def make_plot(items, figfile = '',
xlabel = '', ylabel = '',
x_log = False, y_log = False,
labels = [], title = '',
):
plt.cla()
plt.clf()
fig = plt.figure()
fig.patch.set_facecolor('white')
for i, item in enumerate(items):
label = None
if len(labels) >= i:
label = labels[i]
plt.plot(item, label=label)
if x_log:
plt.xscale('log')
if y_log:
plt.yscale('log')
plt.title(title)
ampl.set_xlabel(xlabel)
ampl.set_ylabel(ylabel)
plt.legend()
if figfile != '':
plt.savefig(figfile)
plt.show()
def histogramOverlay(frames, data, labels, xlabel, ylabel, figfile = '',
x_min = 0, x_max = 2200, xbins = 22, normed = True, y_log = False,
atlas_x = -1, atlas_y = -1, simulation = False,
textlist = []):
xbin = np.arange(x_min, x_max, (x_max - x_min) / xbins)
plt.cla()
plt.clf()
fig = plt.figure()
fig.patch.set_facecolor('white')
zorder_start = -1 * len(data) # hack to get axes on top
for i, datum in enumerate(data):
plt.hist(frames[i][datum], bins = xbin, density = normed,
alpha = 0.5, label=labels[i], zorder=zorder_start + i)
plt.xlim(x_min, x_max)
if y_log:
plt.yscale('log')
ampl.set_xlabel(xlabel)
ampl.set_ylabel(ylabel)
if atlas_x >= 0 and atlas_y >= 0:
ampl.draw_atlas_label(atlas_x, atlas_y, simulation = simulation, fontsize = 18)
drawLabels(fig, atlas_x, atlas_y, simulation, textlist)
fig.axes[0].zorder = len(data)+1 #hack to keep the tick marks up
plt.legend()
if figfile != '':
plt.savefig(figfile)
plt.show()
def lineOverlay(xcenter,
lines,
labels,
xlabel,
ylabel,
figfile='',
x_min=0.1,
x_max=1000,
x_log=True,
y_min=0,
y_max=2,
y_log=False,
linestyles=[],
colorgrouping=-1,
extra_lines=[],
atlas_x=-1,
atlas_y=-1,
simulation=False,
textlist=[]):
plt.cla()
plt.clf()
fig = plt.figure()
fig.patch.set_facecolor('white')
for extra_line in extra_lines:
plt.plot(extra_line[0], extra_line[1], linestyle='--', color='black')
colors = plt.rcParams["axes.prop_cycle"].by_key()["color"]
for i, line in enumerate(lines):
if len(linestyles) > 0:
linestyle = linestyles[i]
else:
linestyle = 'solid'
if colorgrouping > 0:
color = colors[int(np.floor(i / colorgrouping))]
else:
color = colors[i]
plt.plot(xcenter,
line,
label=labels[i],
linestyle=linestyle,
color=color)
if x_log:
plt.xscale('log')
if y_log:
plt.yscale('log')
plt.xlim(x_min, x_max)
plt.ylim(y_min, y_max)
ampl.set_xlabel(xlabel)
ampl.set_ylabel(ylabel)
drawLabels(fig, atlas_x, atlas_y, simulation, textlist)
plt.legend()
if figfile != '':
plt.savefig(figfile)
plt.show()
def roc_plot(xlist, ylist, figfile = '',
xlabel='False positive rate',
ylabel='True positive rate',
x_min = 0, x_max = 1.1, x_log = False,
y_min = 0, y_max = 1.1, y_log = False,
linestyles=[], colorgrouping=-1,
extra_lines=[], labels=[],
atlas_x=-1, atlas_y=-1, simulation=False,
textlist=[], title=''):
plt.cla()
plt.clf()
fig = plt.figure()
fig.patch.set_facecolor('white')
for extra_line in extra_lines:
plt.plot(extra_line[0], extra_line[1], linestyle='--', color='black')
colors = plt.rcParams["axes.prop_cycle"].by_key()["color"]
for i, (x,y) in enumerate(zip(xlist,ylist)):
if len(linestyles) > 0:
linestyle = linestyles[i]
else:
linestyle = 'solid'
if colorgrouping > 0:
color = colors[int(np.floor(i / colorgrouping))]
else:
color = colors[i%(len(colors)-1)]
label = None
if len(labels) > 0:
label = labels[i]
plt.plot(x, y, label = label, linestyle=linestyle, color=color)
if x_log:
plt.xscale('log')
if y_log:
plt.yscale('log')
plt.title(title)
plt.xlim(x_min, x_max)
plt.ylim(y_min, y_max)
ampl.set_xlabel(xlabel)
ampl.set_ylabel(ylabel)
plt.legend()
if figfile != '':
plt.savefig(figfile)
plt.show()
def rocScan(varlist, scan_targets, labels, plotpath, ylabels, data):
'''
Creates a set of ROC curve plots by scanning over the specified variables.
