-
Notifications
You must be signed in to change notification settings - Fork 175
/
Copy pathmodel.py
1271 lines (901 loc) · 51 KB
/
model.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
415
416
417
418
419
420
421
422
423
424
425
426
427
428
429
430
431
432
433
434
435
436
437
438
439
440
441
442
443
444
445
446
447
448
449
450
451
452
453
454
455
456
457
458
459
460
461
462
463
464
465
466
467
468
469
470
471
472
473
474
475
476
477
478
479
480
481
482
483
484
485
486
487
488
489
490
491
492
493
494
495
496
497
498
499
500
501
502
503
504
505
506
507
508
509
510
511
512
513
514
515
516
517
518
519
520
521
522
523
524
525
526
527
528
529
530
531
532
533
534
535
536
537
538
539
540
541
542
543
544
545
546
547
548
549
550
551
552
553
554
555
556
557
558
559
560
561
562
563
564
565
566
567
568
569
570
571
572
573
574
575
576
577
578
579
580
581
582
583
584
585
586
587
588
589
590
591
592
593
594
595
596
597
598
599
600
601
602
603
604
605
606
607
608
609
610
611
612
613
614
615
616
617
618
619
620
621
622
623
624
625
626
627
628
629
630
631
632
633
634
635
636
637
638
639
640
641
642
643
644
645
646
647
648
649
650
651
652
653
654
655
656
657
658
659
660
661
662
663
664
665
666
667
668
669
670
671
672
673
674
675
676
677
678
679
680
681
682
683
684
685
686
687
688
689
690
691
692
693
694
695
696
697
698
699
700
701
702
703
704
705
706
707
708
709
710
711
712
713
714
715
716
717
718
719
720
721
722
723
724
725
726
727
728
729
730
731
732
733
734
735
736
737
738
739
740
741
742
743
744
745
746
747
748
749
750
751
752
753
754
755
756
757
758
759
760
761
762
763
764
765
766
767
768
769
770
771
772
773
774
775
776
777
778
779
780
781
782
783
784
785
786
787
788
789
790
791
792
793
794
795
796
797
798
799
800
801
802
803
804
805
806
807
808
809
810
811
812
813
814
815
816
817
818
819
820
821
822
823
824
825
826
827
828
829
830
831
832
833
834
835
836
837
838
839
840
841
842
843
844
845
846
847
848
849
850
851
852
853
854
855
856
857
858
859
860
861
862
863
864
865
866
867
868
869
870
871
872
873
874
875
876
877
878
879
880
881
882
883
884
885
886
887
888
889
890
891
892
893
894
895
896
897
898
899
900
901
902
903
904
905
906
907
908
909
910
911
912
913
914
915
916
917
918
919
920
921
922
923
924
925
926
927
928
929
930
931
932
933
934
935
936
937
938
939
940
941
942
943
944
945
946
947
948
949
950
951
952
953
954
955
956
957
958
959
960
961
962
963
964
965
966
967
968
969
970
971
972
973
974
975
976
977
978
979
980
981
982
983
984
985
986
987
988
989
990
991
992
993
994
995
996
997
998
999
1000
import numpy as np
from sklearn.model_selection import StratifiedKFold
from pivottablejs import pivot_ui
import warnings
import GPyOpt
import GPy
from IPython.display import Markdown, display, Image
import logging
import copy
import time
from tqdm import tnrange
from models.imputers import missForest, matrix_completion, mean, median
from models.imputers import most_frequent
from models.classifiers import RandomForest, RandomForestClassifier
from models.classifiers import GradientBoosting, GradientBoostingClassifier
from models.classifiers import XGboost, Adaboost, Bagging, BernNaiveBayes
from models.classifiers import BernoulliNB, GaussNaiveBayes
from models.classifiers import MultinomialNaiveBayes, MultinomialNB
from models.classifiers import LogisticReg
from models.classifiers import Percept, DecisionTrees, DecisionTreeClassifier
from models.classifiers import QDA_, LDA_, KNN, LinearSVM, NeuralNet
from models.classifiers import classifier_set_xtr_arg, classifier_get_xtr_arg
from models.preprocessors import Scaler, MinMaxScaler, UniformTransform
from models.preprocessors import GaussianTransform, FeatureNormalizer
from models.preprocessors import GaussProjection, PrincipalComponentAnalysis
from pipelines.basePipeline import basePipeline
import sys
if not sys.warnoptions:
warnings.simplefilter("ignore")
import initpath_ap
initpath_ap.init_sys_path()
import utilmlab
def is_ipython():
try:
__IPYTHON__
return True
except NameError:
return False
class evaluate:
'''
wrapper class: returns the score as set by the user.
