-
Notifications
You must be signed in to change notification settings - Fork 5
/
Copy pathImagene_v3_2v.py
1369 lines (1251 loc) · 73.9 KB
/
Imagene_v3_2v.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
##Imagene.py
##Note: This script is only operable when within the docker container package. Check github for more info on running the package.
##Version 1.02: ##Added if condition to check if the low ratio RMSE plot file exists before writing it to HTML file
##Version 1.03.1: ## a) Changed the conditions for binarization of feature columns in Y_test and Y_pred for AUC calculations. Now, if value<threshold then it gets binarized to 0 else to 1.
## ## b) Changed decision threshold list to [0,0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9,1.0].
## ## c) Edited mp.xlim and ylim parameters for AUC v/s decision_threshold plot to include left=0.0 and right=1.0 values (for xlim).
##Version 1.04: ## Changed the decision threshold list to [0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9].
## ## Changed the if condition in 1038 to match length of the AUC_value_list and decision_threshold_list to >=8. Made the correspoding change in print statement in line 1046 accordingly.
## ## Changed xlim: now left=0.1 and right=0.9
## ## Reverted the condition for binarization of feature columns in Y_test and Y_pred for AUC calculations. Now, if value<=threshold then it gets binarized to 0 else to 1.
##Version 1.05: ##Using pickle module instead of joblib module to load model.pkl
##Version 1.1: ##Adding multiTask lasso and multitask elastic net models
##Version 2.0: Added Feature importances and permutation tests for R-square and AUC. R-square gets calculated using sklearn.metrics.r2_score method.
##Version 2.2: if pVal_adj_method="none" (i.e. p_adjust=p_value) or corr_threshold (aka correlation coefficient threshold) < 0.0 (for ex: -1.0), then no <0.05 filtering on p_adjust value.
##Note: Please monitor github commits for version updates after Version: 2.2.
##Version 2.3: ##ratio=rmse when stdev=0, changed print statement accordingly.
## ##Introduced RandomForest for binary label classification in "BuildModel", and their AUC score calculation (introduced if condition for the same in "evaluate")
##Version 2.4: ##Added DecisionTreeClassifier
##Version 2.5: Add visualize tree for DecisionTreeClassifier.
##Version 2.6: Moved print tree outside evaluate inside buildModel function.
## Fixed couple print statements to prepend the output file paths with str(save_dir) and tagDir, mainly the AUC files from RandomForest and DecisionTreeClassifier operations.
##Version 2.7: Adding supervised neural networks (MultiLayerPerceptron) Model, silenting feature importance and featureSelection based training specifically for that model_type.
##Version 2.8: Adding LogisticRegression Model
##Version 2.9: For grid_search='True', No feature importances or feature selection executed.
##Version 3.0: Setting min AUC=0.75 for permutation calculation (p-value estimation).
##Version 3.1: Setting min AUC=0.75 for permutation calculation (p-value estimation) for classifier models as well.
##Version 3.2: Introducing SVM.
##Version 3.2v: Added 'grid_search' parameter in evaluate call for mode='validate' in line 1229.
##Author: Shrey Sukhadia
#!/usr/bin/python
from cProfile import label
from graphviz import Source
import matplotlib as mpl
mpl.use('Agg')
import os, re, sys, math
import argparse
import numpy as np
import pandas as pd
#from pandas.stats.api import ols
import joblib
import pickle
import configparser
import ast
import argparse
from sklearn.neural_network import MLPClassifier
from sklearn.svm import SVC
from sklearn.feature_selection import SelectFromModel
from sklearn import preprocessing
#from sklearn import tree
from sklearn.preprocessing import MaxAbsScaler
from sklearn.preprocessing import MinMaxScaler
from sklearn.preprocessing import StandardScaler
from sklearn.preprocessing import Normalizer
from sklearn.preprocessing import binarize
from sklearn.model_selection import train_test_split
from sklearn.decomposition import PCA
from sklearn.manifold import TSNE
from sklearn.metrics import confusion_matrix
from sklearn.metrics import accuracy_score
from sklearn.metrics import classification_report
from sklearn.metrics import roc_auc_score
from sklearn.metrics import roc_curve
from sklearn.metrics import precision_recall_curve
from sklearn.metrics import average_precision_score
from sklearn.metrics import auc
from scipy import stats
from scipy.interpolate import make_interp_spline, BSpline
from sklearn.model_selection import cross_val_score
from sklearn.model_selection import GridSearchCV
#model packages
from sklearn import linear_model
from sklearn import tree
from sklearn.linear_model import ElasticNet
from sklearn.linear_model import MultiTaskElasticNet
from sklearn.linear_model import LinearRegression
from sklearn.linear_model import LogisticRegression
from sklearn.ensemble import RandomForestClassifier
from sklearn.tree import DecisionTreeClassifier
from sklearn import metrics
from sklearn.metrics import make_scorer
from sklearn.metrics import mean_squared_error
from sklearn.metrics import r2_score
import seaborn as sb
import matplotlib.pyplot as mp
import scikitplot as skplt
from math import sqrt
import base64
from datetime import datetime
#import statsmodels as statsm
#import statsmodels.api as smapi
#from mpl_toolkits.mplot3d import Axes3D
#import rpy2.robjects as R
from rpy2.robjects.packages import importr
from rpy2.robjects.vectors import FloatVector
##Correlation function establishes correlations between gene and imaging features and calculates FDR adjusted pvalues further to establish statistical significance of those correlations.
