def PlotEfficiency(htest_s, htest_b, saveDir, saveName, saveFormats):
ROOT.gStyle.SetOptStat(0)
canvas = plot.CreateCanvas()
canvas.cd()
canvas.SetLeftMargin(0.145)
canvas.SetRightMargin(0.11)
# Calculate signal and background efficiency vs output
xvalue, eff_s, eff_b, error = CalcEfficiency(htest_s, htest_b)
graph_s = plot.GetGraph(xvalue, eff_s, error, error, error, error)
graph_b = plot.GetGraph(xvalue, eff_b, error, error, error, error)
plot.ApplyStyle(graph_s, ROOT.kBlue)
plot.ApplyStyle(graph_b, ROOT.kRed)
# Calculate significance vs output
h_signif0, h_signif1 = CalcSignificance(htest_s, htest_b)
plot.ApplyStyle(h_signif0, ROOT.kGreen)
plot.ApplyStyle(h_signif1, ROOT.kGreen + 3)
#=== Get maximum of significance
maxSignif0 = h_signif0.GetMaximum()
maxSignif1 = h_signif1.GetMaximum()
maxSignif = max(maxSignif0, maxSignif1)
# Normalize significance
h_signifScaled0 = h_signif0.Clone("signif0")
h_signifScaled0.Scale(1. / float(maxSignif))
h_signifScaled1 = h_signif1.Clone("signif1")
h_signifScaled1.Scale(1. / float(maxSignif))
#Significance: Get new maximum
ymax = max(h_signifScaled0.GetMaximum(), h_signifScaled1.GetMaximum())
for obj in [graph_s, graph_b, h_signifScaled0, h_signifScaled1]:
obj.GetXaxis().SetTitle("Output")
obj.GetYaxis().SetTitle("Efficiency")
obj.SetMaximum(ymax * 1.1)
obj.SetMinimum(0)
#Draw
h_signifScaled0.Draw("HIST")
h_signifScaled1.Draw("HIST SAME")
graph_s.Draw("PL SAME")
graph_b.Draw("PL SAME")
graph = plot.CreateGraph([0.5, 0.5], [0, ymax * 1.1])
graph.Draw("same")
#Legend
leg = plot.CreateLegend(0.50, 0.25, 0.85, 0.45)
leg.AddEntry(graph_s, "Signal Efficiency", "l")
leg.AddEntry(graph_b, "Bkg Efficiency", "l")
leg.AddEntry(h_signifScaled0, "S/#sqrt{S+B}", "l")
leg.AddEntry(h_signifScaled1, "2#times(#sqrt{S+B} - #sqrt{B})", "l")
leg.Draw()
# Define Right Axis (Significance)
signifColor = ROOT.kGreen + 2
rightAxis = ROOT.TGaxis(1, 0, 1, 1.1, 0, 1.1 * maxSignif, 510, "+L")
rightAxis.SetLineColor(signifColor)
rightAxis.SetLabelColor(signifColor)
rightAxis.SetTitleColor(signifColor)
rightAxis.SetTitleOffset(1.25)
rightAxis.SetLabelOffset(0.005)
rightAxis.SetLabelSize(0.04)
rightAxis.SetTitleSize(0.045)
rightAxis.SetTitle("Significance")
rightAxis.Draw()
plot.SavePlot(canvas, saveDir, saveName, saveFormats)
canvas.Close()
return
def main():
ROOT.gStyle.SetOptStat(0)
# Apply tdr style
style = tdrstyle.TDRStyle()
style.setOptStat(False)
style.setGridX(False)
style.setGridY(False)
# Definitions
filename = opts.filename
tfile = ROOT.TFile.Open(filename)
dName = ""
if ("TT" in filename):
dName = "TT"
elif ("QCD" in filename):
dName = "QCD"
#Signal and background branches
signal = uproot.open(filename)["treeS"]
background = uproot.open(filename)["treeB"]
# Input list
inputList = []
inputList.append("TrijetPtDR")
inputList.append("TrijetDijetPtDR")
inputList.append("TrijetBjetMass")
inputList.append("TrijetLdgJetBDisc")
inputList.append("TrijetSubldgJetBDisc")
inputList.append("TrijetBJetLdgJetMass")
inputList.append("TrijetBJetSubldgJetMass")
inputList.append("TrijetMass")
inputList.append("TrijetDijetMass")
inputList.append("TrijetBJetBDisc")
inputList.append("TrijetSoftDrop_n2")
inputList.append("TrijetLdgJetCvsL")
inputList.append("TrijetSubldgJetCvsL")
inputList.append("TrijetLdgJetPtD")
inputList.append("TrijetSubldgJetPtD")
inputList.append("TrijetLdgJetAxis2")
inputList.append("TrijetSubldgJetAxis2")
inputList.append("TrijetLdgJetMult")
inputList.append("TrijetSubldgJetMult")
nInputs = len(inputList)
#Signal and background dataframes
df_signal = signal.pandas.df(inputList)
df_background = background.pandas.df(inputList)
# Number of events to predict the output and plot the top-quark mass
if (opts.plotSignal):
nEvts = len(df_signal.index)
else:
nEvts = len(df_background.index)
