small changes towards pf candidate based tagger

Train/DeepJetTrain_PF.py

@@ -60,13 +60,13 @@
 startlearnrate=0.0005
 from DeepJet_callbacks import DeepJet_callbacks
 
-callbacks=DeepJet_callbacks(stop_patience=100, 
+callbacks=DeepJet_callbacks(stop_patience=300, 
 
                             lr_factor=0.5,
                             lr_patience=3, 
                             lr_epsilon=0.003, 
                             lr_cooldown=6, 
-                            lr_minimum=0.00001, 
+                            lr_minimum=0.000001, 
 
                             outputDir=outputDir)
 useweights=False
@@ -76,8 +76,6 @@
 
 
 from DataCollection import DataCollection
-from TrainData_deepCSV_ST import TrainData_deepCSV_ST
-from TrainData_deepCSV import TrainData_deepCSV
 
 traind=DataCollection()
 traind.readFromFile(inputData)
@@ -118,7 +116,7 @@
 print(traind.getTruthShape()[0])
 from DeepJet_models import Dense_model_broad, Dense_model_broad_flat
 #model = Dense_model_Rec(inputs,traind.getTruthShape()[0],shapes,0.3)
-model = Dense_model_broad_flat(inputs,traind.getTruthShape()[0],shapes,0.1)
+model = Dense_model_broad(inputs,traind.getTruthShape()[0],shapes,0.1)
 #model = Dense_model_broad(inputs,traind.getTruthShape()[0],shapes,0.1)
 print('compiling')
 

convertFromRoot/convertFromRoot.py

@@ -73,6 +73,7 @@ def main(argv=None):
 
     from TrainData import TrainData
     from TrainData_deepCSV import TrainData_deepCSV
+    from TrainData_deepConvCSV import TrainData_deepConvCSV
     from TrainData_deepCMVA import TrainData_deepCMVA
     from TrainData_deepCSV_PF import TrainData_deepCSV_PF,TrainData_deepCSV_miniPF,TrainData_deepCSV_microPF
     from TrainData_deepConvCSV import TrainData_deepConvCSV
@@ -83,6 +84,8 @@ def main(argv=None):
     traind=TrainData
     if Class == 'TrainData_deepCSV':
         traind=TrainData_deepCSV
+    elif Class == 'TrainData_deepConvCSV':
+        traind=TrainData_deepConvCSV
     elif Class == 'TrainData_deepJet_Reg':
         traind=TrainData_deepJet_Reg
     elif Class ==  'TrainData_deepCSV_PF':

modules/DataCollection.py

@@ -5,6 +5,8 @@
 '''
 #from tensorflow.contrib.labeled_tensor import batch
 #from builtins import list
+from __future__ import print_function
+
 from Weighter import Weighter
 from TrainData import TrainData, fileTimeOut
 #for convenience

modules/DeepJet_callbacks.py

@@ -3,6 +3,7 @@
 
 @author: jkiesele
 '''
+from __future__ import print_function
 
 from ReduceLROnPlateau import ReduceLROnPlateau
 from keras.callbacks import Callback, EarlyStopping,History,ModelCheckpoint #, ReduceLROnPlateau # , TensorBoard

modules/DeepJet_models.py

@@ -164,47 +164,35 @@ def Dense_model_broad(Inputs,nclasses,Inputshapes,dropoutRate=-1):
 
     cpf  = Convolution1D(64, 1, kernel_initializer='lecun_uniform',  activation='relu')(Inputs[1])
     cpf = Dropout(dropoutRate)(cpf)
-    cpf  = Convolution1D(64, 1, kernel_initializer='lecun_uniform',  activation='relu')(cpf)
-    cpf = Dropout(dropoutRate)(cpf)
-    cpf  = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(cpf)
-    cpf = Dropout(dropoutRate)(cpf)
     cpf  = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(cpf)
     cpf = Dropout(dropoutRate)(cpf)
     cpf  = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(cpf)
     cpf = Dropout(dropoutRate)(cpf)
-    cpf  = Convolution1D(16, 1, kernel_initializer='lecun_uniform',  activation='relu')(cpf)
+    cpf  = Convolution1D(8, 1, kernel_initializer='lecun_uniform',  activation='relu')(cpf)
     cpf = Flatten()(cpf)
 