def correlation(d_1,d_2,d_1_header,d_2_header, model_type, corr_method, corr_threshold,pVal_adjust_method, tagDir):
pval=dict()
pcorr=dict()
Var1_list=[]; Var2_list=[]; spCorr_list=[]; pvalue_list=[]
for i in d_1_header:
for j in d_2_header:
if(corr_method=="spearman"):
pCorr,pv=stats.spearmanr(d_1[i], d_2[j])
elif(corr_method=="pearson"):
pCorr,pv=stats.pearsonr(d_1[i], d_2[j])
else:
pCorr,pv=stats.spearmanr(d_1[i], d_2[j])
Var1_list.append(i); Var2_list.append(j); spCorr_list.append(pCorr); pvalue_list.append(pv)
rstats = importr('stats')
pAJ_list = rstats.p_adjust(FloatVector(pvalue_list), method = pVal_adjust_method)
Master_df = pd.DataFrame({'Var1': Var1_list, 'Var2': Var2_list, 'correlation': spCorr_list, 'p_value': pvalue_list, 'p_adjust': pAJ_list })
Master_df_sorted=Master_df.sort_values(by=['p_adjust'])
#print(pcorr_PAJ_df_sorted.head(30))
Master_df_sorted.to_csv(str(save_dir)+"All_correlations.csv")
if(pVal_adjust_method=="none" or corr_threshold < 0.0):
Master_df_sorted_sgn=Master_df_sorted
else:
Master_df_sorted_sgn=Master_df_sorted[Master_df_sorted['p_adjust']<0.05]
Master_df_sorted_sgn_filtered=Master_df_sorted_sgn[abs(Master_df_sorted_sgn['correlation'])>corr_threshold]
Master_df_sorted_sgn.to_csv(str(save_dir)+"Significant_correlations.csv")
Master_df_sorted_sgn_filtered.to_csv(str(save_dir)+"Significant_correlations_gt_corr_threshold.csv")
#print(Master_df_sorted_sgn)
List_of_Var1=Master_df_sorted_sgn_filtered["Var1"].tolist()
List_of_Var2=Master_df_sorted_sgn_filtered["Var2"].tolist()
fC_List_Var1=[]
for i in List_of_Var1:
fC=i.split("_")[0]
fC_List_Var1.append(fC)
fC_uniq = sorted(set(fC_List_Var1))
#outfileHTML=open(model_type+".output.html",'w')
image_tag_list=[]
#image_count=0
for i in fC_uniq:
print(i)
#fC_df=Master_df_sorted_sgn[Master_df_sorted_sgn['Var1'].str.match(i, case=True, flags=0)].head(50)
#fC_df=Master_df_sorted_sgn[Master_df_sorted_sgn['Var1'].str.contains(i, regex=False, na=False)].head(50)
fC_df=Master_df_sorted_sgn_filtered[Master_df_sorted_sgn_filtered['Var1'].str.contains(i, regex=False, na=False)]
fC_df_pivot=fC_df.pivot_table(index="Var1",columns="Var2",values="correlation",fill_value=0)
#fC_df_pivot = pd.pivot_table(fC_df, index = "Var1", values = ["Var2", "spearman_correlation"]).stack().reset_index(level = 1)
print("For "+i+" features:")
print(fC_df_pivot)
try:
sb.clustermap(fC_df_pivot)
except ValueError as err:
mp.clf()
sb.heatmap(fC_df_pivot)
if(pVal_adjust_method=="none" or corr_threshold < 0.0):
mp.title("Correlations for "+i+" features", size=16)
else:
mp.title("Top significant correlations (FDR_adjusted_pValue<0.05) for "+i+" features", size=16)
mp.xticks(rotation=90)
mp.yticks(size=7)
mp.savefig(str(save_dir)+''+tagDir+'Correlation_for_'+i+'_features.png',orientation='landscape',dpi=90,bbox_inches='tight')
data_image = open(str(save_dir)+tagDir+'Correlation_for_'+i+'_features.png', 'rb').read().encode('base64').replace('\n', '')
image_tag_list.append('<img src="data:image/png;base64,{0}" style="max-width:50%;">'.format(data_image))
mp.clf()
outfileHTML=open(str(save_dir)+tagDir+model_type+".output.html",'a')
#outfileHTML.write("<h1 style=text-align:center;color:red;>"+"Radiogenomics Analysis Report"+"</h1>"+"\n")
outfileHTML.write("<h2 style=text-align:center;>"+"--------------------------Multivariate Correlations ("+corr_method+" based)-----------------------"+"</h2>"+"\n".join(image_tag_list)+"\n")
outfileHTML.close()
return(sorted(set(List_of_Var1)),sorted(set(List_of_Var2)))
##Reading datasets
def read_dataset(dataset):
if os.path.isfile(dataset):
raw_dataframe = pd.read_csv(dataset, sep=',')
else:
print("{} should be csv/tsv file ".format(dataset))
sys.exit()
dataframe = raw_dataframe.copy()
print("Shape of dataframe:{}".format(dataframe.shape))
return dataframe
##Normalization function
def normal_dataframe(dataframe, norm_type, header, normalize = True):
if normalize:
if norm_type =='min_max':
scaler = MinMaxScaler()
dataframe_scaled = scaler.fit_transform(dataframe)
#return dataframe_scaled
elif norm_type == 'Stand_scaler':
scaler = StandardScaler()
dataframe_scaled = scaler.fit_transform(dataframe)
#return dataframe_scaled
##Removing Zscore as it is redundant to StandScaler
#elif norm_type == 'zscore':
# dataframe_scaled = stats.zscore(dataframe)
# #return dataframe_scaled
elif norm_type == 'MaxAbsScaler':
scaler = MaxAbsScaler()
dataframe_scaled = scaler.fit_transform(dataframe)
#return dataframe_scaled
else:
print("Invalid normalization type/method detected. Skipping normalization. If you wish to normalize this dataset, then correct the normalization_method in the config file and rerun. Proceeding with no normalization")
## The dataframe needs to be converted to numpy to be consistent with the return value when normalization method is detected and normalization actually happens.
#d_array = dataframe.to_records(index='True')
#feature_array = d_array.indices()
#return d_array, feature_array
##Switch back to returning dataframe
return dataframe
##SILENTING conversion of numpy into dataframe as we do want to return numpy array now that the calling of normal_dataframe occurs outside the preprocessing function.
##Switch back to converting numpy into dataframe
dataframe_scaled=pd.DataFrame(data=dataframe_scaled,columns=header)
#feature_array=scaler.feature_names_in_
return dataframe_scaled
else:
#d_array = dataframe.to_records(index='True')
#feature_array = d_array.indices()
#return d_array, feature_array
##Switch back to returning dataframe
return dataframe
##Preprocessing of datasets
def preprocessing(dataframe , label, data_type, label_type, mode, tagDir, checkNA = True):
#try:
outfileHTML=open(str(save_dir)+tagDir+model_type+".output.html",'a')
outfileHTML.write("<h3>"+"No. of "+data_type+" features provided: "+str(len(dataframe.columns)-1)+"</h3>")
if(isinstance(label,pd.DataFrame)):
outfileHTML.write("<h3>"+"No. of "+label_type+" features provided:"+str(len(label.columns)-1)+"</h3>")
if checkNA:
dataframe.replace("", np.nan, inplace=True)
if dataframe.isnull().values.any():
dataframe = dataframe.dropna(axis=1, how='any')
if isinstance(label,pd.DataFrame):
label.replace("", np.nan, inplace=True)
if label.isnull().values.any():
label = label.dropna(axis=1, how='any')
#print label
if isinstance(label,pd.DataFrame):
if dataframe['ID'].equals(label['ID']):
outfileHTML.write("<h3>"+"SampleID check results: 'The SampleIDs match for "+data_type+" and "+label_type+" features'"+"</h3>"+"\n")
#outfileHTML.write("<h3>"+"No. of samples: "+"</h3>"+"\n")
print("The SampleIDs match for "+data_type+" and "+label_type+" features")
#dataframe = dataframe.drop(['ID'],axis = 1)
#label = label.drop(['ID'],axis = 1)
else:
sys.exit("The SampleIDs in data and label vary. Cannot proceed further. Please fix and rerun!")
label = label.drop(['ID'],axis = 1)
label_header=list(label.keys())
else:
label_header="NA"
sampleIDs = dataframe['ID']
outfileHTML.write("<h3>"+"No. of samples: "+str(len(sampleIDs))+"</h3>"+"\n")
dataframe = dataframe.drop(['ID'],axis = 1)
dataframe_header=list(dataframe.keys())
## SILENTING normalization on entire dataset as train and test dataset has to undergo normalizations differently later, thereby preventing leakage of test into train.
