To compile the standalone library requires the ROOT 6 [1] and Tensorflow 1.2 (or newer) [2] packages.
[1] https://root.cern.ch/downloading-root
Follow the instructions to install ROOT on your system
[2] https://www.tensorflow.org/install/install_c
Follow the instructions to install tensorflow for C. GPU support is not necessary unless you will be training new models. For compatibility with the TopTagger model run the following download command in place of step 2 on this page to download tensorflow v1.3.0 instead of the lastest version
TF_TYPE="cpu" # Change to "gpu" for GPU support (not necessary for inferance)
OS="linux" # Change to "darwin" for macOS
TARGET_DIRECTORY="/usr/local" #Change this to install the c-api in other places
curl -L \
"https://storage.googleapis.com/tensorflow/libtensorflow/libtensorflow-${TF_TYPE}-${OS}-x86_64-1.3.0.tar.gz" |
sudo tar -C $TARGET_DIRECTORY -xzAfter all the necessary packages are installed, checkout the TopTagger repository and compile with the following command (with a terminal configured to use root/tensorflow)
#After downloading the code from github or untaring the standalone tarball
cd TopTagger/TopTagger/test
./configure
make -j4
#To install in a non-standard directory add PREFIX=/install/path to the configure command
#If the tensorflow c-api was installed elsewhere than "/usr/local" add TENSORFLOWPATH=[path to c-api] to the configure command
#the command below is optional for system wide install (will require sudo rights)
make installThis command will produce a static and a shared library which contain all the necessary symbols to link the top tagger into other C++ code. This will also compile a standalone example called ``topTaggerTest''.
A basic standalone example using the top tagging code is provided in "TopTagger/test/topTaggerTest.cpp". This is a basic example program which reads in the necessary top tagger inputs from a file (``exampleInputs.root'') and runs the top tagger code. As validation, this prints out the number of top quarks reconstructed in each event as well as some basic properties of each top quark reconstructed. For reference the output of this script can be seen at the end of this readme.
The example executable is compiled along with the standalone library as described above. If the top tagger was not installed system wide, then there is a script "taggerSetup.sh" which is produced by the configure command which must be sourced to set system variables appropriately. A ".csh" version is also proveded.
cd TopTagger/test
#do once per terminal if opencv or the top tagger are not installed system wide
source taggerSetup.sh
#do once in any directory where you will run the top tagger example
getTaggerCfg.sh -t DeepCombined_Example_v1.0.1
#run the example code
./topTaggerTest
A basic standalone example using the top tagging code is provided in "TopTagger/python/TopTagger.py". This test code gives an example of using the top tagger from python. By default it is configured to read from nanoAOD files where the necessary extra variables have been added, but this can also read from the example root file. As validation, this prints out the number of top quarks reconstructed in each event as well as some basic properties of each top quark reconstructed. This file "TopTagger.py" can be imported into other python scripts and the "TopTagger" python class used as a general interface to the top tagger from python. "TopTaggerProducer.py" is a nanoAOD postprocessor module which will add the output of the rerun tagger to the postprocessed files assuming the necessary inputs have been added to the nanoAOD.
The necessary c-api interface to use the tagger in python is compiled along with the standalone library as described above. There is a script "taggerSetup.sh" which is produced by the configure command which must be sourced to set system variables appropriately. A ".csh" version is also proveded.