 
     npf = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu',input_shape=Inputshapes[2])(Inputs[2])
     npf = Dropout(dropoutRate)(npf)
     npf = Convolution1D(16, 1, kernel_initializer='lecun_uniform',  activation='relu')(npf)
     npf = Dropout(dropoutRate)(npf)
-    npf = Convolution1D(16, 1, kernel_initializer='lecun_uniform',  activation='relu')(npf)
-    npf = Dropout(dropoutRate)(npf)
-    npf = Convolution1D(8, 1, kernel_initializer='lecun_uniform',  activation='relu')(npf)
+    npf = Convolution1D(4, 1, kernel_initializer='lecun_uniform',  activation='relu')(npf)
     npf = Flatten()(npf)
 
     vtx = Convolution1D(64, 1, kernel_initializer='lecun_uniform',  activation='relu',input_shape=Inputshapes[3])(Inputs[3])
     vtx = Dropout(dropoutRate)(vtx)
-    vtx = Convolution1D(64, 1, kernel_initializer='lecun_uniform',  activation='relu')(vtx)
-    vtx = Dropout(dropoutRate)(vtx)
-    vtx = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(vtx)
-    vtx = Dropout(dropoutRate)(vtx)
     vtx = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(vtx)
     vtx = Dropout(dropoutRate)(vtx)
     vtx = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(vtx)
     vtx = Dropout(dropoutRate)(vtx)
-    vtx = Convolution1D(16, 1, kernel_initializer='lecun_uniform',  activation='relu')(vtx)
+    vtx = Convolution1D(8, 1, kernel_initializer='lecun_uniform',  activation='relu')(vtx)
     vtx = Flatten()(vtx)
 
 
     x = merge( [Inputs[0],cpf,npf,vtx ] , mode='concat')
 
 
-    x=  Dense(600, activation='relu',kernel_initializer='lecun_uniform')(x)
-    x = Dropout(dropoutRate)(x)
-    x=  Dense(300, activation='relu',kernel_initializer='lecun_uniform')(x)
+    x=  Dense(350, activation='relu',kernel_initializer='lecun_uniform')(x)
     x = Dropout(dropoutRate)(x)
     x=  Dense(100, activation='relu',kernel_initializer='lecun_uniform')(x)
     x = Dropout(dropoutRate)(x)
@@ -214,7 +202,6 @@ def Dense_model_broad(Inputs,nclasses,Inputshapes,dropoutRate=-1):
     x = Dropout(dropoutRate)(x)
     x=  Dense(100, activation='relu',kernel_initializer='lecun_uniform')(x)
     x = Dropout(dropoutRate)(x)
-    x=  Dense(100, activation='relu',kernel_initializer='lecun_uniform')(x)
     predictions = Dense(nclasses, activation='softmax',kernel_initializer='lecun_uniform')(x)
     model = Model(inputs=Inputs, outputs=predictions)
     return model
@@ -282,22 +269,22 @@ def Dense_model_microPF(Inputs,nclasses,Inputshapes,dropoutRate=-1):
 
     from keras.layers.local import LocallyConnected1D
 
-    npf  = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(Inputs[1])
-    npf = Dropout(dropoutRate)(npf)
-    npf  = Convolution1D(8, 1, kernel_initializer='lecun_uniform',  activation='relu')(npf)
-    npf = Dropout(dropoutRate)(npf)
-    npf  = Convolution1D(4, 1, kernel_initializer='lecun_uniform',  activation='relu')(npf)
-    npf = Dropout(dropoutRate)(npf)
-    npf = Flatten()(npf)
+    #npf  = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(Inputs[1])
+    #npf = Dropout(dropoutRate)(npf)
+    #npf  = Convolution1D(8, 1, kernel_initializer='lecun_uniform',  activation='relu')(npf)
+    #npf = Dropout(dropoutRate)(npf)
+    #npf  = Convolution1D(4, 1, kernel_initializer='lecun_uniform',  activation='relu')(npf)
+    #npf = Dropout(dropoutRate)(npf)
+    npf = Flatten()(Inputs[1])
 
