-
Notifications
You must be signed in to change notification settings - Fork 0
/
main.R
1866 lines (1555 loc) · 134 KB
/
main.R
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
gc()
rm(list = ls())
# #### settings for reticulate (interface to Python)
# python_path = '/home/jessica/miniconda3/bin' #insert python path
# setwd('.') #set the working directory
# Sys.setenv(RETICULATE_PYTHON = python_path)
# library(reticulate)
#
# use_python(python = python_path)
# MKL : https://academic.oup.com/bioinformatics/article/34/6/1009/4565592
#Unsupervised multiple kernel learning for heterogeneous data integration
# #### settings to set the conda environment
setwd(".") # set working directory
library(reticulate)
use_condaenv(condaenv="r_4.3") #set path to conda environment
# ####
# Load all required packages (reticulate was loaded before)
library(Boruta)
library(caret)
library(cutpointr)
library(dplyr)
library(doParallel)
library(entropy) # for unsup feat selection via entropy estimation
library(foreach)
library(genieclust)
library(ggplot2)
library(glmnet)
library(kernlab)
library(igraph)
library(impute)
library(intrinsicDimension)
library(intRinsic)
library(limma)
library(loe)
library(missRanger)
library(mixKernel) # for MKL di Mariette and Vollaneaux
library(SNFtool)
library(MOFA2)
library(parallel)
library(PerfMeas)
library(plyr)
library(preprocessCore)
library(randomForest)
library(ranger)
library(Rdimtools)
library(readr)
library(readxl)
library(RMKL) # per supervised multiple kernel learning (install from source file, it is archived)
library(ROCR)
library(R.utils)
library(rsvd)
library(StatMatch)
library(stringr)
library(tsne)
library(umap)
library(tableone)
library(writexl)
source('./analisi_RF_myOpti_ranger_breiman.R')
source('./blocking_ID.R')
source('./cluster_features_par.R')
source('./compute_distance.R')
source('./compute_perf_eval.R')
source('./create_mofa_embedding.R')
source('./estimate_ID_Danco.R')
source('./estimate_ID_twoNN.R')
source('./evaluate_ntree_mtry_ranger_breiman.R')
source('./feature_selection_ranger_breiman.R')
source('./filter_cols_by_p.R')
source('./fix_dim_names.R')
source('./hierarchical_svd_entropy_par.R')
source('./RCUR_selection_par.R')
source('./reduce_data_dimensionality.R')
source('./remove_correlated_par.R')
source('./utilities.R')
#reticulate::source_python('embedding.py')
# defined path to data
dati_rinominati <- file.path('./data')
ext_no_neighs <<- 5
ext_seed <<- 3
dim_red <- list(
"umap" = function(x=NULL, k=NULL, method = 'euclidean', n_neighbors = ext_no_neighs){
set.seed(3)
cat('umap with ', method, ' distance\n')
if ((sum(is.na(x)))>0){
x_dist = compute_distance(x, method = method)
proj_data = umap(x_dist, n_components = k,
n_neighbors = n_neighbors, input="dist", method = 'umap-learn')
}else{
proj_data = umap(x, n_components = k,
n_neighbors = n_neighbors, method = 'umap-learn')
}
return(as.matrix(proj_data$layout))},
"tsne" = function(x=NULL, k=NULL){
set.seed(3)
cat('tsne\n')
initial_dims = ifelse(ncol(x)<(nrow(x)-1), min(k*10, round(ncol(x)*2/3)), (nrow(x)-1))
return(tsne(imputa(x), k = k, initial_dims = initial_dims))
},
"rpca" = function(x=NULL, k=NULL){
cat('RPCA\n')
s <- rpca(imputa(x),k = k,
center = TRUE,
scale = TRUE,
retx = TRUE,
rand = TRUE)
return(s$x)
},
"rcur" = function(x=NULL, k=NULL){
r <- rcur(imputa(x),k = k,
rand = TRUE)
x = x[ , r$C.idx]
cat(row.names(x))
cat('returning mat with dimension:', dim(x), '\n')
return(x)
},
"cur" = function(x=NULL, k=NULL){
r <- rcur(imputa(x),k = k,
rand = FALSE)
x = x[ , r$C.idx]
cat(row.names(x))
cat('returning mat with dimension:', dim(x), '\n')
return(x)
},
"laplacianEigenmaps_dimRedtools" = function(x = NULL, k = NULL, type = c("knn", ext_no_neighs)) {
# type tells how to connect the graph. type proportion means that the 10% of edges are kept
# type = c("knn", n_neighbors) composes knn neighborhoods and guarantess connectivity
# usa gli eigenvectors, quindi k deve per forza essere il minimo tra k, ncol(x)-1, nrow(x)-1
cat('LaplacianEigenmaps\n')
set.seed(3)
xx = do.lapeig(imputa(x), ndim = min(k, nrow(x)-1), type = type)
return(xx$Y)
},
"mofa" = function(x = NULL, k = 15, seed = ext_seed, no_mofa_factors = 15,
sparse_weights = TRUE, sparse_factors = FALSE){
