-
Notifications
You must be signed in to change notification settings - Fork 1
/
shared_functions.py
1708 lines (1195 loc) · 61.8 KB
/
shared_functions.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
#!/usr/bin/env python
# coding: utf-8
# (shared_functions)=
# # Shared functions
#
# This notebook contains functions which are commonly reused in the book, for loading and saving data, fitting and assessing prediction models, or plotting results.
#
# The notebook can be downloaded from GitHub with
#
# ```
# !curl -O https://raw.githubusercontent.com/Fraud-Detection-Handbook/fraud-detection-handbook/main/Chapter_References/shared_functions.ipynb
#
# ```
#
# The notebook can be included in other notebooks using
#
# ```
# %run shared_functions
# ```
#
#
# ## General imports
# In[ ]:
# General
import os
import pandas as pd
import numpy as np
import math
import sys
import time
import pickle
import json
import datetime
import random
#import sklearn
import sklearn
from sklearn import *
get_ipython().run_line_magic('matplotlib', 'inline')
import matplotlib
import matplotlib.pyplot as plt
import seaborn as sns
sns.set_style('darkgrid', {'axes.facecolor': '0.9'})
import graphviz
import xgboost
# For imbalanced learning
import imblearn
import warnings
warnings.filterwarnings('ignore')
# ## Loading and saving data
# ### read_from_files
#
# First use in [Chapter 3, Baseline Feature Transformation](Baseline_Feature_Transformation).
# In[ ]:
# Load a set of pickle files, put them together in a single DataFrame, and order them by time
# It takes as input the folder DIR_INPUT where the files are stored, and the BEGIN_DATE and END_DATE
def read_from_files(DIR_INPUT, BEGIN_DATE, END_DATE):
files = [os.path.join(DIR_INPUT, f) for f in os.listdir(DIR_INPUT) if f>=BEGIN_DATE+'.pkl' and f<=END_DATE+'.pkl']
frames = []
for f in files:
df = pd.read_pickle(f)
frames.append(df)
del df
df_final = pd.concat(frames)
df_final=df_final.sort_values('TRANSACTION_ID')
df_final.reset_index(drop=True,inplace=True)
# Note: -1 are missing values for real world data
df_final=df_final.replace([-1],0)
return df_final
# ### save_object
#
# In[ ]:
#Save oject as pickle file
def save_object(obj, filename):
with open(filename, 'wb') as output:
pickle.dump(obj, output, pickle.HIGHEST_PROTOCOL)
# ## Data preprocessing
# ### scaleData
#
# First use in [Chapter 3, Baseline Fraud Detection System](Baseline_FDS).
# In[ ]:
def scaleData(train,test,features):
scaler = sklearn.preprocessing.StandardScaler()
scaler.fit(train[features])
train[features]=scaler.transform(train[features])
test[features]=scaler.transform(test[features])
return (train,test)