# User few events for testing
if (opts.test):
nEvts = 1000
print "=== Number of events: ", nEvts
#Signal and background datasets
dset_signal = df_signal.values
dset_background = df_background.values
# Concat signal, background datasets
df_list = [df_signal, df_background]
df_all = pandas.concat(df_list)
dataset = df_all.values
# Target (top-quark mass) datasets
dset_target_all = pandas.concat([
signal.pandas.df(["TrijetMass"]),
background.pandas.df(["TrijetMass"])
]).values
dset_target_bkg = background.pandas.df(["TrijetMass"]).values
dset_target_signal = signal.pandas.df(["TrijetMass"]).values
# Signal and background inputs
X_signal = dset_signal[:nEvts, 0:nInputs]
X_background = dset_background[:nEvts, 0:nInputs]
if (opts.plotSignal):
X = dset_signal[:nEvts, 0:nInputs]
target = dset_target_signal[:nEvts, :]
else:
X = dset_background[:nEvts, 0:nInputs]
target = dset_target_bkg[:nEvts, :]
#Load models
lamValues = [0, 1, 5, 10,
20] #, 100, 500] # fixme! should be given as option
colors = [
ROOT.kBlack, ROOT.kBlue, ROOT.kMagenta, ROOT.kOrange, ROOT.kRed,
ROOT.kGreen, ROOT.kOrange + 7
]
canvas = plot.CreateCanvas()
canvas.cd()
ymaxFactor = 1.1
ymax = 0
histoList = []
if (opts.setLogY):
canvas.SetLogy()
ymaxFactor = 2
# load the models with different lambda
for lam in lamValues:
print "Lambda = ", lam
if (lam == 0):
# For labda = 0 load the simple classification (sequential) model
loaded_model = load_model(
'models_16Nov/Model_relu_relu_relu_sigmoid.h5')
else:
loaded_model = load_model(
'models_16Nov/modelANN_32_Adam_0p0008_500_tanh_relu_msle_lam%s.h5'
% (lam))
# Compile the model
loaded_model.compile(loss='binary_crossentropy',
optimizer='adam',
metrics=['acc'])
Y = loaded_model.predict(X, verbose=1)
# Concatenate Y (predicted output) and target (top-quark mass)
# Ymass 0 column: output
# Ymass 1st column: top-quark mass
Ymass = numpy.concatenate((Y, target), axis=1)
# Get selected top candidates (pass the output working point)
Ymass_sel = Ymass[Ymass[:, 0] >=
opts.wp] # Select samples with y > WP (col 0)
massSel = Ymass_sel[:,
1] # Get the top-quark mass (col 1) for the selected samples.
#Plot resutls
nbins = 100
xmin = 0
xmax = 1000
# Change x-axis range when plotting the signal
if (opts.plotSignal):
nbins = 45
xmax = 450
width = float(xmax) / nbins
histo = ROOT.TH1F('histo_lam%s' % (lam), '', nbins, xmin, xmax)
mass_sel = []
print "selected entries:", len(massSel)
for mass in massSel:
histo.Fill(mass)
histoList.append(histo.Clone("histoClone_lam%s" % lam))
del loaded_model
# Create legend
leg = plot.CreateLegend(0.6, 0.67, 0.9, 0.85)
#Legend text
dText = dName
dText = dText.replace("TT", "t#bar{t}")
# Loop over the histograms in the histoList
for i in range(len(histoList)):
leg.AddEntry(histoList[i],
"%s (#lambda = %.0f)" % (dText, lamValues[i]), "f")
# Normalize histograms to unity
histoList[i].Scale(1. / (histoList[i].Integral()))
ymax = max(ymax, histoList[i].GetMaximum())
histoList[i].GetXaxis().SetTitle("m_{top} (GeV)")
histoList[i].GetYaxis().SetTitle("Arbitrary Units / %.0f GeV" %
(width))
plot.ApplyStyle(histoList[i], colors[i])
histoList[i].Draw("HIST same")
for i in range(len(histoList)):
histoList[i].SetMaximum(ymax * ymaxFactor)
leg.Draw("same")
if (opts.plotSignal):
sbText = "truth-matched"
else:
sbText = "unmatched"
# Additional text
tex1 = plot.Text(" %s top candidates" % sbText, 0.85, 0.5)
tex2 = plot.Text(" with output value > %s" % (opts.wp), 0.82, 0.45)
#Draw text
tex1.Draw()
tex2.Draw()
# Draw line to indicate the real value of the top-quark mass
graph = plot.CreateGraph([173., 173.], [0, ymax * ymaxFactor])
graph.Draw("same")
# CMS extra text and lumi text
#plot.CMSText("CMS Preliminary") #Fixme! cmsExtra doesn't work
#cmsText.Draw()
# Output directory
dirName = plot.getDirName(opts.saveDir)