k = min(k, no_mofa_factors)
mofaobj = create_mofa_from_matrix(list(t(x)))
data_opts <- get_default_data_options( mofaobj )
#data_opts$scale_views = TRUE
model_opts <- get_default_model_options( mofaobj )
model_opts$num_factors <- k
# model_opts$spikeslab_factors = sparse_factors
# model_opts$spikeslab_weights = sparse_weights
# model_opts$ard_factors = sparse_factors
# model_opts$ard_weights = sparse_weights
train_opts <- get_default_training_options( mofaobj )
train_opts$seed <- seed
train_opts$convergence_mode <- "slow"
mofaobj <- prepare_mofa( mofaobj,
training_options = train_opts,
model_options = model_opts,
data_options = data_opts )
mofaobj <- run_mofa( mofaobj, outfile = file.path('./', 'MOFA.hdf5'), save_data = FALSE, use_basilisk = TRUE)
Z = mofaobj@expectations$Z # latent factor matrix
Z = Z$group1
return(Z)
}
)
codes = unique(unlist(lapply(lapply(list.files(file.path(dati_rinominati)), str_split_1, pattern ='_'), function(x) x[1])))
#codes = c('BLCA2', 'BRCA1', 'BRCA2', 'KIRC1', 'LUAD2', 'LUSC3', 'OV1', 'PRAD1', 'SKCM1')
# To compare ID and NoID
# main_best_models(str_add_save = 'compNew_speedy_',
# cohort_list = codes,
# use_id_list = list(noID_1 = list(use_id = FALSE, frac_samples = 1, minNumViews = 4),
# noID_2 = list(use_id = FALSE, frac_samples = 2, minNumViews = 4),
# ID = list(use_id = TRUE, frac_samples = NA, minNumViews = 3)),
# just_collect_results = FALSE,
# n_internal_iters = 1, n_external_iters = 15,
# perc_samples = 1,
# ntree = NA, mtry = NA, # use default parameters for RF
# timeout_per_int_iter = 10,
# feature_selection_method_list = 'none',
# unsup_feature_selection_list = c('feature_clustering', 'par_rcur', 'NULL'),
# best_red_list = c('rcur', 'rpca','laplacianEigenmaps_dimRedtools', 'tsne', 'umap', 'NULL'),
# regenerate_DR_data = list(miRna = FALSE, RnaSeq = FALSE, RPPA = FALSE, methy = FALSE), # to regenerate some data
# data_integration_list = c('concatenation', 'SNF_nonorm','mofa', 'mkl', 'mofaPT', 'SNF_nonormPT', 'mklPT'), # se Lenovo 'here we ask to also integrate pt data
# regenerate_integration_list = list('concatenation'=FALSE, 'SNF_nonorm'=FALSE, 'SNF_nonormPT'=FALSE, 'mofa'=FALSE,
# 'mofaPT'=FALSE, 'mkl'=FALSE, 'mklPT'=FALSE),
# recompute_RF = FALSE,
# only_embed_dataset = FALSE, only_integrate_views = FALSE)
# For mofa
# only ID
# main_best_models(str_add_save = 'compNew_speedy_',
# use_id_list = list(ID = list(use_id = TRUE, frac_samples = 3, minNumViews = 3)),
# just_collect_results = FALSE,
# n_internal_iters = 1, n_external_iters = 15,
# perc_samples = 1,
# ntree = NA, mtry = NA, # use default parameters for RF
# feature_selection_method_list = 'none',
# unsup_feature_selection_list = c('feature_clustering', 'par_rcur', 'NULL'),
# best_red_list = c('tsne','rpca','laplacianEigenmaps_dimRedtools', 'umap', 'NULL'),
# regenerate_DR_data = list(miRna = FALSE, RnaSeq = FALSE, RPPA = FALSE, methy = FALSE), # to regenerate some data
# data_integration_list = c('concatenation', 'SNF_nonorm','mofa', 'mkl', 'mofaPT', 'SNF_nonormPT', 'mklPT'), # se Lenovo 'here we ask to also integrate pt data
# regenerate_integration_list = list('concatenation'=FALSE, 'SNF_nonorm'=FALSE, 'SNF_nonormPT'=FALSE, 'mofa'=FALSE,
# 'mofaPT'=FALSE, 'mkl'=FALSE, 'mklPT'=FALSE),
# recompute_RF = FALSE,
# timeout_per_int_iter = 3,
# timeout_per_ext_iter = 3,
# only_embed_dataset = FALSE, only_integrate_views = FALSE)
# Main function for DR+data fusion pipelines
main_best_models <- function(str_add_save = 'BestModels_', data_types = c('miRna', 'RnaSeq', 'RPPA', 'methy'),
maxVars = 30000,
with_pt = TRUE,
cohort_list = codes,
normalize_fun = 'standard_scaler',
use_id_list = list(noID_1 = list(use_id = FALSE, frac_samples = 1, minNumViews = 4),
noID_2 = list(use_id = FALSE, frac_samples = 2, minNumViews = 4),
ID = list(use_id = TRUE, frac_samples = NA, minNumViews = 3)),
# each element of the list above is a list with three elements: use_id, frac_samples, minNunViews
# - frac_samples is related to use_id = FALSE
# when the corresponding value in use_id_list is TRUE --> the ID is used and frac_samples is NA
# OTHERWISE
# when the corresponding value in use_id_list is FALSE --> no ID is used to define the dimensionality of the reduced space