# ## Train/Test splitting strategies
# ### get_train_test_set
#
# First use in [Chapter 3, Baseline Fraud Detection System](Baseline_FDS).
# Sampling ratio added in [Chapter 5, Validation Strategies](Validation_Strategies).
# In[ ]:
def get_train_test_set(transactions_df,
start_date_training,
delta_train=7,delta_delay=7,delta_test=7,
sampling_ratio=1.0,
random_state=0):
# Get the training set data
train_df = transactions_df[(transactions_df.TX_DATETIME>=start_date_training) &
(transactions_df.TX_DATETIME<start_date_training+datetime.timedelta(days=delta_train))]
# Get the test set data
test_df = []
# Note: Cards known to be compromised after the delay period are removed from the test set
# That is, for each test day, all frauds known at (test_day-delay_period) are removed
# First, get known defrauded customers from the training set
known_defrauded_customers = set(train_df[train_df.TX_FRAUD==1].CUSTOMER_ID)
# Get the relative starting day of training set (easier than TX_DATETIME to collect test data)
start_tx_time_days_training = train_df.TX_TIME_DAYS.min()
# Then, for each day of the test set
for day in range(delta_test):
# Get test data for that day
test_df_day = transactions_df[transactions_df.TX_TIME_DAYS==start_tx_time_days_training+
delta_train+delta_delay+
day]
# Compromised cards from that test day, minus the delay period, are added to the pool of known defrauded customers
test_df_day_delay_period = transactions_df[transactions_df.TX_TIME_DAYS==start_tx_time_days_training+
delta_train+
day-1]
new_defrauded_customers = set(test_df_day_delay_period[test_df_day_delay_period.TX_FRAUD==1].CUSTOMER_ID)
known_defrauded_customers = known_defrauded_customers.union(new_defrauded_customers)
test_df_day = test_df_day[~test_df_day.CUSTOMER_ID.isin(known_defrauded_customers)]
test_df.append(test_df_day)
test_df = pd.concat(test_df)
# If subsample
if sampling_ratio<1:
train_df_frauds=train_df[train_df.TX_FRAUD==1].sample(frac=sampling_ratio, random_state=random_state)
train_df_genuine=train_df[train_df.TX_FRAUD==0].sample(frac=sampling_ratio, random_state=random_state)
train_df=pd.concat([train_df_frauds,train_df_genuine])
# Sort data sets by ascending order of transaction ID
train_df=train_df.sort_values('TRANSACTION_ID')
test_df=test_df.sort_values('TRANSACTION_ID')
return (train_df, test_df)
# In[ ]:
def get_train_delay_test_set(transactions_df,
start_date_training,
delta_train=7,delta_delay=7,delta_test=7,
sampling_ratio=1.0,
random_state=0):
# Get the training set data
train_df = transactions_df[(transactions_df.TX_DATETIME>=start_date_training) &
(transactions_df.TX_DATETIME<start_date_training+datetime.timedelta(days=delta_train))]
# Get the delay set data
delay_df = transactions_df[(transactions_df.TX_DATETIME>=start_date_training+datetime.timedelta(days=delta_train)) &
(transactions_df.TX_DATETIME<start_date_training+datetime.timedelta(days=delta_train)+
+datetime.timedelta(days=delta_delay))]
# Get the test set data
test_df = []
# Note: Cards known to be compromised after the delay period are removed from the test set
# That is, for each test day, all frauds known at (test_day-delay_period) are removed
# First, get known defrauded customers from the training set
known_defrauded_customers = set(train_df[train_df.TX_FRAUD==1].CUSTOMER_ID)
# Get the relative starting day of training set (easier than TX_DATETIME to collect test data)
start_tx_time_days_training = train_df.TX_TIME_DAYS.min()
# Then, for each day of the test set
for day in range(delta_test):
# Get test data for that day
test_df_day = transactions_df[transactions_df.TX_TIME_DAYS==start_tx_time_days_training+
delta_train+delta_delay+
day]
# Compromised cards from that test day, minus the delay period, are added to the pool of known defrauded customers
test_df_day_delay_period = transactions_df[transactions_df.TX_TIME_DAYS==start_tx_time_days_training+
delta_train+
day-1]
new_defrauded_customers = set(test_df_day_delay_period[test_df_day_delay_period.TX_FRAUD==1].CUSTOMER_ID)
known_defrauded_customers = known_defrauded_customers.union(new_defrauded_customers)
test_df_day = test_df_day[~test_df_day.CUSTOMER_ID.isin(known_defrauded_customers)]
test_df.append(test_df_day)
test_df = pd.concat(test_df)
# If subsample
if sampling_ratio<1:
train_df_frauds=train_df[train_df.TX_FRAUD==1].sample(frac=sampling_ratio, random_state=random_state)
train_df_genuine=train_df[train_df.TX_FRAUD==0].sample(frac=sampling_ratio, random_state=random_state)
train_df=pd.concat([train_df_frauds,train_df_genuine])
# Sort data sets by ascending order of transaction ID
train_df=train_df.sort_values('TRANSACTION_ID')
test_df=test_df.sort_values('TRANSACTION_ID')
return (train_df, delay_df, test_df)
