def plot_flatness_by_particle(labels,
predictions_dict,
spectator,
spectator_name,
predictions_dict_comparison=None,
names_algorithms=['MVA', 'Baseline'],
weights=None,
bins_number=30,
ignored_sideband=0.1,
thresholds=None,
cuts_values=False,
ncol=1):
plt.figure(figsize=(22, 20))
for n, (name, label) in enumerate(names_labels_correspondence.items()):
plt.subplot(3, 2, n + 1)
mask = labels == label
legends = []
for preds, name_algo in zip(
[predictions_dict, predictions_dict_comparison], names_algorithms):
if preds is None:
continue
probs = preds[label][mask]
if cuts_values:
thresholds_values = cut_values
else:
thresholds_values = [
weighted_quantile(probs,
quantiles=1 - eff / 100.,
sample_weight=None
if weights is None else weights[mask])
for eff in thresholds
]
eff = get_efficiencies(
probs,
spectator[mask],
sample_weight=None if weights is None else weights[mask],
bins_number=bins_number,
errors=True,
ignored_sideband=ignored_sideband,
thresholds=thresholds_values)
for thr in thresholds_values:
eff[thr] = (eff[thr][0], 100 * numpy.array(eff[thr][1]),
100 * numpy.array(eff[thr][2]), eff[thr][3])
plot_fig = ErrorPlot(eff)
plot_fig.xlabel = '{} {}'.format(name, spectator_name)
plot_fig.ylabel = 'Efficiency'
plot_fig.title = name
plot_fig.ylim = (0, 100)
plot_fig.plot(fontsize=22)
plt.xticks(fontsize=12), plt.yticks(fontsize=12)
legends.append(
['{} Eff {}%'.format(thr, name_algo) for thr in thresholds])
plt.legend(numpy.concatenate(legends),
loc='best',
fontsize=12,
framealpha=0.5,
ncol=ncol)
def plot_flatness_particle(labels, predictions_dict, spectator, spectator_name, particle_name,
weights=None, bins_number=30, ignored_sideband=0.1,
thresholds=None, cuts_values=False):
plt.figure(figsize=(18, 22))
for n, (name, label) in enumerate(names_labels_correspondence.items()):
plt.subplot(3, 2, n + 1)
mask = labels == names_labels_correspondence[particle_name]
probs = predictions_dict[label][mask]
mask_signal = labels == label
probs_signal = predictions_dict[label][mask_signal]
if cuts_values:
thresholds_values = cut_values
else:
thresholds_values = [weighted_quantile(probs_signal, quantiles=1 - eff / 100.,
sample_weight=None if weights is None else weights[mask_signal])
for eff in thresholds]
eff = get_efficiencies(probs, spectator[mask],
sample_weight=None if weights is None else weights[mask],
bins_number=bins_number, errors=True, ignored_sideband=ignored_sideband,
thresholds=thresholds_values)
for thr in thresholds_values:
eff[thr] = (eff[thr][0], 100*numpy.array(eff[thr][1]), 100*numpy.array(eff[thr][2]), eff[thr][3])
plot_fig = ErrorPlot(eff)
plot_fig.xlabel = '{} {}'.format(particle_name, spectator_name)
plot_fig.ylabel = 'Efficiency'
plot_fig.title = 'MVA {}'.format(name)
plot_fig.ylim = (0, 100)
plot_fig.plot(fontsize=22)
plt.xticks(fontsize=12), plt.yticks(fontsize=12)
if not cuts_values:
plt.legend(['Signal Eff {}%'.format(thr) for thr in thresholds], loc='best', fontsize=18, framealpha=0.5)
def plot_flatness_by_particle(labels, predictions_dict, spectator, spectator_name, predictions_dict_comparison=None,
names_algorithms=['MVA', 'Baseline'],
weights=None, bins_number=30, ignored_sideband=0.1,
thresholds=None, cuts_values=False, ncol=1):
plt.figure(figsize=(22, 20))
for n, (name, label) in enumerate(names_labels_correspondence.items()):