# In this case the lower-dim (minD) is defined as:
# if (frac_samples >= 1 ) minD = (min(N,D))-1
# else minD = round(min(N,D)/frac_samples)
# - minNumViews is set to all views (4) when use_ID = FALSE, else we consider other combinations
factor_SSS = 1,
# factor_SSS is the coefficient that I use to decide
# whether a view is to big and may benefit from the intermediate reduction.
# If D/N > factor_SSS then I use unsup FS , otherwise I don't
unsup_feature_selection_list = c('feature_clustering', 'par_rcur', 'NULL'), # can't use simple RCUR here because it will select at most min(N,D) variables
best_red_list = c('rpca', 'laplacianEigenmaps_dimRedtools', 'tsne', 'umap', 'NULL'),
rcur_patience = 1000,
no_neighs = ext_no_neighs,
data_integration_list = c('concatenation', 'SNF_nonorm', 'mofa', 'mkl', 'mofaPT', 'SNF_nonormPT', 'mklPT'), # 'here we ask to also integrate pt data
#default MOFA parameters
no_mofa_factors = 15, sparse_weights = TRUE, sparse_factors = FALSE, #mofa default
RF_method_list = c('RF'), #, 'extratrees'),
k_snf = 20, # default SNF parameters: k = number of neighbors
sigma_snf = 0.5, # default SNF parameters: sigma = sigma value in affinity matrix
t_snf = 11, # default SNF parameters: t = number of iterations
partial_views = TRUE,
feature_selection_method_list = c('RF_importance'),
#feature_selection_method may also be c("corr-pearson", "corr-kendall", "corr-spearman", 'none'),
perc_samples = 1,
ratio_train = 0.9,
seed = ext_seed,
outcome_col = 'os',
neg_label = '0',
pos_label = '1',
measures = c("aucpr", "auc", 'sens', 'spec', 'ppv', 'npv', 'acc', 'f'),
opti_meas = 'aucpr',
weights_opti_meas = 1,
n_internal_iters = 21,
n_external_iters = 15,
tuneRF = FALSE,
mtry = NA,
ntree = NA,
class_weights = 'none',
balanced_bootstraps = TRUE,
thr_conf = 0.05,
vis_norm_data = FALSE,
ncores = max(min(my_detectCores(), detectCores()-5),1),
timeout_per_int_iter = 15,
timeout_per_ext_iter = 15,
just_collect_results = FALSE, # to only collect already run results
only_compute_blocking = FALSE, # just compute blocking IDs
only_embed_dataset = FALSE, # just arrive until embedding is done and do not perform integration and classification
only_integrate_views = FALSE, # just arrive until integration and do not perform classification
regenerate_blocking = list(miRna = FALSE, RnaSeq = FALSE, RPPA = FALSE, methy = FALSE),
regenerate_DR_data = list(miRna = FALSE, RnaSeq = FALSE, RPPA = FALSE, methy = FALSE),
regenerate_integration_list = list('concatenation'=FALSE, 'SNF_nonorm'=FALSE, 'SNF_nonormPT'=FALSE, 'mofa'=FALSE,
'mofaPT'=FALSE, 'mkl'=FALSE, 'mklPT'=FALSE),
recompute_RF = FALSE){
str_add_save = paste0(ifelse(str_add_save != '', paste0(str_add_save ,'_'),''),
'useID_', paste(names(use_id_list), collapse = '-'))
clin_vars = c('years_to_birth','gender', 'race', 'ethnicity', 'patient.age_at_initial_pathologic_diagnosis')
# outcome_col <- 'pfi'
center_col <- 'COHORT'
corr_filter = TRUE
dist_fun_twoNN = 'canberra'
perc_points = 0.9
maxit = 11
factor_blocking = 3
max_blocking_runs = list(miRna = 100, RnaSeq = 100, RPPA = 100, methy = 100)
ntry = list(miRna = 21, RnaSeq = 21, RPPA = 21, methy = 21)
no_els_to_monitor = 10
ID_estimator_fun = 'estimate_ID_twoNN'
args_list = list('estimate_ID_danco' = list(list_k = c(4,6,8,12,16), maxD = 100, perc_points = perc_points, maxit = maxit,
ncores = min(ncores, detectCores()-1)) ,
'estimate_ID_twoNN' = list(dist_fun_twoNN = 'canberra', perc_points = perc_points, maxit = maxit,
ncores = min(ncores, detectCores()-1)))
#serve per limitare il numero di variabili tolte dalla p_value_select
limitVar = list(miRna = 350, RnaSeq = 3000, RPPA = 3000, methy = 3000)
#normalize_data_fun = list(miRna = min_max_norm, RnaSeq = min_max_norm, RPPA = NULL, methy = min_max_norm)
normalize_data_fun = list(miRna = my_quantile_normalization,
RnaSeq = my_quantile_normalization,
RPPA = my_quantile_normalization,
methy = my_quantile_normalization
)
dim_split_corr = list(miRna = 505, RnaSeq = 1000, RPPA = 1000, methy = 1000)
cutoff = list(miRna = 0.8, RnaSeq = 0.8, RPPA = 0.8, methy = 0.8)
maxit_corr = list(miRna = 11, RnaSeq = 11, RPPA = 11, methy = 11)
method_corr = list(miRna = 'spearman', RnaSeq = 'spearman', RPPA = 'spearman', methy = 'spearman')
maxDim = list(miRna = NA, RnaSeq = NA, RPPA = NA, methy = NA)
minD = list(miRna = NA, RnaSeq = NA, RPPA = NA, methy = NA)
dim_split_feat_clustering = list(miRna = 75, RnaSeq = 75, RPPA = 75, methy = 75)
maxit_unsup_feature_selection = list(miRna = 5, RnaSeq = 5, RPPA = 5, methy = 5)
in_all_df = NULL
# fracVar = 0.9
# embedded_data_types_fn = file.path('embedded_data_types.Rda')
# explain_data_types_fn = file.path('explain_data_types.Rda')
all_centers_res = data.frame()
all_centers_mega_res = data.frame()
uncompleted = matrix(0, nrow = length(cohort_list)*length(use_id_list), ncol = length(data_integration_list))
not_converged = matrix(0, nrow = length(cohort_list)*length(use_id_list), ncol = length(data_integration_list))
colnames(uncompleted) = data_integration_list
colnames(not_converged) = data_integration_list
rownames(uncompleted) = unlist(lapply(cohort_list,function(x) paste0(x, '_', names(use_id_list))))
rownames(not_converged) = unlist(lapply(cohort_list,function(x) paste0(x, '_', names(use_id_list))))
for (centers in cohort_list){
# if we have more centers, the data of the second, third, ... centers will be aligned to the data of the first center
str_centers = paste(centers, sep = '', collapse = '_')
cat('\n******+++++++++++++++++++++++++++++++++*****\n')
cat('starting to process ', str_centers, '\n')
if (length(centers)==1){ norm_quantiles = FALSE }else{ norm_quantiles = TRUE }
if (!dir.exists('./results')) dir.create('./results')
res_dir = file.path('./results', paste('res', str_centers, sep = '_'))
if (!dir.exists(res_dir)) dir.create(res_dir)
# riikka_dir = file.path(res_dir, 'riikka')
# if (!dir.exists(riikka_dir)) dir.create(riikka_dir)
### files xlsx for saving RF results
selected_file_mean_results_RF = file.path(res_dir, paste(str_add_save, 'mean_results_selected.xlsx', sep =''))
selected_file_mega_results_RF = file.path(res_dir, paste(str_add_save, 'mega_results_selected.xlsx', sep =''))
selected_mean_res_df = data.frame()
selected_mega_res_df = data.frame()
wkl_weights = data.frame()
###################
df_pt = data.frame()
new_names = NULL
for (center_now in centers){
if (file.exists(file.path(dati_rinominati, paste0(center_now, '_', outcome_col,'.rds')))){
df_c = data.frame(readRDS(file = file.path(dati_rinominati, paste0(center_now, '_', outcome_col,'.rds'))))
}else{
warning(paste0('NESSUN OUTCOME TROVATO PER ', center_now, ' - SKIPPING CENTER\n'))
next
}
names(df_c) = outcome_col
df_c[[center_col]] = center_now
df_clin = data.frame(readRDS(file = file.path(dati_rinominati, paste0(center_now, '_clinics.rds'))))
df_clin = df_clin[, names(df_clin) %in% clin_vars]
df_c = merge(df_c, df_clin, by = 'row.names')
