def Fit(OutPre):
V = 'HT_all'
Var = [V]
Utils.stdinfo("Getting the variable from the root file")
ListSamples = DIClasses.Init(Var, Cuts=True) #HT_all
Sampler = SampleHandler.SampleHandler(ListSamples)
Sampler.norm = False #Y axis norm to one
Sampler.valSize = 0.2
Sampler.Split = 'EO'
Sampler.Scale = None #X axis scaling for NN
Sampler.SequenceLength = 0
Sampler.Plots = False
DataSet = Sampler.GetANNInput(verbose=0)
train, test = DataSet.GetInput("Even")
vali = DataSet.vali
PlotService.VarHist(DataSet,
'NLO',
'./plots/',
'test',
V,
r"H_{T} [GeV]",
500,
2200,
30,
Norm=True)
os.system("mv plots/" + V + ".png plots/" + V + "_wo.png")
OutPreSig = OutPre[:, 0]
cut, senSig = FindCut(OutPreSig, test.OutTrue, test.Weights)
print(cut, senSig)
Events = test.Events[OutPreSig > cut]
Weights = test.Weights[OutPreSig > cut]
OutTrue = test.OutTrue[OutPreSig > cut]
MultiClass = test.MultiClass[OutPreSig > cut]
NewSample = DIClasses.DISample(Events, Weights, OutTrue, MultiClass,
vali.LVariables, vali.Names)
PlotService.VarHist(NewSample,
'NLO',
'./plots/',
'test',
V,
r"H_{T} [GeV]",
500,
2200,
30,
Norm=True) #H_{\text{T}}^{\text{all}} [GeV]
def PhyNNCompare(self, layer, LCompareVar):
"""
Comparesion of the neuron output in a given layer to variables that are build by using physics intuition
layer: Which Layer should be used for the comparision?
LCompareVar: List of Variables to which the neuron Output is compared
"""
ListSamples = DIClasses.Init(LCompareVar, Cuts=True)
Sampler = SampleHandler.SampleHandler(ListSamples)
Sampler.norm = False
Sampler.SequenceLength = 0
Sampler.Plots = False
CompareData = Sampler.GetANNInput(verbose=0)
CompareTrain, CompareTest = CompareData.GetInput("Even")
train, test = self.DataSet.GetInput("Even")
input1 = self.Model.input # input placeholder
output1 = [l.output for l in self.Model.layers] # all layer outputs
fun = K.function([input1, K.learning_phase()],
output1) # evaluation function
LayerOutput = fun([train.Events, 1.])
Layer = LayerOutput[layer - 4]
for i in range(Layer.shape[1]):
for j in range(CompareTrain.Events.shape[1]):
NeuronOutput = DIClasses.DISample(
Utils.Transform(Layer[:, i],
'MinMax'), CompareTrain.Weights, None,
CompareTrain.OutTrue, None, None)
CompareVar = DIClasses.DISample(
Utils.Transform(CompareTrain.Events[:, j],
'MinMax'), CompareTrain.Weights, None,
CompareTrain.OutTrue, None, None)
#PlotService.SigBkgHist(NeuronOutput,'Neuron ('+str(i)+') [a.u.]',40,0,1,tag='_Neuron_'+str(i))
#PlotService.SigBkgHist(CompareVar, LCompareVar[j], 40,0,1,tag='LCompareVar'+str(j))
Hist2D = PlotService.SigBkg2D(CompareVar, NeuronOutput,
'Neuron (' + str(i) + ') [a.u.]',
LCompareVar[j], 30, 30, 0, 1, 0,
1)
Utils.stdinfo(
"The mutual information of Neuron {0} and {1} is : {2}".