# ### prequentialSplit
#
# First use in [Chapter 5, Validation Strategies](Validation_Strategies).
# In[ ]:
def prequentialSplit(transactions_df,
start_date_training,
n_folds=4,
delta_train=7,
delta_delay=7,
delta_assessment=7):
prequential_split_indices=[]
# For each fold
for fold in range(n_folds):
# Shift back start date for training by the fold index times the assessment period (delta_assessment)
# (See Fig. 5)
start_date_training_fold = start_date_training-datetime.timedelta(days=fold*delta_assessment)
# Get the training and test (assessment) sets
(train_df, test_df)=get_train_test_set(transactions_df,
start_date_training=start_date_training_fold,
delta_train=delta_train,delta_delay=delta_delay,delta_test=delta_assessment)
# Get the indices from the two sets, and add them to the list of prequential splits
indices_train=list(train_df.index)
indices_test=list(test_df.index)
prequential_split_indices.append((indices_train,indices_test))
return prequential_split_indices
# ## Predictions functions
# ### fit_model_and_get_predictions
#
# First use in [Chapter 3, Baseline Fraud Detection System](Baseline_FDS).
# In[ ]:
def fit_model_and_get_predictions(classifier, train_df, test_df,
input_features, output_feature="TX_FRAUD",scale=True):
# By default, scales input data
if scale:
(train_df, test_df)=scaleData(train_df,test_df,input_features)
# We first train the classifier using the `fit` method, and pass as arguments the input and output features
start_time=time.time()
classifier.fit(train_df[input_features], train_df[output_feature])
training_execution_time=time.time()-start_time
# We then get the predictions on the training and test data using the `predict_proba` method
# The predictions are returned as a numpy array, that provides the probability of fraud for each transaction
start_time=time.time()
predictions_test=classifier.predict_proba(test_df[input_features])[:,1]
prediction_execution_time=time.time()-start_time
predictions_train=classifier.predict_proba(train_df[input_features])[:,1]
# The result is returned as a dictionary containing the fitted models,
# and the predictions on the training and test sets
model_and_predictions_dictionary = {'classifier': classifier,
'predictions_test': predictions_test,
'predictions_train': predictions_train,
'training_execution_time': training_execution_time,
'prediction_execution_time': prediction_execution_time
}
return model_and_predictions_dictionary
# In[ ]:
# ## Performance assessment
# ### card_precision_top_k_day
#
# First use in [Chapter 3, Baseline Fraud Detection System](Baseline_FDS).
# Detailed in [Chapter 4, Precision_top_K_Metrics](Precision_Top_K_Metrics).
# In[ ]:
def card_precision_top_k_day(df_day,top_k):
# This takes the max of the predictions AND the max of label TX_FRAUD for each CUSTOMER_ID,
# and sorts by decreasing order of fraudulent prediction
df_day = df_day.groupby('CUSTOMER_ID').max().sort_values(by="predictions", ascending=False).reset_index(drop=False)
# Get the top k most suspicious cards
df_day_top_k=df_day.head(top_k)
list_detected_compromised_cards=list(df_day_top_k[df_day_top_k.TX_FRAUD==1].CUSTOMER_ID)
# Compute precision top k
card_precision_top_k = len(list_detected_compromised_cards) / top_k
return list_detected_compromised_cards, card_precision_top_k
# ### card_precision_top_k
#
# First use in [Chapter 3, Baseline Fraud Detection System](Baseline_FDS).
# Detailed in [Chapter 4, Precision_top_K_Metrics](Precision_Top_K_Metrics).
# In[ ]:
def card_precision_top_k(predictions_df, top_k, remove_detected_compromised_cards=True):
# Sort days by increasing order
list_days=list(predictions_df['TX_TIME_DAYS'].unique())
list_days.sort()
# At first, the list of detected compromised cards is empty
list_detected_compromised_cards = []
card_precision_top_k_per_day_list = []
nb_compromised_cards_per_day = []
# For each day, compute precision top k
for day in list_days:
df_day = predictions_df[predictions_df['TX_TIME_DAYS']==day]
df_day = df_day[['predictions', 'CUSTOMER_ID', 'TX_FRAUD']]
# Let us remove detected compromised cards from the set of daily transactions
df_day = df_day[df_day.CUSTOMER_ID.isin(list_detected_compromised_cards)==False]
nb_compromised_cards_per_day.append(len(df_day[df_day.TX_FRAUD==1].CUSTOMER_ID.unique()))
detected_compromised_cards, card_precision_top_k = card_precision_top_k_day(df_day,top_k)
card_precision_top_k_per_day_list.append(card_precision_top_k)
# Let us update the list of detected compromised cards
if remove_detected_compromised_cards:
list_detected_compromised_cards.extend(detected_compromised_cards)
# Compute the mean
mean_card_precision_top_k = np.array(card_precision_top_k_per_day_list).mean()
# Returns precision top k per day as a list, and resulting mean
return nb_compromised_cards_per_day,card_precision_top_k_per_day_list,mean_card_precision_top_k
# ### card_precision_top_k_custom
#
# First use in [Chapter 5, Validation Strategies](Validation_Strategies).
# In[ ]:
def card_precision_top_k_custom(y_true, y_pred, top_k, transactions_df):
# Let us create a predictions_df DataFrame, that contains all transactions matching the indices of the current fold
# (indices of the y_true vector)
predictions_df=transactions_df.iloc[y_true.index.values].copy()
predictions_df['predictions']=y_pred
# Compute the CP@k using the function implemented in Chapter 4, Section 4.2
nb_compromised_cards_per_day,card_precision_top_k_per_day_list,mean_card_precision_top_k= card_precision_top_k(predictions_df, top_k)
# Return the mean_card_precision_top_k
return mean_card_precision_top_k
# ### performance_assessment
#
# First use in [Chapter 3, Baseline Fraud Detection System](Baseline_FDS).
# In[ ]:
def performance_assessment(predictions_df, output_feature='TX_FRAUD',
prediction_feature='predictions', top_k_list=[100],
rounded=True):
AUC_ROC = metrics.roc_auc_score(predictions_df[output_feature], predictions_df[prediction_feature])
AP = metrics.average_precision_score(predictions_df[output_feature], predictions_df[prediction_feature])
performances = pd.DataFrame([[AUC_ROC, AP]],
columns=['AUC ROC','Average precision'])
for top_k in top_k_list:
_, _, mean_card_precision_top_k = card_precision_top_k(predictions_df, top_k)
performances['Card Precision@'+str(top_k)]=mean_card_precision_top_k
if rounded:
performances = performances.round(3)
return performances
# ### performance_assessment_model_collection
#
# First use in [Chapter 3, Baseline Fraud Detection System](Baseline_FDS).
# In[ ]:
def performance_assessment_model_collection(fitted_models_and_predictions_dictionary,
transactions_df,
type_set='test',
top_k_list=[100]):
performances=pd.DataFrame()
for classifier_name, model_and_predictions in fitted_models_and_predictions_dictionary.items():
predictions_df=transactions_df
predictions_df['predictions']=model_and_predictions['predictions_'+type_set]
performances_model=performance_assessment(predictions_df, output_feature='TX_FRAUD',
prediction_feature='predictions', top_k_list=top_k_list)
performances_model.index=[classifier_name]
performances=performances.append(performances_model)
return performances
# ### execution_times_model_collection
#
# First use in [Chapter 3, Baseline Fraud Detection System](Baseline_FDS).
# In[ ]:
def execution_times_model_collection(fitted_models_and_predictions_dictionary):
execution_times=pd.DataFrame()
for classifier_name, model_and_predictions in fitted_models_and_predictions_dictionary.items():
execution_times_model=pd.DataFrame()
execution_times_model['Training execution time']=[model_and_predictions['training_execution_time']]
execution_times_model['Prediction execution time']=[model_and_predictions['prediction_execution_time']]
execution_times_model.index=[classifier_name]
execution_times=execution_times.append(execution_times_model)
