def test_slice_assign():
hist = Hist(10, 0, 1)
hist[:] = [i for i in range(len(hist))]
assert all([a.value == b for a, b in zip(hist, range(len(hist)))])
clone = hist.Clone()
# reverse bins
hist[:] = clone[::-1]
assert all([a.value == b.value for a, b in zip(hist, clone[::-1])])
def test_ravel():
hist = Hist2D(3, 0, 1, 4, 0, 1)
for i, bin in enumerate(hist.bins()):
bin.value = i
bin.error = i
rhist = hist.ravel()
assert_equal(list(rhist.y()), list(range(12)))
assert_equal(list(rhist.yerrh()), list(range(12)))
def test_overall_efficiency():
for stat_op in range(0, 8):
Eff = Efficiency(Hist(20, -3, 3), Hist(20, -3, 3))
Eff_1bin = Efficiency(Hist(1, -3, 3), Hist(1, -3, 3))
Eff.SetStatisticOption(stat_op)
Eff_1bin.SetStatisticOption(stat_op)
for i in range(1000):
x = gauss(0, 3.6)
w = uniform(0, 1)
passed = w > 0.5
Eff.Fill(passed, x)
Eff_1bin.Fill(passed, x)
assert_almost_equal(Eff.overall_efficiency(overflow=True)[0],
Eff_1bin.overall_efficiency(overflow=True)[0])
assert_almost_equal(Eff.overall_efficiency(overflow=True)[1],
Eff_1bin.overall_efficiency(overflow=True)[1])
assert_almost_equal(Eff.overall_efficiency(overflow=True)[2],
Eff_1bin.overall_efficiency(overflow=True)[2])
def test_edgesl():
h = Hist([1, 2, 3, 4])
assert_equal(list(h.xedgesl()), [1, 2, 3])
assert_equal(list(h.xedgesl(overflow=True)), [float('-inf'), 1, 2, 3, 4])
assert_equal(h.xedgesl(0), float('-inf'))
assert_equal(h.xedgesl(-1), 4)
assert_equal(h.xedgesl(4), 4)
# wrap around
assert_equal(h.xedgesl(5), float('-inf'))
for i in range(1, h.nbins()):
assert_equal(h.xedgesl(i), i)
def test_edgesh():
h = Hist([1, 2, 3, 4])
assert_equal(list(h.xedgesh()), [2, 3, 4])
assert_equal(list(h.xedgesh(overflow=True)), [1, 2, 3, 4, float('inf')])
assert_equal(h.xedgesh(0), 1)
assert_equal(h.xedgesh(-1), float('inf'))
assert_equal(h.xedgesh(4), float('inf'))
# wrap around
assert_equal(h.xedgesh(5), 1)
for i in range(1, h.nbins()):
assert_equal(h.xedgesh(i), i + 1)
def test_edges():
h = Hist([1, 2, 3, 4])
assert_equal(list(h.xedges()), [1, 2, 3, 4])
assert_equal(list(h.xedges(overflow=True)),
[float('-inf'), 1, 2, 3, 4, float('inf')])
assert_equal(h.xedges(0), float('-inf'))
assert_equal(h.xedges(-1), float('inf'))
assert_equal(h.xedges(5), float('inf'))
# wrap around
assert_equal(h.xedges(6), float('-inf'))
for i in range(1, h.nbins() + 1):
assert_equal(h.xedges(i), i)
def test_edgesl():
h = Hist([1, 2, 3, 4])
assert_equal(list(h.xedgesl()), [1, 2, 3])
assert_equal(list(h.xedgesl(overflow=True)),
[float('-inf'), 1, 2, 3, 4])
assert_equal(h.xedgesl(0), float('-inf'))
assert_equal(h.xedgesl(-1), 4)
assert_equal(h.xedgesl(4), 4)
# wrap around
assert_equal(h.xedgesl(5), float('-inf'))
for i in range(1, h.nbins()):
assert_equal(h.xedgesl(i), i)
def test_edgesh():
h = Hist([1, 2, 3, 4])
assert_equal(list(h.xedgesh()), [2, 3, 4])
assert_equal(list(h.xedgesh(overflow=True)),
[1, 2, 3, 4, float('inf')])
assert_equal(h.xedgesh(0), 1)
assert_equal(h.xedgesh(-1), float('inf'))
assert_equal(h.xedgesh(4), float('inf'))
# wrap around
assert_equal(h.xedgesh(5), 1)
for i in range(1, h.nbins()):
assert_equal(h.xedgesh(i), i + 1)
def test_edges():
h = Hist([1, 2, 3, 4])
assert_equal(list(h.xedges()), [1, 2, 3, 4])
assert_equal(
list(h.xedges(overflow=True)),