row.names(df_c) = df_c$Row.names
# plyr:: rbind.fill perde i rownames!
new_names = c(new_names, row.names(df_c))
df_pt = plyr::rbind.fill(df_pt, df_c[, !(names(df_c) %in% 'Row.names')])
row.names(df_pt) = new_names
df_c = NULL
}
df_pt[[outcome_col]] = factor(df_pt[[outcome_col]], levels = c(neg_label, pos_label))
if (any(grepl('gender', names(df_pt)))) df_pt$gender = factor(df_pt$gender, levels = c('male', 'female'))
if (any(grepl('race', names(df_pt)))) df_pt$race = factor(df_pt$race, levels = c('white', 'black or african american', 'asian'))
if (any(grepl('ethnicity', names(df_pt)))) df_pt$ethnicity = factor(df_pt$ethnicity, levels = c('not hispanic or latino','hispanic or latino'))
for (cc in names(df_pt)){
cat(cc, ' - ')
if (length(unique(df_pt[[cc]]))==1) df_pt = df_pt[, !(names(df_pt) %in% cc)]
}
rm(cc)
df_outcome = data.frame(df_pt[[outcome_col]], row.names = row.names(df_pt))
# dataframe che poi andra' a comporre il dataframe dei dati combinati da passare alla RF
names(df_outcome) = outcome_col
df_pt_use = df_pt[, !(names(df_pt) %in% c(outcome_col, 'COHORT'))]
vars_pt = names(df_pt_use)[!names(df_pt_use) %in% outcome_col]
for (exp_id_no in 1:length(use_id_list)){
exp_id = names(use_id_list)[exp_id_no] # I'll need it later to record the uncompleted or unconverged experiments
use_id = use_id_list[[exp_id]][['use_id']]
frac_samples = use_id_list[[exp_id]][['frac_samples']]
minNumViews = use_id_list[[exp_id]][['minNumViews']]
for (unsup_feature_selection in unsup_feature_selection_list){
ID_string_unsupFS = ifelse(use_id, '_ID', paste0('_noID_', frac_samples)) # when I use the ID the unsup feature selection process is equal for all frac_samples
unsup_feature_selection = paste0(unsup_feature_selection, ID_string_unsupFS)
for (best_red in best_red_list){
ID_string_red = paste0('_', ifelse(use_id, '', 'no'), 'ID_', frac_samples)
if (!use_id){
# if I'm not using the ID the last dimensionality will surely be lower than N; therefore, no need to use par_rcur
if (gsub(unsup_feature_selection, pattern = ID_string_unsupFS, replacement = '') =='par_rcur') next
# when not using the ID to choose the dimensionality only 1-shot DR is chosen to avoid empirically choosing also the intermediate step
if ((gsub(unsup_feature_selection, pattern = ID_string_unsupFS, replacement = '') != 'NULL') & (best_red != 'NULL')) next
}
# if both are null I skip the experiment bacause it is too demanding
if (( gsub(unsup_feature_selection, pattern = ID_string_unsupFS, replacement = '') == 'NULL') & (best_red == 'NULL')) next
# the combo:
# FS = 'rcur' (or 'cur') + FE = 'NULL'
# is equal to the inverse combo:
# FS = 'NULL' + FE = 'rcur' (or cur); if both of them are scheduled I'm jumping the first of them
if ((grepl('cur', gsub(unsup_feature_selection, pattern = ID_string_unsupFS, replacement = '')) & (best_red == 'NULL')) &
any(best_red_list == gsub(unsup_feature_selection, pattern = ID_string_unsupFS, replacement = '')) &
any(gsub(unsup_feature_selection, pattern = ID_string_unsupFS, replacement = '') == 'NULL')) next
if (!(best_red == 'NULL')){
best_red = paste0(best_red, ID_string_red)
}
mat_data_all_names = data.frame('pt' = row.names(df_outcome), row.names = row.names(df_outcome))
# per problemi di psazio non posso tenerle in memoria ma nelle liste seguenti salvo solo la posizione dei df
explain_data_types <- list()
embedded_data_types <- list()
embedded_data_types[['pt']] = df_pt_use
explain_data_types[['pt']] = df_pt_use
if (verbose ==1){
cat('******\n')
cat('start with setting = - unsup feature selection = ', unsup_feature_selection,
' - best_red = ', best_red, ' - use_id = ', use_id, '\n')