return execution_times
# ### get_class_from_fraud_probability
#
# First use in [Chapter 4, Threshold Based Metrics](Threshold_Based_Metrics).
# In[ ]:
# Getting classes from a vector of fraud probabilities and a threshold
def get_class_from_fraud_probability(fraud_probabilities, threshold=0.5):
predicted_classes = [0 if fraud_probability<threshold else 1
for fraud_probability in fraud_probabilities]
return predicted_classes
# ### threshold_based_metrics
#
# First use in [Chapter 4, Threshold Based Metrics](Threshold_Based_Metrics).
# In[ ]:
def threshold_based_metrics(fraud_probabilities, true_label, thresholds_list):
results = []
for threshold in thresholds_list:
predicted_classes = get_class_from_fraud_probability(fraud_probabilities, threshold=threshold)
(TN, FP, FN, TP) = metrics.confusion_matrix(true_label, predicted_classes).ravel()
MME = (FP+FN)/(TN+FP+FN+TP)
TPR = TP/(TP+FN)
TNR = TN/(TN+FP)
FPR = FP/(TN+FP)
FNR = FN/(TP+FN)
BER = 1/2*(FPR+FNR)
Gmean = np.sqrt(TPR*TNR)
precision = 1 # 1 if TP+FP=0
FDR = 1 # 1 if TP+FP=0
if TP+FP>0:
precision = TP/(TP+FP)
FDR=FP/(TP+FP)
NPV = 1 # 1 if TN+FN=0
FOR = 1 # 1 if TN+FN=0
if TN+FN>0:
NPV = TN/(TN+FN)
FOR = FN/(TN+FN)
F1_score = 2*(precision*TPR)/(precision+TPR)
results.append([threshold, MME, TPR, TNR, FPR, FNR, BER, Gmean, precision, NPV, FDR, FOR, F1_score])
results_df = pd.DataFrame(results,columns=['Threshold' ,'MME', 'TPR', 'TNR', 'FPR', 'FNR', 'BER', 'G-mean', 'Precision', 'NPV', 'FDR', 'FOR', 'F1 Score'])
return results_df
# In[ ]:
# ### get_summary_performances
#
# First use in [Chapter 5, Model Selection](Model_Selection).
# In[ ]:
def get_summary_performances(performances_df, parameter_column_name="Parameters summary"):
metrics = ['AUC ROC','Average precision','Card Precision@100']
performances_results=pd.DataFrame(columns=metrics)
performances_df.reset_index(drop=True,inplace=True)
best_estimated_parameters = []
validation_performance = []
test_performance = []
for metric in metrics:
index_best_validation_performance = performances_df.index[np.argmax(performances_df[metric+' Validation'].values)]
best_estimated_parameters.append(performances_df[parameter_column_name].iloc[index_best_validation_performance])
validation_performance.append(
str(round(performances_df[metric+' Validation'].iloc[index_best_validation_performance],3))+
'+/-'+
str(round(performances_df[metric+' Validation'+' Std'].iloc[index_best_validation_performance],2))
)
test_performance.append(
str(round(performances_df[metric+' Test'].iloc[index_best_validation_performance],3))+
'+/-'+
str(round(performances_df[metric+' Test'+' Std'].iloc[index_best_validation_performance],2))
)
performances_results.loc["Best estimated parameters"]=best_estimated_parameters
performances_results.loc["Validation performance"]=validation_performance
performances_results.loc["Test performance"]=test_performance
optimal_test_performance = []
optimal_parameters = []
for metric in ['AUC ROC Test','Average precision Test','Card Precision@100 Test']:
index_optimal_test_performance = performances_df.index[np.argmax(performances_df[metric].values)]
optimal_parameters.append(performances_df[parameter_column_name].iloc[index_optimal_test_performance])
optimal_test_performance.append(
str(round(performances_df[metric].iloc[index_optimal_test_performance],3))+
'+/-'+
str(round(performances_df[metric+' Std'].iloc[index_optimal_test_performance],2))
)