[float('-inf'), 1, 2, 3, 4, float('inf')])
assert_equal(h.xedges(0), float('-inf'))
assert_equal(h.xedges(-1), float('inf'))
assert_equal(h.xedges(5), float('inf'))
# wrap around
assert_equal(h.xedges(6), float('-inf'))
for i in range(1, h.nbins() + 1):
assert_equal(h.xedges(i), i)
def test_overall_efficiency():
for stat_op in range(0, 8):
Eff = Efficiency(Hist(20, -3, 3), Hist(20, -3, 3))
Eff_1bin = Efficiency(Hist(1, -3, 3), Hist(1, -3, 3))
Eff.SetStatisticOption(stat_op)
Eff_1bin.SetStatisticOption(stat_op)
for i in range(1000):
x = gauss(0, 3.6)
w = uniform(0, 1)
passed = w > 0.5
Eff.Fill(passed, x)
Eff_1bin.Fill(passed, x)
assert_almost_equal(
Eff.overall_efficiency(overflow=True)[0],
Eff_1bin.overall_efficiency(overflow=True)[0])
assert_almost_equal(
Eff.overall_efficiency(overflow=True)[1],
Eff_1bin.overall_efficiency(overflow=True)[1])
assert_almost_equal(
Eff.overall_efficiency(overflow=True)[2],
Eff_1bin.overall_efficiency(overflow=True)[2])
def get_weights(y_train, pt_train, options):
print 'getting weights...'
Nbins = 100
LV = rt.TH1D("LV", "LV", Nbins, options.train_pt_cut, 100.)
PU = rt.TH1D("PU", "PU", Nbins, options.train_pt_cut, 100.)
W = rt.TH1D("W", "W", Nbins, options.train_pt_cut, 100.)
y_train = y_train.values
pt_train = pt_train.values
for y in y_train:
if y == 1: LV.Fill(pt_train[i0])
else: PU.Fill(pt_train[i0])
LV.Scale(1. / LV.Integral())
PU.Scale(1. / PU.Integral())
plot_th1([LV, PU])
#make weight in each pT cat
for i0 in range(LV.GetNbinsX()):
if (PU.GetBinContent(i0 + 1) == 0.) or (LV.GetBinContent(i0 + 1)
== 0.):
W.SetBinContent(i0 + 1, 1)
else:
W.SetBinContent(
i0 + 1,
float(LV.GetBinContent(i0 + 1)) /
float(PU.GetBinContent(i0 + 1)))
#make sure weight isn't too big or too small
weightUpper = 5.0
weightLower = 0.2
if W.GetBinContent(i0 + 1) > weightUpper:
W.SetBinContent(i0 + 1) == weightUpper
if W.GetBinContent(i0 + 1) < weightLower:
W.SetBinContent(i0 + 1) == WeightLower
W.Print("all")
#apply weights on training PU samples
candWeights = np.zeros(len(y_train), dtype=float)
for y in y_train:
if y == 1: candWeights[i0] == 1. #don't weight LV
else:
candWeights[i0] = W.GetBinContent(W.FindBin(
pt_train[i0])) #weight PU
return candWeights
def get_data(self):
print 'Build dataset'
lFile = h5py.File(options.infile + '.h5', 'r')
lArr = np.array(lFile.get('Events')[:])
df = pd.DataFrame(data=lArr)
y = df['LV']
x = df.drop(columns=['ecalE', 'genE', 'LV', 'energy', 'pt'])
eta = df['eta']
pt = df['pt']
num_vars = len(x.columns)
msk = np.random.rand(len(df)) < 0.8
msk2 = np.zeros(len(df), dtype=bool)
for i0 in range(len(msk)):
if (msk[i0] == True) and (pt.iloc[i0] > options.train_pt_cut):
msk2[i0] = True
#print msk2[0:1000]
y_train, x_train = y[msk2], x[msk2]
y_test, x_test = y[~msk2], x[~msk2]
print x_test
#propagate eta, pt for ROC curves, weights
eta_train, pt_train = eta[msk2], pt[msk2]
eta_test, pt_test = eta[~msk2], pt[~msk2]
print 'Train on %i PF candidates (satisfying pT > %f) and use %i for validation.' % (
len(y_train), options.train_pt_cut, len(y_test))
return x_train, x_test, y_train, y_test, num_vars, eta_test, pt_test, eta_train, pt_train
# create a simple 1D histogram with 10 constant-width bins between 0 and 1
h_simple = Hist(10, 0, 1)
print(h_simple.name)