}
# dati gi? standardizzati!!
norm_omics = readRDS(file = file.path(dati_rinominati, paste0(centers, '_norm_omics.rds')))
for (str_desc in data_types){
cat('\n\n*****************---------********************\n')
name_data = names(norm_omics)[grepl(paste0('_', str_desc), names(norm_omics), ignore.case = TRUE)]
#str_desc = names(norm_omics)[grepl(str_desc, names(norm_omics), ignore.case = TRUE)]
cat('processing ', str_desc, ' data type\n')
# unsup_feature_selection = unsup_feature_selection
best_red_now = best_red
save_ID_list = list()
save_ID_list[['ID_est']] = NULL
save_ID_list[['sd_ID_est']] = NULL
save_ID_list[['blocking_estimates']] = list()
cat('-------------------------------------------------------\n')
task = 'clean_data'
cat(task, '\n')
# ognuno segue il suo percorso -
my_path_records = list()
my_path_records[[task]] = res_dir
file_data_fn = file.path(my_path_records[[task]], paste(str_desc, '_', task,'.Rda' , sep ='' ))
#################### OPENING AND CLEANING DATA
if (file.exists(file_data_fn)){
load(file_data_fn)
#save(mat_data, file = file_data_fn)
}
else mat_data = norm_omics[[name_data]]
# se ? troppo grosso riduco con entropia (dovrei poi ricomputare la blocking id)
if (ncol(mat_data)>maxVars){
cat('data is hyper-diper high dimensional!\n')
file_data_fn_entropy_reduction = file.path(my_path_records[[task]],
paste(str_desc, '_entropy_filt_after_opening.Rda' , sep ='' ))
if (file.exists(file_data_fn_entropy_reduction)){
cat('loading lower dimensional data \n')
load(file_data_fn_entropy_reduction)
}else{
cat('entropy filtering to reduce dimensionality to ', maxVars, ' variables \n')
mat_data = imputa(mat_data)
entp_feat = apply(mat_data,2, entropy, method="Laplace")
entp_feat = sort(entp_feat, decreasing = TRUE)
mat_data = mat_data[ , match(names(entp_feat)[1:maxVars], colnames(mat_data))]
save(mat_data, file = file_data_fn_entropy_reduction)
}
}
if ((norm_quantiles) & (length(centers)>1)){
file_data_fn = file.path(my_path_records[[task]], paste(str_desc, '_', task, '_quantile_norm.Rda' , sep ='' ))
if (file.exists(file_data_fn)){
load(file = file_data_fn)
# load(file = ID_fn)
# blocking_mat = blocking_estimates[[task]]
cat('LOADED ', str_desc, ' mat_data after ', task, ' (quantile_normalized) = ', dim(mat_data), ' \n')
}else{
mat_data = imputa(mat_data)
#apply quantile norm to target center using the first center in the list of centers as target
target_center = centers[1]
idx_in_pt_df = match(row.names(mat_data), row.names(df_pt))
center_data = df_pt[idx_in_pt_df, center_col]
target_length = NULL
trasf_data = function(x) return(x)
norm_quant_data = TRUE
if (str_desc == "RPPA"){
norm_quant_data = FALSE
}
cat('normalize centers\n')
if (norm_quant_data){
for (cc in 1:ncol(mat_data)){
if (vis_norm_data){
df_s1 = data.frame(data = mat_data[, cc], aes = paste(center_data, '_before'))
h1 = ggplot(df_s1, aes(x = data, color=aes, fill = aes)) + geom_density(alpha = 0.2)+
ggtitle('before normalization')
print(h1)
}
target_data = trasf_data(normalize.quantiles.determine.target(
as.matrix(mat_data[(center_data == target_center),cc]),
target.length = target_length))
mat_data[!(center_data == target_center),cc] = trasf_data(
normalize.quantiles.use.target(as.matrix(mat_data[!(center_data == target_center),cc]), target_data))
if (vis_norm_data){
df_s1 = rbind(df_s1,data.frame(data = mat_data[, cc], aes = (center_data)))
h2 = ggplot(df_s1, aes(x = data, color=aes, fill = aes)) + geom_density(alpha = 0.1)+
ggtitle('after normalization')
print(h2)
}
}
rm(cc)
}
save(mat_data, file = file_data_fn)
cat('SAVED ', str_desc, ' mat_data after ', task, ' (quantile normalized) = ', dim(mat_data), ' \n')
}
}
#################### END OPENING AND CLEANING DATA
ID_fn = file.path(my_path_records[[task]],
paste0('ID_estimates', ifelse(use_id, '_ID', '_noID'), '_' , factor_blocking,'_', str_desc, '.Rda'))
if (use_id){
if (file.exists(ID_fn) & !(regenerate_blocking[[str_desc]])){
load(ID_fn)
############## POI DA TOGLIERE!!!!
# save_ID_list = save_ID_list[!(names(save_ID_list) %in% c('maxDim','minD'))]
# save(save_ID_list, file = ID_fn)
#
cat('********************\nRETRIEVING SAVED BLOCK ESTIMATES \n ')
if (length(save_ID_list[['blocking_estimates']])>0){
blocking_mat = save_ID_list[['blocking_estimates']][[task]]