cd TopTagger/test
#do once per terminal if opencv or the top tagger are not installed system wide
source ../test/taggerSetup.sh
#do once in any directory where you will run the top tagger example
getTaggerCfg.sh -t DeepCombined_Example_v1.0.1
#run the example code
python ../python/TopTagger.py -e -f exampleInputs.root -b slimmedTuple
The top tagging algorithm takes both a collection of all AK4 jets and all AK8 jets (passing basic pt and |eta| acceptance requirments) along with supporting variables. The input variables used for the example program are listed below
//Variables to hold inputs
//AK4 jet variables
//Each entry in these vectors refers to information for 1 AK4 jet
std::vector<TLorentzVector>** AK4JetLV = new std::vector<TLorentzVector>*();
std::vector<double>** AK4JetBtag = new std::vector<double>*();
std::vector<double>** AK4qgMult = new std::vector<double>*();
std::vector<double>** AK4qgPtD = new std::vector<double>*();
std::vector<double>** AK4qgAxis1 = new std::vector<double>*();
std::vector<double>** AK4qgAxis2 = new std::vector<double>*();
std::vector<double>** AK4recoJetschargedHadronEnergyFraction = new std::vector<double>*();
std::vector<double>** AK4recoJetschargedEmEnergyFraction = new std::vector<double>*();
std::vector<double>** AK4recoJetsneutralEmEnergyFraction = new std::vector<double>*();
std::vector<double>** AK4ElectronEnergyFraction = new std::vector<double>*();
std::vector<double>** AK4PhotonEnergyFraction = new std::vector<double>*();
std::vector<double>** AK4recoJetsneutralEnergyFraction = new std::vector<double>*();
std::vector<double>** AK4recoJetsHFHadronEnergyFraction = new std::vector<double>*();
std::vector<double>** AK4recoJetsmuonEnergyFraction = new std::vector<double>*();
std::vector<double>** AK4recoJetsHFEMEnergyFraction = new std::vector<double>*();
std::vector<double>** AK4NeutralHadronMultiplicity = new std::vector<double>*();
std::vector<double>** AK4ChargedHadronMultiplicity = new std::vector<double>*();
std::vector<double>** AK4ElectronMultiplicity = new std::vector<double>*();
std::vector<double>** AK4MuonMultiplicity = new std::vector<double>*();
std::vector<double>** AK4PhotonMultiplicity = new std::vector<double>*();
std::vector<double>** AK4DeepCSVbb = new std::vector<double>*();
std::vector<double>** AK4DeepCSVb = new std::vector<double>*();
std::vector<double>** AK4DeepCSVc = new std::vector<double>*();
std::vector<double>** AK4DeepCSVcc = new std::vector<double>*();
//AK8 jet varaibles
//The elements of each vector refer to one AK8 jet
std::vector<TLorentzVector>** AK8JetLV = new std::vector<TLorentzVector>*();
std::vector<double>** AK8JetTau1 = new std::vector<double>*();
std::vector<double>** AK8JetTau2 = new std::vector<double>*();
std::vector<double>** AK8JetTau3 = new std::vector<double>*();
std::vector<double>** AK8JetSoftdropMass = new std::vector<double>*();
std::vector<std::vector<TLorentzVector>>** AK8SubjetLV = new std::vector<std::vector<TLorentzVector>>*();
std::vector<std::vector<double>>** AK8SubjetBtag = new std::vector<std::vector<double>>*();
std::vector<std::vector<double>>** AK8SubjetMult = new std::vector<std::vector<double>>*();
std::vector<std::vector<double>>** AK8SubjetPtD = new std::vector<std::vector<double>>*();
std::vector<std::vector<double>>** AK8SubjetAxis1 = new std::vector<std::vector<double>>*();
std::vector<std::vector<double>>** AK8SubjetAxis2 = new std::vector<std::vector<double>>*();
In addition to the code output below you may also get some tensorflow warnings about tensorflow not being compiled to tsupport various special CPU instruction sets. This is ok and these messages can be ignored.
--- Reader : Booking "BDTG" of type "BDT" from weights-t2tt850-sm-baseline-nodphi-nomtb-hqu-08112016.xml.
--- MethodBase : Reading weight file: weights-t2tt850-sm-baseline-nodphi-nomtb-hqu-08112016.xml
--- BDTG : Read method "BDTG" of type "BDT"
--- BDTG : MVA method was trained with TMVA Version: 4.2.1
--- BDTG : MVA method was trained with ROOT Version: 6.06/08
--- DataSetInfo : Added class "Signal" with internal class number 0
--- DataSetInfo : Added class "Background" with internal class number 1
--- Reader : Booked classifier "BDTG" of type: "BDT"
--- Reader : Booking "BDTG" of type "BDT" from sdWTag_ttbarTraining_v0.xml.