format(i, LCompareVar[j], self.GetNormedMi(Hist2D)))
def Fit(OutPre):
Var = ['met_met']
stdinfo("Getting the variable from the root file")
ListSamples = Init(Var,Cuts=True) #HT_all
Sampler = SampleHandler.SampleHandler(ListSamples)
Sampler.norm = False #Y axis norm to one
Sampler.valSize = 0.2
Sampler.Split = 'EO'
Sampler.Scale = 'None' #X axis scaling for NN
Sampler.SequenceLength = 0
Sampler.Plots = False
DataSet = Sampler.GetANNInput(verbose=0)
train, test = DataSet.GetInput("Even")
#VarHist(Sampler.ListAnaSetup,'./plots/','test', "HT_all", r"H_{\text{T}}^{\text{all}} [GeV]",500,2000,30,Norm=False)
cut = 0.31 #TODO: Implement best cut search
PlotService.SBVar('./plots/',Var[0],test, r"H_{\text{T}}^{\text{all}} [GeV]",30,0,500,tag='_after_'+str(cut)+'_cut',Norm=False)
Events = test.Events[OutPre > cut]
Weights = test.Weights[OutPre > cut]
OutTrue = test.OutTrue[OutPre > cut]
NewSample = DISample(Events,Weights,None,OutTrue,None)
PlotService.SBVar('./plots/',Var[0],NewSample, r"H_{\text{T}}^{\text{all}} [GeV]",30,0,500,tag='No_cut',Norm=False)
assert 0 == 1 #TODO:Check plot
def Main(self):
#importing the Variable to plot
Samples = 'nomLoose'
Utils.stdinfo("Getting the variable from the root file")
ListSamples = DIClasses.Init([self.Var],Samples,Cuts=True) #HT_all
Sampler = SampleHandler.SampleHandler(ListSamples)
Sampler.norm = False #Y axis norm to one
Sampler.valSize = 0.2
Sampler.Split = 'EO'
Sampler.Scale = None #X axis scaling for NN
Sampler.SequenceLength = 0
Sampler.Plots = False
DataSet = Sampler.GetANNInput(verbose=0)
for Name in self.ModelNames:
self.Name = Name
# Plotting the variable with any NN Score cuts
PlotService.VarHist(DataSet,'NLO','./plots/','All', self.Var, r"H_{T} [GeV]",500,2200,30,Norm=True)
if("Even" in Name):
os.system("mv plots/"+self.Var+".png plots/"+self.Var+"_wo_Even.png")
elif("Odd" in Name):
os.system("mv plots/"+self.Var+".png plots/"+self.Var+"_wo_Odd.png")
#Optimizing a cut based on S/sqrt(S+B) and apllying it to the variable
OutPre = MakePrediction(DataSet)
cut, senSig = FindCut(OutPre,DataSet)
print(cut, senSig)
test = ApllyCuts(test,OutPre[1],cut)
train = ApllyCuts(train,OutPre[0],cut)
vali = ApllyCuts(vali,OutPre[2],cut)
NewDataSet = DIClasses.DIDataSet(test,train,vali)
PlotService.VarHist(NewDataSet,'NLO','./plots/','All', self.Var, r"H_{T} [GeV]",500,2200,30,Norm=True) #H_{\text{T}}^{\text{all}} [GeV]
if("Even" in Name):
os.system("mv plots/"+self.Var+".png plots/"+self.Var+"_Even.png")
elif("Odd" in Name):
os.system("mv plots/"+self.Var+".png plots/"+self.Var+"_Odd.png")
def PhyNNCompare(self, layer, LCompareVar):
"""
Comparesion of the neuron output in a given layer to variables that are build by using physics intuition
layer: Which Layer should be used for the comparision?
LCompareVar: List of Variables to which the neuron Output is compared
"""
ListSamples = DIClasses.Init(LCompareVar, Cuts=True)
Sampler = SampleHandler.SampleHandler(ListSamples)
Sampler.norm = False
Sampler.SequenceLength = 1
Sampler.Plots = False
CompareData = Sampler.GetANNInput(verbose=0)
CompareTrain, CompareTest = CompareData.GetInput("Even")
train, test = self.DataSet.GetInput("Even")
input1 = self.Model.input # input placeholder
output1 = [l.output for l in self.Model.layers] # all layer outputs
fun = K.function([input1, K.learning_phase()],
output1) # evaluation function
LayerOutput = fun([train.Events, 1.])