#if data_normalize_method != 'none':
# print("performing "+data_normalize_method+" normalization for "+data_type+" features")
# outfileHTML.write("<h3>"+"performing "+data_normalize_method+" normalization for "+data_type+" features"+"</h3>"+"\n")
# dataframe = normal_dataframe(dataframe ,data_normalize_method, dataframe_header)
#print(dataframe)
#if isinstance(label,pd.DataFrame):
# if label_normalize_method != 'none':
# #print(label)
# print("performing "+label_normalize_method+" for "+label_type+" features")
# outfileHTML.write("<h3>"+"performing "+label_normalize_method+" for "+label_type+" features"+"</h3>"+"\n")
# label = normal_dataframe(label ,label_normalize_method, label_header)
outfileHTML.close()
#c_=pd.concat([dataframe1, dataframe2], axis=1)
print(dataframe)
#dataframe = dataframe.loc[:, (dataframe!=0).any(axis=0)]; nan_value = float("NaN"); dataframe.replace("", nan_value, inplace=True); dataframe=dataframe.dropna()
#dataframe_header=list(dataframe.keys())
print(dataframe.shape[1])
if isinstance(label,pd.DataFrame):
#label = label.loc[:, (label!=0).any(axis=0)]; label.replace("", nan_value, inplace=True); label=label.dropna()
print(label)
#label_header=list(label.keys())
print(label.shape[1])
return dataframe , label, sampleIDs, label_header, dataframe_header
##Train-test split function
def splitdata(dataframe , label, t_size, mode_, data_normalize_method, label_normalize_method, data_type, label_type, dataframe_header, label_header, tagDir):
outfileHTML=open(str(save_dir)+tagDir+model_type+".output.html",'a')
train, test , Y_train , Y_test = train_test_split(dataframe, label , test_size = t_size)
##Converting numpy arrays to dataframe to perform normalization on them
train=pd.DataFrame(data=train,columns=dataframe_header)
print(train)
#train= train.loc[:, (train!=0).any(axis=0)]; nan_value = float("NaN"); train.replace("", nan_value, inplace=True); train=train.dropna()
test=pd.DataFrame(data=test,columns=dataframe_header)
print(test)
#test= test.loc[:, (test!=0).any(axis=0)]; nan_value = float("NaN"); test.replace("", nan_value, inplace=True); test=test.dropna()
Y_train=pd.DataFrame(data=Y_train,columns=label_header)
print(Y_train)
#Y_train= Y_train.loc[:, (Y_train!=0).any(axis=0)]; nan_value = float("NaN"); Y_train.replace("", nan_value, inplace=True); Y_train=Y_train.dropna()
Y_test=pd.DataFrame(data=Y_test,columns=label_header)
print(Y_test)
#Y_test= Y_test.loc[:, (Y_test!=0).any(axis=0)]; nan_value = float("NaN"); Y_test.replace("", nan_value, inplace=True); Y_test=Y_test.dropna()
#if mode_=="Train":
##Introducing normalizations for TRAIN and TEST datasets.
if data_normalize_method != 'none':
print("performing "+data_normalize_method+" normalization for "+data_type+" features")
##NORMALIZING TRAIN data
outfileHTML.write("<h3>"+"performing "+data_normalize_method+" normalization for "+data_type+" features for TRAIN set"+"</h3>"+"\n")
train= normal_dataframe(train, data_normalize_method, dataframe_header)
##Convert to dataframe with indices
#train_df = pd.DataFrame(train, index = train_features)
#train=train[:, ~np.isnan(train).any(axis=0)]
#train= train.loc[:, (train!=0).any(axis=0)]; nan_value = float("NaN"); train.replace("", nan_value, inplace=True); train=train.dropna()
print("Printing Normalized Train data:")
print(train)
##NORMALIZING TEST set
outfileHTML.write("<h3>"+"performing "+data_normalize_method+" normalization for "+data_type+" features for TEST set"+"</h3>"+"\n")
test= normal_dataframe(test, data_normalize_method, dataframe_header)
#test_df = pd.DataFrame(test, index = test_features)
#test=test[:, ~np.isnan(test).any(axis=0)]
#test= test.loc[:, (test!=0).any(axis=0)]; nan_value = float("NaN"); test.replace("", nan_value, inplace=True); test=test.dropna()
print("Printing Normalized Test data:")
print(test)
if isinstance(label,pd.DataFrame):
if label_normalize_method != 'none':
print("performing "+label_normalize_method+" for "+label_type+" features")
##NORMALIZING TRAINING label
outfileHTML.write("<h3>"+"performing "+label_normalize_method+" for "+label_type+" features for TRAIN set"+"</h3>"+"\n")
Y_train = normal_dataframe(Y_train, label_normalize_method, label_header)
#Y_train_df = pd.DataFrame(Y_train, index = Y_train_features)
#Y_train=Y_train[:, ~np.isnan(Y_train).any(axis=0)]
#Y_train= Y_train.loc[:, (Y_train!=0).any(axis=0)]; nan_value = float("NaN"); Y_train.replace("", nan_value, inplace=True); Y_train=Y_train.dropna()
print("Printing Normalized Train label:")
print(Y_train)
##NORMALIZING TEST label
outfileHTML.write("<h3>"+"performing "+label_normalize_method+" for "+label_type+" features for TEST set"+"</h3>"+"\n")
Y_test = normal_dataframe(Y_test, label_normalize_method, label_header)
#Y_test_df = pd.DataFrame(Y_test, index = Y_test_features)
#Y_test=Y_test[:, ~np.isnan(Y_test).any(axis=0)]
#Y_test= Y_test.loc[:, (Y_test!=0).any(axis=0)]; nan_value = float("NaN"); Y_test.replace("", nan_value, inplace=True); Y_test=Y_test.dropna()
print("Printing Normalized Test label:")
print(Y_test)
outfileHTML.write("<h1 style=text-align:center;color:purple>"+"-------------------------------Number of Samples for Training and Testing---------------------------------"+"</h1>"+"\n")
outfileHTML.write("<h3>"+"No. of samples for training:{}".format(len(train))+"</h3>"+"\n")
outfileHTML.write("<h3>"+"No. of samples for test:{}".format(len(test))+"</h3>"+"\n")
outfileHTML.close()
print("Trainig data:{} , Testing data:{} ".format(len(train) ,len(test)))
#if mode_=="validate":
# print("Custom test data:{} ".format(len(test)))
test.to_csv(str(save_dir)+""+tagDir+"Test_dataset.csv",sep="\t"); Y_test.to_csv(str(save_dir)+""+tagDir+"Y_test_dataset.csv",sep="\t"); Y_train.to_csv(str(save_dir)+""+tagDir+"Y_train_dataset.csv",sep="\t"); train.to_csv(str(save_dir)+""+tagDir+"Train_dataset.csv",sep="\t");
return train , Y_train , test , Y_test
##Build models, multiple options of modeltypes accepted from user per the method list below