 
 
-    x = merge( [Inputs[0],npf,Inputs[2] ] , mode='concat')
+    x = merge( [Inputs[0],npf] , mode='concat')
 
 
-    x=  Dense(150, activation='relu',kernel_initializer='lecun_uniform')(x)
+    x=  Dense(250, activation='relu',kernel_initializer='lecun_uniform')(x)
     x = Dropout(dropoutRate)(x)
-    x=  Dense(100, activation='relu',kernel_initializer='lecun_uniform')(x)
+    x=  Dense(200, activation='relu',kernel_initializer='lecun_uniform')(x)
     x = Dropout(dropoutRate)(x)
     x=  Dense(100, activation='relu',kernel_initializer='lecun_uniform')(x)
     x = Dropout(dropoutRate)(x)
@@ -314,34 +301,32 @@ def Dense_model_ConvCSV(Inputs,nclasses,Inputshape,dropoutRate=0.25):
     Inputs similar to 2016 training, but with covolutional layers on each track and sv
     """
 
-    a  = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(Inputs[1])
+    a  = Convolution1D(8, 1, kernel_initializer='lecun_uniform',  activation='relu')(Inputs[1])
     a = Dropout(dropoutRate)(a)
-    a  = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(a)
+    a  = Convolution1D(8, 1, kernel_initializer='lecun_uniform',  activation='relu')(a)
     a = Dropout(dropoutRate)(a)
     a  = Convolution1D(8, 1, kernel_initializer='lecun_uniform',  activation='relu')(a)
     a = Dropout(dropoutRate)(a)
     a=Flatten()(a)
 
-    b  = Convolution1D(1, 1, kernel_initializer='lecun_uniform',  activation='relu')(Inputs[2])
-    b = Dropout(dropoutRate)(b)
-    b =Flatten()(b)
-
-    c  = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(Inputs[3])
+    c  = Convolution1D(8, 1, kernel_initializer='lecun_uniform',  activation='relu')(Inputs[2])
     c = Dropout(dropoutRate)(c)
-    c  = Convolution1D(32, 1, kernel_initializer='lecun_uniform',  activation='relu')(c)
+    c  = Convolution1D(8, 1, kernel_initializer='lecun_uniform',  activation='relu')(c)
     c = Dropout(dropoutRate)(c)
     c  = Convolution1D(8, 1, kernel_initializer='lecun_uniform',  activation='relu')(c)
     c = Dropout(dropoutRate)(c)
     c=Flatten()(c)
 
-    x = merge( [Inputs[0],a,b,c] , mode='concat')
+    x = merge( [Inputs[0],a,c] , mode='concat')
 
     x = Dense(100, activation='relu',kernel_initializer='lecun_uniform')(x)
     x = Dropout(dropoutRate)(x)
     x = Dense(100, activation='relu',kernel_initializer='lecun_uniform')(x)
     x = Dropout(dropoutRate)(x)
     x = Dense(100, activation='relu',kernel_initializer='lecun_uniform')(x)
     x = Dropout(dropoutRate)(x)
+    x = Dense(100, activation='relu',kernel_initializer='lecun_uniform')(x)
+    x = Dropout(dropoutRate)(x)
     x=  Dense(100, activation='relu',kernel_initializer='lecun_uniform')(x)
     predictions = Dense(nclasses, activation='softmax',kernel_initializer='lecun_uniform')(x)
     model = Model(inputs=Inputs, outputs=predictions)

modules/TrainData.py

@@ -4,6 +4,7 @@
 @author: jkiesele
 '''
 
+from __future__ import print_function
 
 from Weighter import Weighter
 

modules/TrainData_deepCSV_PF.py

@@ -351,42 +351,22 @@ def __init__(self):
         TrainData.__init__(self)
 