performances_results.loc["Optimal parameter(s)"]=optimal_parameters
performances_results.loc["Optimal test performance"]=optimal_test_performance
return performances_results
# ### model_selection_performances
#
# First use in [Chapter 5, Model Selection](Model_Selection).
# In[ ]:
def model_selection_performances(performances_df_dictionary,
performance_metric='AUC ROC'):
# Note: max_depth of 50 is similar to None
default_parameters_dictionary={
"Decision Tree": 50,
"Logstic Regression": 1,
"Random Forest": "100/50",
"XGBoost": "100/0.1/2"
}
mean_performances_dictionary={
"Default parameters": [],
"Best validation parameters": [],
"Optimal parameters": []
}
std_performances_dictionary={
"Default parameters": [],
"Best validation parameters": [],
"Optimal parameters": []
}
# For each model class
for model_class, performances_df in performances_df_dictionary.items():
# Get the performances for the default paramaters
default_performances=performances_df[performances_df['Parameters summary']==default_parameters_dictionary[model_class]]
default_performances=default_performances.round(decimals=3)
mean_performances_dictionary["Default parameters"].append(default_performances[performance_metric+" Test"].values[0])
std_performances_dictionary["Default parameters"].append(default_performances[performance_metric+" Test Std"].values[0])
# Get the performances for the best estimated parameters
performances_summary=get_summary_performances(performances_df, parameter_column_name="Parameters summary")
mean_std_performances=performances_summary.loc[["Test performance"]][performance_metric].values[0]
mean_std_performances=mean_std_performances.split("+/-")
mean_performances_dictionary["Best validation parameters"].append(float(mean_std_performances[0]))
std_performances_dictionary["Best validation parameters"].append(float(mean_std_performances[1]))
# Get the performances for the boptimal parameters
mean_std_performances=performances_summary.loc[["Optimal test performance"]][performance_metric].values[0]
mean_std_performances=mean_std_performances.split("+/-")
mean_performances_dictionary["Optimal parameters"].append(float(mean_std_performances[0]))
std_performances_dictionary["Optimal parameters"].append(float(mean_std_performances[1]))
# Return the mean performances and their standard deviations
return (mean_performances_dictionary,std_performances_dictionary)
# In[ ]:
def model_selection_performances(performances_df_dictionary,
performance_metric='AUC ROC',
model_classes=['Decision Tree',
'Logistic Regression',
'Random Forest',
'XGBoost'],
default_parameters_dictionary={
"Decision Tree": 50,
"Logistic Regression": 1,
"Random Forest": "100/50",
"XGBoost": "100/0.1/3"
}):
mean_performances_dictionary={
"Default parameters": [],
"Best validation parameters": [],
"Optimal parameters": []
}
std_performances_dictionary={
"Default parameters": [],
"Best validation parameters": [],
"Optimal parameters": []
}
# For each model class
for model_class in model_classes:
performances_df=performances_df_dictionary[model_class]
# Get the performances for the default paramaters
default_performances=performances_df[performances_df['Parameters summary']==default_parameters_dictionary[model_class]]
default_performances=default_performances.round(decimals=3)
mean_performances_dictionary["Default parameters"].append(default_performances[performance_metric+" Test"].values[0])
std_performances_dictionary["Default parameters"].append(default_performances[performance_metric+" Test Std"].values[0])
# Get the performances for the best estimated parameters
performances_summary=get_summary_performances(performances_df, parameter_column_name="Parameters summary")
mean_std_performances=performances_summary.loc[["Test performance"]][performance_metric].values[0]
mean_std_performances=mean_std_performances.split("+/-")
mean_performances_dictionary["Best validation parameters"].append(float(mean_std_performances[0]))
std_performances_dictionary["Best validation parameters"].append(float(mean_std_performances[1]))
# Get the performances for the boptimal parameters
mean_std_performances=performances_summary.loc[["Optimal test performance"]][performance_metric].values[0]
mean_std_performances=mean_std_performances.split("+/-")
mean_performances_dictionary["Optimal parameters"].append(float(mean_std_performances[0]))
std_performances_dictionary["Optimal parameters"].append(float(mean_std_performances[1]))
# Return the mean performances and their standard deviations
return (mean_performances_dictionary,std_performances_dictionary)
