-
Notifications
You must be signed in to change notification settings - Fork 5
/
fitIAK3D.R
2262 lines (1930 loc) · 107 KB
/
fitIAK3D.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
#####################################################################
### this version includes site-specific random effects (could be something other than site, eg soil order)
### if siteIDData is given, then columns of the created X are used as fixed and random (site-specific) random effects
### eg if X had just evi90 as column, then model would be
### response = a + evi90 + ASite + BSite * evi90
### where ASite is N(0,sdA) and BSite is N(0,sdB), ASite, BSite,... independent
### and lmer formula would be
### response ~ evi90 + (1 | siteID) + (0 + evi90 | siteID)
### not sure if will be useful, prob only for small datasets, and only with small number of covariates (think allows > 1)
#####################################################################
fitIAK3D <- function(xData , dIData , zData , covsData , modelX , modelx = 'matern' , nud = 0.5 ,
sdfdType_cd1 = 0 , sdfdType_cxd0 = 0 , sdfdType_cxd1 = 0 , cmeOpt = 0 , sdfdKnots = NULL , minRange = NA , maxRange = NA , prodSum = TRUE ,
lnTfmdData = FALSE , useReml = TRUE , optionsModelX = list('reduceXAfterInitFit' = FALSE) , compLikMats = list('compLikOptn' = 0) , namePlot = NA ,
lmmFit = list() , rqrBTfmdPreds = TRUE , parsInit = NULL , attachBigMats = TRUE , testMCGrad = NULL , nCores = min(detectCores() , 8) , siteIDData = NULL){
if(!is.null(siteIDData)){
if(compLikMats$compLikOptn != 0){ stop('Error - only exact likelihood with site-specific random effects!') }else{}
if(identical(modelX , 'cubist') | (is.list(modelX) && identical(modelX$type , 'cubist'))){ stop('Error - do not use cubist model with site-specific random effects!') }else{}
if(lnTfmdData){ stop('Error - do not use lnTfmdData = TRUE with site-specific random effects!') }else{}
### make sure it is a character...
siteIDData <- as.character(siteIDData)
}else{}
#############################################
### if modelX is passed in as character, initialise it in this function...
### for gam2, SCALED covs get added to covsData
#############################################
tmp <- initialiseModelX(xData = xData , dIData = dIData , zData = zData , covsData = covsData ,
modelX = modelX , optionsModelX = optionsModelX)
modelX <- tmp$modelX
covsData <- tmp$covsData
rm(tmp)
#############################################
### check options for reducing gam model...
#############################################
if(optionsModelX$reduceXAfterInitFit){
if(identical(modelX$type , 'gam2')){
if(is.null(optionsModelX$alpha)){ stop('Error - enter alpha for reducing cubist model!') }else{}
if(is.null(optionsModelX$optStat4Drop)){
stop('Error - enter optionsModelX$optStat4Drop for reduceXAfterInitFit with gam2 models.')
}else{}
if(is.null(optionsModelX$maxnSpatVars)){
stop('Error - enter optionsModelX$maxnSpatVars for reduceXAfterInitFit with gam2 models.')
}else{}
if(is.null(optionsModelX$maxnKnotsd)){
stop('Error - enter optionsModelX$maxnKnotsd for reduceXAfterInitFit with gam2 models.')
}else{}
if(is.null(optionsModelX$maxnKnotss)){
stop('Error - enter optionsModelX$maxnKnotss for reduceXAfterInitFit with gam2 models.')
}else{}
}else if(identical(modelX$type , 'cubist')){
if(is.null(optionsModelX$alpha)){ stop('Error - enter alpha for reducing cubist model!') }else{}
}else{
stop('Error - the option to reduceXAfterInitFit is only coded for gam2 models.')
}
}else{}
#############################################
### check MC info...
#############################################
if(is.null(testMCGrad)){
if(compLikMats$compLikOptn == 0){
if((length(zData) < 1000) & Sys.info()[1] == 'Windows'){
### small dataset, using windows, prob not worth using parallel version of gradient
testMCGrad <- FALSE
}else{
testMCGrad <- TRUE
}
}else{
### for CL methods, nll fn is parallelised, so don't use parallel grad fn.
testMCGrad <- FALSE
}
}else{}
# if(!identical(class(xData) , 'matrix')){ stop('Stopping - for fitIAKD3D function, enter xData as a matrix') }else{}
# if(!identical(class(dIData) , 'matrix')){ stop('Stopping - for fitIAKD3D function, enter dIData as a matrix with 2 columns') }else{}
########################################################
### if xData or dIData were dataframes, convert to matrices here.
### and make sure all are numeric...
########################################################
if(!is.matrix(xData)){
xData <- as.matrix(xData)
}else{}
if(!is.matrix(dIData)){
dIData <- as.matrix(dIData)
}else{}
xCopy <- matrix(NA , nrow(xData) , ncol(xData))
for (i in 1:ncol(xData)){ xCopy[,i] <- as.numeric(xData[,i]) }
xData <- xCopy
remove(xCopy)
dICopy <- matrix(NA , nrow(dIData) , ncol(dIData))
for (i in 1:ncol(dIData)){ dICopy[,i] <- as.numeric(dIData[,i]) }
dIData <- dICopy
remove(dICopy)
#########################################################
### error check for duplicated xData,dIData data, if no measurement error being included...
#########################################################
if(cmeOpt == 0){
iDuplicated <- which(duplicated(cbind(xData , dIData)))
if(length(iDuplicated) > 0){ stop(paste0('Error - the data at positions ' , iDuplicated , ' are duplicated!')) }else{}
}else{}
#########################################################
### error check for any NAs; removing them if found...
#########################################################
iNA <- which(is.na(zData) | rowSums(is.na(covsData)) > 0)
if(length(iNA) > 0){
print('Attention! Some NAs found in data ; removing them.')
xData <- xData[-iNA,drop = FALSE]
dIData <- dIData[-iNA,drop = FALSE]
covsData <- covsData[-iNA,drop = FALSE]
zData <- zData[-iNA]
if(!is.null(siteIDData)){
siteIDData <- siteIDData[-iNA]
}else{}
}else{}
#########################################################
### round depths to nearest cm (assuming depths were inputted in m)
#########################################################
dIData <- round(dIData , digits = 2)
if(min(dIData[,2] - dIData[,1]) <= 0){ stop('Error - some data entered with lower depth less than or equal to upper depth. Check profiles!') }else{}
################################################
### if compLikMats is not null, fit assuming additive on input scale (ie put lnTfmdData = FALSE)
### ie approx involves approx by arithmetic averaging on log-transformed data + comp lik.
### but allow rqrBTfmdPreds to remain as inputted
################################################
if(compLikMats$compLikOptn > 0){
if (lnTfmdData){
lnTfmdData <- FALSE
print('Attention - averaging assumed to occur on the transformed scale.')
}else{}
}else{}
#######################################
### for Normal data, fit beta and cxd[d=0] automatically
#######################################
### for lnNormal data, fit beta by newton-raphson
#######################################
if(lnTfmdData){
### in this case, only allowed reml if depth not in fixed effects,
### so assume has to be ml...
if(useReml){
print('Attention! When underlying point-support variable assumed lognormal, cannot use REML because beta parameters contribute non-linearly to likelihood function!')
print('Therefore switching option to use ML.')
useReml <- F
}else{}
}else{}
###################################################
### if using Eidsvik, make sure ML, not REML...
###################################################
if(compLikMats$compLikOptn == 2 || compLikMats$compLikOptn == 3){
if(useReml){
print('Attention - do not useReml with Eidsvik CL - appropriate for ML only! Changing option here.')
useReml <- FALSE
}else{}
}else{}
################################################
### if not prodSum, will be product covariance function
### in which case, put sdfdType_cd1 <- -9
################################################
if(!prodSum){
if (!is.element(modelx , c('matern' , 'wendland' , 'spherical'))){ stop('Error - for product covariance, specify matern or wendland for modelx!') }else{}
sdfdType_cd1 <- -9
}else{}
################################################
### if using gam2, all knot info will be in modelX$listfefdKnots...
### other info will be in modelX$incInts and modelX$colnamesXcns
################################################
# if(identical(modelX$type , 'gam2')){
if((!is.numeric(modelX)) && identical(modelX$type , 'gam2')){
if(length(optionsModelX$allKnotsd) > 0){ stop('Error - if using gam2 for trend, enter knot info in modelX$listfefdKnots (you can use function makelistfefdKnots to do this); do not use allKnotsd.\n
Note, modelX should also set modelX$incInts and modelX$comnamesXcns')}else{}
}else{}
################################################
### for sdfd as spline fn (if sdfdType > 0) set knots if not given...
################################################
if(is.null(sdfdKnots)){
sdfdKnots <- setKnots4sdfd(dIData , sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1)
}else{
if(sdfdType_cd1 > 0 | sdfdType_cxd0 > 0 | sdfdType_cxd1 > 0){
if(is.null(sdfdKnots$bdryKnots) | ((sdfdType_cd1 > 0) & (length(sdfdKnots$intKnots_cd1) != sdfdType_cd1)) |
((sdfdType_cxd0 > 0) & (length(sdfdKnots$intKnots_cxd0) != sdfdType_cxd0)) |
((sdfdType_cxd1 > 0) & (length(sdfdKnots$intKnots_cxd1) != sdfdType_cxd1))){
stop('Error - sdfdKnots incorrectly defined for sdfdType - check this!')
}else{}
}else{}
}
######################################################
### set up fixed-effect design matrices...
######################################################
if((!is.numeric(modelX)) && identical(modelX$type , 'gam2')){
tmp <- makeXvX_gam2(covData = covsData , dIData = dIData , listfefdKnots = modelX$listfefdKnots , incInts = modelX$incInts , intMthd = modelX$intMthd , colnamesXcns = modelX$colnamesX , nDiscPts = 1000 , lnTfmdData = lnTfmdData)
XData <- tmp$X
vXU <- tmp$vXU
iU <- tmp$iU
namesX <- tmp$namesX
XLims <- NA
}else{
tmp <- makeXvX(covData = covsData , dIData = dIData , modelX = modelX , allKnotsd = optionsModelX$allKnotsd , opt_dSpline = optionsModelX$opt_dSpline , nDiscPts = 1000 , lnTfmdData = lnTfmdData)
XData <- tmp$X
vXU <- tmp$vXU
iU <- tmp$iU
namesX <- tmp$namesX
XLims <- tmp$XLims
}
remove(tmp) ; gc(verbose = FALSE)
############################################################
### check if XData has a lot of zeros, and if so, make it sparse...
############################################################
# propnX0 <- sum(XData == 0) / (ncol(XData) * nrow(XData))
# if(propnX0 > 0.8){
# XData <- Matrix(XData)
# }else{}
# if(length(vXU) > 1){
# propnvXU0 <- sum(vXU == 0) / (ncol(vXU) * nrow(vXU))
# if(propnvXU0 > 0.8){
# vXU <- Matrix(vXU)
# }else{}
# }else{}
#
######################################################
### select columns of X for site-specific random effects...
### only include the spatial covariates. dIMidPts will be site specific because of iak cov fn.
######################################################
if(!is.null(siteIDData)){
icols4ssre <- setdiff(seq(ncol(XData)) , which(substr(colnames(XData) , 1 , 8) == 'dIMidPts'))
colnames4ssre <- colnames(XData)[icols4ssre]
rm(icols4ssre)
}else{
colnames4ssre <- NULL
}
######################################################
### setup design matrices for iak3d covariance matrix...
######################################################
if (compLikMats$compLikOptn == 0){
# setupMats <- setupIAK3D(xData , dIData , nDscPts = 0)
setupMats <- setupIAK3D(xData , dIData , nDscPts = 0 ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 , sdfdKnots = sdfdKnots ,
siteIDData = siteIDData , XData = XData , colnames4ssre = colnames4ssre)
}else{
# setupMats <- setupIAK3D_CL(xData , dIData , nDscPts = 0 , compLikMats = compLikMats)
setupMats <- setupIAK3D_CL(xData , dIData , nDscPts = 0 , compLikMats = compLikMats ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 , sdfdKnots = sdfdKnots)
}
######################################################
### set sensible initial parameters and evaluate nll...
######################################################
verboseOptim <<- T
if(is.null(lmmFit$parBnds)){
parBnds <- setParBndsIAK3D(xData = xData , dIData = dIData , setupMats = setupMats , compLikMats = compLikMats , modelx = modelx , minRange = minRange , maxRange = maxRange)
}else{
parBnds <- lmmFit$parBnds
}
if(is.null(lmmFit$pars)){
if(is.null(parsInit)){
parsInit <- setInitsIAK3D(xData = xData , dIData = dIData , zData = zData , XData = XData , vXU = vXU , iU = iU , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , lnTfmdData = lnTfmdData , lmmFit = lmmFit , compLikMats = compLikMats)$pars
### for exact lik, run through setInits function again, with the C from above used to give better initial beta (then better pars)...
updateInits <- FALSE
if(updateInits & compLikMats$compLikOptn == 0){
parsBTfmd <- readParsIAK3D(pars = parsInit , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , parBnds = parBnds , lnTfmdData = lnTfmdData)
initC <- setCIAK3D(parsBTfmd = parsBTfmd , modelx = modelx ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , setupMats = setupMats)$C
parsInit <- setInitsIAK3D(xData = xData , dIData = dIData , zData = zData , XData = XData , vXU = vXU , iU = iU , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , lnTfmdData = lnTfmdData , lmmFit = lmmFit , compLikMats = compLikMats , initC = initC)$pars
rm(initC)
}else{}
}else{}
verboseOptim <<- F
}else{
parsInit <- lmmFit$pars
### should check here that we have right number of parameters...
}
if(is.null(lmmFit$fitRange)){
fitRange <- setFitRangeIAK3D(pars = parsInit , modelx = modelx ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , lnTfmdData = lnTfmdData)
}else{
fitRange <- lmmFit$fitRange
}
##################################################
### the lmer setup bit...
### add to parsInit, fitRange
##################################################
if(!is.null(siteIDData)){
print('ATTENTION!!!!! SITE_SPECIFIC RANDOM EFFECT!')
parsInit <- c(parsInit , rep(0 , length(setupMats$listStructs4ssre)))
fitRange <- rbind(fitRange , cbind(rep(-20 , length(setupMats$listStructs4ssre)) , rep(Inf , length(setupMats$listStructs4ssre))))
}else{}
######################################################
### optimize parameters...
######################################################
if(is.null(lmmFit$pars)){
### print out init pars and nll...
print('At initial parameters...')
start_time <- Sys.time()
verboseOptim <<- T
if (compLikMats$compLikOptn == 0){
nllInit <- nllIAK3D(pars = parsInit , zData = zData , XData = XData , vXU = vXU , iU = iU , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , lnTfmdData = lnTfmdData , rtnAll = F)
}else{
nllInit <- nllIAK3D_CL(pars = parsInit , zData = zData , XData = XData , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , compLikMats = compLikMats , rtnAll = F , nCores = nCores)
}
verboseOptim <<- F
end_time <- Sys.time()
nllTime <- end_time - start_time
### and fit...
if (compLikMats$compLikOptn == 0){
if(testMCGrad){
print('Testing whether parallel or series version of gradient function is quicker...')
start_time <- Sys.time()
gr <- gradnllIAK3D(pars = parsInit , zData = zData , XData = XData , vXU = vXU , iU = iU , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , lnTfmdData = lnTfmdData)
end_time <- Sys.time()
grTime <- end_time - start_time
start_time <- Sys.time()
gr.mc <- gradnllIAK3D.mc(pars = parsInit , zData = zData , XData = XData , vXU = vXU , iU = iU , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , lnTfmdData = lnTfmdData , nCores = nCores)
end_time <- Sys.time()
grTime.mc <- end_time - start_time
if(grTime.mc < grTime){
print('Multi-core version of grad was quicker, so will use this to compute gradients of nll using all cores for parameter fitting (see function gradnllIAK3D.mc to change this).')
gr4Optim <- gradnllIAK3D.mc
}else{
print('Single-core version of grad was quicker, so will use this to compute gradients of nll for parameter fitting.')
gr4Optim <- gradnllIAK3D
}
}else{
gr4Optim <- gradnllIAK3D
}
print('Now fitting...')
parsFit <- optimIt(par = parsInit , fn = nllIAK3D , gr = gr4Optim , fitRange = fitRange ,
zData = zData , XData = XData , vXU = vXU , iU = iU , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , lnTfmdData = lnTfmdData , nCores = nCores)
### all args taken from this env now.
# parsFit <- optimIt(par = parsInit , fn = nllIAK3D , gr = gr4Optim , fitRange = fitRange)
}else{
if(testMCGrad){
print('Testing whether parallel or series version of gradient function is quicker...')
start_time <- Sys.time()
gr_CL <- gradnllIAK3D_CL(pars = parsInit , zData = zData , XData = XData , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , compLikMats = compLikMats , nCores = nCores)
end_time <- Sys.time()
grTime <- end_time - start_time
start_time <- Sys.time()
gr_CL <- gradnllIAK3D_CL.mc(pars = parsInit , zData = zData , XData = XData , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , compLikMats = compLikMats , nCores = nCores)
end_time <- Sys.time()
grTime.mc <- end_time - start_time
if(grTime.mc < grTime){
print('Multi-core version of grad was quicker, so will use this to compute gradients of nll using all cores for parameter fitting (see function gradnllIAK3D_CL.mc to change this).')
gr4Optim <- gradnllIAK3D_CL.mc
}else{
print('Single-core version of grad was quicker, so will use this to compute gradients of nll for parameter fitting.')
gr4Optim <- gradnllIAK3D_CL
}
}else{
gr4Optim <- gradnllIAK3D_CL
}
print('Now fitting...')
parsFit <- optimIt(par = parsInit , fn = nllIAK3D_CL , gr = gr4Optim , fitRange = fitRange ,
zData = zData , XData = XData , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , compLikMats = compLikMats , nCores = nCores)
### args now taken from this env.
# parsFit <- optimIt(par = parsInit , fn = nllIAK3D_CL , gr = gr4Optim , fitRange = fitRange)
}
### note that parameters have now been fitted...
print('Parameters fitted...')
}else{
### or to just take the values in inits, which were the transformed values taken from the inputted lmmFit...
parsFit <- list('par' = parsInit)
}
verboseOptim <<- T
if (compLikMats$compLikOptn == 0){
lmmFit <- nllIAK3D(pars = parsFit$par , zData = zData , XData = XData , vXU = vXU , iU = iU , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , lnTfmdData = lnTfmdData , rtnAll = T , attachBigMats = attachBigMats)
}else{
lmmFit <- nllIAK3D_CL(pars = parsFit$par , zData = zData , XData = XData , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , compLikMats = compLikMats , rtnAll = T , attachBigMats = attachBigMats , nCores = nCores)
}
verboseOptim <<- F
### lost colnames somewhere?!
if((!is.numeric(modelX)) && identical(modelX$type , 'cubist')){
colnames(XData) <- modelX$namesX
}else{}
#######################################################################
### if optionsModelX$reduceXAfterInitFit with a gam2 model
### then remove columns of X (using stepBackXcns algorihtm) and refit IAK3D model.
#######################################################################
if(optionsModelX$reduceXAfterInitFit && ((identical(modelX$type , 'gam2') | identical(modelX$type , 'cubist')))){
if(identical(modelX$type , 'gam2')){
print('Reducing the initial gam2 model by removing redundant columns...')
