def plot_features(self):
"""Plot the features"""
self.msg_svc.INFO(
"DL : Plotting features comparing top quarks and anti-quarks. ")
top = self.df.loc[self.df['target'] == self.metadata['t_target']]
tbar = self.df.loc[self.df['target'] == self.metadata['tbar_target']]
filename = self.metadata['file'].split('/')[-1].split('.')[0].rstrip(
'\n')
processed_features = []
for hi, feature in enumerate(self.features2plot):
eventlevel = False
if feature.startswith(
't_'): # specific top properties in dataframe
feature = feature[2:]
eventlevel = True
elif feature.startswith(
'tbar_'): # specific tbar properties in dataframe
feature = feature[5:]
eventlevel = True
else:
eventlevel = False # single object in dataframe (use 'target' to distinguish)
if feature in processed_features: continue
else: processed_features.append(feature)
if 'btag' in feature:
x_label = self.text_dicts[feature]['label'].format(
self.metadata['btag_wkpt'])
else:
x_label = self.text_dicts[feature]['label']
hist = HepPlotter("histogram", 1)
hist.ratio_plot = False
hist.binning = self.text_dicts[feature]['bins']
hist.stacked = False
hist.logplot = False
hist.x_label = x_label
hist.y_label = "Events"
hist.format = self.image_format
hist.saveAs = self.output + "/hist_" + feature + "_" + self.date
hist.ATLASlabel = 'top left'
hist.ATLASlabelStatus = 'Simulation Internal'
hist.numLegendColumns = 1
hist.extra_text.Add(self.processlabel_args[filename]['label'],
coords=[0.03, 0.80])
hist.initialize()
multiply = 1.
if feature.endswith('_m_ttbar') or feature == 'pt':
multiply = 1e-3
if eventlevel:
hist.Add(self.df['t_' + feature],
name=feature + '_top',
linecolor='r',
color='r',
draw='step',
label='Large-R Jet (top)')
hist.Add(self.df['tbar_' + feature],
name=feature + '_tbar',
linecolor='b',
color='b',
draw='step',
label='Large-R Jet (anti-top)')
else:
hist.Add(top[feature].multiply(multiply),
name=feature + '_top',
linecolor='r',
color='r',
draw='step',
label='Large-R Jet (top)')
hist.Add(tbar[feature].multiply(multiply),
name=feature + '_tbar',
linecolor='b',
color='b',
draw='step',
label='Large-R Jet (anti-top)')
p = hist.execute()
hist.savefig()
## Correlation Matrices of Features (top/antitop) ##
corrmat_df_top = top[self.features].corr()
corrmat_df_tbar = tbar[self.features].corr()
names = ["top", "tbar"]
namelabels = [r"t correlations", r"$\bar{\text{t}}$ correlations"]
fontProperties = {'family': 'sans-serif'}
opts = {'cmap': plt.get_cmap("bwr"), 'vmin': -1, 'vmax': +1}
for c, corrmat in enumerate([corrmat_df_top, corrmat_df_tbar]):
fig, ax = plt.subplots()
# hide the upper part of the triangle
#mask = np.zeros_like(corrmat, dtype=np.bool) # return array of zeros with same shape as corrmat
#mask[np.tril_indices_from(mask)] = True
#corrmat_mask = np.ma.array(corrmat, mask=mask)
heatmap1 = ax.pcolor(corrmat, **opts)
cbar = plt.colorbar(heatmap1, ax=ax)
cbar.ax.set_yticklabels(
[i.get_text().strip('$') for i in cbar.ax.get_yticklabels()],
**fontProperties)
labels = corrmat.columns.values
labels = [i.replace('_', '\_') for i in labels]
# shift location of ticks to center of the bins
ax.set_xticks(np.arange(len(labels)) + 0.5, minor=False)
ax.set_yticks(np.arange(len(labels)) + 0.5, minor=False)