# If the name is not specified, a UUID is used so that ROOT never complains
# about two histograms having the same name.
# Alternatively you can specify the name (and the title or any other style
# attributes) in the constructor:
h_simple = Hist(10, -4, 12, name='my hist', title='Some Data',
drawstyle='hist',
legendstyle='F',
fillstyle='/')
# fill the histogram
for i in range(1000):
# all ROOT CamelCase methods are aliased by equivalent snake_case methods
# so you can call fill() instead of Fill()
h_simple.Fill(random.gauss(4, 3))
# easily set visual attributes
h_simple.linecolor = 'blue'
h_simple.fillcolor = 'green'
h_simple.fillstyle = '/'
# attributes may be accessed in the same way
print(h_simple.name)
print(h_simple.title)
print(h_simple.markersize)
# plot
def train(self):
x_train, x_test, y_train, y_test, num_vars, eta_test, pt_test, eta_train, pt_train = self.get_data(
)
Nbatch = 500
Nepoch = 10
model = self.build_model(num_vars)
model.summary()
scaler = StandardScaler().fit(x_train)
x_train = scaler.transform(x_train)
callbacks = all_callbacks(
stop_patience=1000,
lr_factor=0.5,
lr_patience=10,
lr_epsilon=0.000001,
lr_cooldown=2,
lr_minimum=0.0000001,
outputDir=
'/uscms_data/d3/jkrupa/pf_studies/CMSSW_10_5_0_pre2/src/CaloTrigNN/CaloNtupler/test/h5_files'
)
print 'Fit model...'
if options.use_weights:
weights = get_weights(y_train, pt_train, options)
history = model.fit(x_train,
y_train,
epochs=Nepoch,
batch_size=Nbatch,
callbacks=callbacks.callbacks,
validation_split=0.0,
sample_weight=weights)
else:
history = model.fit(x_train,
y_train,
epochs=Nepoch,
batch_size=Nbatch,
callbacks=callbacks.callbacks,
validation_split=0.0)
#https://hackernoon.com/simple-guide-on-how-to-generate-roc-plot-for-keras-classifier-2ecc6c73115a
y_pred = model.predict(x_test).ravel()
#inclusive
fpr, tpr, thresholds = roc_curve(y_test, y_pred)
#kinematic binning
if options.inc:
lpT = [1., 10000.0]
leta = [1.7, 3.0]
else:
lpT = [1., 5., 10., 20., 10000.0]
leta = [1.7, 2.0, 2.5, 3.0]
if options.makeroc:
for i0 in range(len(lpT) - 1):
for i1 in range(len(leta) - 1):
make_roc_curve(y_pred, y_test, eta_test, pt_test, lpT[i0],
lpT[i0 + 1], leta[i1], leta[i1 + 1],
options)
frozen_graph = freeze_session(
K.get_session(),
output_names=[out.op.name for out in model.outputs])
tf.train.write_graph(frozen_graph,
"h5_files",
"tf_model.pb",
as_text=False)
print_model_to_json(
model,
'/uscms_data/d3/jkrupa/pf_studies/CMSSW_10_5_0_pre2/src/CaloTrigNN/CaloNtupler/test/h5_files/model.json'
)
model.save_weights(
'/uscms_data/d3/jkrupa/pf_studies/CMSSW_10_5_0_pre2/src/CaloTrigNN/CaloNtupler/test/h5_files/dense_model_weights.h5'
)
json_string = model.to_json()
from rootpy.plotting import Hist, Canvas, Legend, set_style
from rootpy.plotting.contrib.quantiles import qqgraph
from rootpy.extern.six.moves import range
set_style('ATLAS')
c = Canvas(width=1200, height=600)
c.Divide(2, 1, 1e-3, 1e-3)
rand = ROOT.TRandom3()
h1 = Hist(100, -5, 5, name="h1", title="Histogram 1",
linecolor='red', legendstyle='l')
h2 = Hist(100, -5, 5, name="h2", title="Histogram 2",
linecolor='blue', legendstyle='l')
for ievt in range(10000):