# # check whether a plateau has been found or, instead, there is no redundancy in the features
# converged = FALSE
# dim_convergence = NA
# dim_lt_1std = NA
converged = save_ID_list$blocking_estimates$convergence
dim_convergence = save_ID_list$blocking_estimates$dim_convergence
dim_lt_1std = save_ID_list$blocking_estimates$dim_lt_1std
# if (nrow(blocking_mat)>no_els_to_monitor){ # should already be!
# # compute the std of the last 5 elements. When the std falls belo 0.1 stop computation and take the last block id as the last dataset id
#
# for (j in (no_els_to_monitor+1):nrow(blocking_mat)){
# last_els = blocking_mat[(j-no_els_to_monitor):j,'block_id']
# std_last_els = sd(last_els, na.rm = TRUE)
# lag = 3
# sum_deriv = sum(diff(last_els, lag = lag)/lag) # per la derivate non faccio il modulo perchè mi va bene se scende e poi risale
# if (verbose == 1) cat('std_last_els = ', std_last_els, ' - sum_deriv = ' , sum_deriv, '\n')
# if ((std_last_els < 0.25) | (sum_deriv < 0.05)){
# if (verbose == 1) cat('convergence reached at ', j, ' steps, that is when the block has dimension ', blocking_mat[j, 'no_vars_in_block'], '\n')
# converged = TRUE
# dim_convergence = blocking_mat[j, 'no_vars_in_block']
# break
# }
# }
# }
if (converged){
####################### CHOOSING INTERMEDIATE DIMENSION
if (verbose ==1) cat('BLOCKING ANALYSIS FOR ', str_desc, '-', paste(centers, collapse = '+'),'CONVERGED - setting dim_convergence as maxDim\n')
target = blocking_mat[nrow(blocking_mat), 'mean_id']
target_sd = blocking_mat[nrow(blocking_mat), 'total_sd']
lower_bound = target_sd/2
idx_block = which((blocking_mat[, 'mean_id']> (target-lower_bound )) &
(blocking_mat[, 'mean_id'] < (target+lower_bound )))[1]
dim_lt_1std = blocking_mat[idx_block, 'no_vars_in_block']
# intermediate dimension is the max between the intermediate dimension estimated
# with blocking-distribution analysis and
# e the number of cases = N
cat('dim_at_lt_1std\tdim_convergence\tmin(dim)\n')
cat(dim_lt_1std, '\t', dim_convergence, '\t', min(dim(mat_data))-1, '\n')
maxDim[[str_desc]] = max(dim_convergence, nrow(mat_data)-1) # anyhow, I do not want it to be less than the sample size
}else{
dim_convergence = NA
dim_lt_1std = NA
cat('BLOCKING ANALYSIS FOR ', str_desc, '-', paste(centers, collapse = '+'), ' did not converge; no FS applied\n' )
ID_orig = save_ID_list$ID_est[1]
sd_ID_orig = save_ID_list$sd_ID_est[1]
target = round(ID_orig, digits = 2)
target_sd = round(sd_ID_orig, digits = 2)
maxDim[[str_desc]] = ncol(mat_data) # do not apply intermediate reduction
}
# save_ID_list$blocking_estimates$convergence = converged
# save_ID_list$blocking_estimates$dim_convergence = dim_convergence
# save_ID_list$blocking_estimates$dim_lt_1std = dim_lt_1std
#
# save(save_ID_list, file = ID_fn)
####### plot blocking_ID
# centers_str = paste(centers, sep ='_')
# png(filename = paste0('C:/DATI/Anacleto/similarities_PNet/missSNF/data_TCGA/AARisultatiAnalizzati_per_articolo/images/',
# centers_str, '_', str_desc, '.png'))
#
# df_blocking = data.frame(blocking_mat[, match(c('no_vars_in_block', 'block_id', 'mean_id', 'sd_block_id' , 'total_sd'), colnames(blocking_mat))] )
# df_blocking = rbind(rep(0,ncol(df_blocking)), df_blocking)
# upper = df_blocking[nrow(df_blocking), 'mean_id'] - df_blocking[nrow(df_blocking), 'total_sd']
# lower = df_blocking[nrow(df_blocking), 'mean_id'] + df_blocking[nrow(df_blocking), 'total_sd']
#
# names(df_blocking) = c('Lj', paste0('block_id (std) = ', round(df_blocking[nrow(df_blocking), 'block_id'],0),' (', round(df_blocking[nrow(df_blocking), 'sd_block_id'],1), ')'),
# paste0('cumulative mean (pooled std) = ', round(df_blocking[nrow(df_blocking), 'mean_id'],0),' (', round(df_blocking[nrow(df_blocking), 'total_sd'],1), ')'),
# 'sd_block_id', 'total_sd')
#
#
# df_block_id = cbind('block-IDs (std.)', df_blocking[, c(1,2,4)])
# df_cumulative = cbind('cumulative mean (total std.)', df_blocking[, c(1,3,5)])
# names(df_block_id) = c('estimate', 'Lj', 'ID', 'std(ID)')
# names(df_cumulative) = c('estimate', 'Lj', 'ID', 'std(ID)')
#
# df_ggplot = rbind( df_block_id, df_cumulative )
# df_ggplot$upper = upper
# df_ggplot$lower = lower
#
#
# colors_to_use = NULL
# out2 = ggplot(df_ggplot, aes(x=Lj, y=ID, col=estimate)) + geom_line() +geom_point()
# out2 = out2 + geom_errorbar(aes(ymin= .data[['ID']]-.data[['std(ID)']],
# ymax= .data[['ID']]+.data[['std(ID)']], colour = estimate),
# width=5,position=position_dodge(0.05))
#
# # out2 <- out2 + geom_ribbon(aes(ymin= upper,
# # ymax= lower), alpha = 0.1, colour = NA)
# out2 = out2 + labs(title =
# paste0('ID estimation: block analysis for ', str_desc, '\n(', centers_str,') after preprocessing'),
# x = 'number of randomly selected features',
# y = 'block-ID with cumulative mean and total std. error', color = 'Legend')
#
#
# if (converged){
# vdf = data.frame(xintercept_name = c('dim at convergence', 'dim at 1 std away from ID estimate'),
# xintercept = c(dim_convergence,dim_lt_1std), stringsAsFactors = FALSE)
# out2 = out2+geom_vline(mapping = aes(xintercept = xintercept, colour = xintercept_name),
# data = vdf, show.legend = TRUE)
#
# # out2 = out2+geom_vline(xintercept = dim_convergence, show.legend = TRUE, colour="red")
# # out2 = out2+geom_vline(xintercept = dim_lt_1std, show.legend = TRUE, colour = "grey")
# # out2 = out2 + annotate("text", dim_convergence, 1, hjust = -.01,
# # label = 'dim at convergence') + annotate("text", dim_lt_1std, dim_convergence, 1, hjust = -.01,
# # label = 'dim at lt 1 std')
# }
#
# print(out2)
# dev.off()
# ############### end plotting
}else{
ID_orig = save_ID_list$ID_est[1]
sd_ID_orig = save_ID_list$sd_ID_est[1]
target = round(ID_orig, digits = 2)
target_sd = round(sd_ID_orig, digits = 2)
maxDim[[str_desc]] = ncol(mat_data) # do not apply intermediate reduction
}
}else{
cat('saving ID estimates in path ', ID_fn, '\n')
ids_twonn = estimate_ID_twoNN(mat_data,
dist_fun_twoNN,
maxit = maxit, perc_points = perc_points)
cat('estimated ids on ', str_desc, ' = ', ids_twonn[['id']], '\n')
save_ID_list[['df_estimates']] = rbind(data.frame(k = 2, IDs = ids_twonn[['id']],
low = ids_twonn[['id']]-ids_twonn[['sd_id']],
up = ids_twonn[['id']]+ids_twonn[['sd_id']],
ID_method = ID_estimator_fun))
save_ID_list[['ID_est']][task] = ids_twonn[['id']]
save_ID_list[['sd_ID_est']][task] = ids_twonn[['sd_id']]
# data_norm = standardNormalization(mat_data)
# dist_mat = (dist2(as.matrix(mat_data),as.matrix(mat_data)))^(1/2)
# h = hist(dist_mat[upper.tri(dist_mat, diag=FALSE)])
# ent = entropy(h$density)
ID_orig = ids_twonn[['id']]
sd_ID_orig = ids_twonn[['sd_id']]
if (round(ncol(mat_data) / nrow(mat_data),1) > factor_SSS){
# if we are in the case of small-sample-size
#BLOCKING ID
blocking_mat = blocking_ID(mat = mat_data,
ID_estimator_fun = ID_estimator_fun,
args_ID = args_list[[ID_estimator_fun]],
save_path = my_path_records[[task]],
task = task,
str_desc = str_desc,
ID_orig = ids_twonn[['id']],
factor = factor_blocking,
max_blocking_runs = max_blocking_runs[[str_desc]],
ntry = ntry[[str_desc]],
no_els_to_monitor = no_els_to_monitor,
verbose = 0,
ncores = min(ncores, detectCores()-1))
save_ID_list[['blocking_estimates']][[task]] = blocking_mat
save(save_ID_list, file = ID_fn)
################## END blocking ID
# check whether a plateau has been found or, instead, there is no redundancy in the features
converged = FALSE
dim_convergence = NA
dim_lt_1std = NA
if (nrow(blocking_mat)>no_els_to_monitor){ # should already be!