 
-        self.addBranches(['jet_pt', 'jet_eta','nCpfcand','nNpfcand','nsv','npv'])
-
-
-
-
-        self.addBranches(['Npfcan_pt',
-                          'Npfcan_ptrel',
-                          'Npfcan_erel',
-                          'Npfcan_phirel',
-                          'Npfcan_etarel',
-                          'Npfcan_deltaR',
-                              'Npfcan_isGamma',
-                              'Npfcan_HadFrac',
-                              'Npfcan_drminsv',
-                              ],
-                             15)
-
-
-
-
-        #b-tag vars
-        self.addBranches([ 'TagVarCSV_jetNSecondaryVertices', 
+        self.addBranches(['jet_pt', 'jet_eta','nNpfcand',#'npv',
+                           'TagVarCSV_jetNSecondaryVertices', 
                            'TagVarCSV_trackSumJetEtRatio', 'TagVarCSV_trackSumJetDeltaR', 
                            'TagVarCSV_vertexCategory', 'TagVarCSV_trackSip2dValAboveCharm', 
                            'TagVarCSV_trackSip2dSigAboveCharm', 'TagVarCSV_trackSip3dValAboveCharm', 
                            'TagVarCSV_trackSip3dSigAboveCharm', 'TagVarCSV_jetNSelectedTracks', 
                            'TagVarCSV_jetNTracksEtaRel'])
-        
+
         self.addBranches(['TagVarCSVTrk_trackJetDistVal',
                               'TagVarCSVTrk_trackPtRel', 
                               'TagVarCSVTrk_trackDeltaR', 
                               'TagVarCSVTrk_trackPtRatio', 
                               'TagVarCSVTrk_trackSip3dSig', 
                               'TagVarCSVTrk_trackSip2dSig', 
                               'TagVarCSVTrk_trackDecayLenVal'],
-                             7)
+                             6)
 
 
         self.addBranches(['TagVarCSV_trackEtaRel'],4)
@@ -403,6 +383,17 @@ def __init__(self):
 
 
 
+        self.addBranches(['Npfcan_pt',
+                          'Npfcan_ptrel',
+                          'Npfcan_erel',
+                          'Npfcan_phirel',
+                          'Npfcan_etarel',
+                          'Npfcan_deltaR',
+                              'Npfcan_isGamma',
+                              'Npfcan_HadFrac',
+                              'Npfcan_drminsv',
+                              ],
+                             15)
 
 
     def readFromRootFile(self,filename,TupleMeanStd, weighter):
@@ -426,20 +417,16 @@ def readFromRootFile(self,filename,TupleMeanStd, weighter):
         # split for convolutional network
 
         x_global = MeanNormZeroPad(filename,TupleMeanStd,
-                                   [self.branches[0]],
-                                   [self.branchcutoffs[0]],self.nsamples)
+                                   self.branches[0:4],
+                                   self.branchcutoffs[0:4],self.nsamples)
 
 
         x_npf = MeanNormZeroPadParticles(filename,TupleMeanStd,
-                                   self.branches[1],
-                                   self.branchcutoffs[1],self.nsamples)
+                                   self.branches[4],
+                                   self.branchcutoffs[4],self.nsamples)
 
 
 
-        x_global_csv = MeanNormZeroPad(filename,TupleMeanStd,
-                                   self.branches[2:],
-                                   self.branchcutoffs[2:],self.nsamples)
-
 
 
         print('took ', sw.getAndReset(), ' seconds for mean norm and zero padding (C module)')
@@ -471,7 +458,6 @@ def readFromRootFile(self,filename,TupleMeanStd, weighter):
             weights=weights[notremoves > 0]
             x_global=x_global[notremoves > 0]
             x_npf=x_npf[notremoves > 0]
-            x_global_csv=x_global_csv[notremoves > 0]
             alltruth=alltruth[notremoves > 0]
 
         newnsamp=x_global.shape[0]
@@ -481,7 +467,7 @@ def readFromRootFile(self,filename,TupleMeanStd, weighter):
         print(x_global.shape,self.nsamples)
 
         self.w=[weights]
-        self.x=[x_global,x_npf,x_global_csv]
+        self.x=[x_global,x_npf]
         self.y=[alltruth]