Layer = LayerOutput[layer]
LCompareVar = CompareData.LVariables
print(pearsonr(train.Events[:, 0], CompareTrain.Events[:, 0]))
return RedVarList
def Lasso(train, test): #Selecting features using Lasso (L1) regularisation
print(len(train.Events[0]))
sfm = SelectFromModel(MultiTaskLassoCV())
sfm.fit(train.Events[:270010], test.Events)
trainE = sfm.transform(train.Events[:270010])
testE = sfm.transform(test.Events)
print(len(testE[0]))
# Basic Sample informations
ListSamples = Init('Five')
# Data Preparation
Sampler = SampleHandler.SampleHandler(ListSamples)
Sampler.norm = 'ZScore'
Sampler.valSize = 0.2
Sampler.Split = 'EO'
Sampler.Scale = False
train, test, vali = Sampler.GetANNInput()
# Lasso(train,test)
ANNSetup = DIANNSetup('80', 'test_model.h5', '8192', 'PyKeras')
RedFeature = SBS(4, ListSamples[0].VarName, ANNSetup)
RedFeature.Main(train, test, vali)
print(RedFeature.Removed)
print(RedFeature.FinalScore)
print(RedFeature.FinalList)
print(RedFeature.scores)
import DIClasses
import PlotService
import HighScoreResults
import numpy as np
from tensorflow import keras
Samples = 'nomLoose' # 'nomLoose' All samples, else List of Samples needed
ModelName = 'LSTM'
# Basic Sample informations
ListSamples = DIClasses.Init(ModelName,Samples,Cuts=True)
# Data Preparation
Mode = 'Fast'
if(Samples != 'nomLoose'):
Mode = 'Save'
Sampler = SampleHandler.SampleHandler(ListSamples,mode=Mode+ModelName)
Sampler.norm = False # Y axis norm to one
Sampler.valSize = 0.2 # Size of the validation sample
Sampler.Split = 'EO' # Use EO (even odd splitting)
Sampler.Scale = 'ZScoreLSTM' # Kind of Norm use ZScoreLTSM
Sampler.SequenceLength = 7 # Length of the Sequence of a Bach (at the momement this controlls amount of jets)
Sampler.Plots = False # Should the plots be done?
DataSet = Sampler.GetANNInput()
class EvalRNN:
def __init__(self, SavePath, ModelName, DataSet):
"""
print("Neurons: {}".format(Neurons))
print("Epochs: {0}".format(Epochs))
print("Batch: {0}".format(Batch))
print("Dropout: {0}".format(Dropout))
print("Regu: {0}".format(Regu))
LearnRate.Print()
ListSamples = DIClasses.Init(
ModelName, Samples, Cuts=True
) # Initiate the setup to import the samples (Cuts string of singal region cuts or True for standard definition)
GPU = True # Enable for GPU training
Mode = 'Fast' # Fast, Slow or Save
if (Samples != 'nomLoose'):
Mode = 'Save'
Sampler = SampleHandler.SampleHandler(ListSamples, mode=Mode +
ModelName) # Initiate the Sampler
Sampler.norm = False # Y axis norm to one
Sampler.valSize = 0.2 # Size of the validation sample
Sampler.Split = 'EO' # Use EO (even odd splitting)
Sampler.Scale = 'ZScoreLSTM' # Kind of Norm use ZScoreLTSM
Sampler.SequenceLength = 7 # Length of the (time) sequence of a bach (#jets per event)
Sampler.Plots = 'NLO' # Which sample should be used for the signal (LO or NLO), no plots => False
if (GPU != True):
tf.config.optimizer.set_jit(True)
os.environ['CUDA_VISIBLE_DEVICES'] = '-1'
DeviceTyp = device_lib.list_local_devices()
DeviceTyp = str(DeviceTyp[0])
DeviceTyp = DeviceTyp[DeviceTyp.find("device_type:") +
14:DeviceTyp.find("device_type:") + 17]
Utils.stdwar("Running on {0} device!".format(DeviceTyp))
# ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ #
Samples = 'nomLoose' # 'nomLoose' All samples, else List of Samples needed
ListSamples = DIClasses.Init(
ModelName, Samples,
Cuts=True) # Feeds the vairables to function Init defined in DIClasses.py
GPU = False # Enables GPU training
Mode = 'Slow' # Fast, Slow or Save
if Samples != 'nomLoose': # Fast and Save modes only available for nomLoose
Mode = 'Slow'
Sampler = SampleHandler.SampleHandler(
ListSamples, mode=Mode + ModelName
) # Creates instance of class SampleHandler defined in SamplerHandler.py
Sampler.NormFlag = False # y axis to one
Sampler.valSize = 0.2 # Size of the validation sample
Sampler.Split = 'EO' # Use EO (even odd splitting)
Sampler.Plots = False # either False, 'LO' or 'NLO's
DataSet = Sampler.GetANNInput()
SavePath = './ANNOutput/FNN/' # NN output save path
# ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ #
# #
# ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ ~ #
if GPU != True: