forked from dominiquegodin/ATLAS-e-ID
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathplots_KM.py
414 lines (350 loc) · 17.1 KB
/
plots_KM.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
264
265
266
267
268
269
270
271
272
273
274
275
276
277
278
279
280
281
282
283
284
285
286
287
288
289
290
291
292
293
294
295
296
297
298
299
300
301
302
303
304
305
306
307
308
309
310
311
312
313
314
315
316
317
318
319
320
321
322
323
324
325
326
327
328
329
330
331
332
333
334
335
336
337
338
339
340
341
342
343
344
345
346
347
348
349
350
351
352
353
354
355
356
357
358
359
360
361
362
363
364
365
366
367
368
369
370
371
372
373
374
375
376
377
378
379
380
381
382
383
384
385
386
387
388
389
390
391
392
393
394
395
396
397
398
399
400
401
402
403
404
405
406
407
408
409
410
411
412
413
414
import numpy as np, h5py, sys, time
import matplotlib; matplotlib.use('Agg')
#import matplotlib; matplotlib.use('pdf')
import matplotlib.style as style
import matplotlib.pyplot as plt
import matplotlib.colors as colors
from matplotlib import pylab
from sklearn import metrics
import os, math, pickle
def get_LLH(data, y_true):
eff_class0, eff_class1 = [],[]
for wp in ['p_LHTight', 'p_LHMedium', 'p_LHLoose']:
y_class0 = data[wp][y_true == 0]
y_class1 = data[wp][y_true == 1]
eff_class0.append( np.sum(y_class0 == 0)/len(y_class0) )
eff_class1.append( np.sum(y_class1 == 0)/len(y_class1) )
return eff_class0, eff_class1
def plot_distributions_KM(y_true, y_prob, var_name='',output_dir='outputs/',postfix=''):
if var_name=='': var_name='distributions'
file_name=output_dir+'/'+var_name+postfix+'.png'
print('plots_KM: saving test sample distributions in:', file_name)
if var_name=='distributions':
probs_class0 = 100*y_prob[:,0][ y_true==0 ]
probs_class1 = 100*y_prob[:,0][ y_true==1 ]
bins = np.arange(0, 100, 0.1)
pylab.xlim(-0.5,100.5)
plt.xticks(np.arange(0,101,step=10))
xstring= 'Signal Probability (%)'
ystring='Distribution (% per '+str(100/len(bins))+'% bin)'
legloc='upper center'
else:
probs_class0 = y_prob[ y_true==0 ]
probs_class1 = y_prob[ y_true==1 ]
bins = np.arange(-2.5, 2.5, 0.1)
ystring='arbitrary unit'
legloc='upper right'
pass
if var_name=="eta":
bins = np.arange(-2.5, 2.5, 0.1)
pylab.xlim(-2.5,2.5)
plt.xticks(np.arange(-2.5,2.5,step=0.1))
xstring='Eta'
ystring='arbitrary unit'
elif var_name=="pt":
bins = np.arange(0, 100, 1)
pylab.xlim(-2.5,2.5)
plt.xticks(np.arange(0,100,step=5))
xstring='Pt [GeV]'
ystring='arbitrary unit'
pass
weights_class0 = len(probs_class0)*[100/len(probs_class0)]
weights_class1 = len(probs_class1)*[100/len(probs_class1)]
histtype ='step'
plt.figure(figsize=(12,8))
pylab.grid(True)
pylab.hist( probs_class0, bins=bins, label='Signal',
facecolor='blue', histtype=histtype, weights=weights_class0 )
pylab.hist( probs_class1, bins=bins, label='Background',
facecolor='red', histtype=histtype, weights=weights_class1 )
plt.xlabel(xstring,fontsize=20)
plt.ylabel(ystring,fontsize=20)
plt.legend(title=postfix[1:],loc=legloc, fontsize=15, numpoints=3)
plt.savefig(file_name)
plt.close()