--- MethodBase : Reading weight file: sdWTag_ttbarTraining_v0.xml
--- BDTG : Read method "BDTG" of type "BDT"
--- BDTG : MVA method was trained with TMVA Version: 4.2.1
--- BDTG : MVA method was trained with ROOT Version: 6.06/02
--- DataSetInfo : Added class "Signal" with internal class number 0
--- DataSetInfo : Added class "Background" with internal class number 1
--- Reader : Booked classifier "BDTG" of type: "BDT"
Event #: 1
N tops: 1
Top properties: N constituents: 3, Pt: 436.5, Eta: 1.398, Phi: -2.267
Constituent properties: Constituent type: 1, Pt: 227.8, Eta: 1.278, Phi: -2.403
Constituent properties: Constituent type: 1, Pt: 157.5, Eta: 1.708, Phi: -2.060
Constituent properties: Constituent type: 1, Pt: 56.7, Eta: 0.547, Phi: -2.300
Event #: 2
N tops: 2
Top properties: N constituents: 3, Pt: 474.9, Eta: 1.177, Phi: 1.374
Constituent properties: Constituent type: 1, Pt: 337.1, Eta: 1.151, Phi: 1.275
Constituent properties: Constituent type: 1, Pt: 90.4, Eta: 1.486, Phi: 1.778
Constituent properties: Constituent type: 1, Pt: 56.4, Eta: 0.484, Phi: 1.334
Top properties: N constituents: 3, Pt: 272.0, Eta: 0.053, Phi: -0.078
Constituent properties: Constituent type: 1, Pt: 210.2, Eta: -0.008, Phi: 0.092
Constituent properties: Constituent type: 1, Pt: 63.0, Eta: 0.602, Phi: -0.334
Constituent properties: Constituent type: 1, Pt: 20.0, Eta: -1.014, Phi: -1.451
Event #: 3
N tops: 0
Event #: 4
N tops: 1
Top properties: N constituents: 3, Pt: 222.1, Eta: -0.617, Phi: -2.966
Constituent properties: Constituent type: 1, Pt: 87.0, Eta: 0.195, Phi: -2.245
Constituent properties: Constituent type: 1, Pt: 86.7, Eta: -0.302, Phi: 2.751
Constituent properties: Constituent type: 1, Pt: 84.2, Eta: -1.258, Phi: -3.096
Event #: 5
N tops: 1
Top properties: N constituents: 3, Pt: 223.0, Eta: -0.556, Phi: -1.532
Constituent properties: Constituent type: 1, Pt: 135.7, Eta: -0.938, Phi: -1.428
Constituent properties: Constituent type: 1, Pt: 57.7, Eta: -0.136, Phi: -1.023
Constituent properties: Constituent type: 1, Pt: 56.5, Eta: 0.418, Phi: -2.376
Event #: 6
N tops: 1
Top properties: N constituents: 1, Pt: 628.1, Eta: -0.831, Phi: 2.056
Constituent properties: Constituent type: 3, Pt: 628.1, Eta: -0.831, Phi: 2.056
Event #: 7
N tops: 0
Event #: 8
N tops: 1
Top properties: N constituents: 3, Pt: 424.4, Eta: 0.958, Phi: 2.787
Constituent properties: Constituent type: 1, Pt: 261.0, Eta: 0.838, Phi: 2.855
Constituent properties: Constituent type: 1, Pt: 128.4, Eta: 1.153, Phi: 2.411
Constituent properties: Constituent type: 1, Pt: 53.4, Eta: 0.738, Phi: -2.913
Event #: 9
N tops: 0
Event #: 10
N tops: 1
Top properties: N constituents: 3, Pt: 414.0, Eta: -0.250, Phi: -2.257
Constituent properties: Constituent type: 1, Pt: 256.7, Eta: -0.340, Phi: -2.015
Constituent properties: Constituent type: 1, Pt: 135.9, Eta: 0.088, Phi: -2.681
Constituent properties: Constituent type: 1, Pt: 41.4, Eta: -0.626, Phi: -2.398