ax.set_xticklabels(labels,
fontProperties,
fontsize=18,
minor=False,
ha='right',
rotation=70)
ax.set_yticklabels(labels,
fontProperties,
fontsize=18,
minor=False)
text_args = {
'fontsize': 16,
'ha': 'left',
'va': 'bottom',
'transform': ax.transAxes
}
## ATLAS Label + Signal name
ax.text(0.02, 1.00, r"\textbf{\textit{ATLAS}} Simulation Internal",
**text_args)
ax.text(
0.03, 0.93,
"{0}, {1}".format(self.processlabel_args[filename]['label'],
namelabels[c]), **text_args)
## Energy Label
text_args['ha'] = 'right'
ax.text(0.99, 1.00, r"$\sqrt{\text{s}}$ = 13 TeV", **text_args)
plt.savefig(self.output + "/correlations_{0}_{1}.{2}".format(
names[c], self.date, self.image_format),
format=self.image_format,
dpi=300,
bbox_inches='tight')
plt.close()
return
def plot_score(self):
"""Plot the features"""
betterColors = hpt.betterColors()['linecolors']
filename = self.metadata['file'].split('/')[-1].split('.')[0].rstrip(
'\n')
# Plot all k-fold cross-validation results
for i, (train_X, train_Y, test_X, test_Y) in enumerate(
zip(self.train_data['X'], self.train_data['Y'],
self.test_data['X'], self.test_data['Y'])):
hist = HepPlotter("histogram", 1)
hist.ratio_plot = True
hist.y_ratio_label = "Test/Train"
hist.normed = True
hist.binning = [0.05 * j for j in range(21)]
hist.stacked = False
hist.logplot = False
hist.x_label = "DNN Score"
hist.y_label = "Events"
hist.format = self.image_format
hist.label_size = 14
hist.saveAs = self.output + "/hist_DNNscore_kfold{0}_{1}".format(
i, self.date)
hist.ATLASlabel = 'top left'
hist.ATLASlabelStatus = 'Simulation Internal'
hist.numLegendColumns = 1
hist.extra_text.Add(self.processlabel_args[filename]['label'],
coords=[0.03, 0.80],
fontsize=14)
hist.initialize()
top_train_scores = self.train_scores[i][train_Y == 1]
tbar_train_scores = self.train_scores[i][train_Y == 0]
top_test_scores = self.test_scores[i][test_Y == 1]
tbar_test_scores = self.test_scores[i][test_Y == 0]
## Train
index = i * 2
top_color = 'r' #betterColors[index]
tbar_color = 'b' #betterColors[index+1]
hist.Add(top_train_scores,
name='score_top_train_' + str(i),
linecolor=top_color,
color=top_color,
linewidth=2,
draw='step',
label='Large-R Jet (top) Train ' + str(i),
ratio_den=True,
ratio_num=False,
ratio_partner='score_top_test_' + str(i))
hist.Add(tbar_train_scores,
name='score_tbar_train_' + str(i),
linecolor=tbar_color,
color=tbar_color,
linewidth=2,
draw='step',
label='Large-R Jet (anti-top) Train ' + str(i),
ratio_den=True,
ratio_num=False,
ratio_partner='score_tbar_test_' + str(i))
## Test
hist.Add(top_test_scores,
name='score_top_test_' + str(i),
linecolor=top_color,
color=top_color,
draw='stepfilled',
label='Large-R Jet (top) Test ' + str(i),
alpha=0.5,
linewidth=0,
ratio_num=True,
ratio_den=False,
ratio_partner='score_top_train_' + str(i))
hist.Add(tbar_test_scores,
name='score_tbar_test_' + str(i),
linecolor=tbar_color,
color=tbar_color,
draw='stepfilled',
label='Large-R Jet (anti-top) Test ' + str(i),
alpha=0.5,
linewidth=0,
ratio_num=True,
ratio_den=False,
ratio_partner='score_tbar_train_' + str(i))
p = hist.execute()
hist.savefig()
## Calculation of the rejection
## use percentile (set above) to calculate at specific efficiency
eff_value = np.percentile(top_test_scores, self.percentile)