'''
def __init__(self):
self.m_metric = None
self.m_metric_allowed = ['aucprc', 'aucroc']
def set_metric(self, metric):
assert metric in self.m_metric_allowed
self.m_metric = metric
def get_metric(self):
assert self.m_metric is not None
return self.m_metric
def score(self, y_test, y_pred, y_pred_proba):
assert y_test is not None
assert y_pred_proba is not None
if self.m_metric == 'aucprc':
score_val = self.average_precision_score(y_test, y_pred_proba)
elif self.m_metric == 'aucroc':
score_val = self.roc_auc_score(y_test, y_pred_proba)
else:
assert 0
logger.debug('evaluate:{:0.5f}'.format(score_val))
return score_val
def score_proba(self, y_test, y_pred_proba):
assert y_test is not None
assert y_pred_proba is not None
return self.score(y_test, None, y_pred_proba)
def roc_auc_score(self, y_test, y_pred_proba):
return utilmlab.evaluate_auc(y_test, y_pred_proba)[0]
def average_precision_score(self, y_test, y_pred_proba):
return utilmlab.evaluate_auc(y_test, y_pred_proba)[1]
def accuracy_score(self, y_true, y_pred, y_pred_proba):
y_hat = np.argmax(y_pred_proba, axis=1)
return sum(y_true == y_hat) / float(len(y_true))
def set_xtr_arg(xtr):
global model_xtr_arg
model_xtr_arg = xtr
classifier_set_xtr_arg(xtr)
def show_xtr_arg():
print(model_xtr_arg)
print(classifier_get_xtr_arg())
def get_xtr_arg():
return model_xtr_arg
def printmd(string):
global gis_ipython
if gis_ipython:
display(Markdown(string))
logger.info('{}'.format(string))
class AutoPrognosis_Classifier:
def __init__(
self,
CV=5,
num_iter=100,
kernel_freq=10,
ensemble=True,
ensemble_size=10,
Gibbs_iter=100,
burn_in=50,
num_components=3,
is_nan=True,
metric='aucroc',
acquisition_type='LCB',
**kwargs):
eva.set_metric(metric)
self.is_pred_proba = True
self.acquisition_type = acquisition_type
self.is_nan = is_nan
self.num_iter = num_iter
self.x_opt = 0
self.CV = CV
self.Gibbs_iter = Gibbs_iter
self.burn_in = burn_in
self.num_components = num_components
self.kernel_freq = kernel_freq
self.ensemble_size = ensemble_size
self.ensemble = ensemble
self.model_ = []
self.models_ = []
self.scores_ = []
self.model_names = ['Random Forest', 'Gradient Boosting', 'XGBoost', 'Adaboost', 'Bagging', 'Bernoulli Naive Bayes',
'Gauss Naive Bayes', 'Multinomial Naive Bayes', 'Logistic Regression','Perceptron',
'Decision Trees','QDA','LDA','KNN','Linear SVM', 'Neural Network']
self.initialize_hyperparameter_map()
self.name = 'autoprognosis_clf'
self.exe_start_time = time.time()
def initialize_hyperparameter_map(self):
self.modind = [0,1,2,3,4,5,7,8,9,10,13,14,15] # indexes of models with hyperparam
self.noHyp = [6,11,12] # indexes of models without hyperparam
self.hypMAP = [[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]]
def get_model(self,dom_,comp_no,comp_map,x_next):
imputers_ = [
[],
[missForest()],
[mean()],
[median()],
[most_frequent()]
]
preprocessors_= [
[],
[MinMaxScaler()],
[UniformTransform()],
[PrincipalComponentAnalysis()]
]
if get_xtr_arg() < 1:
imputers_.append([matrix_completion()])
# WA: bagging with base estimator cannot handle negative values
preprocessors_.append([Scaler()])
preprocessors_.append([GaussProjection()])
preprocessors_.append([GaussianTransform()])
preprocessors_.append([FeatureNormalizer()])
select_imp = np.random.randint(
1 if self.is_nan else 0,
len(imputers_)-1)
select_pre = np.random.randint(0, len(preprocessors_)-1)
# Create hyper-parameter dictionary
#----------------------------------
base_ = [DecisionTreeClassifier(), RandomForestClassifier(), BernoulliNB(), MultinomialNB(),
GradientBoostingClassifier()]
LR_solvs = ['newton-cg', 'lbfgs', 'liblinear', 'sag', 'saga']
ppens = ['l2','l1','elasticnet']
dtcr = ['gini','entropy']
knnalgs = ['auto','ball_tree','kd_tree','brute']
knnwgs = ['uniform','distance']
svmpen = ['l2','l1']
nnsolvers = ['lbfgs','sgd','adam']
nnactiv = ['identity','logistic','tanh','relu']
crit_ = ['gini','entropy']
# Unpack model names
#----------------------------------
domain_list = [dom_[k]['name'] for k in range(len(dom_))]
pivot_ = domain_list.index('classifier')
mdl_index = comp_map[comp_no][int(x_next[pivot_])]
# Create a list of model instantiations
#--------------------------------------
if mdl_index==0:
model = RandomForest(n_estimators=int(x_next[domain_list.index('RandomForest.ntrees')]),
criterion=crit_[int(x_next[domain_list.index('RandomForest.criterion')])])
elif mdl_index==1:
model = GradientBoosting(n_estimators=int(x_next[domain_list.index('GradientBoosting.n_estimators')]),
learning_rate=x_next[domain_list.index('GradientBoosting.learning_rate')],
max_depth=int(x_next[domain_list.index('GradientBoosting.max_depth')]))
elif mdl_index==2:
model = XGboost(n_estimators=int(x_next[domain_list.index('XGboost.n_estimators')]),
learning_rate=x_next[domain_list.index('XGboost.learning_rate')],
max_depth=int(x_next[domain_list.index('XGboost.max_depth')]))
elif mdl_index==3:
model = Adaboost(n_estimators=int(x_next[domain_list.index('Adaboost.n_estimators')]),
learning_rate=x_next[domain_list.index('Adaboost.learning_rate')])
elif mdl_index==4:
model = Bagging(n_estimators=int(x_next[domain_list.index('Bagging.n_estimators')]),
max_samples=x_next[domain_list.index('Bagging.max_samples')],
max_features=1.0,#x_next[domain_list.index('Bagging.max_features')],
base_estimator=base_[int(x_next[domain_list.index('Bagging.base_estimator')])])
elif mdl_index==5:
model = BernNaiveBayes(alpha=x_next[domain_list.index('BernoullinNaiveBayes.alpha')])
elif mdl_index==6:
model = GaussNaiveBayes()
elif mdl_index==7:
model = MultinomialNaiveBayes(alpha=x_next[domain_list.index('MultinomialNaiveBayes.alpha')])
elif mdl_index==8:
model = LogisticReg(C=x_next[domain_list.index('LogisticRegression.C')],
solver=LR_solvs[int(x_next[domain_list.index('LogisticRegression.solver')])],
max_iter=int(x_next[domain_list.index('LogisticRegression.max_iter')]))
elif mdl_index==9:
model = Percept(alpha=x_next[domain_list.index('Perceptron.alpha')],
penalty=ppens[int(x_next[domain_list.index('Perceptron.penalty')])])
elif mdl_index==10:
model = DecisionTrees(criterion=dtcr[int(x_next[domain_list.index('DecisionTrees.criterion')])])
elif mdl_index==11:
model = QDA_()
elif mdl_index==12:
model = LDA_()
elif mdl_index==13:
model = KNN(p=int(x_next[domain_list.index('KNN.p')]),
algorithm=knnalgs[int(x_next[domain_list.index('KNN.algorithm')])],
weights=knnwgs[int(x_next[domain_list.index('KNN.weights')])],
n_neighbors=int(x_next[domain_list.index('KNN.n_neighbors')]))
elif mdl_index==14:
model = LinearSVM(penalty=svmpen[int(x_next[domain_list.index('LinearSVM.penalty')])])
elif mdl_index==15:
model = NeuralNet(num_layers=int(x_next[domain_list.index('NeuralNet.num_layers')]),
num_units=int(x_next[domain_list.index('NeuralNet.num_units')]),
solver=nnsolvers[int(x_next[domain_list.index('NeuralNet.solver')])],
activation=nnactiv[int(x_next[domain_list.index('NeuralNet.activation')])])
model_list_base = imputers_[select_imp] + preprocessors_[select_pre] + [model]
if nmax_model:
model_list = model_list_base[-nmax_model:]
else:
# auto
model_list = list()
if self.is_nan:
model_list.append(model_list_base[0]) # add imputer
model_list.append(model_list_base[-1]) # add clf
model = basePipeline(model_list=model_list)
return model
def get_opt_domain(self):
assert 0
# Define domains for classifier sets
base_models_ = [RandomForest(),GradientBoosting(),XGboost(),Adaboost(),Bagging(),BernNaiveBayes(),
GaussNaiveBayes(),MultinomialNaiveBayes(),LogisticReg(),Percept(),DecisionTrees(),
QDA_(),LDA_(),KNN(),LinearSVM(),NeuralNet()]
domain_ = [{'name': 'classifier', 'type': 'categorical',
'domain': (0,1,2,3,4,5,6,7,8,9,10,11,12,13,14,15),'dimensionality': 1}]
for bmodel_ in base_models_:
domain_ = domain_ + bmodel_.get_hyperparameter_space()
dim_ = len(domain_)
return domain_, dim_
def evaluate_CV_objective(self, X_in, Y_in, modraw_):
mod_back = modraw_
#mod_back.fit(X_in, Y_in)
rval_eva = evaluate_clf(X_in.copy(), Y_in.copy(), copy.deepcopy(modraw_), n_folds = self.CV)
logger.info('CV_objective:{}'.format(rval_eva))
f = -1*rval_eva[0][0]
return f, mod_back, rval_eva[1]
#----------------------
def load_initial_model(self):
# [RF 0 (1,2),GBM 1 (3,4,5),XGB 2 (6,7,8),Ada 3 (9,10),Bag 4 (11,12,13,14), BNB 5 (15),
# GNB 6 [], MNB 7 (16), LR 8 (17,18,19), Perc 9 (20,21), DT 10 (22), QDA_ 11 (),
# LDA_ 12 (), KNN 13 (23,24,25,26), LSVM 14 (27), NN 15 (28,29,30,31)]
init_assigns = [[1,0,0],[1,0,0],[1,0,0],[1,0,0],[1,0,0],[1,0,0],[1,0,0],[1,0,0],
[0,1,0],[0,1,0],[0,1,0],[0,1,0],[0,1,0],[0,1,0],[0,1,0],[0,1,0],
[0,0,1],[0,0,1],[0,0,1],[0,0,1],[0,0,1],[0,0,1],[0,0,1],[0,0,1],[0,0,1],[0,0,1],[0,0,1],
[0,1,0],[0,1,0],[0,1,0],[0,1,0]]
X_inits = [np.array([[1,100,6,0.1,100,6,0.1,100,0,2,100,0,0]]),
np.array([[0,1,1,100,1,1,0,100,1]]),
np.array([[0,1,0,0,0,30,0,1,0,1,0,100]])]
return X_inits, init_assigns
def BO_(self,domains_,X_step,Y_step):
bo_steps = []
x_next = []
GP_ = []
for u in range(len(domains_)):
# GPyOpt.methods.BayesianOptimization(
# None,
# domain=domains_[u],
# X=X_step[u],
# Y=Y_step[u],
# acquisition_type='LCB')
bo_steps.append(GPyOpt.methods.BayesianOptimization(
f=None, domain=domains_[u], X=X_step[u],
Y=Y_step[u],
acquisition_type=self.acquisition_type,
model_type='GP', exact_feval=True,
cost_withGradients=None
))
x_next.append(bo_steps[-1].suggest_next_locations()[0])
GP_.append(bo_steps[-1].model.model)
return x_next, GP_
def BO_ens(self,domains_,X_step,Y_step,constr_):
bo_steps = []
x_next = []
GP_ = []
for u in range(len(domains_)):
bo_steps.append(GPyOpt.methods.BayesianOptimization(
f=None,
domain=domains_[u],
constraints=constr_,
X=X_step[u],
Y=Y_step[u],
acquisition_type=self.acquisition_type,
model_type='GP',
exact_feval=True,
cost_withGradients=None
))
x_next.append(bo_steps[-1].suggest_next_locations()[0])
GP_.append(bo_steps[-1].model.model)
return x_next, GP_
def fit(self, X, Y):
# -------------------------------------------
# Initial decomposition and hyper-parameters
# -------------------------------------------
X_inits, self._assigns = self.load_initial_model()
domains_, dummy_, self.compons_ = get_clustered_domains(self.modind,self.hypMAP,self.noHyp,self._assigns)
self.domains_ = domains_
# -------------------------------------------
# Obtain initial BO objective
# -------------------------------------------
Y_inits = [np.array([self.evaluate_CV_objective(X.copy(), Y.copy(), self.get_model(domains_[k],k,self.compons_,X_inits[k][0]))[0]]).reshape((1,1)) for k in range(len(X_inits))]
X_step = X_inits
Y_step = Y_inits
self.scores_ = list(Y_inits[0])
eva_prop_ = list([{
'name': 'initial',
eva.get_metric(): -float(self.scores_[0])
}]) # element 0
assert len(self.scores_) == len(eva_prop_)
# -------------------------------------------
# Initialize BO iterations counter
# -------------------------------------------
current_iter = 0
time_start_fit = time.time()
assert len(self.model_) == 0
for i in tnrange(self.num_iter, desc='BO progress'):
x_next, GP_ = self.BO_(self.domains_,X_step,Y_step)
self.GP_ = GP_
y_next = []
for u in range(self.num_components):
self.models_.append(self.get_model(self.domains_[u],u,self.compons_,x_next[u]))
rval_model = self.get_model(self.domains_[u],u,self.compons_,x_next[u])
y_next_, modb_, eva_prp = self.evaluate_CV_objective(X.copy(), Y.copy(), rval_model)
eva_prp['iter'] = current_iter
eva_prp['component_idx'] = u
eva_prp['hyperparameter_properties'] = rval_model.get_properties()
eva_prp['model'] = str(rval_model)
y_next.append(y_next_)
self.models_.append(modb_)
X_step[u] = np.vstack((X_step[u], x_next[u]))
Y_step[u] = np.vstack((Y_step[u], y_next_))
self.scores_.append(y_next_)
eva_prop_.append(eva_prp)
assert len(self.scores_) == len(eva_prop_)
current_iter += 1
time_fit = time.time() - time_start_fit
time_fold = (self.num_iter) * (time_fit/current_iter)
time_iter_info = '{:0.0f}s ({:0.0f}s) ({:0.0f}s)'.format(
time_fit,
time_fit/current_iter,
time_fold)
### Update Kernel
if (np.mod(i,self.kernel_freq) == 0) and (i != 0):
msg_ = "Iteration number: " + str(current_iter) + " " + time_iter_info + " ||--------------------------- ** Updating kernel ** ---------------------------|| "
logger.warning('{}'.format(msg_))
self.merged_domain_, self.X_merged, self.Y_merged = merge_domains(self.domains_, X_step, Y_step, self.compons_)
Decomposed_kern, Gposter_ = self.Kernel_decomposition([self.X_merged], [self.Y_merged], [self.merged_domain_])
self.Gposter_ = Gposter_
self.kernel_merged_domain_ = self.merged_domain_
self.domains_ , dummy_dims, self.compons_ = get_clustered_domains(self.modind, self.hypMAP, self.noHyp, Gposter_)
X_step, Y_step = split_domain(self.domains_, self.merged_domain_, self.compons_, self.X_merged, self.Y_merged)
self.X_step = X_step
self.Y_step = Y_step
##########
else:
display_msg = " "
for u in range(self.num_components):
display_msg = display_msg + "[[" + self.get_model(self.domains_[u],u,self.compons_,x_next[u]).name + "]]" + ", "
best_ = str(np.min(np.array([np.min(self.GP_[u].Y) for u in range(self.num_components)])))
msg_ = "Iteration number: " + str(current_iter) + " " + time_iter_info + ", Current pipelines: " + display_msg + "BO objective: " + str(best_)
logger.warning('{}'.format(msg_))
self.X_step = X_step
self.Y_step = Y_step
#self.x_opts_ = [X_step[u][np.argmin(self.GP_[u].Y)] for u in range(self.num_components)]
#self.y_opts_ = np.array([np.min(self.GP_[u].Y) for u in range(self.num_components)])
#best_comp = np.argmin(self.y_opts_)
#self.x_opt = self.x_opts_[best_comp]
#self.model = self.get_model(self.domains_[best_comp],best_comp,self.compons_,self.x_opt)
self.merged_domain_, self.X_merged, self.Y_merged = merge_domains(self.domains_, X_step, Y_step, self.compons_)
dummy_x, final_GPmodel = self.BO_([self.merged_domain_],[self.X_merged], [self.Y_merged])
self.GPmodel_ = final_GPmodel
self.x_opt = self.X_merged[np.argmin(final_GPmodel[0].Y)]
self.model = self.get_model(self.merged_domain_, 0, [list(range(len(self.modind)+len(self.noHyp)))], self.x_opt)
printmd("\r{0}".format("**The best model is: **"+self.model.name))
self.model.fit(X.copy(), Y.copy())
if self.ensemble:
self.ensemble_score = 0
printmd("\r{0}".format(" |||| Now building the ensemble..."))
self.ensemble_models, self.ensemble_weights = self.get_ensemble_(X.copy(), Y.copy(), CV_=5)
#self.ensemble_models, self.ensemble_weights = self.build_ensemble_(X, Y, CV_=5)
printmd("\r{0}".format("**Ensemble: **"+ str([self.ensemble_models[k].name for k in range(len(self.ensemble_models))])))
printmd("\r{0}".format("**Ensemble weights: **"+ str(self.ensemble_weights)))
best_score = float(np.min(np.array(self.scores_)))
logger.info('score:{:0.3f} {} ensemble score:{} {}'.format(
best_score,
self.model.name,
str(self.ensemble_score),
str([self.ensemble_models[k].name for k in range(len(self.ensemble_models))])))
if self.ensemble_score < best_score:
printmd("\r{0}".format("**The ensemble helps!**"))
else:
printmd("\r{0}".format("**The ensemble did not help.**"))
return eva_prop_
#----------------------
def predict(self,X):
pred = self.model.predict(X)
#W = self.get_weights()
#bigpreds_ = [self.models_[k].predict(X)*W[k] for k in range(self.num_iter)]
pred_ens = pred#np.sum(np.array(bigpreds_),axis=0)
return pred, pred_ens
#----------------------
def get_ensemble(self,X_, Y_label):
ensemble_size = self.ensemble_size
BO_num_iter = 30
x_opts_ = self.X_merged[np.argsort(np.array(self.GPmodel_[0].Y),axis=0)[0:ensemble_size]]
pulled_models_ = [self.get_model(self.merged_domain_, 0, [list(range(len(self.modind)+len(self.noHyp)))], x_opts_[mm_].reshape(x_opts_[mm_].shape[1],)) for mm_ in range(len(x_opts_))]
for k in range(len(pulled_models_)):
pulled_models_[k].fit(X_.copy(),Y_label)
ens_domain = [{'name': 'W'+str(k), 'type': 'continuous', 'domain': (0,1),'dimensionality': 1} for k in range(ensemble_size)]
X_inits = (1/ensemble_size)*np.ones((1,ensemble_size))
assert 0
Y_inits = evaluate_ensemble(X_, Y_label, pulled_models_, X_inits, 'AUC')
X_step_ens = X_inits
Y_step_ens = np.array([Y_inits]).reshape(1,1)
y_next = []
for i in range(BO_num_iter):
x_next, GP_ens = self.BO_([ens_domain], [X_step_ens], [Y_step_ens])
x_next = x_next/np.sum(x_next)
y_next_ = evaluate_ensemble(X_, Y_label, pulled_models_, x_next, 'AUC')
y_next.append(y_next_)
X_step_ens = np.vstack((X_step_ens, x_next))
Y_step_ens = np.vstack((Y_step_ens, y_next_))
ensemble_weights = X_step_ens[np.argmin(GP_ens[0].Y)]
return pulled_models_, ensemble_weights
def get_ensemble_(self,X_, Y_label, CV_):
ensemble_size = self.ensemble_size
BO_num_iter = 30
x_opts_ = self.X_merged[np.argsort(np.array(self.GPmodel_[0].Y),axis=0)[0:ensemble_size]]
ens_models_ = [self.get_model(self.merged_domain_, 0, [list(range(len(self.modind)+len(self.noHyp)))], x_opts_[mm_].reshape(x_opts_[mm_].shape[1],)) for mm_ in range(len(x_opts_))]
pulled_models_ = [[self.get_model(self.merged_domain_, 0, [list(range(len(self.modind)+len(self.noHyp)))], x_opts_[mm_].reshape(x_opts_[mm_].shape[1],))]*CV_ for mm_ in range(len(x_opts_))]
skf_ = StratifiedKFold(n_splits=CV_)
X_test_ = []
Y_test_ = []
X_train_ = []
Y_train_ = []
for train_index, test_index in skf_.split(X_, Y_label):
X_train_.append(X_.loc[X_.index[train_index]])
Y_train_.append(Y_label.loc[Y_label.index[train_index]])
X_test_.append(X_.loc[X_.index[test_index]])
Y_test_.append(Y_label.loc[Y_label.index[test_index]])
for k in range(len(pulled_models_)):
ens_models_[k].fit(X_.copy(), Y_label.copy())
ee = 0
pulled_models_init = list()
for idx in range(CV_):
pulled_models_init.append(copy.deepcopy(pulled_models_[k][idx]))
for uu in range(CV_):
# ee = 0
pulled_models_[k][uu] = pulled_models_init[uu]
pulled_models_[k][uu].fit(X_train_[uu].copy(),Y_train_[uu].copy())
ee += 1
ens_domain = [{'name': 'W'+str(k), 'type': 'continuous', 'domain': (0,1),'dimensionality': 1} for k in range(ensemble_size)]
X_inits = (1/ensemble_size)*np.ones((1,ensemble_size))
Y_inits = evaluate_ensemble_(X_test_, Y_test_, pulled_models_, X_inits, 'AUC')
X_step_ens = X_inits
Y_step_ens = np.array([Y_inits]).reshape(1,1)
y_next = [Y_inits]
for i in range(BO_num_iter):
x_next, GP_ens = self.BO_([ens_domain], [X_step_ens], [Y_step_ens])
x_next = x_next/np.sum(x_next)
y_next_ = evaluate_ensemble_(X_test_, Y_test_, pulled_models_, x_next, 'AUC')
y_next.append(y_next_)
X_step_ens = np.vstack((X_step_ens, x_next))
Y_step_ens = np.vstack((Y_step_ens, y_next_))
self.ensemble_score = np.min(np.array(y_next))
ensemble_weights = X_step_ens[np.argmin(GP_ens[0].Y)]
return ens_models_, ensemble_weights
def build_ensemble_(self,X_, Y_label, CV_):
ensemble_size = self.ensemble_size
BO_num_iter = 50
#x_opts_ = self.X_merged[np.unique(np.argsort(np.array(self.GPmodel_[0].Y),axis=0),return_index=True)[1][0:ensemble_size]]
x_opts_ = self.X_merged[np.argsort(np.array(self.GPmodel_[0].Y),axis=0)[0:ensemble_size]]
ens_models_ = [self.get_model(self.merged_domain_, 0, [list(range(len(self.modind)+len(self.noHyp)))], x_opts_[mm_].reshape(x_opts_[mm_].shape[1],)) for mm_ in range(len(x_opts_))]
pulled_models_ = [[self.get_model(self.merged_domain_, 0, [list(range(len(self.modind)+len(self.noHyp)))], x_opts_[mm_].reshape(x_opts_[mm_].shape[1],))]*CV_ for mm_ in range(len(x_opts_))]
#ens_models_ = [self.get_model(self.merged_domain_, 0, [list(range(len(self.modind)+len(self.noHyp)))], x_opts_[mm_]) for mm_ in range(len(x_opts_))]
#pulled_models_ = [[self.get_model(self.merged_domain_, 0, [list(range(len(self.modind)+len(self.noHyp)))], x_opts_[mm_])]*CV_ for mm_ in range(len(x_opts_))]
skf_ = StratifiedKFold(n_splits=CV_)
X_test_ = []
Y_test_ = []
X_train_ = []
Y_train_ = []
for train_index, test_index in skf_.split(X_, Y_label):
X_train_.append(X_.loc[X_.index[train_index]])
Y_train_.append(Y_label.loc[Y_label.index[train_index]])
X_test_.append(X_.loc[X_.index[test_index]])
Y_test_.append(Y_label.loc[Y_label.index[test_index]])
for k in range(len(pulled_models_)):
ens_models_[k].fit(X_, Y_label)
for uu in range(CV_):
ee = 0
pulled_models_[k][ee].fit(X_train_[ee],Y_train_[ee])
ee += 1
ens_domain = [{'name': 'W'+str(k), 'type': 'continuous', 'domain': (0,1),'dimensionality': 1} for k in range(ensemble_size)]
X_inits = (1/ensemble_size)*np.ones((1,ensemble_size))
Y_inits = evaluate_ensemble_(X_test_, Y_test_, pulled_models_, X_inits, 'AUC')
X_step_ens = X_inits
Y_step_ens = np.array([Y_inits]).reshape(1,1)
y_next = []
for i in range(BO_num_iter):
x_next, GP_ens = self.BO_ens([ens_domain], [X_step_ens], [Y_step_ens], get_ensemble_constraints(ensemble_size))
x_next = x_next/np.sum(x_next)
y_next_ = evaluate_ensemble_(X_test_.copy(), Y_test_.copy(), pulled_models_, x_next, 'AUC')
y_next.append(y_next_)
X_step_ens = np.vstack((X_step_ens, x_next))
Y_step_ens = np.vstack((Y_step_ens, y_next_))
ensemble_weights = X_step_ens[np.argmin(GP_ens[0].Y)]
return ens_models_, ensemble_weights
def get_weights(self):
#mu_ = self.GP_.model.model.predict(self.GP_.model.model.X)[0]
#vr_ = self.GP_.model.model.predict(self.GP_.model.model.X)[1]
#M_ = -np.repeat(mu_, len(mu_), axis=1) + np.transpose(np.repeat(mu_, len(mu_), axis=1))
#V_ = np.sqrt(np.repeat(vr_, len(vr_), axis=1) + np.transpose(np.repeat(vr_, len(vr_), axis=1)))
#Args_ = norm.cdf(np.divide(M_,V_))
#W_ = [np.prod(Args_[u, [x for x in range(len(Args_[u,:])) if x != u]]) for u in range(len(Args_))]
W = np.array([-1*(self.scores_[k]<0)*self.scores_[k] for k in range(len(self.scores_))])
W_ = (W-np.min(W))/np.sum((W-np.min(W)))
return W_
def Kernel_decomposition(self,dataX,dataY,domn_):
num_hyperparams = len(domn_[0])-1
M = self.num_components
alphas_ = [10]*M
init_assign = [np.argmax(self._assigns[k])+1 for k in range(len(self._assigns))]
Late_assign = [init_assign]
Gibbs_iters = self.Gibbs_iter
Burn_in = self.burn_in
GibbsPoster = []
dummy_x, GPmodel = self.BO_(domn_, dataX, dataY)
for g in range(Gibbs_iters):
for h in range(num_hyperparams):
Gumbel_dummy = np.random.gumbel(size=M)
Temp_assigns = np.repeat(np.array(Late_assign[-1]), M, axis=0).reshape(num_hyperparams,M).T
Temp_assigns[:,h] = np.array(list(range(M)))+1
Gram_matxs = [get_GPy_logLikelhood(dataX[0],dataY[0],get_Gibbs_kernel_(GPmodel[0], M, list(Temp_assigns[m,:]))) for m in range(M)]
A_m = [[np.sum((Temp_assigns[k,:]==m+1)*1) for m in range(M)] for k in range(M)]
Gumbal_max_vec = Gumbel_dummy + Gram_matxs + 0.5*(num_hyperparams+1)*np.log(2*np.pi) + np.log(np.diag(np.matrix(A_m))+alphas_)
z_assign = np.argmax(Gumbal_max_vec)
Temp_Late_assign = Late_assign[-1].copy()
Temp_Late_assign[h] = z_assign + 1
Late_assign.append(Temp_Late_assign)
Late_assign_ = Late_assign[Burn_in:][::num_hyperparams]
GibbsPoster = [[len(np.where(np.array(Late_assign_)[:,k]==m+1)[0])/len(np.array(Late_assign_)[:,k]) for m in range(M)] for k in range(num_hyperparams)]
final_assignments = [int(np.argmax(np.array(GibbsPoster[u]))+1) for u in range(len(GibbsPoster))]
M_ = len(np.unique(final_assignments))
Decomposed_kern = get_Gibbs_kernel_(GPmodel[0],M_,final_assignments)
return Decomposed_kern, GibbsPoster
# Visualization
def visualize_data(self,X):
return pivot_ui(X)
def APReport(self):
report_d = dict()
best_score = float(np.min(np.array(self.scores_)))
report_d['best_score_single_pipeline'] = -best_score
report_d['model_names_single_pipeline'] = self.model.name
report_d['ensemble_score'] = -float(self.ensemble_score)
report_d['ensemble_pipelines'] = [self.ensemble_models[u].name for u in range(self.ensemble_size)]
report_d['ensemble_pipelines_weight'] = [self.ensemble_weights[u] for u in range(self.ensemble_size)]
report_d['optimisation_metric'] = eva.get_metric()
report_d['hyperparameter_properties'] = self.model.get_properties()
report_d['acquisition_type'] = self.acquisition_type
logger.info('score:{:0.3f} {} ensemble score:{:0.3f} {}'.format(
best_score,
self.model.name,
self.ensemble_score,
str([self.ensemble_models[k].name for k in range(len(self.ensemble_models))])))
if self.ensemble_score < best_score:
logger.info("**The ensemble helps!**")
else:
logger.info("**The ensemble did not help.**")
printmd("\r{0}".format("***Ensemble Report***"))
printmd("\r{0}".format("**----------------------**"))
for u in range(self.ensemble_size):
printmd("\r{0}".format("**Rank"+ str(u) +": "+ self.ensemble_models[u].name+", Ensemble weight: "+ str(self.ensemble_weights[u]) +"**")) # print picture and colored text
printmd("\r{0}".format("**----------------------**"))
print(self.ensemble_models[u].__dict__)
logger.info('ensemble_models dict:{}'.format(self.ensemble_models[u].__dict__))
printmd("\r{0}".format("**_____________________________________________**"))
printmd("\r{0}".format(self.ensemble_models[u].explained))
if self.ensemble_models[u].image_name is not None:
i = Image(filename=self.ensemble_models[u].image_name,width=self.ensemble_models[u].image_size[0],height=self.ensemble_models[u].image_size[1])
display(i,width=self.ensemble_models[u].image_size[0],height=self.ensemble_models[u].image_size[1])
else:
logger.debug('warning:no image_name')
printmd("\r{0}".format("**----------------------**"))
printmd("\r{0}".format("***Kernel Report***"))
report_d['kernel_members'] = dict()
for k in range(self.num_components):
report_d['kernel_members'][k] = [self.model_names[self.compons_[k][m]] for m in range(len(self.compons_[k]))]
printmd("\r{0}".format("**Component"+" "+str(k)+"**"))
printmd("\r{0}".format("**Members:"+" "+str([self.model_names[self.compons_[k][m]] for m in range(len(self.compons_[k]))])+"**"))
print(self.GP_[k].kern)
logger.info('GP kern:{}'.format(self.GP_[k].kern))
return report_d
# In[141]:
def predict_ens(X, models_, W):
bigpreds_ = [models_[k].predict(X)*W[0][k] for k in range(len(models_))]
pred_ens = np.sum(np.array(bigpreds_),axis=0)
return pred_ens
def evaluate_ensemble(X, Y, models_, W, metric):
assert 0
score_ens = -1*roc_auc_score(Y,predict_ens(X, models_, W))
return score_ens
#----------------------
def predict_ens_(X, models_, W):
