As promised, here is an update with a weighted partonic bin of 15-20 GeV/c & this is still a "proof-of-principle" sort of deal. Below is the process for this analysis:

code can be found in /star/u/rsalinas

• ./embedding.1.sh 
  • This produces the TTree required for the embedding and contains pJetPtmatched, dJetPt, pJetWeight,...
• Grabbed the output file to my local machine (NNPembedding_tree.pt15_20.root)
• Passed NNPembeddingt_tree.pt15_20.root to TMVARegpract.C to do regression 
• Saved output file 
• Used self-made script to plot these outputs (will be uploaded to rsalinas on RCF) 

Plots 
Unweighted 

• This is the regression w/o passing the regression tree containing the pJetWeight. As seen, it is not expected that the behavior of the partonic bin doesn't fall smoothly.
• The pad on the left is the regression for various different methods w/ an unexpected cut in the regression at 5 GeV/c. We speculate that this is due to the detector jets themselves having a cut at that threshold. 
• The right pad is a comparison of the unweighted particle matched Jet pt and the weighted particle matched Jet pt. At this current state, I'm ambivalent towards shape of the weighted particle jet Pt spectrum due to not having much experience in how it "should appear". I would refrain from making any contention that I am unsure of (for now).

Weighted

• This regression concerns itself with the weighted particle Jet pt via passing the TTree holding the weights for the events to the TMVA regression. I expect the pad on the left to have the regression not match the pJetMtx due to pJetMtx not being weighted when plotted. 
• Alas, this brings me back to the title of this blog entry; the long and winding regression. There are several things here I am not sure are being done correctly:
  • Why is the regression (on the right pad) several orders of magnitude less than the weighted distribution?
  • Is the weighted distribution being done correctly? This was done by saving the partonic weight for the matched particle jet, saving it to a branch (along with all the other variables) and calling tree->Draw("pJetPtMtx",pJetWeight") which is analogous to filling a TH1 with a weight (th1->Fill(pJetPtMtx,pJetWeight).
  • How is the weight tree being passed to the TMVA being handled?
• An encouraging thought may aid me in the ailing regression road; the regression output MLP appears to have the same shape as the weighted pJetMtx (right pad) w/ just an issue of several orders of magnitude. This may be an easy fix (hopefully) 
•