Event #: 11
N tops: 1
Top properties: N constituents: 3, Pt: 444.4, Eta: -0.569, Phi: 3.114
Constituent properties: Constituent type: 1, Pt: 367.1, Eta: -0.500, Phi: 2.977
Constituent properties: Constituent type: 1, Pt: 73.6, Eta: -0.889, Phi: -2.445
Constituent properties: Constituent type: 1, Pt: 25.6, Eta: -0.055, Phi: -3.127
Event #: 12
N tops: 1
Top properties: N constituents: 3, Pt: 180.0, Eta: -0.287, Phi: 1.145
Constituent properties: Constituent type: 1, Pt: 141.7, Eta: -0.229, Phi: 1.437
Constituent properties: Constituent type: 1, Pt: 37.9, Eta: -0.010, Phi: 0.887
Constituent properties: Constituent type: 1, Pt: 32.0, Eta: -0.569, Phi: -0.184
Event #: 13
N tops: 1
Top properties: N constituents: 3, Pt: 212.3, Eta: 0.297, Phi: -1.735
Constituent properties: Constituent type: 1, Pt: 172.2, Eta: 0.249, Phi: -1.297
Constituent properties: Constituent type: 1, Pt: 54.2, Eta: -0.497, Phi: -2.536
Constituent properties: Constituent type: 1, Pt: 38.9, Eta: 1.049, Phi: -2.805
Event #: 14
N tops: 1
Top properties: N constituents: 3, Pt: 164.5, Eta: 1.983, Phi: -2.734
Constituent properties: Constituent type: 1, Pt: 135.4, Eta: 1.743, Phi: -2.528
Constituent properties: Constituent type: 1, Pt: 48.5, Eta: 1.023, Phi: -2.983
Constituent properties: Constituent type: 1, Pt: 21.8, Eta: 2.641, Phi: 1.216
Event #: 15
N tops: 0
Event #: 16
N tops: 2
Top properties: N constituents: 3, Pt: 232.8, Eta: 0.381, Phi: -2.796
Constituent properties: Constituent type: 1, Pt: 218.7, Eta: 0.454, Phi: -2.650
Constituent properties: Constituent type: 1, Pt: 43.7, Eta: -0.260, Phi: 1.801
Constituent properties: Constituent type: 1, Pt: 24.4, Eta: -0.025, Phi: -2.305
Top properties: N constituents: 3, Pt: 219.0, Eta: -0.721, Phi: -1.208
Constituent properties: Constituent type: 1, Pt: 165.8, Eta: -0.590, Phi: -1.427
Constituent properties: Constituent type: 1, Pt: 54.8, Eta: -0.675, Phi: -0.903
Constituent properties: Constituent type: 1, Pt: 20.3, Eta: -1.137, Phi: 0.122
Event #: 17
N tops: 2
Top properties: N constituents: 3, Pt: 332.1, Eta: -0.661, Phi: -1.792
Constituent properties: Constituent type: 1, Pt: 172.8, Eta: -0.978, Phi: -1.621
Constituent properties: Constituent type: 1, Pt: 110.1, Eta: -0.073, Phi: -2.054
Constituent properties: Constituent type: 1, Pt: 55.5, Eta: -0.524, Phi: -1.810
Top properties: N constituents: 3, Pt: 194.3, Eta: -1.203, Phi: 2.916
Constituent properties: Constituent type: 1, Pt: 157.4, Eta: -0.889, Phi: 2.990
Constituent properties: Constituent type: 1, Pt: 36.0, Eta: -1.900, Phi: -3.015
Constituent properties: Constituent type: 1, Pt: 24.3, Eta: -0.674, Phi: 1.492
Event #: 18
N tops: 1
Top properties: N constituents: 3, Pt: 272.1, Eta: 0.708, Phi: 0.539
Constituent properties: Constituent type: 1, Pt: 131.5, Eta: 0.935, Phi: 0.947
Constituent properties: Constituent type: 1, Pt: 112.5, Eta: 0.546, Phi: -0.084