def plot_ROC_curves(test_sample, y_true, y_prob, ROC_type, postfix='',output_dir='outputs/'):
#print('plots_KM:', output_dir)
file_name = output_dir
#if postfix!='':file_name+='differential/'
if not os.path.isdir(file_name): os.mkdir(file_name)
file_name+= 'ROC'+str(ROC_type)+'_curve'+postfix+'.png'
print('plots_KM: saving test sample ROC'+str(ROC_type)+' curve in: ', file_name)
eff_class0, eff_class1 = get_LLH(test_sample, y_true)
#y_prob = y_prob[np.logical_or(y_true==0, y_true==1)]
#y_true = y_true[np.logical_or(y_true==0, y_true==1)]
#FalsePositveRate, TruePositveRate
fpr, tpr, threshold = metrics.roc_curve(y_true, y_prob[:,0], pos_label=0)
signal_ratio = len(y_true[y_true==0])/len(y_true)
accuracy = tpr*signal_ratio + (1-fpr)*(1-signal_ratio)
best_tpr, best_fpr = tpr[np.argmax(accuracy)], fpr[np.argmax(accuracy)]
colors = [ 'red', 'blue', 'green' ]
labels = [ 'LLH tight', 'LLH medium', 'LLH loose' ]
plt.figure(figsize=(12,8))
pylab.grid(True)
axes = plt.gca()
axes.xaxis.set_ticks(np.arange(0, 101, 10))
plt.xlabel('Signal Efficiency (%)',fontsize=20)
if ROC_type == 1:
plt.xlim([0, 100])
plt.ylim([0, 100.5])
axes.yaxis.set_ticks(np.arange(0, 101, 10))
plt.ylabel('Background Rejection (%)',fontsize=20)
plt.text(22, 34, 'AUC: '+str(format(metrics.auc(fpr,tpr),'.4f')),
{'color': 'black', 'fontsize': 22}, va="center", ha="center")
val = plt.plot(100*tpr, 100*(1-fpr), label='Signal vs Bkg', color='#1f77b4')
plt.scatter( 100*best_tpr, 100*(1-best_fpr), s=30, marker='D', c=val[0].get_color(),
label="{0:<16s} {1:>3.2f}%".format('Best Accuracy:',100*max(accuracy)) )
for LLH in zip( eff_class0, eff_class1, colors, labels ):
plt.scatter( 100*LLH[0], 100*(1-LLH[1]), s=40, marker='o', c=LLH[2], label='('+\
str( format(100*LLH[0],'.1f'))+'%, '+str( format(100*(1-LLH[1]),'.1f') )+\
')'+r'$\rightarrow$'+LLH[3] )
plt.legend(loc='lower left', fontsize=15, numpoints=3)
plt.savefig(file_name)
if ROC_type == 2:
pylab.grid(False)
len_0 = len(fpr[fpr==0])
x_min = min(60, 10*np.floor(10*eff_class0[0]))
y_max = 10000
if fpr[np.argwhere(np.diff(np.sign(tpr-x_min/100)))[0][0]]>0 and eff_class1[0]>0:
y_max = 100*np.ceil(max(1/fpr[np.argwhere(np.diff(np.sign(tpr-x_min/100)))[0][0]], 1/eff_class1[0])/100)
elif fpr[np.argwhere(np.diff(np.sign(tpr-x_min/100)))[0][0]]>0:
y_max = 100*np.ceil(1/fpr[np.argwhere(np.diff(np.sign(tpr-x_min/100)))[0][0]]/100)
elif eff_class1[0]>0:
y_max = 100*np.ceil(1/eff_class1[0]/100)
pass
plt.xlim([x_min, 100])
plt.ylim([1, y_max])
LLH_scores = [1/fpr[np.argwhere(tpr>=value)[0]] for value in eff_class0
if fpr[np.argwhere(tpr>=value)[0]]!=0]
#LLH_scores = [10,10,10]#[1/fpr[np.argwhere(tpr >= value)[0]] for value in eff_class0]
for n in np.arange(len(LLH_scores)):
axes.axhline(LLH_scores[n], xmin=(eff_class0[n]-x_min/100)/(1-x_min/100), xmax=1,
ls='--', linewidth=0.5, color='#1f77b4')
if eff_class0[n]>0 and eff_class1[n]>0:
axes.axvline(100*eff_class0[n], ymin=abs(1/eff_class1[n]-1)/(plt.yticks()[0][-1]-1),
ymax=abs(LLH_scores[n]-1)/(plt.yticks()[0][-1]-1), ls='--', linewidth=0.5, color='#1f77b4')
for val in LLH_scores:
plt.text(100.2, val, str(int(val)), {'color': '#1f77b4', 'fontsize': 10}, va="center", ha="left")
axes.yaxis.set_ticks( np.append([1],plt.yticks()[0][1:]) )
plt.ylabel('1/(Background Efficiency)',fontsize=20)
val = plt.plot(100*tpr[len_0:], 1/fpr[len_0:], label='Signal vs Bkg ' + postfix[1:]+(" %d e"%len(y_true)), color='#1f77b4',)
x_array = tpr[len_0:]
y_array = 1/fpr[len_0:]
plt.scatter( 100*best_tpr, 1/best_fpr, s=30, marker='D', c=val[0].get_color(),
label="{0:<15s} {1:>3.2f}%".format('Best Accuracy:',100*max(accuracy)) )
for LLH in zip( eff_class0, eff_class1, colors, labels ):
if LLH[0]>0 and LLH[1]>0:
plt.scatter( 100*LLH[0], 1/LLH[1], s=40, marker='o', c=LLH[2], label='('+\
str(format(100*LLH[0],'.1f'))+'%, '+str(format(1/LLH[1],'.0f'))+\
')'+r'$\rightarrow$'+LLH[3] )
plt.legend(loc='upper right', fontsize=15, numpoints=3)
plt.savefig(file_name)
if ROC_type == 3:
best_threshold = threshold[np.argmax(accuracy)]
plt.xlim([0, 100])
plt.ylim([max(50, 10*np.round(min(10*accuracy))),10*np.ceil(10*max(accuracy))])
plt.xlabel('Discrimination Threshold (%)',fontsize=30)
plt.ylabel('Accuracy (%)',fontsize=30)
val = plt.plot(100*threshold[1:], 100*accuracy[1:], color='#1f77b4')
#plt.plot( 100*threshold[1:], 100*tpr[1:], color='r')
#plt.plot( 100*threshold[1:], 100*(1-fpr[1:]), color='g')
std_accuracy = test_accuracy(y_true, y_prob)
std_threshold = np.argwhere(np.diff(np.sign(accuracy-std_accuracy))).flatten()
#std_threshold = np.argwhere(accuracy >= std_accuracy)[0].flatten()
plt.scatter( [ 50], #100*threshold[std_threshold[-1]] ],
[ 100*accuracy [std_threshold[-1]] ],
s=40, marker='o', c=val[0].get_color(),
label="{0:<17s} {1:>2.2f}%".format('Accuracy at 50%:',100*accuracy[std_threshold[-1]]) )
plt.scatter( 100*best_threshold, 100*max(accuracy), s=40, marker='D', c=val[0].get_color(),
label="{0:<16s} {1:>8.2f}%".format('Best Accuracy:',100*max(accuracy)) )
plt.legend(loc='lower center', fontsize=15, numpoints=3)
plt.savefig(file_name)
pass
plt.close()
#print([x_array.size, y_array.size])
return [x_array, y_array]
def differential_plots(test_LLH, y_true, y_prob, boundaries, bin_indices,varname='pt',output_dir='outputs/',evalLLH=False,makeOutput=True):
roc_curves = dict()
roc_curves["roc_inclusive"]= plot_ROC_curves(test_LLH, y_true, y_prob, ROC_type=2,output_dir=output_dir+"/differential/")
tmp_idx=0
x_centers = list()
x_errs = list()
y_prob_sig = y_prob[y_true==0]
sigEffs = [0.7,0.8,0.9]
globalCuts = list()
bkg_rejs_fEff = {}
bkg_errs_fEff = {}
bkg_rejs_gEff = {}
bkg_errs_gEff = {}
sig_effs_gEff = {}
sig_errs_gEff = {}
for sigEff in sigEffs:
bkg_rejs_fEff.update({sigEff:[]})
bkg_errs_fEff.update({sigEff:[]})
bkg_rejs_gEff.update({sigEff:[]})
bkg_errs_gEff.update({sigEff:[]})
sig_effs_gEff.update({sigEff:[]})
sig_errs_gEff.update({sigEff:[]})
pCut = np.percentile(y_prob_sig,(1-sigEff)*100,axis=0) [0] #global cut
globalCuts.append(pCut)
pass
for bin_idx in bin_indices:
if bin_idx.size==0:
tmp_idx+=1
continue
pass
fill_rej=False
pfix ="_"+varname+"%d" % tmp_idx
if tmp_idx!=0: pfix+="_Lo%.2f" % boundaries[tmp_idx-1] #lo