if(compLikMats$compLikOptn == 0){
tmp <- stepBackXcns(Xcns = XData , zData = zData ,
iAX = lmmFit$iCX / lmmFit$cxdhat , iAz = lmmFit$iC %*% zData / lmmFit$cxdhat ,
alpha = optionsModelX$alpha , optStat4Drop = optionsModelX$optStat4Drop ,
maxnSpatVars = optionsModelX$maxnSpatVars , maxnKnotsd = optionsModelX$maxnKnotsd , maxnKnotss = optionsModelX$maxnKnotss)
}else{
print('ATTENTION - STEP BACK ALGORITHM WITH COMPOSITE LIKELIHOOD NEEDS CHECKING!')
tmp <- stepBackXcns(Xcns = XData , zData = zData , XziAXz = lmmFit$XziAXz_APPROX ,
alpha = optionsModelX$alpha , optStat4Drop = optionsModelX$optStat4Drop ,
maxnSpatVars = optionsModelX$maxnSpatVars , maxnKnotsd = optionsModelX$maxnKnotsd , maxnKnotss = optionsModelX$maxnKnotss)
}
modelX$colnamesX <- colnames(tmp$Xcns)
modelX$type <- 'gam2' # just making sure!
tmp <- makeXvX_gam2(covData = covsData , dIData = dIData , listfefdKnots = modelX$listfefdKnots , incInts = modelX$incInts , intMthd = modelX$intMthd , colnamesXcns = modelX$colnamesX , nDiscPts = 1000 , lnTfmdData = lnTfmdData)
XData <- tmp$X
vXU <- tmp$vXU
iU <- tmp$iU
namesX <- tmp$namesX
XLims <- NA
}else if(identical(modelX$type , 'cubist')){
print('Reducing the initial cubist model by removing redundant columns...')
if(compLikMats$compLikOptn == 0){
tmp <- refineXIAK3D(X = XData , z = zData , profID = NULL , alpha = optionsModelX$alpha , iAX = lmmFit$iCX / lmmFit$cxdhat , iAz = lmmFit$iC %*% zData / lmmFit$cxdhat)
namesRemove <- tmp$namesRemove
ipRemove <- tmp$ipRemove
tmp <- removeColinPreds(X = XData , cubistModel = modelX , dataFit = covsData , ipRmvd = ipRemove , reasonRmv = 1)
XData <- tmp$X
modelX <- tmp$cubistModel
modelX$namesX <- colnames(XData)
modelX$type <- 'cubist'
tmp <- makeXvX(covData = covsData , dIData = dIData , modelX = modelX , allKnotsd = optionsModelX$allKnotsd , opt_dSpline = optionsModelX$opt_dSpline , nDiscPts = 1000 , lnTfmdData = lnTfmdData)
XData <- tmp$X
vXU <- tmp$vXU
iU <- tmp$iU
namesX <- tmp$namesX
XLims <- tmp$XLims
}else{
stop('Not sure the code works to reduce cubist X with approximate likelihood.')
}
}else{}
print(paste0('Refitting with the reduced ' , modelX$type , ' model...'))
if (compLikMats$compLikOptn == 0){
parsFit <- optimIt(par = parsInit , fn = nllIAK3D , gr = gr4Optim , fitRange = fitRange ,
zData = zData , XData = XData , vXU = vXU , iU = iU , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , lnTfmdData = lnTfmdData , nCores = nCores)
verboseOptim <<- T
print('Parameters refitted...')
lmmFit <- nllIAK3D(pars = parsFit$par , zData = zData , XData = XData , vXU = vXU , iU = iU , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml ,
lnTfmdData = lnTfmdData , rtnAll = T , attachBigMats = attachBigMats)
}else{
parsFit <- optimIt(par = parsInit , fn = nllIAK3D_CL , gr = gr4Optim , fitRange = fitRange ,
zData = zData , XData = XData , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml , compLikMats = compLikMats , nCores = nCores)
verboseOptim <<- T
print('Parameters refitted...')
