-
Notifications
You must be signed in to change notification settings - Fork 1
/
DecomposingTest_10D.py
289 lines (241 loc) · 10.5 KB
/
DecomposingTest_10D.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
#!/usr/bin/env python
'''
This python script can be used to reproduce the results on 10D distributions
on the article Experiments using machine learning to approximate likelihood ratios
for mixture models (ACAT 2016).
'''
__author__ = "Pavez J. <juan.pavezs@alumnos.usm.cl>"
import ROOT
import numpy as np
from sklearn import svm, linear_model
from sklearn.externals import joblib
from sklearn.metrics import roc_curve, auc
from sklearn.ensemble import GradientBoostingClassifier
import sys
import os.path
import pdb
import matplotlib.pyplot as plt
from mpl_toolkits.mplot3d import Axes3D
from mlp import MLPTrainer
from make_data import makeData, makeModelND, makeModelPrivateND,\
makeModel
from utils import printMultiFrame, printFrame, saveFig, loadData,\
makeROC, makeSigBkg, makePlotName
from train_classifiers import trainClassifiers, predict
from decomposed_test import DecomposedTest
from xgboost_wrapper import XGBoostClassifier
'''
A simple example for the work on the section
5.4 of the paper 'Approximating generalized
likelihood ratio test with calibrated discriminative
classifiers' by Kyle Cranmer
'''
def evalC1C2Likelihood(test,c0,c1,dir='/afs/cern.ch/user/j/jpavezse/systematics',
workspace='workspace_DecomposingTestOfMixtureModelsClassifiers.root',
c1_g='',model_g='mlp',use_log=False,true_dist=False,vars_g=None,clf=None,
verbose_printing=False):
f = ROOT.TFile('{0}/{1}'.format(dir,workspace))
w = f.Get('w')
f.Close()
if true_dist == True:
vars = ROOT.TList()
for var in vars_g:
vars.Add(w.var(var))
x = ROOT.RooArgSet(vars)
else:
x = None
score = ROOT.RooArgSet(w.var('score'))
if use_log == True:
evaluateRatio = test.evaluateLogDecomposedRatio
post = 'log'
else:
evaluateRatio = test.evaluateDecomposedRatio
post = ''
npoints = 25
csarray = np.linspace(0.01,0.2,npoints)
cs2array = np.linspace(0.1,0.4,npoints)
testdata = np.loadtxt('{0}/data/{1}/{2}/{3}_{4}.dat'.format(dir,model_g,c1_g,'test','F1'))
decomposedLikelihood = np.zeros((npoints,npoints))
trueLikelihood = np.zeros((npoints,npoints))
c1s = np.zeros(c1.shape[0])
c0s = np.zeros(c1.shape[0])
pre_pdf = []
pre_dist = []
pre_pdf.extend([[],[]])
pre_dist.extend([[],[]])
for k,c0_ in enumerate(c0):
pre_pdf[0].append([])
pre_pdf[1].append([])
pre_dist[0].append([])
pre_dist[1].append([])
for j,c1_ in enumerate(c1):
if k <> j:
f0pdf = w.function('bkghistpdf_{0}_{1}'.format(k,j))
f1pdf = w.function('sighistpdf_{0}_{1}'.format(k,j))
outputs = predict('{0}/model/{1}/{2}/{3}_{4}_{5}.pkl'.format(dir,model_g,c1_g,
'adaptive',k,j),testdata,model_g=model_g,clf=clf)
f0pdfdist = np.array([test.evalDist(score,f0pdf,[xs]) for xs in outputs])
f1pdfdist = np.array([test.evalDist(score,f1pdf,[xs]) for xs in outputs])
pre_pdf[0][k].append(f0pdfdist)
pre_pdf[1][k].append(f1pdfdist)
else:
pre_pdf[0][k].append(None)
pre_pdf[1][k].append(None)
if true_dist == True:
f0 = w.pdf('f{0}'.format(k))
f1 = w.pdf('f{0}'.format(j))
if len(testdata.shape) > 1:
f0dist = np.array([test.evalDist(x,f0,xs) for xs in testdata])
f1dist = np.array([test.evalDist(x,f1,xs) for xs in testdata])
else:
f0dist = np.array([test.evalDist(x,f0,[xs]) for xs in testdata])
f1dist = np.array([test.evalDist(x,f1,[xs]) for xs in testdata])
pre_dist[0][k].append(f0dist)
pre_dist[1][k].append(f1dist)
# Evaluate Likelihood in different c1[0] and c1[1] values
for i,cs in enumerate(csarray):
for j, cs2 in enumerate(cs2array):
c1s[:] = c1[:]
c1s[0] = cs
c1s[1] = cs2
c1s[2] = 1.-cs-cs2
decomposedRatios,trueRatios = evaluateRatio(w,testdata,
x=x,plotting=False,roc=False,c0arr=c0,c1arr=c1s,true_dist=true_dist,
pre_evaluation=pre_pdf,
pre_dist=pre_dist)
if use_log == False:
decomposedLikelihood[i,j] = np.log(decomposedRatios).sum()
trueLikelihood[i,j] = np.log(trueRatios).sum()
else:
decomposedLikelihood[i,j] = decomposedRatios.sum()
trueLikelihood[i,j] = trueRatios.sum()
decomposedLikelihood = decomposedLikelihood - decomposedLikelihood.min()
X,Y = np.meshgrid(csarray, cs2array)
decMin = np.unravel_index(decomposedLikelihood.argmin(), decomposedLikelihood.shape)
min_value = [csarray[decMin[0]],cs2array[decMin[1]]]
if verbose_printing == True:
saveFig(X,[Y,decomposedLikelihood,trueLikelihood],makePlotName('comp','train',type='multilikelihood'),labels=['composed','true'],contour=True,marker=True,dir=dir,marker_value=(c1[0],c1[1]),print_pdf=True,min_value=min_value)
if true_dist == True:
trueLikelihood = trueLikelihood - trueLikelihood.min()
trueMin = np.unravel_index(trueLikelihood.argmin(), trueLikelihood.shape)
return [[csarray[trueMin[0]],cs2array[trueMin[1]]], [csarray[decMin[0]],cs2array[decMin[1]]]]
else:
return [[0.,0.],[csarray[decMin[0]],cs2array[decMin[1]]]]
def fitCValues(test,c0,c1,dir='/afs/cern.ch/user/j/jpavezse/systematics',
c1_g='',model_g='mlp',true_dist=False,vars_g=None,
workspace='workspace_DecomposingTestOfMixtureModelsClassifiers.root',
use_log=False, clf=None):
if use_log == True:
post = 'log'
else:
post = ''
n_hist_c = 200
keys = ['true','dec']
c1_ = dict((key,np.zeros(n_hist_c)) for key in keys)
c1_values = dict((key,np.zeros(n_hist_c)) for key in keys)
c2_values = dict((key,np.zeros(n_hist_c)) for key in keys)
fil2 = open('{0}/fitting_values_c1c2{1}.txt'.format(dir,post),'w')
for i in range(n_hist_c):
makeData(vars_g, c0,c1, num_train=200000,num_test=500,no_train=True,
workspace=workspace,dir=dir,c1_g=c1_g)
if i == 0:
verbose_printing = True
else:
verbose_printing = False
((c1_true,c2_true),(c1_dec,c2_dec)) = evalC1C2Likelihood(test,c0,c1,dir=dir,
c1_g=c1_g,model_g=model_g, true_dist=true_dist,vars_g=vars_g,
workspace=workspace,use_log=use_log,clf=clf, verbose_printing=verbose_printing)
print '2: {0} {1} {2} {3}'.format(c1_true, c1_dec, c2_true, c2_dec)
fil2.write('{0} {1} {2} {3}\n'.format(c1_true, c1_dec, c2_true, c2_dec))
fil2.close()
def plotCValues(test,c0,c1,dir='/afs/cern.ch/user/j/jpavezse/systematics',
c1_g='',model_g='mlp',true_dist=False,vars_g=None,
workspace='workspace_DecomposingTestOfMixtureModelsClassifiers.root',
use_log=False):
if use_log == True:
post = 'log'
else:
post = ''
n_hist_c = 200
keys = ['true','dec']
c1_values = dict((key,np.zeros(n_hist_c)) for key in keys)
c2_values = dict((key,np.zeros(n_hist_c)) for key in keys)
c1_2 = np.loadtxt('{0}/fitting_values_c1c2{1}.txt'.format(dir,post))
c1_values['true'] = c1_2[:,0]
c1_values['dec'] = c1_2[:,1]
c2_values['true'] = c1_2[:,2]
c2_values['dec'] = c1_2[:,3]
saveFig([],[c1_values['true'],c1_values['dec']],