# compute the std of the last 5 elements. When the std falls belo 0.1 stop computation and take the last block id as the last dataset id
for (j in (no_els_to_monitor+1):nrow(blocking_mat)){
last_els = blocking_mat[(j-no_els_to_monitor):j,'block_id']
std_last_els = sd(last_els, na.rm = TRUE)
lag = 3
sum_deriv = sum(diff(last_els, lag = lag)/lag) # per la derivate non faccio il modulo perchè mi va bene se scende e poi risale
if (verbose == 1) cat('std_last_els = ', std_last_els, ' - sum_deriv = ' , sum_deriv, '\n')
if ((std_last_els < 0.25) | (sum_deriv < 0.05)){
if (verbose == 1) cat('convergence reached at ', j, ' steps, that is when the block has dimension ', blocking_mat[j, 'no_vars_in_block'], '\n')
converged = TRUE
dim_convergence = blocking_mat[j, 'no_vars_in_block']
break
}
}
}
if (converged){
####################### CHOOSING INTERMEDIATE DIMENSION
target = blocking_mat[nrow(blocking_mat), 'mean_id']
target_sd = blocking_mat[nrow(blocking_mat), 'total_sd']
lower_bound = target_sd/2
idx_block = which((blocking_mat[, 'mean_id']> (target-lower_bound )) &
(blocking_mat[, 'mean_id'] < (target+lower_bound )))[1]
dim_lt_1std = blocking_mat[idx_block, 'no_vars_in_block']
# intermediate dimension is the max between the intermediate dimension estimated
# with blocking-distribution analysis and
# e the number of cases = N
cat('dim_at_lt_1std\tdim_convergence\tmin(dim)\n')
cat(dim_lt_1std, '\t', dim_convergence, '\t', min(dim(mat_data))-1, '\n')
maxDim[[str_desc]] = max(dim_convergence, nrow(mat_data)-1) # anyhow, I do not want it to be less than the sample size
}else{
dim_convergence = NA
dim_lt_1std = NA
cat('BLOCKING ANALYSIS FOR ', str_desc, '-', paste(centers, collapse = '+'), ' did not converge; no FS applied' )
ID_orig = save_ID_list$ID_est[1]
sd_ID_orig = save_ID_list$sd_ID_est[1]
target = round(ID_orig, digits = 2)
target_sd = round(sd_ID_orig, digits = 2)
maxDim[[str_desc]] = ncol(mat_data) # do not apply intermediate reduction
}
save_ID_list$blocking_estimates$convergence = converged
save_ID_list$blocking_estimates$dim_convergence = dim_convergence
save_ID_list$blocking_estimates$dim_lt_1std = dim_lt_1std
save(save_ID_list, file = ID_fn)
####### plot blocking_ID
centers_str = paste(centers, sep ='_')
if (!dir.exists('./results/images/')) dir.create('./results/images/')
png(filename = paste0('./results/images/',
centers_str, '_', str_desc, '.png'))
df_blocking = data.frame(blocking_mat[, match(c('no_vars_in_block', 'block_id', 'mean_id', 'sd_block_id' , 'total_sd'), colnames(blocking_mat))] )
df_blocking = rbind(rep(0,ncol(df_blocking)), df_blocking)
upper = df_blocking[nrow(df_blocking), 'mean_id'] - df_blocking[nrow(df_blocking), 'total_sd']
lower = df_blocking[nrow(df_blocking), 'mean_id'] + df_blocking[nrow(df_blocking), 'total_sd']
names(df_blocking) = c('Lj', paste0('block_id (std) = ', round(df_blocking[nrow(df_blocking), 'block_id'],0),' (', round(df_blocking[nrow(df_blocking), 'sd_block_id'],1), ')'),
paste0('cumulative mean (pooled std) = ', round(df_blocking[nrow(df_blocking), 'mean_id'],0),' (', round(df_blocking[nrow(df_blocking), 'total_sd'],1), ')'),
'sd_block_id', 'total_sd')
df_block_id = cbind('block-IDs (std.)', df_blocking[, c(1,2,4)])
df_cumulative = cbind('cumulative mean (total std.)', df_blocking[, c(1,3,5)])
names(df_block_id) = c('estimate', 'Lj', 'ID', 'std(ID)')
names(df_cumulative) = c('estimate', 'Lj', 'ID', 'std(ID)')
df_ggplot = rbind( df_block_id, df_cumulative )
df_ggplot$upper = upper
df_ggplot$lower = lower
colors_to_use = NULL
out2 = ggplot(df_ggplot, aes(x=Lj, y=ID, col=estimate)) + geom_line() +geom_point()
out2 = out2 + geom_errorbar(aes(ymin= .data[['ID']]-.data[['std(ID)']],
ymax= .data[['ID']]+.data[['std(ID)']], colour = estimate),
width=5,position=position_dodge(0.05))
# out2 <- out2 + geom_ribbon(aes(ymin= upper,
# ymax= lower), alpha = 0.1, colour = NA)
out2 = out2 + labs(title =
paste0('ID estimation: block analysis for ', str_desc, '\n(', centers_str,') after preprocessing'),
x = 'number of randomly selected features',
y = 'block-ID with cumulative mean and total std. error', color = 'Legend')
if (converged){
vdf = data.frame(xintercept_name = c('dim at convergence', 'dim at 1 std away from ID estimate'),