def BuildModel(train , Y_train , test , Y_test , method, params, cv_par, scoring_par, gridsearch, param_grid, select_label_var_list, select_data_var_list, data_type, label_type, featureSelFrmModel_flag, tagDir, trainmodel):
'''
initializing model and training the model
'''
outfileHTML=open(str(save_dir)+tagDir+model_type+".output.html",'a')
outfileHTML.write("<h2 style=text-align:center;color:blue>"+"--------------------------Model Summary-----------------------"+"</h2>"+"\n")
if method in ['DecisionTree','LinearRegression', 'LinearModel' , 'LASSO', 'multiTaskLASSO', 'multiTaskLinearModel', 'RandomForest', 'DecisionTreeClassifier', 'MLPClassifier', 'LogisticRegression', 'SVC']:
if gridsearch == 'True':
#param_grid_keys=list(param_grid.keys())
if 'cv' in param_grid.keys():
cv_grid=param_grid['cv']
del param_grid['cv']
else:
cv_grid=None
if 'scoring' in param_grid.keys():
scoring_grid=param_grid['scoring']
del param_grid['scoring']
else:
scoring_grid=None
#for i in param_grid_keys:
# if re.search('cv',i):
# cv_grid = param_grid[i]
# del param_grid[i]
# elif re.search('scoring',i):
# scoring_grid = param_grid[i]
# del param_grid[i]
# else:
# continue
if method == 'LASSO':
if gridsearch == 'True':
try:
print(" Starting grid search for LASSO")
model = GridSearchCV(linear_model.Lasso(), param_grid=param_grid,cv=cv_grid,scoring=scoring_grid)
#print(model.get_params)
except:
print("Grid search status:{}".format(grid_search))
else:
#model = linear_model.Lasso(alpha=alpha_lasso)
model = linear_model.Lasso(**params)
#print(model.get_params)
if method == 'DecisionTree':
if gridsearch == 'True':
try:
print("starting grid search for Decision Tress")
model = GridSearchCV(tree.DecisionTreeRegressor(), param_grid=param_grid, scoring=scoring_grid, cv=cv_grid)
except:
print("Grid search status:{}".format(grid_search))
else:
model = tree.DecisionTreeRegressor(**params)
#print(model.get_params)
if method == 'LinearRegression':
if gridsearch == 'True':
try:
print("starting grid search for Linear Regression")
model = GridSearchCV(LinearRegression(), param_grid=param_grid, scoring=scoring_grid, cv=cv_grid)
except:
print("Grid search status:{}".format(grid_search))
else:
model = LinearRegression(**params)
#print(model.get_params)
if method == 'LinearModel':
if gridsearch == 'True':
#random_state=param_grid['random_state']
#del new_param_grid['random_state']
try:
print("Starting grid search for ElasticNet")
model = GridSearchCV(ElasticNet(), param_grid=param_grid , cv = cv_grid , scoring=scoring_grid)
except:
print("Grid search status:{}".format(grid_search))
else:
model = ElasticNet(**params)
#print(model.get_params)
if method == 'multiTaskLASSO':
if gridsearch == 'True':
try:
print(" Starting grid search for MultiTaskLASSO")
model = GridSearchCV(linear_model.MultiTaskLasso(), param_grid=param_grid,cv=cv_grid,scoring=scoring_grid)
#print(model.get_params)
except:
print("Grid search status:{}".format(grid_search))
else:
#model = linear_model.Lasso(alpha=alpha_lasso)
model = linear_model.MultiTaskLasso(**params)
#print(model.get_params)
if method == 'multiTaskLinearModel':
if gridsearch == 'True':
#random_state=param_grid['random_state']
#del new_param_grid['random_state']
try:
print("Starting grid search for MultiTaskElasticNet")
model = GridSearchCV(MultiTaskElasticNet(), param_grid=param_grid , cv = cv_grid , scoring=scoring_grid)
except:
print("Grid search status:{}".format(grid_search))
else:
model = MultiTaskElasticNet(**params)
if method == 'RandomForest':
if gridsearch == 'True':
#random_state=param_grid['random_state']
#del new_param_grid['random_state']
try:
print("Starting grid search for RandomForest")
model = GridSearchCV(RandomForestClassifier(), param_grid=param_grid , cv = cv_grid , scoring=scoring_grid)
except:
print("Grid search status:{}".format(grid_search))
else:
#model = MultiTaskElasticNet(**params)
model = RandomForestClassifier(**params)
if method == 'DecisionTreeClassifier':
if gridsearch == 'True':
#random_state=param_grid['random_state']
#del new_param_grid['random_state']
try:
print("Starting grid search for DecisionTreeClassifier")
model = GridSearchCV(DecisionTreeClassifier(), param_grid=param_grid , cv = cv_grid , scoring=scoring_grid)
except:
print("Grid search status:{}".format(grid_search))
else:
#model = MultiTaskElasticNet(**params)
model = DecisionTreeClassifier(**params)
if method == 'MLPClassifier':
if gridsearch == 'True':
#random_state=param_grid['random_state']
#del new_param_grid['random_state']
try:
print("Starting grid search for MLPClassifier")
model = GridSearchCV(MLPClassifier(), param_grid=param_grid , cv = cv_grid , scoring=scoring_grid)
except:
print("Grid search status:{}".format(grid_search))
else:
#model = MultiTaskElasticNet(**params)
model = MLPClassifier(**params)
if method == 'LogisticRegression':
if gridsearch == 'True':
try:
print("starting grid search for Logistic Regression")
model = GridSearchCV(LogisticRegression(), param_grid=param_grid, scoring=scoring_grid, cv=cv_grid)
except:
print("Grid search status:{}".format(grid_search))
else:
model = LogisticRegression(**params)
if method == 'SVC':
if gridsearch == 'True':
try:
print("starting grid search for Logistic Regression")
model = GridSearchCV(SVC(), param_grid=param_grid, scoring=scoring_grid, cv=cv_grid)
except:
print("Grid search status:{}".format(grid_search))
else:
model = SVC(**params)
else:
print("options are :DecisionTree, LinearRegression, LinearModel, LASSO, LinearModel (aka ElasticNet), multiTaskLinearModel, multiTaskLASSO, RandomForest, DecisionTreeClassifier, MLPClassifier, LogisticRegression, SVC")
if trainmodel:
#try:
'''
performing training
'''
## ADDING SelectFromModel for Feature Selection using model and using only those features for training and testing further (change featureSelFrmModel_flag to sfm_flag later)
if(featureSelFrmModel_flag==1):
column_headers__=train.columns
selector = SelectFromModel(estimator=model).fit(train, Y_train)
##Skipping the transform as it yields numpy array. Rather getting an array of features with "True or False" for selection. Extracting headers and then selecting only those features from train and test dataframes below.
