def test_train(test, train, y_test, y_train, labels=[], bins=25, node=0, plot_dir=None, weight_test=None, weight_train=None):
ks = {}
fig, ax = plt.subplots(1,1,figsize=(10,10))
h = {}
for i, label in enumerate(labels):
_ks, _p = scipy.stats.kstest(
train[:,node][(y_train==i)],
test[:,node][(y_test==i)]
)
#_ks, _p = -1, -1
ks[label] = (_p, _ks)
h[label+'_test'] = Hist1D(test[:,node][(y_test==i)], bins=bins, weights=weight_test[(y_test==i)]).normalize()
h[label+'_train'] = Hist1D(train[:,node][(y_train==i)], bins=bins, label=label+' (p=%.2f, KS=%.2f)'%(_p, _ks), weights=weight_train[(y_train==i)]).normalize()
h[label+'_test'].plot(color=colors[i], histtype="step", ls='--', linewidth=2)
h[label+'_train'].plot(color=colors[i], histtype="step", linewidth=2)
if plot_dir:
finalizePlotDir(plot_dir)
fig.savefig("{}/score_node_{}.png".format(plot_dir, node))
fig.savefig("{}/score_node_{}.pdf".format(plot_dir, node))
return ks
def test_train_cat(test, train, y_test, y_train, labels=[], n_cat=5, plot_dir=None, weight_test=None, weight_train=None):
ks = {}
bins = [x-0.5 for x in range(n_cat+1)]
fig, ax = plt.subplots(1,1,figsize=(10,10))
h = {}
for i, label in enumerate(labels):
_ks, _p = scipy.stats.kstest(
train.argmax(axis=1)[(y_train==i)],
test.argmax(axis=1)[(y_test==i)]
)
#_ks, _p = -1, -1
ks[label] = (_p, _ks)
h[label+'_test'] = Hist1D(test.argmax(axis=1)[(y_test==i)], bins=bins, weights=weight_test[(y_test==i)]).normalize()
h[label+'_train'] = Hist1D(train.argmax(axis=1)[(y_train==i)], bins=bins, label=label+' (p=%.2f, KS=%.2f)'%(_p, _ks), weights=weight_train[(y_train==i)]).normalize()
h[label+'_test'].plot(color=colors[i], histtype="step", ls='--', linewidth=2)
h[label+'_train'].plot(color=colors[i], histtype="step", linewidth=2)
ax.set_ylabel('a.u.')
ax.set_xlabel('category')
ax.set_ylim(0,1/n_cat*5)
if plot_dir:
finalizePlotDir(plot_dir)
fig.savefig("{}/categories.png".format(plot_dir))
fig.savefig("{}/categories.pdf".format(plot_dir))
return ks
def saveFig( fig, ax, rax, path, name, scale='linear', shape=False, y_max=-1 ):
outdir = os.path.join(path,scale)
finalizePlotDir(outdir)
ax.set_yscale(scale)
ax.set_ylabel('Events')
if scale == 'linear':
if y_max<0 or True:
pass
else:
ax.set_ylim(0, 1 if shape else 1.2*y_max)
else:
if y_max<0 and not shape:
pass
else:
ax.set_ylim(0.000005 if shape else 0.05, 3 if shape else 300*y_max)
#if scale == 'log':
# ax.set_ylim(y_min, y_max)
#else:
# ax.set_ylim(0, y_max)
# #if shape:
# # ax.yaxis.set_ticks(np.array([10e-4,10e-3,10e-2,10e-1,10e0]))
# #else:
# # ax.yaxis.set_ticks(np.array([10e-2,10e-1,10e0,10e1,10e2,10e3,10e4,10e5,10e6]))
handles, labels = ax.get_legend_handles_labels()
new_labels = []
WHcount = 0
for handle, label in zip(handles, labels):
#print (handle, label)
try:
#new_labels.append(my_labels[label])
if WHcount>0 and label=='WH':
new_labels.append('WH (1000,0)')
handle.set_color('#000000')
else:
new_labels.append(my_labels[label])
if not label=='pseudodata':
handle.set_color(colors[label])
if label=='WH':
WHcount+=1
except:
pass
if rax:
plt.subplots_adjust(hspace=0)
rax.set_ylabel('Obs./Pred.')