plt.subplot(3, 2, n + 1)
mask =labels == label
legends = []
for preds, name_algo in zip([predictions_dict, predictions_dict_comparison], names_algorithms):
if preds is None:
continue
probs = preds[label][mask]
if cuts_values:
thresholds_values = cut_values
else:
thresholds_values = [weighted_quantile(probs, quantiles=1 - eff / 100.,
sample_weight=None if weights is None else weights[mask])
for eff in thresholds]
eff = get_efficiencies(probs, spectator[mask],
sample_weight=None if weights is None else weights[mask],
bins_number=bins_number, errors=True, ignored_sideband=ignored_sideband,
thresholds=thresholds_values)
for thr in thresholds_values:
eff[thr] = (eff[thr][0], 100*numpy.array(eff[thr][1]), 100*numpy.array(eff[thr][2]), eff[thr][3])
plot_fig = ErrorPlot(eff)
plot_fig.xlabel = '{} {}'.format(name, spectator_name)
plot_fig.ylabel = 'Efficiency'
plot_fig.title = name
plot_fig.ylim = (0, 100)
plot_fig.plot(fontsize=22)
plt.xticks(fontsize=12), plt.yticks(fontsize=12)
legends.append(['{} Eff {}%'.format(thr, name_algo) for thr in thresholds])
plt.legend(numpy.concatenate(legends), loc='best', fontsize=12, framealpha=0.5, ncol=ncol)
def plot_flatness_particle(labels,
predictions_dict,
spectator,
spectator_name,
particle_name,
weights=None,
bins_number=30,
ignored_sideband=0.1,
thresholds=None,
cuts_values=False):
plt.figure(figsize=(18, 22))
for n, (name, label) in enumerate(names_labels_correspondence.items()):
plt.subplot(3, 2, n + 1)
mask = labels == names_labels_correspondence[particle_name]
probs = predictions_dict[label][mask]
mask_signal = labels == label
probs_signal = predictions_dict[label][mask_signal]
if cuts_values:
thresholds_values = cut_values
else:
thresholds_values = [
weighted_quantile(probs_signal,
quantiles=1 - eff / 100.,
sample_weight=None
if weights is None else weights[mask_signal])
for eff in thresholds
]
eff = get_efficiencies(
probs,
spectator[mask],
sample_weight=None if weights is None else weights[mask],
bins_number=bins_number,
errors=True,
ignored_sideband=ignored_sideband,
thresholds=thresholds_values)
for thr in thresholds_values:
eff[thr] = (eff[thr][0], 100 * numpy.array(eff[thr][1]),
100 * numpy.array(eff[thr][2]), eff[thr][3])
plot_fig = ErrorPlot(eff)
plot_fig.xlabel = '{} {}'.format(particle_name, spectator_name)
plot_fig.ylabel = 'Efficiency'
plot_fig.title = 'MVA {}'.format(name)
plot_fig.ylim = (0, 100)
plot_fig.plot(fontsize=22)
plt.xticks(fontsize=12), plt.yticks(fontsize=12)
if not cuts_values:
plt.legend(['Signal Eff {}%'.format(thr) for thr in thresholds],
loc='best',
fontsize=18,
framealpha=0.5)
def flatness_eta_figure(proba,
proba_baseline,
eta,
track_name,
particle_name,
save_path=None,
show=False):
"""
Plot signal efficiency vs pseudo rapidity figure.
Parameters
----------
proba : array_like
Predicted probabilities with array shape = [n_samples].
probas_baseline : array_like
Baseline predicted probabilities with array shape = [n_samples].
eta : array_like
Pseudo rapidity values with array shape = [n_samples].
track_name : string
The track name.
particle_name : string
The particle name.
save_path : string
Path to a directory where the figure will saved. If None the figure will not be saved.
show : boolean
If true the figure will be displayed.