#train=selector.transform(train)##The output is numpy array without headers.
##Get an array of "True or False" for features. True means selected, False means not-selected. Selection happens through feature importances yield by the model using SelectFromModel function above.
feature_selected_or_not_=selector.get_support()
##Make a dataframe of that array which has the header as names of each feature.
fsd=pd.DataFrame(data=feature_selected_or_not_)
fsd=pd.DataFrame.transpose(fsd)
fsd.columns=column_headers__
print(fsd)
##Drop the features that are "False", i.e. not selected.
fsd=fsd.drop(columns=fsd.columns[(fsd == False).any()])
print("These are the features selected by SelectFromModel function")
print(fsd)
fsd.to_csv(str(save_dir)+""+tagDir+'_'+model_type+'_features_selected.txt')
feature_headers__=fsd.columns
##Converting numpy array to dataframe with headers for selected features.
#train=pd.DataFrame(data=train,columns=feature_headers__)##no need to generate dataframe from numpy array anymore
train=train[feature_headers__]
print("This is the train set post feature selection")
print(train)
##Selecting same features in test as well.
print("This is the test set post feature selection")
test=test[feature_headers__]
print(test)
#outfileHTML=open(model_type+".output.html",'a')
#outfileHTML.write("<h1>"+"--------------------------Model Summary-----------------------"+"</h1>"+"\n")
outfileHTML.write("<h3>"+"Model Type : "+method+"</h3>"+"\n")
if gridsearch == 'True':
grid_result = model.fit(train ,Y_train)
outfileHTML.write("<h4>"+"Grid Search Metrics"+"</h4>"+"\n")
#outfileHTML.write("<h3>"+"Best Parameters: "+"</h3>"+"\n")
#outfileHTML.write("<h3>"+ "scoring="+str(scoring_grid)+"</h3>"+"\n")
outfileHTML.write("\n"+"<h5>"+'Best Score : '+str(grid_result.best_score_)+"</h5>"+"\n")
else:
scores = cross_val_score(model.fit(train ,Y_train), train , Y_train, cv=cv_par , scoring = scoring_par)
outfileHTML.write("<h4>"+"Cross Validation Metrics:"+"</h4>"+"\n")
outfileHTML.write("<h5>"+"Parameters: cv="+str(cv_par)+" scoring="+str(scoring_par)+"</h5>"+"\n")
outfileHTML.write("<h5>"+"Cross validation score:{}".format(-1*scores.mean())+"</h5>"+"\n")
store_params=model.get_params();
outfileHTML.write("<h3>"+"Model Parameters:"+"</h3>"+"\n")
for i in store_params.keys():
outfileHTML.write("<h4>"+str(i)+":"+str(store_params[i])+"</h4>")
#Y_pred_train = model.predict(train)
#print("MSE of Train set:{}".format(metrics.mean_squared_error(Y_train, Y_pred_train)))
#Y_pred = model.predict(test)
#print("MSE of Test set:{}".format(metrics.mean_squared_error(Y_test, Y_pred)))
outfileHTML.close()
evaluate(model,train,Y_train,select_label_var_list,'train_eval',model_type, data_type, label_type, tagDir, gridsearch)
print(Y_test)
returned_label_resulted=evaluate(model,test,Y_test,select_label_var_list,'test_eval',model_type, data_type, label_type, tagDir, gridsearch)
##Permut each label coulmn of the test dataset 100 times and validate the model
for l in returned_label_resulted:
for n in range(1,21):
Y_test[[l]]=np.random.permutation(Y_test[[l]])
evaluate(model, test, Y_test , select_label_var_list, 'validation'+'_permut_'+str(n)+'_'+str(l), model_type, data_type, label_type, tagDir, gridsearch)
#except:
#print(" Issue with model training")
##Print tree for train
return model
##Evaluate the models for test or validation.
def evaluate(model , test , Y_test, select_label_var_list, prefix, model_type, data_type, label_type, tagDir, gridsearch):
'''
evalauting the model preformance
'''
outfileHTML=open(str(save_dir)+tagDir+model_type+".output.html",'a')
Y_pred = model.predict(test)
print(Y_test)
print(Y_pred)
if(prefix=="train_eval"):
heading="Model evaluation for Train data"
color="red"
elif(prefix=="test_eval"):
heading="Model evaluation for Test data"
color="green"
elif(prefix=="validation" or re.search("validation",prefix)):
heading="Model evaluation for Validation data"
color="black"
outfileHTML.write("<h2 style=text-align:center;color:"+color+">"+"----------------"+heading+"--------------------"+"</h2>"+"\n")
outfileHTML.write("<h3>"+"Min Square Error for the Model"+"</h3>"+"\n")
outfileHTML.write("<h4>"+"MSE of "+prefix+" set:{}".format(metrics.mean_squared_error(Y_test, Y_pred))+"</h4>")
#Converting numpy.ndarray to dataframes
column_dict_Y_test=dict()
column_dict_Y_predict=dict()
#for i in label_header:
# column_dict_Y_test.update({i: Y_test[:, label_header.index(i)]})
# column_dict_Y_predict.update({i: Y_predict[:, label_header.index(i)]})
#Y_test_df=pd.DataFrame(column_dict_Y_test)
#Y_pred_df=pd.DataFrame(column_dict_Y_predict)
Y_test_df = pd.DataFrame(data=Y_test, columns=select_label_var_list)
#print Y_test_df
Y_pred_df = pd.DataFrame(data=Y_pred, columns=select_label_var_list)
#print Y_pred_df
##CALCULATE the RATIO of RMSE to Orignal Stdev for each label-feature
ratio_low_dict=dict()
ratio_high_dict=dict()
ratio_dict=dict()
mean_dict=dict()
rmse_dict=dict()
r2_score_dict=dict()
std_dict=dict()
n_rows=(len(select_label_var_list)/5)+1
n_cols=5
n_plots=len(select_label_var_list)
print("performing further calculations for "+prefix)
for i in select_label_var_list:
Y_test_c=pd.DataFrame.to_numpy(Y_test_df[[i]])