One set is created for each target (neural net score dataset).
varlist: a list of rocVar instances to scan over
scan_targets: a list of neural net score datasets to use
labels: a list of target names (strings); must be the same length as scan_targets
'''
for target, target_label in zip(scan_targets, labels):
for v in varlist:
# prepare matplotlib figure
plt.cla()
plt.clf()
fig = plt.figure()
fig.patch.set_facecolor('white')
plt.plot([0, 1], [0, 1], 'k--')
for binning, label in zip(v.selections, v.labels):
# first generate ROC curve
x, y, t = roc_curve(
ylabels[data.test & binning][:, 1],
target[data.test & binning],
drop_intermediate=False,
)
var_auc = auc(x, y)
plt.plot(x,
y,
label=label + ' (area = {:.3f})'.format(var_auc))
plt.title('ROC Scan of ' + target_label + ' over ' + v.latex)
plt.xlim(0, 1.1)
plt.ylim(0, 1.1)
ampl.set_xlabel('False positive rate')
ampl.set_ylabel('True positive rate')
plt.legend()
plt.savefig(plotpath + 'roc_scan_' + target_label + '_' + v.name +
'.pdf')
plt.show()
def rocScan(varlist,
scan_targets,
labels,
ylabels,
data,
plotpath='',
x_min=0.,
x_max=1.0,
y_min=0.0,
y_max=1.0,
x_log=False,
y_log=False,
rejection=False,
x_label='False positive rate',
y_label='True positive rate',
linestyles=[],
colorgrouping=-1,
extra_lines=[],
atlas_x=-1,
atlas_y=-1,
simulation=False,
textlist=[]):
'''
Creates a set of ROC curve plots by scanning over the specified variables.
One set is created for each target (neural net score dataset).
varlist: a list of rocVar instances to scan over
scan_targets: a list of neural net score datasets to use
labels: a list of target names (strings); must be the same length as scan_targets
'''
rocs = buildRocs(varlist, scan_targets, labels, ylabels, data)
for target_label in labels:
for v in varlist:
# prepare matplotlib figure
plt.cla()
plt.clf()
fig = plt.figure()
fig.patch.set_facecolor('white')
plt.plot([0, 1], [0, 1], 'k--')
for label in v.labels:
# first generate ROC curve
x = rocs[target_label + label]['x']
y = rocs[target_label + label]['y']
var_auc = auc(x, y)
if not rejection:
plt.plot(x,
y,
label=label + ' (area = {:.3f})'.format(var_auc))
else:
plt.plot(y,
1. / x,
label=label + ' (area = {:.3f})'.format(var_auc))
# plt.title('ROC Scan of '+target_label+' over '+v.latex)
if x_log:
plt.xscale('log')
if y_log:
plt.yscale('log')
plt.xlim(x_min, x_max)
plt.ylim(y_min, y_max)
ampl.set_xlabel(x_label)
ampl.set_ylabel(y_label)
plt.legend()
drawLabels(fig, atlas_x, atlas_y, simulation, textlist)
if plotpath != '':
plt.savefig(plotpath + 'roc_scan_' + target_label + '_' +
v.name + '.pdf')
plt.show()
-bkg_err,
color='black',
alpha=0.3)
ratio = (data - bkg) / bkg
ratio_ax.errorbar(bin_centers,
ratio,
yerr=(np.sqrt(data) * (ratio / data)),
fmt='ko')
out_of_range = np.where(ratio > 1, 1, np.where(ratio < -1, -1, np.NaN))
ratio_ax.plot(bin_centers, out_of_range, marker=CARETUP, color='paper:red')
ratio_ax.set_ylabel(r"$\frac{\textsf{Data} - \textsf{Bkg}}{\textsf{Bkg}}$",
fontsize=12)
ratio_ax.set_ylim((-1, 1))
ratio_ax.yaxis.set_minor_locator(AutoMinorLocator())
ax.yaxis.set_minor_locator(AutoMinorLocator())
atlas.set_xlabel(var_labels[args.var], ax=ratio_ax)
ax.set_ylim((0, ax.get_ylim()[1]))
atlas.set_ylabel('Events', ax=ax)
region = args.region
if region == 'sig':
region = 'signal'
region = region.capitalize()
atlas.draw_atlas_label(
0.3,
0.97,
ax=ax,
status='int',
energy='13 TeV',
lumi=24,
rwgt_func = FitFunction(bin_centers, ratio, ratio_err)
ax = plt.subplot(len(rwgt_vars), 1, i + 1)
ax.errorbar(bin_centers, ratio, yerr=ratio_err, fmt='o')
rwgt_funcs[var] = rwgt_func
x = np.linspace(min_var, max_var, 1000)
y = rwgt_func.fit(x)
extreme = np.max(np.abs(y - 1))
done = extreme < 0.05
ax.plot(x, y)
ax.plot(rwgt_func.orig_x, rwgt_func.vals, 'g:')
for spine in ax.spines.values():
spine.set_color('paper:green' if done else 'paper:red')
spine.set_linewidth(2)
atlas.set_xlabel(f"{rwgt_vars[var]} (max. dev.: {extreme:.4f})",
ax=ax)
all_done.append(done)
all_rwgt_funcs.append(rwgt_funcs)
plt.tight_layout()
plt.subplots_adjust(top=0.95)
plt.suptitle(f"Iteration {iter_num} (f = {f:.3f})", fontsize=15)
pdf.savefig()
plt.close()
iter_num += 1
if np.all(all_done) or iter_num >= 100:
print("\nDone. Optimizing f one final time...")
sf = np.ones(bkg.shape[0])
ttbar_2tag_sf = ttbar_2tag.mc_sf.values
if len(all_rwgt_funcs) != 0:
for rwgt_dict in all_rwgt_funcs:
for k, v in rwgt_dict.items():