xintercept = c(dim_convergence,dim_lt_1std), stringsAsFactors = FALSE)
out2 = out2+geom_vline(mapping = aes(xintercept = xintercept, colour = xintercept_name),
data = vdf, show.legend = TRUE)
# out2 = out2+geom_vline(xintercept = dim_convergence, show.legend = TRUE, colour="red")
# out2 = out2+geom_vline(xintercept = dim_lt_1std, show.legend = TRUE, colour = "grey")
# out2 = out2 + annotate("text", dim_convergence, 1, hjust = -.01,
# label = 'dim at convergence') + annotate("text", dim_lt_1std, dim_convergence, 1, hjust = -.01,
# label = 'dim at lt 1 std')
}
print(out2)
dev.off()
# ############### end plotting
}else{
# there is no suspect of small-sample-size
# cat(str_desc, ': number of columns ( D = ', ncol(mat_data),') is comparable to number of samples ( N = ',
# nrow(mat_data), ' --> D/N = )', round(ncol(mat_data)/nrow(mat_data), 1) , ' - no need for intermediate step\n')
#
warning(str_desc, ': number of columns ( D = ', ncol(mat_data),') is comparable to number of samples ( N = ',
nrow(mat_data), ' --> D/N = )', round(ncol(mat_data)/nrow(mat_data), 1) ,
' - no need for intermediate step\n', immediate. = TRUE)
target = round(ID_orig, digits = 2)
target_sd = round(sd_ID_orig, digits = 2)
maxDim[[str_desc]] = ncol(mat_data) # do not apply intermediate reduction
}
} # end if (file.exists(ID_fn))
# If there is no feature extraction I keep only the most representative features
if ((gsub(best_red_now, pattern = ID_string_red, replacement='') =='NULL') |
(grepl('cur', best_red_now))) minD[[str_desc]] = maxDim[[str_desc]]
else minD[[str_desc]] = ceiling(frac_samples*target +(target_sd*3))
}else{ # NOT USING THE ID
cat('NOT USING ID\n')
blocking_mat = NULL
target = ifelse(frac_samples==1, min(dim(mat_data))-1, round(min(dim(mat_data))/frac_samples))
target_sd = 0
maxDim[[str_desc]] = target # either apply FS or FE; anyhow go down to target
minD[[str_desc]] = target
} # end if (use_id)
if (only_compute_blocking) next
if (!(gsub(pattern = ID_string_unsupFS, replacement = '', unsup_feature_selection) == 'NULL')){
old_path = my_path_records[[task]]
str_add_fs = ''
maxDim_now = maxDim[[str_desc]]
if (use_id){
cat('FOR ', str_desc, 'I should select ', maxDim[[str_desc]], ' features by FS!!\n')
}else{
cat('no ID reduction maxDim == ncol(mat_data)?\n', maxDim_now == ncol(mat_data))
}
task = unsup_feature_selection
my_path_records[[task]] = file.path(old_path, task)
if (!file.exists(my_path_records[[task]])){ dir.create(my_path_records[[task]])}
file_data_fn = file.path(my_path_records[[task]], paste(str_desc, str_add_fs, '.Rda' , sep ='' ))
cat('-------------------------------------------------------\n')
cat(task, '\n')
if (file.exists(file_data_fn) & (!regenerate_DR_data[[str_desc]])){
load(file = file_data_fn)
cat('loading ', task, ' data with dim', dim(mat_data), '\n')
}else{
cat(task, 'until dimension is greater than ', maxDim_now, '\n')
if (maxDim_now == ncol(mat_data)){
warning(str_desc, ": intermediate dimenson - ", maxDim_now, " == ", ncol(mat_data),
" - mat dimension ! mat_data remains the same!\n", immediate. = TRUE)
# cat(str_desc, ": intermediate dimenson - ", maxDim_now, " == ", ncol(mat_data),
# " - mat dimension ! mat_data remains the same!\n")
}else if (gsub(unsup_feature_selection, pattern = ID_string_unsupFS, replacement = '') == 'feature_clustering'){
cat("feature clustering\n")
mat_data = cluster_features_par(mat_data, dim_split = 1000, maxD = maxDim_now,
maxit = ifelse(use_id, 3, 1),
method = method_corr[[str_desc]],
ncores = min(ncores, detectCores()-1))
}else if(gsub(unsup_feature_selection, pattern = ID_string_unsupFS, replacement = '') == 'svd_entropy'){
cat("parallel svd selection\n")
mat_data = hierarchical_svd_entropy_par(mat_data,
dim_split = 10000,
maxD = maxDim_now,
maxit = maxit_unsup_feature_selection[[str_desc]],
ncores = min(ncores, detectCores()-1))
}else if (gsub(unsup_feature_selection, pattern = ID_string_unsupFS, replacement = '') == 'par_rcur'){
cat("parallel RCUR selection\n")
mat_data = RCUR_selection_par(mat_data, patience = rcur_patience, maxD = maxDim_now,
maxit = maxit_unsup_feature_selection[[str_desc]])
}else if (gsub(unsup_feature_selection, pattern = ID_string_unsupFS, replacement = '') == 'rcur'){