Y_pred_c=pd.DataFrame.to_numpy(Y_pred_df[[i]])
#if(i=='module1'):
# print Y_test_c
# print Y_pred_c
rmse=sqrt(metrics.mean_squared_error(Y_test_c, Y_pred_c))
r2score=r2_score(Y_test_c, Y_pred_c)
mean=np.mean(Y_test_c)
stdev=np.std(Y_test_c)
if(stdev==0):
if(mean==0):
print(" The feature column "+i+" has mean and stdev values as zero. Its rmse is "+ str(rmse) + ". Not considering it for rmse:stdev ratio calculation")
ratio="NA"
else:
if(rmse!=0):
#print("Note:The feature column "+i+" has a non-zero mean and stdev is zero. It seems the feature column is monotonic. Its rmse is "+str(rmse)+" .Not considering the feature for rmse:stdev ratio calculation")
#ratio="NA"
print("Note:The feature column "+i+" has a non-zero mean and stdev is zero. It seems the feature column is monotonic. Its rmse is "+str(rmse)+" .Considering rmse:stdev ratio = rmse")
ratio=rmse
elif(rmse==0):
#print("Note:The feature column "+i+" has mean, stdev and rmse as zero. Hence excluding it from rmse:stdev ratio calculation")
#ratio="NA"
print("Note:The feature column "+i+" has mean, stdev and rmse as zero. Hence rmse:stdev ratio =rmse")
ratio=rmse
else:
ratio=abs(rmse)/abs(stdev)
#print i+"\t"+str(mean)
#rmse_n_mean_df.update()
mean_dict.update({i:mean})
rmse_dict.update({i:rmse})
r2_score_dict.update({i:r2score})
std_dict.update({i:stdev})
if(ratio=="NA"):
ratio=-1.0
elif(ratio <= 1.0):
ratio_low_dict.update({i:ratio})
elif(ratio > 1.0):
ratio_high_dict.update({i:ratio})
#ratio_dict.update({i:ratio})
#mean_for_rmse_high_dict.update({i:mean})
#print("HIGH RMSE for "+i+" :{}".format(rmse))
#mse_high_file.write("HIGH MSE for "+i+" :{}".format(mse))
#mse_high_file.write("\n")
ratio_dict.update({i:ratio})
label_header_low_ratio=list(ratio_low_dict.keys())
mean_header=list(mean_dict.keys())
mean_df=pd.DataFrame.from_dict(mean_dict,orient='index',columns=['Observed Mean'])
#mean_df.to_csv(str(save_dir)+prefix+"_"+model_type+"_mean.csv")
std_header=list(std_dict.keys())
std_df=pd.DataFrame.from_dict(std_dict,orient='index',columns=['Observed Stdev'])
#std_df.to_csv(str(save_dir)+prefix+"_"+model_type+"_std.csv")
rmse_header=list(rmse_dict.keys())
rmse_df=pd.DataFrame.from_dict(rmse_dict,orient='index',columns=['RMSE between observed and predicted values'])
ratio_header=list(ratio_dict.keys())
ratio_df=pd.DataFrame.from_dict(ratio_dict,orient='index',columns=['Ratio_of_RMSE_and_Stdev'])
r2_score_header=list(r2_score_dict.keys())
r2_score_df=pd.DataFrame.from_dict(r2_score_dict,orient='index',columns=['r2_score'])
#rmse_df.to_csv(str(save_dir)+prefix+"_"+model_type+"_rmse.csv")
#print rmse_header
#print mean_header
rmse_n_mean_df = rmse_df.merge(mean_df, how='outer', left_index=True, right_index=True)
rmse_n_mean_n_std_df = rmse_n_mean_df.merge(std_df,how='outer', left_index=True, right_index=True)
rmse_n_mean_n_std_n_ratio_df = rmse_n_mean_n_std_df.merge(ratio_df,how='outer', left_index=True, right_index=True)
rmse_n_mean_n_std_n_ratio_n_r2_score_df = rmse_n_mean_n_std_n_ratio_df.merge(r2_score_df,how='outer', left_index=True, right_index=True)
count=0
Only_ratio_n_mean_df=rmse_n_mean_n_std_n_ratio_df[["Observed Mean","Ratio_of_RMSE_and_Stdev"]]
#Only_ratio_n_mean_df.plot.scatter("Observed Mean","Ratio_of_RMSE_and_Stdev")
Only_ratio_n_mean_df.plot.bar()
#for e in list(rmse_n_mean_n_std_n_ratio_df.index.values):
# mp.scatter(rmse_n_mean_n_std_n_ratio_df.loc(e,"Observed Mean"), rmse_n_mean_n_std_n_ratio_df.loc(e,"Ratio_of_RMSE_and_Stdev"), label=e, marker=count)
# count=count+1
# if(count==11):
# count=0
rmse_n_mean_n_std_n_ratio_df.to_csv(str(save_dir)+""+tagDir+prefix+'_'+model_type+'_rmse_mean_std_and_ratio.csv')
rmse_n_mean_n_std_n_ratio_n_r2_score_df.to_csv(str(save_dir)+""+tagDir+prefix+'_'+model_type+'_rmse_mean_std_and_ratio_and_r2_score.csv')
#if len(list(rmse_n_mean_n_std_n_ratio_df.index.values)) <= 40:
# mp.legend(loc=(1.04,0))
#else:
# mp.legend(bbox_to_anchor=(1.04, 1.04, 2.04, 2.04), loc='upper left', ncol=2, mode="expand")
mp.xticks(rotation=90)
mp.ylabel('Ratio_of_RMSE_and_Stdev')
mp.xlabel(label_type+" Features")
mp.title("Observed Mean and Ratio_of_RMSE_and_Stdev",size=12)
mp.legend(loc=(1.04,0.5))
#mp.savefig(str(save_dir)+'test_result_plots_high_rmse.png',orientation='landscape',dpi=200,bbox_inches='tight')
mp.savefig(str(save_dir)+""+tagDir+prefix+'_'+model_type+'_rmse_mean_std_and_ratio.png',bbox_inches='tight')
mp.clf()
Only_ratio_n_r2_score_df=rmse_n_mean_n_std_n_ratio_n_r2_score_df[["Ratio_of_RMSE_and_Stdev","r2_score"]]
Only_ratio_n_r2_score_df.plot.bar()
mp.xticks(rotation=90)
mp.ylabel('r2_score')
mp.xlabel(label_type+" Features")
mp.title("Ratio_of_RMSE_and_Stdev and r2_score",size=12)
mp.legend(loc=(1.04,0.5))
#mp.savefig(str(save_dir)+'test_result_plots_high_rmse.png',orientation='landscape',dpi=200,bbox_inches='tight')
mp.savefig(str(save_dir)+""+tagDir+prefix+'_'+model_type+'_rmse_std_and_ratio_and_r2_score.png',bbox_inches='tight')
mp.clf()
Y_test_low_ratio_df=Y_test_df[label_header_low_ratio]
Y_pred_low_ratio_df=Y_pred_df[label_header_low_ratio]
count=0
for c in Y_pred_low_ratio_df.columns:
mp.scatter(Y_test_low_ratio_df[c], Y_pred_low_ratio_df[c], label=c, marker=count)
count=count+1
if(count==11):
count=0
