Dear expert,
We are contacting you because we are attempting to regress a given function using TMVA package and we obtain a wiered result.
We are currently trying to regress the mass of the following system: lepton, jet, jet and neutrino (ljjv).
The input variables we are considering are:
- ljj (pt,eta,energy,mass,pz)
- Met/jj.Pt()
The target function is the following distribution:
As you can see, the distribution consists of two parts: one part peaks at 130GeV, the other part peaks at 160 GeV.
Out of our expectation, the regression gives us a distribution with single peak at 150:
We've checked the following two cases:
- target to distribution which peaks at 130
- target to distribution which peaks at 160
In both cases, the regression peaks at correct place:
In the previous plots, we’ve been using BDTG method.
We suspected that the possible issue could be related to the binning of target variables and we change the nCut parameters from 20 to 4 and 100 respectively, we got similar results in both cases.
Also, we tried the other methods (KNN LD FDA_GA MLP) but we got the similar results again.
Did we miss anything?
We’d be very appreciated if you could give us some feedbacks or suggestions to help us improving the regression.
Sincerely thanks,
Binghuan