tbar_wrong = tbar_test_scores[tbar_test_scores >= eff_value]
rejection = len(tbar_wrong) * 1.0 / len(tbar_test_scores)
self.rejections.append(rejection)
self.rejection = {
'mean': np.mean(self.rejections),
'std': np.std(self.rejections)
}
return
def drawSyst(self, name=[], symmetrized=None, one_sided=False):
"""
Draw single systematic with nominal
@param name name(s) for histogram
@param symmetrized Values from symmetrized uncertainties
@param one_sided Boolean for one sided systematic or not
"""
if name[0].endswith("up"):
systname = name[0][:-2].replace("_", "-")
elif name[0].endswith("down"):
systname = name[0][:-4].replace("_", "-")
systname = systname.split("xleptonicT-mmerged-boostedcomb-")[1]
h_nominal = self.systData['nominal'][
'data'] # data (histogram bins values)
b_nominal = self.systData['nominal'][
'center'] # dummy values to get binning right
hist = HepPlotter("histogram", 1)
hist.ratio_plot = True # plot a ratio of things [Data/MC]
hist.ratio_type = "ratio" # "ratio"
hist.stacked = False # stack plots
hist.rebin = self.rebin
hist.logplot = False # plot on log scale
hist.x_label = self.x_labels[self.variable]['label']
hist.y_label = "Events"
hist.extra_text = systname + '\n ' + self.sampleName
hist.binning = self.systData['nominal']['bins']
hist.numLegendColumns = 1
hist.y_ratio_label = "Syst/Nom"
hist.lumi = '14.7' # in /fb
hist.format = 'png' # file format for saving image
hist.saveAs = self.outpath + "h_syst_" + self.sampleName + "_" + systname # save figure with unique name
hist.CMSlabel = 'top left' # 'top left', 'top right'; hack code for something else
hist.CMSlabelStatus = 'Simulation Internal' # ('Simulation')+'Internal' || 'Preliminary'
hist.initialize()
## Regular uncertainties
up = self.systData[name[0]]['center']
upData = self.systData[name[0]]['data']
hist.Add(up,
weights=upData,
name=systname + " UP",
label="UP",
linecolor='r',
color='r',
linestyle='dotted',
draw='step',
ratio_num=True,
ratio_den=False,
ratio_partner="nominal")
if not one_sided:
down = self.systData[name[1]]['center']
downData = self.systData[name[1]]['data']
hist.Add(down,
weights=downData,
name=systname + " DOWN",
label="DOWN",
linecolor='b',
color='b',
linestyle='dotted',
draw='step',
ratio_num=True,
ratio_den=False,
ratio_partner="nominal")
## Symmetrized uncertainties
if symmetrized is not None:
# - same binning as 'up' systematic
hist.Add(up,
weights=h_nominal + symmetrized,
name=systname + " UP Symm.",
label="UP Symm.",
linecolor='r',
linestyle='solid',
draw='step',
color='r',
ratio_num=True,
ratio_den=False,
ratio_partner="nominal")
hist.Add(up,
weights=h_nominal - symmetrized,
name=systname + " DOWN Symm.",
label="DOWN Symm.",
linecolor='b',
linestyle='solid',
draw='step',
color='b',
ratio_num=True,
ratio_den=False,
ratio_partner="nominal")
## nominal
uncertainty_hists = [
systname + " UP", systname + " DOWN", systname + " UP Symm.",
systname + " DOWN Symm."
]
hist.Add(b_nominal,
weights=h_nominal,
name="nominal",
label="nominal",
linecolor='k',
draw='step',
linestyle='solid',
ratio_num=False,
ratio_den=True,
ratio_partner=uncertainty_hists)
p = hist.execute() # can do something with p, if needed
hist.savefig() # save and close the figure
return