mp.xlabel('Actual_values')
mp.ylabel('Predicted_values')
mp.title("Actual_values v/s Predicted Values - for features with Low RMSE:Actual_Stdev",size=9)
#mp.legend(loc=(1.04,0))
#mp.legend(bbox_to_anchor=(0., 1.02, 1., .102), loc='lower left',
# ncol=2, mode="expand")
if len(label_header_low_ratio) <= 40:
mp.legend(loc=(1.04,0))
else:
mp.legend(bbox_to_anchor=(1.04, 1.04, 2.04, 2.04), loc='upper left', ncol=2, mode="expand")
#mp.savefig(str(save_dir)+'test_result_plots_low_rmse.png',orientation='landscape',dpi=200,bbox_inches='tight')
mp.savefig(str(save_dir)+""+tagDir+prefix+'_'+model_type+'_test_result_plots_low_ratio.png',bbox_inches='tight')
mp.clf()
label_header_high_ratio=list(ratio_high_dict.keys())
Y_test_high_ratio_df=Y_test_df[label_header_high_ratio]
Y_pred_high_ratio_df=Y_pred_df[label_header_high_ratio]
count=0
for d in Y_pred_high_ratio_df.columns:
mp.scatter(Y_test_high_ratio_df[d], Y_pred_high_ratio_df[d], label=d, marker=count)
count=count+1
if(count==11):
count=0
mp.xlabel('Actual_values')
mp.ylabel('Predicted_values')
mp.title("Actual_values v/s Predicted Values - for features with high RMSE:Actual_Stdev",size=9)
mp.legend(loc=(1.04,0))
#mp.savefig(str(save_dir)+'test_result_plots_high_rmse.png',orientation='landscape',dpi=200,bbox_inches='tight')
mp.savefig(str(save_dir)+""+tagDir+prefix+'_'+model_type+'_test_result_plots_high_ratio.png',bbox_inches='tight')
mp.clf()
#Merging mean_df and rmse_dict
if(len(ratio_low_dict)>1):
ratio_low_df=pd.DataFrame.from_dict(ratio_low_dict,orient='index',columns=['RMSE/Stdev'])
#mean_for_rmse_low_df=pd.DataFrame.from_dict(mean_for_rmse_low_dict,orient='index',columns=['Mean'])
#mse_df=pd.DataFrame({'Label_features':label_list, 'MSE':mse_val})
#print(mse_df)
#mse_df.set_index('Label_features')
ratio_low_df.to_csv(str(save_dir)+""+tagDir+prefix+"_"+model_type+"_Labels_with_Low_Ratio.csv")
#mean_for_rmse_low_df.to_csv(str(save_dir)+prefix+"_"+model_type+"_Original_Mean_for_Labels_with_Low_RMSE.csv")
#rmse_low_n_mean_df = rmse_low_df.merge(mean_for_rmse_low_df, how='outer', left_index=True, right_index=True)
#mp.plot(rmse_low_df)
ratio_low_df.plot(kind='bar')
mp.ylabel('RMSE/Stdev')
mp.xlabel('Label Features')
mp.xticks(rotation=90)
if len(label_header_low_ratio) <= 40:
mp.xticks(size=5)
else:
mp.xticks(size=3)
##mp.title(i,size=1)
mp.title("Low RMSE/Stdev for the label features",size=12)
mp.savefig(str(save_dir)+""+tagDir+prefix+'_'+model_type+'_Low_Ratio_plot.png',orientation='landscape',dpi=100,bbox_inches='tight')
mp.clf()
#pd.plotting.scatter_matrix(rmse_low_n_mean_df)
#mp.title("Comparing labels with Low RMSEs against their actual Mean values")
#mp.savefig(str(save_dir)+prefix+'_'+model_type+'_Labels_w_Low_RMSE_and_actual_Means_plot.png',orientation='landscape',dpi=100,bbox_inches='tight')
#mp.clf()
else:
print("Only 1 key:value pair in ratio_low_dict, so not proceeding with its plotting")
if(len(ratio_high_dict)>1):
ratio_high_df=pd.DataFrame.from_dict(ratio_high_dict,orient='index',columns=['RMSE/Stdev'])
#mean_for_rmse_high_df=pd.DataFrame.from_dict(mean_for_rmse_high_dict,orient='index',columns=['RMSE'])
#mse_df=pd.DataFrame({'Label_features':label_list, 'MSE':mse_val})
#print(mse_df)
#mse_df.set_index('Label_features')
ratio_high_df.to_csv(str(save_dir)+""+tagDir+prefix+"_"+model_type+"_Labels_with_High_Ratio.csv")
#mean_for_rmse_high_df.to_csv(str(save_dir)+prefix+"_"+model_type+"_Original_Mean_for_Labels_with_High_RMSE.csv")
#rmse_high_n_mean_df = rmse_high_df.merge(mean_for_rmse_high_df, how='outer', left_index=True, right_index=True)
mp.plot(ratio_high_df)
mp.ylabel('RMSE/Stdev')
mp.xlabel('Label Features')
mp.xticks(rotation=90)
mp.xticks(size=4)
mp.title("High RMSE/Stdev for the label features",size=12)
mp.savefig(str(save_dir)+""+tagDir+prefix+'_'+model_type+'_High_Ratio_plot.png',orientation='landscape',dpi=100,bbox_inches='tight')
mp.clf()
#pd.plotting.scatter_matrix(rmse_high_n_mean_df)
#mp.title("Comparing labels with High RMSEs against their actual Mean values")
#mp.savefig(str(save_dir)+prefix+'_'+model_type+'_Labels_w_High_RMSE_and_actual_Means_plot.png',orientation='landscape',dpi=100,bbox_inches='tight')
#mp.clf()
else:
print("Only 1 key:value pair in ratio_high_dict, so not proceeding with its plotting")
if(prefix=="train_eval"):
heading="Model evaluation for Train data"
elif(prefix=="test_eval"):
heading="Model evaluation for Test data"
elif(prefix=="validation"):
heading="Model evaluation for Validation data"
outfileHTML.write("<h4 stype=text-align:center;color:brown>"+"No. of features showing LOW 'RMSE/Stdev' (<=1.0): "+"\n"+str(len(label_header_low_ratio))+"</h4>")
outfileHTML.write("<h5>"+"All such features with their Low 'RMSE/Stdev' values could be found in output file: "+prefix+"_"+model_type+"_Labels_with_Low_Ratio.csv"+"</h4>"+"\n")
outfileHTML.write("<h4 stype=text-align:center;color:brown>"+"No. of features showing HIGH 'RMSE/Stdev' (>1.0): "+"\n"+str(len(label_header_high_ratio))+"</h4>")
outfileHTML.write("<h5>"+"All such features with their High 'RMSE/Stdev' values could be found in output file: "+prefix+"_"+model_type+"_Labels_with_High_Ratio.csv"+"</h4>"+"\n"+"\n")
outfileHTML.write("<h3>"+heading+" for label features showing Low 'RMSE/Stdev' (<=1.0)"+"</h3>"+"\n")