# ## Model selection
# ### prequential_grid_search
#
# First use in [Chapter 5, Validation Strategies](Validation_Strategies).
# In[ ]:
def prequential_grid_search(transactions_df,
classifier,
input_features, output_feature,
parameters, scoring,
start_date_training,
n_folds=4,
expe_type='Test',
delta_train=7,
delta_delay=7,
delta_assessment=7,
performance_metrics_list_grid=['roc_auc'],
performance_metrics_list=['AUC ROC'],
n_jobs=-1):
estimators = [('scaler', sklearn.preprocessing.StandardScaler()), ('clf', classifier)]
pipe = sklearn.pipeline.Pipeline(estimators)
prequential_split_indices=prequentialSplit(transactions_df,
start_date_training=start_date_training,
n_folds=n_folds,
delta_train=delta_train,
delta_delay=delta_delay,
delta_assessment=delta_assessment)
grid_search = sklearn.model_selection.GridSearchCV(pipe, parameters, scoring=scoring, cv=prequential_split_indices, refit=False, n_jobs=n_jobs)
X=transactions_df[input_features]
y=transactions_df[output_feature]
grid_search.fit(X, y)
performances_df=pd.DataFrame()
for i in range(len(performance_metrics_list_grid)):
performances_df[performance_metrics_list[i]+' '+expe_type]=grid_search.cv_results_['mean_test_'+performance_metrics_list_grid[i]]
performances_df[performance_metrics_list[i]+' '+expe_type+' Std']=grid_search.cv_results_['std_test_'+performance_metrics_list_grid[i]]
performances_df['Parameters']=grid_search.cv_results_['params']
performances_df['Execution time']=grid_search.cv_results_['mean_fit_time']
return performances_df
# ### model_selection_wrapper
#
# First use in [Chapter 5, Model Selection](Model_Selection).
# In[ ]:
def model_selection_wrapper(transactions_df,
classifier,
input_features, output_feature,
parameters,
scoring,
start_date_training_for_valid,
start_date_training_for_test,
n_folds=4,
delta_train=7,
delta_delay=7,
delta_assessment=7,
performance_metrics_list_grid=['roc_auc'],
performance_metrics_list=['AUC ROC'],
n_jobs=-1):
# Get performances on the validation set using prequential validation
performances_df_validation=prequential_grid_search(transactions_df, classifier,
input_features, output_feature,
parameters, scoring,
start_date_training=start_date_training_for_valid,
n_folds=n_folds,
expe_type='Validation',
delta_train=delta_train,
delta_delay=delta_delay,
delta_assessment=delta_assessment,
performance_metrics_list_grid=performance_metrics_list_grid,
performance_metrics_list=performance_metrics_list,
n_jobs=n_jobs)
# Get performances on the test set using prequential validation
performances_df_test=prequential_grid_search(transactions_df, classifier,
input_features, output_feature,
parameters, scoring,
start_date_training=start_date_training_for_test,
n_folds=n_folds,
expe_type='Test',
delta_train=delta_train,
delta_delay=delta_delay,
delta_assessment=delta_assessment,
performance_metrics_list_grid=performance_metrics_list_grid,
performance_metrics_list=performance_metrics_list,
n_jobs=n_jobs)
# Bind the two resulting DataFrames
performances_df_validation.drop(columns=['Parameters','Execution time'], inplace=True)
performances_df=pd.concat([performances_df_test,performances_df_validation],axis=1)