h1.Fill(rand.Gaus(0, 0.8))
h2.Fill(rand.Gaus(0, 1))
pad = c.cd(1)
h1.Draw('hist')
h2.Draw('hist same')
leg = Legend([h1, h2], pad=pad, leftmargin=0.5,
topmargin=0.11, rightmargin=0.05,
textsize=20)
leg.Draw()
pad = c.cd(2)
def test_slice_assign_error():
hist = Hist(10, 0, 1)
hist[:] = [(i, i / 2.) for i in range(len(hist))]
assert all([a.value == b for a, b in zip(hist, range(len(hist)))])
assert all([a.error == b / 2. for a, b in zip(hist, range(len(hist)))])
def test_slice_assign_bad():
hist = Hist(10, 0, 1)
hist[:] = range(len(hist) + 1)
def test_slice_assign_bad():
hist = Hist(10, 0, 1)
hist[:] = range(len(hist) + 1)
def load_analysis(self, inputs, res_T):
cHW = res_T[inputs.split('/')[-1]][0]
tcHW = res_T[inputs.split('/')[-1]][1]
xsec = res_T[inputs.split('/')[-1]][2]
# Create chain of root trees
chain1 = ROOT.TChain("Delphes")
chain1.Add(inputs)
# Create object of class ExRootTreeReader
treeReader = ROOT.ExRootTreeReader(chain1)
numberOfEntries = treeReader.GetEntries()
# create new root file
root_name = 'cHW_{}_tcHW_{}.root'.format(cHW, tcHW)
csv_name = 'cHW_{}_tcHW_{}.csv'.format(cHW, tcHW)
f = root_open(root_name, "recreate")
tree = Tree("cHW_{}_tcHW_{}".format(cHW, tcHW))
tree.create_branches({
'PT_l1': 'F',
'PT_l2': 'F',
'PT_ll': 'F',
'Cos_lZ': 'F',
'DPHI_ll': 'F',
'PT_j1': 'F',
'PT_j2': 'F',
'PT_b1': 'F',
'PT_b2': 'F',
'Eta_H': 'F',
'phi_H': 'F',
'M_H': 'F',
'Cos_Hb1': 'F',
'PT_H': 'F',
'PT_ZH': 'F',
'M_Z': 'F',
'M_ZH': 'F',
'cHW': 'F',
'tcHW': 'F',
'xsec': 'F'
})
# Get pointers to branches used in this analysis
branchJet = treeReader.UseBranch("Jet")
branchElectron = treeReader.UseBranch("Electron")
branchMuon = treeReader.UseBranch("Muon")
branchPhoton = treeReader.UseBranch("Photon")
branchMET = treeReader.UseBranch("MissingET")
# Loop over all events
for entry in range(0, numberOfEntries):
# Load selected branches with data from specified event
treeReader.ReadEntry(entry)
muons = []
for n in xrange(branchMuon.GetEntries()):
muons.append(branchMuon.At(n))
if len(muons) >= 2:
muons = sorted(branchMuon,
key=lambda Muon: Muon.P4().Pt(),
reverse=True)
else:
continue
missing = sorted(branchMET,
key=lambda MisingET: MisingET.MET,
reverse=True)
muon1 = muons[0]
muon2 = muons[1]
Muon1 = ROOT.TLorentzVector()
Muon2 = ROOT.TLorentzVector()
Muon1.SetPtEtaPhiE(muon1.P4().Pt(),
muon1.P4().Eta(),
muon1.P4().Phi(),
muon1.P4().E())
Muon2.SetPtEtaPhiE(muon2.P4().Pt(),
muon2.P4().Eta(),
muon2.P4().Phi(),
muon2.P4().E())
met = ROOT.TLorentzVector()
met.SetPtEtaPhiE(missing[0].P4().Pt(), missing[0].P4().Eta(),
missing[0].P4().Phi(), missing[0].P4().E())
bjato1 = ROOT.TLorentzVector()
bjato2 = ROOT.TLorentzVector()
jato1 = ROOT.TLorentzVector()
jato2 = ROOT.TLorentzVector()
####################################################################################
bjets, ljets = [], []
for n in xrange(branchJet.GetEntries()):
if branchJet.At(n).BTag == 1:
bjets.append(branchJet.At(n))
else:
ljets.append(branchJet.At(n))
if len(bjets) >= 2:
bjets = sorted(bjets,
key=lambda BJet: BJet.P4().Pt(),
reverse=True)
else:
continue
ljets = sorted(ljets, key=lambda Jet: Jet.P4().Pt(), reverse=True)
