Adding the mvajj and mvajj_true probabilities

[AlessandroTarabini]

As I reported in my presentation (cfg slide 8 of this presentation), there is the need to compute the probabilities that were previously labelled as mvajj and mvajj_true. This was done using MelaAnalytics, how can we do it now?

[pianonick411]

Hi @AlessandroTarabini,

mavjj(_true) is computed in this line in MELAAnalyitics. It calls computeDijetConvBW. I am not sure of the exact physical meaning, but the computation is easy to recreate in our framework.

To perform this computation, we will need two booleans. One will be similar to isPM4l, which should state that we use self.MELA.computeDijetConvBW() as the function that gets the probability. The other boolean should specify whether or not we are computing mvajj or mvajj_true. This happens to be the only argument that computeDijetConvBW can take in the python interface, as the rest is inherited from the MELA object, similar to the PM4l computation.

[namapane]

Thanks @pianonick411 ! A small comment for @AlessandroTarabini : in order not to have 2 separate parameters (with 4 combinations, one of which does not make sense) I would suggest to add a single parameter with three values (eg "None", "mvajj", "mvajj_true", but you may have better ideas).

[AlessandroTarabini]

Thanks @pianonick411!
@namapane We need two flags, because it might be that one wants mvajj_true but not mvajj, but the two flags are not supposed to be combined.

I created a branch linked to this issue and pushed the commit.

I have tested the discriminant with W+H, VBF, and ggH, here the output:

Plot from Run2:

Qualitatively looks ok, but probably to really understand if we are in synch with Run2 probabilities we have to run on Run2 NanoAODv15 and check event-by-event.

If it is ok fro you, I will open the MR

[namapane]

@namapane We need two flags, because it might be that one wants mvajj_true but not mvajj, but the two flags are not supposed to be combined.

I'm not sure of what you mean with "are not supposed to be combined", but a single string can accommodate specifying none of the two, only the first, only the second, or even both if relevant. More in general, I am a bit worried of seeing that we are creating a setup with a dozen of boolean variables, with a huge number of technically possible combinations, most of which are probably meaningless (is ispmavjj==True and ispAux==True meaningful? And ispmvajj_true==True and computeprop==True? etc.) and I think we should try to handle this before it becomes a mess.

More importantly, do we really need a separate variable for these special cases, in addition to the "nominal" probablity variable? If, so the flags are misnamed, they should be called "addpmvajj" and "addpmvajj_True".
Otrherwise, the result should stored in the existing "nominal" probability variable, somehow like "Prod" and "Dec" do.

[AlessandroTarabini]

Ciao Nicola,
I agree that we start having a lot of flags and there are some combinations that will likely never be used.
Regarding the "mvajj" probabilities, yes, we need these variables in addition to the nominal one. The D_VH probabilities need the nominal, the mvajj, and the mvajj_true probabilities, they should be multiplied together, but perhaps for other discriminants you only need some of them.

One thing we can do to reduce the number of flags is to include these options directly inside the name. For instance, now the probabilities needed for the D_ZH discriminant are defined here. Since "ispmavjj": True and "ispmvajj_true": True, we get the three probabilities needed for the computation of the discriminant. We could triple the entries in the pyfragment and changing the name "Name": "HadZH_SIG_ghz1_1_JHUGen_JECNominal", "Name": "HadZH_SIG_ghz1_1_JHUGen_JECNominal_mvajj", "Name": "HadZH_SIG_ghz1_1_JHUGen_JECNominal_mvajj_true", and then adding if statements checking the content of the name. But I do not know if it is really a "cleaner" solution

[namapane]

Thanks @AlessandroTarabini , if this is the case I would keep things as is, but change the flag names to addpmvajj and addpmvajj_True consistently with addPAux (meaning that is changes what is stored in the "nominal" variable while add specifies to add additional variables).