# Save the plot
saveName = opts.saveName
if (opts.saveName == "TopMassANN"):
saveName = "TopMassANN_%s_%s" % (dName, sbText.replace("-", "_"))
# Save the histogram
plot.SavePlot(canvas, dirName, saveName)
def PlotAndWriteJSON(signal, bkg, saveDir, saveName, jsonWr, saveFormats,
**kwargs):
resultsDict = {}
resultsDict["signal"] = signal
resultsDict["background"] = bkg
normalizeToOne = False
if "normalizeToOne" in kwargs:
normalizeToOne = kwargs["normalizeToOne"]
# Create canvas
ROOT.gStyle.SetOptStat(0)
canvas = plot.CreateCanvas()
canvas.cd()
hList = []
gList = []
yMin = 100000
yMax = -1
xMin = 0.0
xMax = 1.0
nBins = 50
xTitle = "DNN output"
yTitle = "Entries"
log = True
if "log" in kwargs:
log = kwargs["log"]
if "xTitle" in kwargs:
xTitle = kwargs["xTitle"]
if "yTitle" in kwargs:
yTitle = kwargs["yTitle"]
elif "output" in saveName.lower():
xTitle = "Entries"
elif "efficiency" in saveName.lower():
xTitle = "Efficiency"
elif "significance" in saveName.lower():
xTitle = "Significance"
else:
pass
if "xMin" in kwargs:
xMin = kwargs['xMin']
if "xMax" in kwargs:
xMax = kwargs['xMax']
if "nBins" in kwargs:
xBins = kwargs['nBins']
# For-loop:
for i, key in enumerate(resultsDict.keys(), 0):
h = ROOT.TH1F(key, '', nBins, xMin, xMax)
for j, x in enumerate(resultsDict[key], 0):
h.Fill(x)
try:
yMin = min(x[0], yMin)
except:
pass
# Save maximum
yMax = max(h.GetMaximum(), yMax)
# Customise & append to list
plot.ApplyStyle(h, i + 1)
if normalizeToOne:
if "ymax" in kwargs:
yMax = kwargs["yMax"]
else:
yMax = 1.0
if "yMin" in kwargs:
yMin = kwargs["yMin"]
else:
yMin = 1e-4
if h.Integral() > 0.0:
h.Scale(1. / h.Integral())
hList.append(h)
if yMin <= 0.0:
yMin = 100
if log:
canvas.SetLogy()
# For-loop: All histograms
for i, h in enumerate(hList, 0):
h.SetMinimum(yMin * 0.85)
h.SetMaximum(yMax * 1.15)
h.GetXaxis().SetTitle(xTitle)
h.GetYaxis().SetTitle(yTitle)
if i == 0:
h.Draw("HIST")
else:
h.Draw("HIST SAME")
# Create legend
leg = plot.CreateLegend(0.6, 0.75, 0.9, 0.85)
for h in hList:
leg.AddEntry(h, h.GetName(), "l")
leg.Draw()
plot.SavePlot(canvas, saveDir, saveName, saveFormats)
canvas.Close()
# Create TGraph
for h in hList:
gList.append(convertHistoToGaph(h))
# Write the Tgraph into the JSON file
for gr in gList:
gName = "%s_%s" % (saveName, gr.GetName())
jsonWr.addGraph(gName, gr)
return
def PlotAndWriteJSON(signal, bkg, saveDir, saveName, jsonWr, saveFormats):
resultsDict = {}
resultsDict["signal"] = signal
resultsDict["background"] = bkg
# Create canvas
ROOT.gStyle.SetOptStat(0)
canvas = plot.CreateCanvas()
canvas.cd()
hList = []
gList = []
yMin = 100000
yMax = -1
# For-loop:
for i, key in enumerate(resultsDict.keys(), 0):
h = ROOT.TH1F(key, '', 50, 0.0, 1.0)
for j, x in enumerate(resultsDict[key], 0):
h.Fill(x)
try:
yMin = min(x[0], yMin)
except:
pass
# Save maximum
yMax = max(h.GetMaximum(), yMax)
# Customise & append to list
plot.ApplyStyle(h, i + 1)
hList.append(h)
if yMin <= 0.0:
yMin = 100
canvas.SetLogy()
# For-loop: All histograms
for i, h in enumerate(hList, 0):
h.SetMinimum(yMin * 0.85)
h.SetMaximum(yMax * 1.15)
h.GetXaxis().SetTitle("DNN output")
if "output" in saveName.lower():
h.GetYaxis().SetTitle("Entries")
elif "efficiency" in saveName.lower():
h.GetYaxis().SetTitle("Efficiency")
elif "significance" in saveName.lower():
h.GetYaxis().SetTitle("Significance")
else:
pass
if i == 0:
h.Draw("HIST")
else:
h.Draw("HIST SAME")
# Create legend
leg = plot.CreateLegend(0.6, 0.75, 0.9, 0.85)
for h in hList:
leg.AddEntry(h, h.GetName(), "l")
leg.Draw()
plot.SavePlot(canvas, saveDir, saveName, saveFormats)
canvas.Close()
# Create TGraph
for h in hList:
gList.append(convertHistoToGaph(h))
# Write the Tgraph into the JSON file
for gr in gList:
gName = "%s_%s" % (saveName, gr.GetName())
jsonWr.addGraph(gName, gr)
return