data_image1 = open(str(save_dir)+tagDir+prefix+'_'+model_type+'_test_result_plots_low_ratio.png', 'rb').read().encode('base64').replace('\n', '')
img_tag1 = '<img src="data:image/png;base64,{0}">'.format(data_image1)
outfileHTML.write(img_tag1+"\n")
if(os.path.exists(""+tagDir+prefix+'_'+model_type+'_Low_Ratio_plot.png')):
data_image2 = open(str(save_dir)+tagDir+prefix+'_'+model_type+'_Low_Ratio_plot.png', 'rb').read().encode('base64').replace('\n', '')
img_tag2 = '<img src="data:image/png;base64,{0}">'.format(data_image2)
outfileHTML.write(img_tag2+"\n")
data_image3 = open(str(save_dir)+tagDir+prefix+'_'+model_type+'_rmse_mean_std_and_ratio.png', 'rb').read().encode('base64').replace('\n', '')
img_tag3 = '<img src="data:image/png;base64,{0}">'.format(data_image3)
outfileHTML.write(img_tag3+"\n")
outfileHTML.write('Content-type: text/html\n\n'+""+"\n"+'<link href="default.css" rel="stylesheet" type="text/css" />')
decision_thresholds=[0.1,0.2,0.3,0.4,0.5,0.6,0.7,0.8,0.9]
##Adding feature-importance calculations
FI_handle=open(str(save_dir)+tagDir+prefix+"_"+"FeatureImportances.txt",'w+')
X_features=list(test.columns)
Y_features=list(Y_test.columns)
if(gridsearch == 'True'):
FI_handle.close()
elif(model_type == "MLPClassifier" or model_type == "SVC"):
FI_handle.close()
elif(model_type!="DecisionTree" and model_type!="RandomForest" and model_type!="DecisionTreeClassifier"):
importance = model.coef_
FI_handle.write("Y_Features, X_features_scored:"+str(X_features)+"\n")
for u,k in enumerate(importance):
FI_handle.write(str(Y_features[u])+","+str(k)+"\n")
FI_handle.close()
else:
importance = model.feature_importances_
FI_handle.write("Features, Score"+"\n")
for u,k in enumerate(importance):
FI_handle.write(str(X_features[u])+","+str(k)+"\n")
FI_handle.close()
##CALCULATE AUCs BASED on the decision thresholds list above.
AUC_fH=open(str(save_dir)+tagDir+prefix+"_"+"AUC_values.txt",'w+')
AUC_fH.write("label"+"\t"+"AUC_value"+"\t"+"decision_threshold"+"\n")
label_resulted=[]
if(model_type == 'RandomForest' or model_type == 'DecisionTreeClassifier' or model_type == 'MLPClassifier' or model_type == 'LogisticRegression'):
for l in select_label_var_list:
print l
fpr, tpr, thresholds = roc_curve(Y_test_df[[l]],Y_pred_df[[l]])
AUC_value=roc_auc_score(Y_test_df[[l]],Y_pred_df[[l]])
if(AUC_value>0.75 and r2_score_dict[l]>0.25):
if l not in label_resulted:
label_resulted.append(l)
AUC_fH.write(l+"\t"+str(AUC_value)+"\n")
if(math.isnan(float(AUC_value))):
print "###AUC_value is nan#######"
print fpr; print tpr; print "#################"
else:
mp.xlabel('False Positive Rate')
mp.ylabel('True Positive Rate')
#mp.ylabel('AUC')
mp.ylim([0.0, 1.05])
mp.xlim([0.0,1.0])
#mp.xticks([0.0,0.2,0.4,0.6,0.8,1.0])
mp.plot(fpr,tpr,color="darkorange",lw=2,label="ROC curve (area = %0.2f)" % AUC_value)
mp.title(' Receiver Operating Characteristic curve for'+l,fontsize=15)
mp.legend(loc="lower right")
mp.savefig(str(save_dir)+tagDir+prefix+'_'+model_type+'_AUC'+l+'.png',orientation='landscape',dpi=100,bbox_inches='tight')
mp.clf()
data_image4 = open(str(save_dir)+tagDir+prefix+'_'+model_type+'_AUC'+l+'.png', 'rb').read().encode('base64').replace('\n', '')
img_tag4 = '<img src="data:image/png;base64,{0}">'.format(data_image4)
outfileHTML.write(img_tag4+"\n")
AUC_fH.close()
outfileHTML.write('Content-type: text/html\n\n'+""+"\n"+'<link href="default.css" rel="stylesheet" type="text/css" />')
#fig = mp.figure(figsize=(25,20))
#tree_= plot_tree(model,
# feature_names=X_features,
# class_names=Y_features,
# filled=True)
#fig.savefig(prefix+'_'+model_type+"decistion_tree.png")
#dotfile = open(str(save_dir)+tagDir+prefix+"_dtree.dot", 'w')
#tree.export_graphviz(model, out_file = dotfile, feature_names = test.columns)
#dotfile.close()
#os.system("dot -Tpng "+str(save_dir)+tagDir+prefix+"_dtree.dot"+" -o "+str(save_dir)+str(tagDir)+str(prefix)+"_dtree.png")
#tree.export_graphviz(model, out_file=prefix+'_'+model_type+'_tree.dot')
return label_resulted
for l in select_label_var_list:
print l
#fig, ax_=mp.subplots(1,1,figsize=(9,9))
decision_threshold_list=[]
AUC_value_list=[]
#print Y_test_df[l]
#print Y_test_df[[l]]
for i in decision_thresholds:
print i
Y_test_df_n=binarize(pd.DataFrame.to_numpy(abs(Y_test_df[[l]])),threshold=i)
#Y_pred_df_n=binarize(pd.DataFrame.to_numpy(abs(Y_pred_df[[l]])),threshold=i)
Y_pred_df_n=Y_pred_df[[l]]##Keeping Ypredict to be continuous,i.e. as it is.
if(np.all((Y_test_df_n==0))):
print("For decision threshold "+str(i)+":")
print("Seems all values for label column "+l+" are zero. Hence not considering it for decision_threshold vs AUC plot")
else:
#fpr, tpr, thresholds = roc_curve(Y_test_df[[l]], Y_pred_df[[l]])
fpr, tpr, thresholds = roc_curve(Y_test_df_n,Y_pred_df_n)
print thresholds
print fpr
print tpr
#print auc(fpr,tpr)
#print l; print i
AUC_value=auc(fpr,tpr)
if(AUC_value>0.75 and r2_score_dict[l]>0.25):
if l not in label_resulted:
label_resulted.append(l)
AUC_fH.write(l+"\t"+str(AUC_value)+"\t"+str(i)+"\n")
if(math.isnan(float(AUC_value))):
print "###AUC_value is nan#######"
print fpr; print tpr; print "#################"
continue
AUC_value_list.append(AUC_value)