorenzo

[siragoni]

Many thanks,
yes, that is exactly the case.
Unfortunately, it seems like it is not possible to describe the weights in a 2D fashion (thus taking care of the correlations from the get and go), and one would have to adopt two 1D descriptions i.e. one for cos(theta) and another for phi.
I seem to now understand that such an approach is not accepted by Roofit…
Thanks for your replies,
Simone

[moneta]

Hi,

To be clear, this is not a RooFit limitation, but a limitation of your analysis method that you cannot get the acceptance as function of the two variables, but only its projections.
The correct procedure is to have a weight for each (x,y) point of your data and this can be modeled in RooFit

Lorenzo