if tmp_idx!=len(bin_indices)-1: pfix+="_Hi%.2f" % boundaries[tmp_idx] #hi
if evalLLH: pfix+='LLH'
if tmp_idx!=0 and tmp_idx!=len(bin_indices)-1:
x_center = (boundaries[tmp_idx-1] + boundaries[tmp_idx])/2
x_centers.append(x_center)
x_err = boundaries[tmp_idx]-x_center
x_errs.append(x_err)
fill_rej=True
pass
new_test_labels=y_true.take(bin_idx)
new_y_prob =y_prob.take(bin_idx,axis=0)
new_test_LLH=dict()
#for llh in test_LLH:
for llh in ['p_LHTight', 'p_LHMedium', 'p_LHLoose']:
#print(llh)
new_test_LLH[llh]=test_LLH[llh][bin_idx]
pass
if not (len(new_y_prob)==len(new_test_labels) and len(new_test_labels)==len(new_test_LLH['p_LHTight'])):
print("data size for data, label, llh= ",len(new_y_prob),len(new_test_labels),len(new_test_LLH['p_LHTight']))
if not(~np.isnan(new_y_prob).any() and ~np.isinf(new_y_prob).any()): print("Nan or Inf detected")
roc_name="roc"+pfix
if "LLH" in roc_name:
roc_name=roc_name[0:len(roc_name)-3]
pass
roc_curves[roc_name] = plot_ROC_curves(new_test_LLH, new_test_labels, new_y_prob, ROC_type=2, postfix=pfix,output_dir=output_dir+'/differential/')
#plot_distributions_KM(new_test_labels,new_y_prob,output_dir=output_dir+'differential/',postfix=pfix)
if fill_rej:
fill_bkg_rejs_f(bkg_rejs_fEff,bkg_errs_fEff,
new_y_prob,new_test_labels,sigEffs)
fill_info_g (bkg_rejs_gEff,bkg_errs_gEff,
sig_effs_gEff,sig_errs_gEff,
new_y_prob,new_test_labels,sigEffs,globalCuts)
tmp_idx+=1
pass
# print(x_centers)
# print(x_errs)
# print(bkg_rejs_fEff[.7])
# print(bkg_errs_fEff[.7])
# print()
# print(bkg_rejs_fEff[.8])
# print(bkg_errs_fEff[.8])
# print()
# print(bkg_rejs_fEff[.9])
# print(bkg_errs_fEff[.9])
#
plot_rej_vsX_curves(x_centers,x_errs, bkg_rejs_fEff,bkg_errs_fEff, sigEffs,varname,output_dir,boundaries[-1],cType="Flat" ,makeOutput=True,evalLLH=evalLLH)
plot_rej_vsX_curves(x_centers,x_errs, bkg_rejs_gEff,bkg_errs_gEff, sigEffs,varname,output_dir,boundaries[-1],cType="GlobB",makeOutput=True,evalLLH=evalLLH)
plot_rej_vsX_curves(x_centers,x_errs, sig_effs_gEff,sig_errs_gEff, sigEffs,varname,output_dir,boundaries[-1],cType="GlobS",makeOutput=True,evalLLH=evalLLH)
#print(roc_curves)
if makeOutput:
outfilename = output_dir+"roc"
if evalLLH: outfilename+= "LLH"
outfilename+="_vs_"+varname+".pkl"
print('Writing ROC pickle file:', outfilename)
pickle.dump(roc_curves,open(outfilename, 'wb'))
return
def plot_rej_vsX_curves(x_centers,x_errs,
bkg_rejs,bkg_errs,
sigEffs,varname,output_dir,x_up,
cType='Flat',makeOutput=False,evalLLH=False):
plt.close()
style.use('classic')
#ax = plt.figure().add_subplot(111)
errGraphs = dict()
for sigEff in sigEffs:
plt.errorbar(np.asarray(x_centers), np.asarray(bkg_rejs[sigEff]),
xerr=np.asarray(x_errs), yerr= np.asarray(bkg_errs[sigEff]),
marker='o',capsize=2, linestyle='None',fillstyle='none')
data_points = [np.asarray(x_centers),
np.asarray(bkg_rejs[sigEff]),
np.asarray(x_errs),
np.asarray(bkg_errs[sigEff])]
errGraphs[cType+"_%d"%(sigEff*100)]=data_points
pass
if varname.find("eta")>=0:
pylab.xlim(-2.5,2.5)
plt.xticks(np.arange(-2.5,2.6,step=0.5))
elif varname.find("pt")>=0:
#pylab.ylim(0,1800)
pylab.xlim(0,x_up+20)
plt.xticks(np.arange(0,x_up,step=50))
pass
ystring='Bkg-rejection'
if cType.find('GlobS')!=-1: ystring='Sig-efficiency'
plt.ylabel(ystring,fontsize=15)
if varname.find('pt')!=-1: plt.xlabel(varname+' [GeV]',fontsize=15)
else: plt.xlabel(varname, fontsize=15)
title_str = cType
if evalLLH: title_str+= "LLH"
title_str += " efficiency plot. sig-eff: "
plt.title(title_str,fontweight='bold') #plt.title(cType+" efficiency plot. sig-eff: ",fontweight='bold')
#KM: some sort of legend
plt.text(0.75, 1.02, '70%', transform=plt.gca().transAxes, color='b', fontsize=15)
plt.text(0.85, 1.02, '80%', transform=plt.gca().transAxes, color='g', fontsize=15)
plt.text(0.95, 1.02, '90%', transform=plt.gca().transAxes, color='r', fontsize=15)
output_name=output_dir+'/'+"rej_vs_"
if cType.find('GlobS')!=-1: output_name=output_dir+'/'+"eff_vs_"
output_name+=varname+"_"+cType
if evalLLH: output_name+="LLH"
output_name+=".png" #output_name+=varname+"_"+cType+".png"
print(output_name)
plt.savefig(output_name)
#plt.close()
if makeOutput: #pickle.dump(errGraphs, open(output_dir+cType+'_graphs.pickle', 'wb'))
outfilename=output_dir+'/'+cType
if evalLLH: outfilename+="LLH"
outfilename+='_graphs.pkl' #outfilename=output_dir+cType+'_graphs.pkl'
print('Writing pickle file:', outfilename)
pickle.dump(errGraphs,open(outfilename, 'wb'))
pass
return
def fill_bkg_rejs_f(bkg_rejs_fEff,bkg_errs_fEff,new_y_prob,new_test_labels,sigEffs):
for sigEff in sigEffs:
new_y_prob_sig = new_y_prob[new_test_labels==0]
new_y_prob_bkg = new_y_prob[new_test_labels==1]
#KM: get percentile cut value
pCut = np.percentile(new_y_prob_sig,(1-sigEff)*100,axis=0) [0] #70% from right side --> 1 - sig_eff
binaryClassified_bkg = new_y_prob_bkg[:,0] > pCut
bkgRej = 0
bkgRejErr = 0
if binaryClassified_bkg.sum()>0:
bkgRej = binaryClassified_bkg.size / binaryClassified_bkg.sum() #inverted bkg-eff as rejection
bkgRejErr = math.sqrt(bkgRej * (bkgRej-1) / binaryClassified_bkg.sum())
pass
bkg_rejs_fEff[sigEff].append(bkgRej)
bkg_errs_fEff[sigEff].append(bkgRejErr)
pass
return
def fill_info_g(bkg_rejs_gEff,bkg_errs_gEff,
sig_effs_gEff,sig_errs_gEff,
new_y_prob,new_test_labels,sigEffs,globalCuts):
for sigEff_target,globCut in zip(sigEffs,globalCuts):
new_y_prob_sig = new_y_prob[new_test_labels==0]
new_y_prob_bkg = new_y_prob[new_test_labels==1]
binaryClassified_bkg = new_y_prob_bkg[:,0] > globCut
binaryClassified_sig = new_y_prob_sig[:,0] > globCut
bkgRej = 0
bkgRejErr = 0
sigEff = 0
sigEffErr = 0
if binaryClassified_bkg.sum()>0:
bkgRej = binaryClassified_bkg.size / binaryClassified_bkg.sum() #inverted bkg-eff as rejection
sigEff = binaryClassified_sig.sum()/ binaryClassified_sig.size #sig-eff
bkgRejErr = math.sqrt(bkgRej * (bkgRej-1) / binaryClassified_bkg.sum())
sigEffErr = math.sqrt(sigEff * (1-sigEff) / binaryClassified_sig.size )
pass
bkg_rejs_gEff[sigEff_target].append(bkgRej)
bkg_errs_gEff[sigEff_target].append(bkgRejErr)
sig_effs_gEff[sigEff_target].append(sigEff)
sig_errs_gEff[sigEff_target].append(sigEffErr)
pass
return