# And return as a single DataFrame
return performances_df
# ### kfold_cv_with_classifier
#
# First use in [Chapter 6, Cost-sensitive learning](Cost_Sensitive_Learning).
# In[ ]:
def kfold_cv_with_classifier(classifier,
X,
y,
n_splits=5,
strategy_name="Basline classifier"):
cv = sklearn.model_selection.StratifiedKFold(n_splits=n_splits, shuffle=True, random_state=0)
cv_results_=sklearn.model_selection.cross_validate(classifier,X,y,cv=cv,
scoring=['roc_auc',
'average_precision',
'balanced_accuracy'],
return_estimator=True)
results=round(pd.DataFrame(cv_results_),3)
results_mean=list(results.mean().values)
results_std=list(results.std().values)
results_df=pd.DataFrame([[str(round(results_mean[i],3))+'+/-'+
str(round(results_std[i],3)) for i in range(len(results))]],
columns=['Fit time (s)','Score time (s)',
'AUC ROC','Average Precision','Balanced accuracy'])
results_df.rename(index={0:strategy_name}, inplace=True)
classifier_0=cv_results_['estimator'][0]
(train_index, test_index) = next(cv.split(X, y))
train_df=pd.DataFrame({'X1':X[train_index,0],'X2':X[train_index,1], 'Y':y[train_index]})
test_df=pd.DataFrame({'X1':X[test_index,0],'X2':X[test_index,1], 'Y':y[test_index]})
return (results_df, classifier_0, train_df, test_df)
# ## Plotting
# ### get_tx_stats
#
# First use in [Chapter 3, Baseline Fraud Detection System](Baseline_FDS).
# In[ ]:
# Compute the number of transactions per day, fraudulent transactions per day and fraudulent cards per day
def get_tx_stats(transactions_df, start_date_df="2018-04-01"):
#Number of transactions per day
nb_tx_per_day=transactions_df.groupby(['TX_TIME_DAYS'])['CUSTOMER_ID'].count()
#Number of fraudulent transactions per day
nb_fraudulent_transactions_per_day=transactions_df.groupby(['TX_TIME_DAYS'])['TX_FRAUD'].sum()
#Number of fraudulent cards per day
nb_compromised_card_per_day=transactions_df[transactions_df['TX_FRAUD']==1].groupby(['TX_TIME_DAYS']).CUSTOMER_ID.nunique()
tx_stats=pd.DataFrame({"nb_tx_per_day":nb_tx_per_day,
"nb_fraudulent_transactions_per_day":nb_fraudulent_transactions_per_day,
"nb_compromised_cards_per_day":nb_compromised_card_per_day})
tx_stats=tx_stats.reset_index()
start_date = datetime.datetime.strptime(start_date_df, "%Y-%m-%d")
tx_date=start_date+tx_stats['TX_TIME_DAYS'].apply(datetime.timedelta)
tx_stats['tx_date']=tx_date
return tx_stats
# ### get_template_tx_stats
#
# First use in [Chapter 3, Baseline Fraud Detection System](Baseline_FDS).
# In[ ]:
# Plot the number of transactions per day, fraudulent transactions per day and fraudulent cards per day
def get_template_tx_stats(ax ,fs,
start_date_training,
title='',
delta_train=7,
delta_delay=7,
delta_test=7,
ylim=300):
ax.set_title(title, fontsize=fs*1.5)
ax.set_ylim([0, ylim])
ax.set_xlabel('Date', fontsize=fs)
ax.set_ylabel('Number', fontsize=fs)
plt.yticks(fontsize=fs*0.7)
plt.xticks(fontsize=fs*0.7)
ax.axvline(start_date_training+datetime.timedelta(days=delta_train), 0,ylim, color="black")
ax.axvline(start_date_training+datetime.timedelta(days=delta_train+delta_delay), 0, ylim, color="black")
ax.text(start_date_training+datetime.timedelta(days=2), ylim-20,'Training period', fontsize=fs)
ax.text(start_date_training+datetime.timedelta(days=delta_train+2), ylim-20,'Delay period', fontsize=fs)
ax.text(start_date_training+datetime.timedelta(days=delta_train+delta_delay+2), ylim-20,'Test period', fontsize=fs)
# ### get_template_roc_curve
#
# First use in [Chapter 4, Threshold Free Metrics](Threshold_Free_Metrics).
# In[ ]:
def get_template_roc_curve(ax, title,fs,random=True):
ax.set_title(title, fontsize=fs)
ax.set_xlim([-0.01, 1.01])
ax.set_ylim([-0.01, 1.01])
ax.set_xlabel('False Positive Rate', fontsize=fs)
ax.set_ylabel('True Positive Rate', fontsize=fs)
if random:
ax.plot([0, 1], [0, 1],'r--',label="AUC ROC Random = 0.5")