# logger.info('predict_ens_{}'.format(models_))
# for k in range(len(models_)):
# pred = models_[k].predict(X)
# logger.info('pred {}'.format(pred.shape))
# logger.info('W0 {}'.format(W[0]))
bigpreds_ = [models_[k].predict(X)*W[0][k] for k in range(len(models_))]
pred_ens = np.sum(np.array(bigpreds_),axis=0)
return pred_ens
def evaluate_ensemble_(X, Y, models_, W, metric):
logger.info('evaluate_ensemble_ {} w:{}'.format(metric, W))
pulmodels_ = [[models_[k][m] for k in range(len(models_))] for m in range(len(models_[0]))]
score_ens_roc = np.mean(np.array([-1*eva.roc_auc_score (np.array(Y[m]).reshape((len(Y[m]),1)),predict_ens_(X[m], pulmodels_[m], W)) for m in range(len(pulmodels_))]))
score_ens_metric = np.mean(np.array([-1*eva.score_proba(np.array(Y[m]).reshape((len(Y[m]),1)),predict_ens_(X[m], pulmodels_[m], W)) for m in range(len(pulmodels_))]))
if False:
logger.info('old score (roc)')
score_ens = score_ens_roc
else:
logger.info('new score (via cli)')
score_ens = score_ens_metric
logger.info('-evaluate_ensemble_ {:0.4f} {:0.4f} {:0.4f}'.format(score_ens, score_ens_roc, score_ens_metric))
return score_ens
#----------------------
def get_ensemble_constraints(ens_size):
stems_ = 'x[:,0]'
for uu_ in range(ens_size-1):
stems_ = stems_ + ' + x[:,' + str(uu_+1) + ']'
cnstrs_ = '(' + stems_ + ')'
ens_constraints = [{'name': 'constr_1','constraint': cnstrs_ + ' - 1 - 0.0001'},
{'name': 'constr_2','constraint': '1 - ' + cnstrs_ + ' - 0.0001'}]
return ens_constraints
def evaluate_clf(X, Y, model_input, n_folds, visualize=False):
metric_ = np.zeros(n_folds)
indx = 0
skf = StratifiedKFold(n_splits=n_folds)
score_roc_lst = list()
score_prc_lst = list()
is_pred_proba = True
if hasattr(model_input, 'get_is_pred_proba'):
is_pred_proba = model_input.get_is_pred_proba()
for train_index, test_index in skf.split(X, Y):
X_train = X.loc[X.index[train_index]].copy()
Y_train = Y.loc[Y.index[train_index]].copy()
X_test = X.loc[X.index[test_index]].copy()
Y_test = Y.loc[Y.index[test_index]].copy()
assert set(Y_train) == set(Y_test)
nnan = 0
model = copy.deepcopy(model_input)
if is_pred_proba:
logger.info('+fit {} {}'.format(X_train.shape,list(set(np.ravel(Y_train)))))
model.fit(X_train, Y_train)
preds = model.predict(X_test)
nnan = sum(np.ravel(np.isnan(preds)))
# metric_[indx] = roc_auc_score(Y_test, preds)
# metric_[indx] = average_precision_score(Y_test, preds)
if nnan:
logger.info('nan preds:{} ratio:{} ratio:{}'.format(
nnan,
nnan/float(len(np.ravel(np.isnan(preds)))),
nnan/len(preds)
))
logger.info(
'warning: nan in predictions, cannot calculate score'
': use low score instead, model:{}'.format(model.name))
score_roc, score_prc = 0.5, 0.0
elif not is_pred_proba:
logger.info(
'warning: clf has no probabilities'
': use low score instead, model:{}'.format(model.name))
score_roc, score_prc = 0.5, 0.0
else:
# score_roc = metrics.roc_auc_score(Y_test, preds)
# score_prc = metrics.average_precision_score(Y_test, preds)
score_roc = eva.roc_auc_score(Y_test, preds)
score_prc = eva.average_precision_score(Y_test, preds)
score_roc_lst.append(score_roc)
score_prc_lst.append(score_prc)
metric_to_score = {
'aucprc': score_prc,
'aucroc': score_roc
}
metric_[indx] = eva.score_proba(Y_test, preds) if is_pred_proba and not nnan else metric_to_score[eva.get_metric()]
logger.info('{} metric_{}'.format(indx, metric_))
if visualize:
printmd("\r{0}".format("**Cross-validation score: **"+ str(metric_[indx])))
print("---------------------------------------------------------")
indx += 1
Output = (metric_.mean(), 1.96*np.std(metric_)/np.sqrt(len(metric_)))
if visualize:
printmd("\r{0}".format("**Final Cross-validation score: **"+ str(Output)))
print("---------------------------------------------------------")
roc = float(np.mean(score_roc_lst))
prc = float(np.mean(score_prc_lst))
logger.info(' evaluate_clf::aucroc {:0.4f} #({}) {}'.format(
roc,
len(score_roc_lst),
model.name))
logger.info(' evaluate_clf::aucprc {:0.4f} #({}) {}'.format(
prc,
len(score_prc_lst),
model.name
))
eva_prop = {
'aucprc': prc,