rax.set_ylim(0.5,1.5)
ax.legend(title='',ncol=2,handles=handles, labels=new_labels, frameon=False)
fig.text(0.15, 0.995, '$\\bf{CMS}$', fontsize=20, horizontalalignment='left', verticalalignment='bottom', transform=ax.transAxes )
fig.text(0.30, 1., '$\\it{Simulation}$', fontsize=14, horizontalalignment='left', verticalalignment='bottom', transform=ax.transAxes )
fig.text(0.8, 1., '13 TeV', fontsize=14, horizontalalignment='left', verticalalignment='bottom', transform=ax.transAxes )
fig.savefig(os.path.join(outdir, "{}.pdf".format(name)))
fig.savefig(os.path.join(outdir, "{}.png".format(name)))
def get_cat_plot(X, y, labels=[], n_cat=5, plot_dir=None, weight=None):
ks = {}
bins = [x-0.5 for x in range(n_cat+1)]
fig, ax = plt.subplots(1,1,figsize=(10,10))
h = {}
for i, label in enumerate(labels):
h[label+'_train'] = Hist1D(X.argmax(axis=1)[(y==i)], bins=bins, label=label, weights=weight[(y==i)])
h[label+'_train'].plot(color=colors[i], histtype="step", linewidth=2)
ax.set_ylabel('a.u.')
ax.set_xlabel('category')
ax.set_ylim(0,200)
if plot_dir:
finalizePlotDir(plot_dir)
fig.savefig("{}/abs_categories.png".format(plot_dir))
fig.savefig("{}/abs_categories.pdf".format(plot_dir))
def makePlot(output,
histo,
axis,
bins=None,
data=[],
normalize=True,
log=False,
save=False,
axis_label=None,
ratio_range=None,
upHists=[],
downHists=[],
shape=False,
ymax=False,
new_colors=colors,
new_labels=my_labels,
order=None,
signals=[],
omit=[],
lumi=60.0,
binwnorm=None,
overlay=None,
use_label=True,
y_axis_label='Events'):
if save:
finalizePlotDir('/'.join(save.split('/')[:-1]))
mc_sel = re.compile(
'(?!(%s))' %
('|'.join(data + omit))) if len(data + omit) > 0 else re.compile('')
data_sel = re.compile('|'.join(data))
bkg_sel = re.compile(
'(?!(%s))' %
('|'.join(data + signals + omit))) if len(data + signals +
omit) > 0 else re.compile('')
if histo is None:
processes = [p[0] for p in output.values().keys() if not p[0] in data]
histogram = output.copy()
else:
processes = [
p[0] for p in output[histo].values().keys() if not p[0] in data
]
histogram = output[histo].copy()
histogram = histogram.project(axis, 'dataset')
if overlay: overlay = overlay.project(axis, 'dataset')
if bins:
histogram = histogram.rebin(axis, bins)
if overlay: overlay = overlay.rebin(axis, bins)
y_max = histogram[bkg_sel].sum("dataset").values(overflow='over')[()].max()
print(histogram[bkg_sel].sum("dataset").values(overflow='over')[()].max())
MC_total = histogram[bkg_sel].sum("dataset").values(
overflow='over')[()].sum()
Data_total = 0
if data:
Data_total = histogram[data_sel].sum("dataset").values(
overflow='over')[()].sum()
#observation = histogram[data[0]].sum('dataset').copy()
#first = True
#for d in data:
# print (d)
# if not first:
# observation.add(histogram[d].sum('dataset'))
# print ("adding")
# first = False
print("Data:", round(Data_total, 0), "MC:", round(MC_total, 2))
if normalize and data_sel:
scales = {process: Data_total / MC_total for process in processes}
histogram.scale(scales, axis='dataset')
else:
scales = {}
if shape:
scales = {
process: 1 / histogram[process].sum("dataset").values(
overflow='over')[()].sum()
for process in processes
}
histogram.scale(scales, axis='dataset')
if data:
fig, (ax, rax) = plt.subplots(2,
1,
figsize=(10, 10),
gridspec_kw={"height_ratios": (3, 1)},
sharex=True)
else:
fig, ax = plt.subplots(1, 1, figsize=(10, 10))
if signals:
for sig in signals:
ax = hist.plot1d(histogram[sig],
overlay="dataset",
ax=ax,
stack=False,
overflow='over',
clear=False,
line_opts=line_opts,
fill_opts=None,
binwnorm=binwnorm)
if overlay:
ax = hist.plot1d(overlay,
overlay="dataset",
ax=ax,
stack=False,
overflow='over',
clear=False,
line_opts=line_opts,
fill_opts=None,
binwnorm=binwnorm)
if shape:
ax = hist.plot1d(histogram[bkg_sel],
overlay="dataset",
ax=ax,
stack=False,
overflow='over',
clear=False,
line_opts=line_opts,
fill_opts=None,
binwnorm=binwnorm)
else:
ax = hist.plot1d(histogram[bkg_sel],
overlay="dataset",
ax=ax,
stack=True,
overflow='over',
clear=False,
line_opts=None,
fill_opts=fill_opts,
order=(order if order else processes),
binwnorm=binwnorm)
if data:
ax = hist.plot1d(histogram[data_sel].sum("dataset"),
ax=ax,
overflow='over',
error_opts=data_err_opts,
clear=False,
binwnorm=binwnorm)
#ax = hist.plot1d(observation, ax=ax, overflow='over', error_opts=data_err_opts, clear=False)
hist.plotratio(
num=histogram[data_sel].sum("dataset"),
denom=histogram[bkg_sel].sum("dataset"),
ax=rax,
error_opts=data_err_opts,
denom_fill_opts=
None, # triggers this: https://github.com/CoffeaTeam/coffea/blob/master/coffea/hist/plot.py#L376
guide_opts={},
unc='num',
#unc=None,
overflow='over')
handles, labels = ax.get_legend_handles_labels()
updated_labels = []
for handle, label in zip(handles, labels):
try:
if label is None or label == 'None':
updated_labels.append("Observation")
handle.set_color('#000000')
else:
updated_labels.append(new_labels[label])
handle.set_color(new_colors[label])
except:
pass
if data:
if ratio_range:
rax.set_ylim(*ratio_range)
else:
rax.set_ylim(0.1, 1.9)
rax.set_ylabel('Obs./Pred.')
if axis_label:
rax.set_xlabel(axis_label)
ax.set_xlabel(axis_label)
ax.set_ylabel(y_axis_label)
if not binwnorm:
if not shape:
addUncertainties(ax,
axis,
histogram,
bkg_sel,
[output[histo + '_' + x] for x in upHists],
[output[histo + '_' + x] for x in downHists],
overflow='over',
rebin=bins,
ratio=False,
scales=scales)
if data:
addUncertainties(rax,
axis,
histogram,
bkg_sel,
[output[histo + '_' + x] for x in upHists],
[output[histo + '_' + x] for x in downHists],
overflow='over',
rebin=bins,
ratio=True,
scales=scales)
if log:
ax.set_yscale('log')
y_mult = 1.7 if not log else 100
if ymax:
ax.set_ylim(0.01, ymax)
else:
y_max = y_max * y_mult * (Data_total /
MC_total) if data else y_max * y_mult
ax.set_ylim(0.01, y_max if not shape else 2)
#if binwnorm: ax.set_ylim(0.5)
ax.legend(
loc='upper right',
ncol=2,
borderaxespad=0.0,
labels=updated_labels,
handles=handles,
)
plt.subplots_adjust(hspace=0)
if use_label:
if len(data) > 0:
fig.text(0.0,
0.995,
'$\\bf{CMS}$ Preliminary',
fontsize=25,
horizontalalignment='left',
verticalalignment='bottom',
transform=ax.transAxes)
else:
fig.text(0.0,
0.995,
'$\\bf{CMS}$ Simulation',
fontsize=25,
horizontalalignment='left',
verticalalignment='bottom',
transform=ax.transAxes)
fig.text(0.6,
0.995,
r'$%.1f\ fb^{-1}$ (13 TeV)' % (lumi),
fontsize=25,
horizontalalignment='left',
verticalalignment='bottom',
transform=ax.transAxes)
if normalize:
fig.text(0.55,
0.65,
'Data/MC = %s' % round(Data_total / MC_total, 2),
fontsize=20,
horizontalalignment='left',
verticalalignment='bottom',
transform=ax.transAxes)
if save:
#finalizePlotDir(outdir)
fig.savefig("{}.pdf".format(save))
fig.savefig("{}.png".format(save))
#fig.savefig(save)
print("Figure saved in:", save)
def makePlot(output, histo, axis, bins=None, mc_sel=bkgonly, data_sel='MuonEG', normalize=True, log=False, save=False, axis_label=None, ratio_range=None, upHists=[], downHists=[], shape=False, ymax=False, new_colors=colors, new_labels=my_labels, order=None):
if save:
finalizePlotDir( '/'.join(save.split('/')[:-1]) )
if histo is None:
processes = [ p[0] for p in output.values().keys() if not p[0]=='MuonEG' ]
