Top mass in VHbb ntuples

[silviodonato]

@capalmer85 @scooperstein
I think it would be very useful to add the leptonic top mass and the hadronic top mass in VHbb ntuples.

[jpata]

do you mean the generated top? That's already there (GenTop_mass, GenTop_decayMode) https://github.com/vhbb/cmssw/blob/vhbbHeppy80X/VHbbAnalysis/Heppy/test/vhbb.py#L133

[silviodonato]

No, I don't. I was meaning the reconstructed top mass to reject TT in W(lv)H(bb) and Z(vv)H(bb). The leptonic top mass is defined including the lepton, MET, and the closest b-jet.

[jpata]

ok nevermind. I had added the GenTop_decayMode relatively recently, that's why I was wondering if you are looking for that.

[scooperstein]

Hi Silvio,

Sorry for the late reply. We calculate the leptonic top mass using code borrowed from the top group.

https://github.com/capalmer85/AnalysisTools/blob/master/plugins/VHbbAnalysis.cc#L1586

I would be happy to help add the calculation of this variable to the Heppy code, although I am not really familiar with adding significant additions like this to the Heppy code so I think it would be much more efficient if someone with more experience could help me add it. Would that be possible?

Best, Stephane

[arizzi]

anyone working on this?

[capalmer85]

Stephane asked for help in his last email. I don't think anyone offered him any. :(

I think Stephane basically needs to know where to put the code (we already have it written in our framework) and perhaps a pushing recipe if it is complicated.

Can you or someone else offer those two pointers?

On Tue, Jan 10, 2017 at 8:35 AM, arizzi notifications@github.com wrote:

anyone working on this?

— You are receiving this because you were mentioned. Reply to this email directly, view it on GitHub https://github.com/vhbb/cmssw/issues/483#issuecomment-271576685, or mute the thread https://github.com/notifications/unsubscribe-auth/AFKqhwgbP00X5u0QqvI4iFh3Ove2pxjoks5rQ4kNgaJpZM4JLKfw .

[arizzi]

the code I see is in c++, can you make it in python ? consider that all variables you use are already in form of p4 in heppy and they are named e.g. event.selectedLeptons or stuff like that...

the final location doesn;t matter we can cut and paste it anywhere, but you can implement a computeTopMass function in VHbbAnalyzer.py and then call it in the main function "process"

[scooperstein]

okay thanks for the info. I will start with implementing a computeTopMass(lep 4-vector, MET 4-vector, b-jet 4-vector) function in python and I'll put it somewhere in VHbbAnalyzer.py. If it is not clear to me where to go from there I may consult someone more experienced with Heppy development to help with the final incorporation into the ntupler. I'm working on something else for tomorrow but I am intending to have this together either tomorrow or Thursday.

[scooperstein]

I adapted the C++ code from the top group that we had incorporated into the WH(bb) framework into a self-contained python script.

https://github.com/scooperstein/PrincetonAnalysisTools/blob/master/scripts/getTopMass.py

Like we discussed at the meeting today, it is now a matter of incorporating this into Heppy. Maybe we can iterate further via email to discuss how is best to do that.

[arizzi]

@scooperstein I thought it was just few lines of python to write, not 200 lines of math operations! What was supposed to be pythonized was the computeTopMass function that you pointed us to (I thought that was it and nothing else was needed, I did not notice it was calling other functions in the same file). It could be quite slow to make all that math in python indeed (beside you spent a day converting it I imagine!!). Apologies for the misunderstanding, now we can use this as it is if it is not too slow, otherwise we can use the pythonized computeTopMass calling external c++ functions.

[scooperstein]

Sorry for the misunderstanding. Well we now have the necessary pieces written in both python and C++. :) Please let me know how you think we should proceed. I can quickly provide the same pieces in C++ if needed.

[arizzi]

we can probably do some speed test of the python solution, if ok we go with it, otherwise we can put the C++ class in src, make a dictionary and use it the same way as SoftActivity and ColorFlow classes

[scooperstein]

Hi Andrea,

I did a test calling this python computeTopMass() function on 1 million randomly generated sets of lep, met, jet1, jet2 vectors and I get:

time python getTopMass.py

213.181u 17.460s 3:39.47 105.0%

maybe you or someone else can comment on whether this is reasonable. I am happy to perform any other tests if requested.

[RemKamal]

Hi all, did it get to vhbb? The version of the script that was posted way above, is okay, if I am running on low size samples, but gives ' Problem (math domain error)', if I use all available stats. I guess this python math exception comes from some negative number(or complex?), which screws some math function. Have you seen anything like this? Thanks.

[scooperstein]

Hi,

As far as I understand this has not been integrated into Heppy, since it was not viewed as a priority for the V25 campaign. Yes indeed there are some rare cases where it was getting domain errors due to negative inputs to power(). If you change it slightly to include a protection like

if (r+sqrt(Delta) > 0): s = complex(pow((r+sqrt(Delta)),(1./3)),0) else: s = complex(-pow(abs((r+sqrt(Delta))),(1./3)),0) if (r-sqrt(Delta) > 0): t = complex(pow((r-sqrt(Delta)),(1./3)),0) else: t = complex(-pow(abs((r-sqrt(Delta))),(1./3)),0)

that should fix it.

[arizzi]

it did not enter V25

On Thu, Feb 9, 2017 at 3:09 PM, RemKamal notifications@github.com wrote:

Hi all, did it get to vhbb? The version of the script that was posted way above, is okay, if I am running on low size samples, but gives ' Problem (math domain error)', if I use all available stats. I guess this python math exception comes from some negative number(or complex?), which screws some math function. Have you seen anything like this? Thanks.

— You are receiving this because you commented. Reply to this email directly, view it on GitHub https://github.com/vhbb/cmssw/issues/483#issuecomment-278650743, or mute the thread https://github.com/notifications/unsubscribe-auth/AEyiltb2EFFjZ23RrZuUuqQjaie5NQPlks5rax3-gaJpZM4JLKfw .

[RemKamal]

@scooperstein thanks, I will try!

[RemKamal]

As math.pow(x, y) says: If both x and y are finite, x is negative, and y is not an integer then pow(x, y) is undefined, and raises ValueError.

For example: r ,Delta = -3033136.74138, 8.63740408607e+12 a, b, c, d, q, rho, theta = 1.0 -135.135850433 -18677.4330836 7090404.49876 -8254.88859146 0.0 0.0 <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< Problem (math domain error)

r ,Delta = -940610.687614, 8.79093981695e+11 a, b, c, d, q, rho, theta = 1.0 281.726948841 21112.0456248 2207484.79714 -1781.54853655 0.0 0.0 <<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<<< Problem (math domain error)