try:
jato1.SetPtEtaPhiE(ljets[0].P4().Pt(), ljets[0].P4().Eta(),
ljets[0].P4().Phi(), ljets[0].P4().E())
except IndexError:
tree.PT_j1 = -999
try:
jato2.SetPtEtaPhiE(ljets[1].P4().Pt(), ljets[1].P4().Eta(),
ljets[1].P4().Phi(), ljets[1].P4().E())
except IndexError:
tree.PT_j2 = -999
####################################################################################
bjato1.SetPtEtaPhiE(bjets[0].P4().Pt(), bjets[0].P4().Eta(),
bjets[0].P4().Phi(), bjets[0].P4().E())
bjato2.SetPtEtaPhiE(bjets[1].P4().Pt(), bjets[1].P4().Eta(),
bjets[1].P4().Phi(), bjets[1].P4().E())
###################################################################################################
if 95 < (bjato1 + bjato2).M() < 135:
tree.PT_l1 = Muon1.Pt()
tree.PT_l2 = Muon2.Pt()
tree.PT_ll = (Muon1 + Muon2).Pt()
tree.PT_b1 = bjato1.Pt()
tree.PT_b2 = bjato2.Pt()
tree.PT_j1 = jato1.Pt()
tree.PT_j2 = jato2.Pt()
Z = ROOT.TLorentzVector()
H = ROOT.TLorentzVector()
ZH = ROOT.TLorentzVector()
Z = (Muon1 + Muon2)
H = (bjato1 + bjato2)
ZH = Z + H
tree.phi_H = H.Phi()
tree.PT_ZH = ZH.Pt()
tree.M_ZH = ZH.M()
tree.PT_H = H.Pt()
tree.Eta_H = H.Eta()
tree.M_H = H.M()
tree.M_Z = Z.M()
tree.DPHI_ll = np.abs(Muon1.DeltaPhi(Muon2))
########################## boosted objects ############################################
Ztob = ROOT.TLorentzVector()
Ztob.SetPxPyPzE(Z.Px(), Z.Py(), Z.Pz(), Z.E())
Zboost = ROOT.TVector3()
Zboost = Ztob.BoostVector()
v = Zboost.Unit()
Muon1.Boost(-Zboost)
Htob = ROOT.TLorentzVector()
Htob.SetPxPyPzE(H.Px(), H.Py(), H.Pz(), H.E())
Hboost = ROOT.TVector3()
Hboost = Htob.BoostVector()
ang = Hboost.Unit()
bjato1.Boost(-Hboost)
tree.Cos_Hb1 = np.cos(bjato1.Angle(ang))
tree.Cos_lZ = np.cos(Muon1.Angle(v))
##########################################################################################
tree.cHW = cHW
tree.tcHW = tcHW
tree.xsec = xsec
tree.Fill()
tree.write()
f.close()
#create the csv output
to_convert = root2array(root_name, "cHW_{}_tcHW_{}".format(cHW, tcHW))
df_conv = pd.DataFrame(to_convert)
df_conv.to_csv(csv_name,
index=False,
header=df_conv.keys(),
mode='w',
sep=' ')
### move everything
if not os.path.exists('500GeV_res'):
os.makedirs('500GeV_res')
os.makedirs('500GeV_res/roots')
os.makedirs('500GeV_res/csv')
shutil.move(root_name, '500GeV_res/roots')
shutil.move(csv_name, '500GeV_res/csv')
from rootpy.extern.six.moves import range
try:
kwargs = {}
for arg in extra:
name, value = arg.lstrip('--').split('=')
kwargs[name] = value
except ValueError:
print("specify style parameters with --name=value")
try:
style = get_style(args.style, **kwargs)
except ValueError:
print('Invalid style: `{0}`. Using the `ATLAS` style.'.format(args.style))
style = get_style('ATLAS')
# Use styles as context managers. The selected style will only apply
# within the following context:
with style:
c = Canvas()
hpx = Hist(100, -4, 4, name="hpx", title="This is the px distribution")
# generate some random data
ROOT.gRandom.SetSeed()
for i in range(25000):
hpx.Fill(ROOT.gRandom.Gaus())
hpx.GetXaxis().SetTitle("random variable [unit]")
hpx.GetYaxis().SetTitle("#frac{dN}{dr} [unit^{-1}]")
hpx.SetMaximum(1000.)