histogram = output.copy()
else:
processes = [ p[0] for p in output[histo].values().keys() if not p[0]=='MuonEG' ]
histogram = output[histo].copy()
histogram = histogram.project(axis, 'dataset')
if bins:
histogram = histogram.rebin(axis, bins)
y_max = histogram[mc_sel].sum("dataset").values(overflow='over')[()].max()
MC_total = histogram[mc_sel].sum("dataset").values(overflow='over')[()].sum()
Data_total = 0
if data_sel:
Data_total = histogram[data_sel].sum("dataset").values(overflow='over')[()].sum()
print ("Data:", round(Data_total,0), "MC:", round(MC_total,2))
if normalize and data_sel:
scales = { process: Data_total/MC_total for process in processes }
histogram.scale(scales, axis='dataset')
else:
scales = {}
if shape:
scales = { process: 1/histogram[process].sum("dataset").values(overflow='over')[()].sum() for process in processes }
histogram.scale(scales, axis='dataset')
if data_sel:
fig, (ax, rax) = plt.subplots(2,1,figsize=(10,10), gridspec_kw={"height_ratios": (3, 1)}, sharex=True)
else:
fig, ax = plt.subplots(1,1,figsize=(10,10) )
if shape:
ax = hist.plot1d(histogram[mc_sel], overlay="dataset", ax=ax, stack=False, overflow='over', clear=False, line_opts=line_opts, fill_opts=None)
else:
ax = hist.plot1d(histogram[mc_sel], overlay="dataset", ax=ax, stack=True, overflow='over', clear=False, line_opts=None, fill_opts=fill_opts, order=(order if order else processes))
if data_sel:
ax = hist.plot1d(histogram[data_sel], overlay="dataset", ax=ax, overflow='over', error_opts=data_err_opts, clear=False)
hist.plotratio(
num=histogram[data_sel].sum("dataset"),
denom=histogram[mc_sel].sum("dataset"),
ax=rax,
error_opts=data_err_opts,
denom_fill_opts=None, # triggers this: https://github.com/CoffeaTeam/coffea/blob/master/coffea/hist/plot.py#L376
guide_opts={},
unc='num',
#unc=None,
overflow='over'
)
handles, labels = ax.get_legend_handles_labels()
updated_labels = []
for handle, label in zip(handles, labels):
#print (label)
try:
updated_labels.append(new_labels[label])
if not label=='MuonEG':
handle.set_color(new_colors[label])
except:
pass
if data_sel:
if ratio_range:
rax.set_ylim(*ratio_range)
else:
rax.set_ylim(0.1,1.9)
rax.set_ylabel('Obs./Pred.')
if axis_label:
rax.set_xlabel(axis_label)
ax.set_xlabel(axis_label)
ax.set_ylabel('Events')
if not shape:
addUncertainties(ax, axis, histogram, mc_sel, [output[histo+'_'+x] for x in upHists], [output[histo+'_'+x] for x in downHists], overflow='over', rebin=bins, ratio=False, scales=scales)
if data_sel:
addUncertainties(rax, axis, histogram, mc_sel, [output[histo+'_'+x] for x in upHists], [output[histo+'_'+x] for x in downHists], overflow='over', rebin=bins, ratio=True, scales=scales)
if log:
ax.set_yscale('log')
y_mult = 1.3 if not log else 100
if ymax:
ax.set_ylim(0.01, ymax)
else:
ax.set_ylim(0.01,y_max*y_mult if not shape else 2)
ax.legend(
loc='upper right',
ncol=2,
borderaxespad=0.0,
labels=updated_labels,
handles=handles,
)
plt.subplots_adjust(hspace=0)
fig.text(0.0, 0.995, '$\\bf{CMS}$ Preliminary', fontsize=20, horizontalalignment='left', verticalalignment='bottom', transform=ax.transAxes )
fig.text(0.8, 0.995, '13 TeV', fontsize=20, horizontalalignment='left', verticalalignment='bottom', transform=ax.transAxes )
if normalize:
fig.text(0.55, 0.65, 'Data/MC = %s'%round(Data_total/MC_total,2), fontsize=20, horizontalalignment='left', verticalalignment='bottom', transform=ax.transAxes )
if save:
#finalizePlotDir(outdir)
fig.savefig("{}.pdf".format(save))
fig.savefig("{}.png".format(save))
#fig.savefig(save)
print ("Figure saved in:", save)