Skip to content

MadAnalysis 5 - Plugin for interpreting expert mode analyses

License

Notifications You must be signed in to change notification settings

MadAnalysis/ma5_expert

Folders and files

NameName
Last commit message
Last commit date

Latest commit

 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 
 

Repository files navigation

MadAnalysis 5 Interpreter For Expert Mode

EPJC arxiv

CI

MadAnalysis 5 output interpreter for expert mode. Parses the cutflow and histogram collections and constructs it with an interactable interface.

Installation

pip install ma5-expert

Outline

Examples can be found in examples folder.

Cutflow Collection

  • Parse all the signal regions and construct an object-base, interactable cutflow.

  • Write combined LaTeX tables for different samples.

  • Compare samples and construct validation tables which allow you to calculate the difference of the relative efficiencies for each given sample with respect to a reference sample.

  • Compare signal and background samples and calculate the figure of merit.

  • Possibility to include experimentally available cutflow data and compare it against MadAnalysis 5 cutflow output.

  • Calculate Monte Carlo uncertainty per cut on the fly

  • Simple cutflow:

CutFlowCollection needs CutFlows path of your sample in MadAnalysis 5 Analysis folder. We provide an ma5 directory in examples folder so we will go through and the code using that. Parsing a cutflow simply requires the path of the CutFlows folder and optionally xsection [pb], lumi [1/fb] and/or Nevents. Note that xsec overwrites the number of events option, if provided number of events are always calculated using the cross section.

import ma5_expert as ma5

sample = ma5.cutflow.Collection(
        "docs/examples/mass1000005_300.0_mass1000022_60.0_mass1000023_250.0_xs_5.689/Output/SAF/defaultset/atlas_susy_2018_31/Cutflows",
        xsection = 5.689, lumi = 139.
)

Here the first input is the path of the CutFlows folder and the rest are simply cross section and luminosity information. One can see the signal regions by simply printing the keys of the CutFlowCollection object;

print(sample.SRnames)
# Output: 
# ['SRC_28', 'SRA_M', 'SRA_L', 'SRA_H', 'SRA', 'SRC', 'SRB', 'SRC_26', 'SRC_24', 'SRC_22']

Each signal region is accessible through CutFlowCollection object. For instance one can get the names of the cuts applied in one of the signal regions.

print(sample.SRA.CutNames)
# Output: 
# ['Initial', '$N_{lep} = 0$', '$N_{j} \\geq 6$', '$N_{b} \\geq 4$', 
# '$\\slashed{E}_T > 350$ [GeV]', '$min(\\Delta\\phi(j_{1-4},\\slashed{E}_T))>0.4$ [rad]', 
# '$\\tau^h$ veto', '$p^{b_1}_T > 200$ [GeV]', '$\\Delta R_{max}(b,b)>2.5$', 
# '$\\Delta R_{max-min}(b,b)<2.5$', '$m(h_{cand})>80$ [GeV]', '$m_{eff} > 1$ [TeV]']

Or simply print the entire cutflow;

print(sample.SRA)
# Output: 
# * SRA :
#  * Initial :
#   - Number of Entries    : 200000
#   - Number of Events     : 790771.000 ± 0.000(ΔMC)
#   - Cut & Rel Efficiency : 1.000, 1.000
#  * $N_{lep} = 0$ :
#   - Number of Entries    : 156651
#   - Number of Events     : 499908.962 ± 609.064(ΔMC)
#   - Cut & Rel Efficiency : 0.632, 0.632
#  * $N_{j} \geq 6$ :
#   - Number of Entries    : 65546
#   - Number of Events     : 209971.179 ± 362.184(ΔMC)
#   - Cut & Rel Efficiency : 0.266, 0.420
#  * $N_{b} \geq 4$ :
#   - Number of Entries    : 19965
#   - Number of Events     : 63883.202 ± 123.205(ΔMC)
#   - Cut & Rel Efficiency : 0.081, 0.304
#  * $\slashed{E}_T > 350$ [GeV] :
#   - Number of Entries    : 191
#   - Number of Events     : 755.117 ± 1.688(ΔMC)
#   - Cut & Rel Efficiency : 0.001, 0.012
#  * $min(\Delta\phi(j_{1-4},\slashed{E}_T))>0.4$ [rad] :
#   - Number of Entries    : 72
#   - Number of Events     : 284.658 ± 0.636(ΔMC)
#   - Cut & Rel Efficiency : 0.000, 0.377
#  * $\tau^h$ veto :
#   - Number of Entries    : 68
#   - Number of Events     : 268.850 ± 0.601(ΔMC)
#   - Cut & Rel Efficiency : 0.000, 0.944
#  * $p^{b_1}_T > 200$ [GeV] :
#   - Number of Entries    : 33
#   - Number of Events     : 130.474 ± 0.292(ΔMC)
#   - Cut & Rel Efficiency : 0.000, 0.485
#  * $\Delta R_{max}(b,b)>2.5$ :
#   - Number of Entries    : 25
#   - Number of Events     : 98.836 ± 0.221(ΔMC)
#   - Cut & Rel Efficiency : 0.000, 0.758
#  * $\Delta R_{max-min}(b,b)<2.5$ :
#   - Number of Entries    : 25
#   - Number of Events     : 98.836 ± 0.221(ΔMC)
#   - Cut & Rel Efficiency : 0.000, 1.000
#  * $m(h_{cand})>80$ [GeV] :
#   - Number of Entries    : 10
#   - Number of Events     : 39.543 ± 0.088(ΔMC)
#   - Cut & Rel Efficiency : 0.000, 0.400
#  * $m_{eff} > 1$ [TeV] :
#   - Number of Entries    : 10
#   - Number of Events     : 39.543 ± 0.088(ΔMC)
#   - Cut & Rel Efficiency : 0.000, 1.000

As can be seen, it shows number of entries (MonteCarlo events), number of events (lumi weighted), cut efficiency and relative efficiency. The error in number of events is the MonteCarlo uncertainty.