"""
thresholds = numpy.percentile(proba, 100 - numpy.array([20, 50, 80]))
thresholds_baseline = numpy.percentile(proba_baseline,
100 - numpy.array([20, 50, 80]))
eff = get_efficiencies(proba,
eta,
bins_number=30,
errors=True,
ignored_sideband=0.005,
thresholds=thresholds)
eff_baseline = get_efficiencies(proba_baseline,
eta,
bins_number=30,
errors=True,
ignored_sideband=0.005,
thresholds=thresholds_baseline)
for i in thresholds:
eff[i] = (eff[i][0], 100. * eff[i][1], 100. * eff[i][2], eff[i][3])
for i in thresholds_baseline:
eff_baseline[i] = (eff_baseline[i][0], 100. * eff_baseline[i][1],
100. * eff_baseline[i][2], eff_baseline[i][3])
eff_total = OrderedDict()
num = len(eff) + len(eff_baseline)
for i in range(len(eff)):
v = eff[eff.keys()[i]]
v_baseline = eff_baseline[eff_baseline.keys()[i]]
eff_total[num] = v
eff_total[num - 1] = v_baseline
num += -2
plot_fig = ErrorPlot(eff_total)
plot_fig.ylim = (0, 100)
plot_fig.plot(new_plot=True, figsize=(10, 7))
labels = [
'Eff model = 20 %', 'Eff baseline = 20 %', 'Eff model = 50 %',
'Eff baseline = 50 %', 'Eff model = 80 %', 'Eff baseline = 80 %'
]
plt.legend(labels, loc='best', prop={'size': 10}, framealpha=0.5, ncol=3)
plt.xlabel(track_name + ' ' + particle_name + ' Pseudo Rapidity', size=15)
plt.xticks(size=15)
plt.ylabel('Efficiency / %', size=15)
plt.yticks(size=15)
plt.title('Flatness_SignalMVAEffVPseudoRapidity_' + track_name + ' ' +
particle_name,
size=15)
if save_path != None:
plt.savefig(save_path + "/" + 'Flatness_SignalMVAEffVPseudoRapidity_' +
track_name + '_' + particle_name + ".png")
if show == True:
plt.show()
plt.clf()
plt.close()
def flatness_eta_figure(proba, proba_baseline, eta, track_name, particle_name, save_path=None, show=False):
"""
Plot signal efficiency vs pseudo rapidity figure.
Parameters
----------
proba : array_like
Predicted probabilities with array shape = [n_samples].
probas_baseline : array_like
Baseline predicted probabilities with array shape = [n_samples].
eta : array_like
Pseudo rapidity values with array shape = [n_samples].
track_name : string
The track name.
particle_name : string
The particle name.
save_path : string
Path to a directory where the figure will saved. If None the figure will not be saved.
show : boolean
If true the figure will be displayed.
"""
thresholds = numpy.percentile(proba, 100 - numpy.array([20, 50, 80]))
thresholds_baseline = numpy.percentile(proba_baseline, 100 - numpy.array([20, 50, 80]))
eff = get_efficiencies(proba,
eta,
bins_number=30,
errors=True,
ignored_sideband=0.005,
thresholds=thresholds)
eff_baseline = get_efficiencies(proba_baseline,
eta,
bins_number=30,
errors=True,
ignored_sideband=0.005,
thresholds=thresholds_baseline)
for i in thresholds:
eff[i] = (eff[i][0], 100. * eff[i][1], 100. * eff[i][2], eff[i][3])
for i in thresholds_baseline:
eff_baseline[i] = (eff_baseline[i][0], 100. * eff_baseline[i][1], 100. * eff_baseline[i][2], eff_baseline[i][3])
eff_total = OrderedDict()
num = len(eff) + len(eff_baseline)
for i in range(len(eff)):
v = eff[eff.keys()[i]]
v_baseline = eff_baseline[eff_baseline.keys()[i]]
eff_total[num] = v
eff_total[num - 1] = v_baseline
num += -2
plot_fig = ErrorPlot(eff_total)
plot_fig.ylim = (0, 100)
plot_fig.plot(new_plot=True, figsize=(10,7))
labels = ['Eff model = 20 %', 'Eff baseline = 20 %',
'Eff model = 50 %', 'Eff baseline = 50 %',
'Eff model = 80 %', 'Eff baseline = 80 %']
plt.legend(labels, loc='best',prop={'size':10}, framealpha=0.5, ncol=3)
plt.xlabel(track_name + ' ' + particle_name + ' Pseudo Rapidity', size=15)
plt.xticks(size=15)
plt.ylabel('Efficiency / %', size=15)
plt.yticks(size=15)
plt.title('Flatness_SignalMVAEffVPseudoRapidity_' + track_name + ' ' + particle_name, size=15)
if save_path != None:
plt.savefig(save_path + "/" + 'Flatness_SignalMVAEffVPseudoRapidity_' + track_name + '_' + particle_name + ".png")
if show == True:
plt.show()
plt.clf()
plt.close()