makePlotName('c1c2','train',type='c1_hist{0}'.format(post)),hist=True,
axis=['signal weight'],marker=True,marker_value=c1[0],
labels=['true','composed'],x_range=[0.,0.2],dir=dir,
model_g=model_g,title='Histogram for estimated values signal weight',print_pdf=True)
saveFig([],[c2_values['true'],c2_values['dec']],
makePlotName('c1c2','train',type='c2_hist{0}'.format(post)),hist=True,
axis=['bkg. weight'],marker=True,marker_value=c1[1],
labels=['true','composed'],x_range=[0.1,0.4],dir=dir,
model_g=model_g,title='Histogram for estimated values bkg. weight',print_pdf=True)
if __name__ == '__main__':
# Setting the classifier to use
model_g = None
classifiers = {'svc':svm.NuSVC(probability=True),'svr':svm.NuSVR(),
'logistic': linear_model.LogisticRegression(),
'bdt':GradientBoostingClassifier(n_estimators=100, learning_rate=1.0,
max_depth=5, random_state=0),
'mlp':MLPTrainer(n_hidden=40, L2_reg=0.0001),
'xgboost': XGBoostClassifier(num_class=2, nthread=4, silent=1,
num_boost_round=100, eta=0.5, max_depth=4)}
clf = None
if (len(sys.argv) > 1):
model_g = sys.argv[1]
clf = classifiers.get(sys.argv[1])
if clf == None:
model_g = 'logistic'
clf = classifiers['logistic']
print 'Not found classifier, Using logistic instead'
# parameters of the mixture model
c0 = np.array([.0,.3, .7])
c1 = np.array([.1,.3, .7])
c1_g = ''
c0 = c0/c0.sum()
c1[0] = sys.argv[2]
if c1[0] < 0.01:
c1_g = "%.3f"%c1[0]
else:
c1_g = "%.2f"%c1[0]
c1[0] = (c1[0]*(c1[1]+c1[2]))/(1.-c1[0])
c1 = c1 / c1.sum()
verbose_printing = True
dir = '/afs/cern.ch/user/j/jpavezse/systematics'
workspace_file = 'workspace_DecomposingTestOfMixtureModelsClassifiers.root'
# features
vars_g = ['x0','x1','x2','x3','x4','x5','x6','x7','x8','x9']
ROOT.gROOT.SetBatch(ROOT.kTRUE)
ROOT.RooAbsPdf.defaultIntegratorConfig().setEpsRel(1E-15)
ROOT.RooAbsPdf.defaultIntegratorConfig().setEpsAbs(1E-15)
# Set this value to False if only final plots are needed
verbose_printing = True
if (len(sys.argv) > 3):
print 'Setting seed: {0} '.format(sys.argv[3])
ROOT.RooRandom.randomGenerator().SetSeed(int(sys.argv[3]))
np.random.seed(int(sys.argv[3]))
# make private mixture model
makeModelPrivateND(vars_g=vars_g,c0=c0,c1=c1,workspace=workspace_file,dir=dir,
model_g=model_g,verbose_printing=verbose_printing,load_cov=True)
# make sintetic data to train the classifiers
makeData(vars_g=vars_g,c0=c0,c1=c1,num_train=100000,num_test=50000,
workspace=workspace_file,dir=dir, c1_g=c1_g, model_g='mlp')
# train the pairwise classifiers
trainClassifiers(clf,3,dir=dir, model_g=model_g,
c1_g=c1_g ,model_file='adaptive')
# class which implement the decomposed method
test = DecomposedTest(c0,c1,dir=dir,c1_g=c1_g,model_g=model_g,
input_workspace=workspace_file, verbose_printing = verbose_printing,
dataset_names=['0','1','2'], clf=clf if model_g=='mlp' else None)
test.fit(data_file='test',importance_sampling=False, true_dist=True,vars_g=vars_g)
test.computeRatios(true_dist=True,vars_g=vars_g,use_log=False)
# compute likelihood for c0[0] and c0[1] values
fitCValues(test,c0,c1,dir=dir,c1_g=c1_g,model_g=model_g,true_dist=True,vars_g=vars_g,
workspace=workspace_file,use_log=False, clf=clf if model_g=='mlp' else None)
plotCValues(test,c0,c1,dir=dir,c1_g=c1_g,model_g=model_g,true_dist=True,vars_g=vars_g,
workspace=workspace_file,use_log=False)