It is also possible to access practical information

print(sample.SRA.isAlive)
# Output: True

which simply checks the number of entries in the final cut. Hence one can remove the SRs which does not have any statistics;

alive = sample.get_alive()
print(f"Number of cutflows survived : {len(alive)},\nNames of the cutflows : { ', '.join([x.id for x in alive]) }")
# Output: 
# Number of cutflows survived : 8,
# Names of the cutflows : SRA_M, SRA_L, SRA_H, SRA, SRC, SRB, SRC_24, SRC_22

Each cut is accessible through the interface;

fifth = sample.SRA[5]
print(f"Efficiency : {fifth.eff:.3f}, Relative MC efficiency {fifth.mc_rel_eff:.3f}, number of events {fifth.Nevents:.1f}, sum of weights {fifth.sumW:.3f}")
# Output: 
# Efficiency : 0.0004, Relative MC efficiency 0.377, number of events 284.7, sum of weights 0.008

One can also construct independent signal regions for sake of comparisson with Ma5 results;

SRA_presel = [319.7,230.5,192.3,87.9,45.1,20.9,19.3,18.2,17.6,15.0,13.7]

ATLAS = ma5.cutflow.Collection() 

ATLAS.addSignalRegion('SRA',   ma5.SRA.CutNames,   SRA_presel+[13.7])
ATLAS.addSignalRegion('SRA_L', ma5.SRA_L.CutNames, SRA_presel+[0.4])
ATLAS.addSignalRegion('SRA_M', ma5.SRA_M.CutNames, SRA_presel+[6.4])
ATLAS.addSignalRegion('SRA_H', ma5.SRA_H.CutNames, SRA_presel+[7.0])

where all properties shown above applies to this new object as well.

back to top

Histogram Collection

  • Parse all the histograms available in the Histos.saf file into interactable histogram object.
import ma5_expert as ma5

collection = ma5.histogram.Collection(
        "docs/examples/mass1000005_300.0_mass1000022_60.0_mass1000023_250.0_xs_5.689/Output/SAF/defaultset/atlas_susy_2018_31/Histograms/histos.saf",
        xsection = 5.689, lumi = 139.
)

print(collection)
# Collection of 6 histograms from `examples/mass1000005_300.0_mass1000022_60.0_mass1000023_250.0_xs_5.689/Output/SAF/defaultset/atlas_susy_2018_31/Histograms/histos.saf`
#    * SRA_Meff: [ nbin: 11, min: 800.00, max: 3000.00 ]
#    * SRA_Mh: [ nbin: 12, min: 0.00, max: 480.00 ]
#    * SRB_PTj1: [ nbin: 9, min: 50.00, max: 950.00 ]
#    * SRB_MhAvg: [ nbin: 16, min: 50.00, max: 450.00 ]
#    * SRC_MET: [ nbin: 13, min: 200.00, max: 1500.00 ]
#    * SRC_Sig: [ nbin: 19, min: 17.00, max: 36.00 ]

Extract the plotting information:

xbins, bins, weights = collection.lumi_histogram("SRA_Mh")
plt.hist(xbins, bins=bins, weights=weights)
plt.xlabel("$M_{h}\ {\\rm [GeV]}$")
plt.ylabel("${\\rm Number\ of\ events}$")
plt.xlim([min(bins), max(bins)])
plt.show()

SRA_Mh

back to top

Integration to Public Analysis Database through MadAnalysis 5

ma5-expert is capable of running MadAnalysis sub-modules through a backend manager. Desired MadAnalysis backend can be set via

import ma5_expert as ma5
ma5.BackendManager.set_madanalysis_backend("/PATH/TO/MADANALYSIS5")

This will initiate the MadAnalysis backend to be used. Then one can use the reinterpretation tools such as exclusion limit computation, externally. One can initiate PAD interface via

interface = ma5.pad.PADInterface(
    sample_path="ma5_expert/docs/examples/mass1000005_300.0_mass1000022_60.0_mass1000023_250.0_xs_5.689",
    dataset_name="defaultset"
)

where sample_path is the main location of the analysis which has been held, and dataset_name is the name of the dataset which corresponds to the particular folder name under sample_path + /Outputs/SAF/. Then results can be computed via

results = interface.compute_exclusion("atlas_susy_2018_31", 5.689, ma5.backend.PADType.PADForSFS)

Note that the given example only computes for atlas_susy_2018_31 and this analysis has been held under PADForSFS which is indicated via PADType. This simply tells function where to look to find corresponding info file, which assumes that PADForSFS has been installed. The value 5.689 sets the cross section in pb.

back to top

Citation

Developed for arXiv:2006.09387

@article{Araz:2020lnp,
    author = "Araz, Jack Y. and Fuks, Benjamin and Polykratis, Georgios",
    title = "{Simplified fast detector simulation in MADANALYSIS 5}",
    eprint = "2006.09387",
    archivePrefix = "arXiv",
    primaryClass = "hep-ph",
    doi = "10.1140/epjc/s10052-021-09052-5",
    journal = "Eur. Phys. J. C",
    volume = "81",
    number = "4",
    pages = "329",
    year = "2021"
}

back to top

TODO

  • Overall Ma5 Analysis parser

  • Add theoretical uncertainties