Constituent properties: Constituent type: 1, Pt: 61.6, Eta: 0.047, Phi: 0.760
Event #: 19
N tops: 2
Top properties: N constituents: 3, Pt: 481.0, Eta: -0.052, Phi: -0.334
Constituent properties: Constituent type: 1, Pt: 279.4, Eta: -0.122, Phi: -0.593
Constituent properties: Constituent type: 1, Pt: 122.6, Eta: 0.468, Phi: -0.114
Constituent properties: Constituent type: 1, Pt: 101.5, Eta: -0.480, Phi: 0.121
Top properties: N constituents: 3, Pt: 136.3, Eta: -0.450, Phi: -1.407
Constituent properties: Constituent type: 1, Pt: 114.3, Eta: -0.658, Phi: -1.665
Constituent properties: Constituent type: 1, Pt: 62.4, Eta: 0.456, Phi: -0.719
Constituent properties: Constituent type: 1, Pt: 24.8, Eta: -0.469, Phi: 2.172
Event #: 20
N tops: 0
Event #: 21
N tops: 2
Top properties: N constituents: 3, Pt: 344.7, Eta: 0.010, Phi: 2.483
Constituent properties: Constituent type: 1, Pt: 247.4, Eta: 0.031, Phi: 2.342
Constituent properties: Constituent type: 1, Pt: 76.6, Eta: -0.489, Phi: 2.821
Constituent properties: Constituent type: 1, Pt: 29.1, Eta: 1.013, Phi: 2.817
Top properties: N constituents: 3, Pt: 206.9, Eta: 0.677, Phi: -0.472
Constituent properties: Constituent type: 1, Pt: 131.2, Eta: 0.623, Phi: -0.268
Constituent properties: Constituent type: 1, Pt: 57.4, Eta: -0.335, Phi: -0.762
Constituent properties: Constituent type: 1, Pt: 25.6, Eta: 1.901, Phi: -0.877
Event #: 22
N tops: 1
Top properties: N constituents: 2, Pt: 322.5, Eta: 1.844, Phi: 1.379
Constituent properties: Constituent type: 3, Pt: 282.1, Eta: 1.779, Phi: 1.597
Constituent properties: Constituent type: 1, Pt: 77.1, Eta: 1.597, Phi: 0.464
Event #: 23
N tops: 0
Event #: 24
N tops: 0
Event #: 25
N tops: 0
Event #: 26
N tops: 2
Top properties: N constituents: 1, Pt: 435.0, Eta: -0.313, Phi: 1.447
Constituent properties: Constituent type: 3, Pt: 435.0, Eta: -0.313, Phi: 1.447
Top properties: N constituents: 3, Pt: 172.6, Eta: -0.164, Phi: 1.443
Constituent properties: Constituent type: 1, Pt: 95.7, Eta: 0.046, Phi: 2.006
Constituent properties: Constituent type: 1, Pt: 67.4, Eta: -0.459, Phi: 0.837
Constituent properties: Constituent type: 1, Pt: 38.5, Eta: -0.023, Phi: 1.104
Event #: 27
N tops: 0
Event #: 28
N tops: 1
Top properties: N constituents: 1, Pt: 614.9, Eta: -0.303, Phi: 0.893
Constituent properties: Constituent type: 3, Pt: 614.9, Eta: -0.303, Phi: 0.893
Event #: 29
N tops: 1
Top properties: N constituents: 3, Pt: 322.5, Eta: -1.178, Phi: -3.138
Constituent properties: Constituent type: 1, Pt: 276.3, Eta: -1.035, Phi: -3.025
Constituent properties: Constituent type: 1, Pt: 46.2, Eta: -1.717, Phi: -2.991
Constituent properties: Constituent type: 1, Pt: 38.0, Eta: -0.260, Phi: 1.633
Event #: 30
N tops: 1
Top properties: N constituents: 2, Pt: 260.2, Eta: 0.081, Phi: 1.268
Constituent properties: Constituent type: 3, Pt: 218.5, Eta: 0.083, Phi: 1.508
Constituent properties: Constituent type: 1, Pt: 70.7, Eta: 0.042, Phi: 0.444