hpx.Draw()
wait()
from random import gauss
from rootpy.io import root_open
from rootpy.tree import Tree, TreeChain
from rootpy.plotting import Hist
from rootpy.extern.six.moves import range
# Make two files, each with a Tree called "test"
print("Creating test tree in chaintest1.root")
f = root_open("chaintest1.root", "recreate")
tree = Tree("test")
branches = {'x': 'F', 'y': 'F', 'z': 'F', 'i': 'I'}
tree.create_branches(branches)
for i in range(10000):
tree.x = gauss(.5, 1.)
tree.y = gauss(.3, 2.)
tree.z = gauss(13., 42.)
tree.i = i
tree.fill()
# Make a histogram of x when y > 1
hist1 = Hist(100, -10, 10, name='hist1')
tree.Draw('x', 'y > 1', hist=hist1)
hist1.SetDirectory(0) # memory resident
print("The first tree has {0:f} entries where y > 1".format(hist1.Integral()))
tree.write()
f.close()
h1 = Hist(100,
-5,
5,
name="h1",
title="Histogram 1",
linecolor='red',
legendstyle='l')
h2 = Hist(100,
-5,
5,
name="h2",
title="Histogram 2",
linecolor='blue',
legendstyle='l')
for ievt in range(10000):
h1.Fill(rand.Gaus(0, 0.8))
h2.Fill(rand.Gaus(0, 1))
pad = c.cd(1)
h1.Draw('hist')
h2.Draw('hist same')
leg = Legend([h1, h2],
pad=pad,
leftmargin=0.5,
topmargin=0.11,
rightmargin=0.05,
textsize=20)
leg.Draw()
'M_H': 'F',
'MT_W': 'F',
'Cos_Hb1': 'F',
'PT_H': 'F',
})
# Get pointers to branches used in this analysis
branchJet = treeReader.UseBranch("Jet")
branchElectron = treeReader.UseBranch("Electron")
branchMuon = treeReader.UseBranch("Muon")
branchPhoton = treeReader.UseBranch("Photon")
branchMET = treeReader.UseBranch("MissingET")
####################################################################
# Loop over all events
for entry in range(0, numberOfEntries):
# Load selected branches with data from specified event
treeReader.ReadEntry(entry)
##########################################################################################################
eletrons = sorted(branchElectron, key=lambda Electron: Electron.PT, reverse=True)
missing = sorted(branchMET, key=lambda MisingET: MisingET.MET, reverse=True)
elec1 = eletrons[0]
eletron1 = ROOT.TLorentzVector()
eletron1.SetPtEtaPhiE(elec1.PT,elec1.Eta,elec1.Phi,elec1.P4().E())
met = ROOT.TLorentzVector()
met.SetPtEtaPhiE(missing[0].P4().Pt(),missing[0].P4().Eta(),missing[0].P4().Phi(),missing[0].P4().E())
bjato1 = ROOT.TLorentzVector()
bjato2 = ROOT.TLorentzVector()
jato1 = ROOT.TLorentzVector()
jato2 = ROOT.TLorentzVector()
####################################################################################
ROOT.gROOT.SetBatch()
set_style('ATLAS')
np.random.seed(0)
random.seed(0)
# create an example TTree dataset
class Sample(TreeModel):
x = FloatCol()
y = FloatCol()
with root_open('sample.root', 'recreate'):
# generate toy data in a TTree
tree = Tree('sample', model=Sample)
for i in range(1000):
tree.x = gauss(0, 1)
tree.y = gauss(0, 1)
tree.Fill()
tree.write()
# read in the TTree as a NumPy array
array = root2array('sample.root', 'sample')
if os.path.exists('bootstrap.gif'):
os.remove('bootstrap.gif')
canvas = Canvas(width=500, height=400)
hist = Hist2D(10, -3, 3, 10, -3, 3, drawstyle='LEGO2')
output = root_open('bootstrap.root', 'recreate')