lmmFit <- nllIAK3D_CL(pars = parsFit$par , zData = zData , XData = XData , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , setupMats = setupMats , parBnds = parBnds , useReml = useReml ,
compLikMats = compLikMats , rtnAll = T , attachBigMats = attachBigMats , nCores = nCores)
}
verboseOptim <<- F
}else{}
#######################################################################
if((!is.numeric(modelX)) && identical(modelX$type , 'cubist')){
colnames(XData) <- modelX$namesX
}else if((!is.numeric(modelX)) && identical(modelX$type , 'gam2')){
colnames(XData) <- modelX$colnamesX
}else{
# print(colnames(XData))
# print(modelX)
print('Should make sure that XData has col names here!')
}
############################################
### also attach everything that was used to call the function to lmmFit...
############################################
lmmFit <- attachSettings2lmmFit(lmmFit = lmmFit , xData = xData , dIData = dIData , zData = zData , covsData = covsData ,
modelX = modelX , namesX = namesX , modelx = modelx , nud = nud , XLims = XLims ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 , cmeOpt = cmeOpt ,
prodSum = prodSum , sdfdKnots = sdfdKnots , minRange = minRange , maxRange = maxRange , setupMats = setupMats , lnTfmdData = lnTfmdData ,
useReml = useReml , optionsModelX = optionsModelX , parBnds = parBnds , fitRange = fitRange , compLikMats = compLikMats , siteIDData = siteIDData , colnames4ssre = colnames4ssre)
if (!is.na(namePlot)){
##############################################
### a quick prediction routine...
### predict through the profile for all data locations and for one distant location (with average of covsData)
##############################################
######################################################
### set up fixed-effect design matrices...
######################################################
# dIPred <- cbind(seq(0 , 1.99 , 0.01) , seq(0.01 , 2 , 0.01))
# dIPred <- cbind(seq(0 , 1.98 , 0.02) , seq(0.02 , 2 , 0.02))
# dIPred <- cbind(seq(0 , 1.9 , 0.1) , seq(0.1 , 2 , 0.1))
# dIPred <- cbind(seq(0 , 14.98 , 0.02) , seq(0.02 , 15 , 0.02))
if(parBnds$maxd < 10){
dIPred <- cbind(seq(0 , parBnds$maxd-0.02 , 0.02) , seq(0.02 , parBnds$maxd , 0.02))
}else{
dTMP <- seq(0 , parBnds$maxd , length = 101)
dIPred <- cbind(dTmp[1:(length(dTMP)-1)] , dTMP[-1])
}
### below 1m, only do every 5th one of these depth intervals...
iKeep1 <- which(dIPred[,1] < 1)
iKeep2 <- which(dIPred[,1] >= 1)
if(length(iKeep2) > 0){ iKeep2 <- iKeep2[seq(1,length(iKeep2),5)] }else{}
dIPred <- dIPred[c(iKeep1,iKeep2),,drop=FALSE]
rm(iKeep1,iKeep2)
dIPred <- round(dIPred , digits = 2)
ndIPred <- nrow(dIPred)
iTmp <- which(!duplicated(xData))
maxnProfPlot <- 100
if(length(iTmp) > maxnProfPlot){
### take a subsample of 200 profiles to plot...
### so that repeatable, don't make it random...
iTmp2 <- round(seq(1 , length(iTmp) , length(iTmp) / maxnProfPlot))
iTmp <- iTmp[iTmp2]
profNamesPlot <- paste0('Profile ' , iTmp2)
}else{
profNamesPlot <- paste0('Profile ' , seq(length(iTmp)))
}
covsPred <- covsData[iTmp,,drop=FALSE]
xPred <- xData[iTmp,,drop=FALSE]
if(is.null(siteIDData)){
siteID4Predict <- NULL
}else{
siteID4Predict <- c(siteIDData[iTmp] , 'a_far_away_site')
}
### add a distant location to the prediction locations...
xPredDistant <- matrix(c(9E99 , 9E99) , 1 , 2)
# covsPredDistant <- colMeansAndModes(covsPred)
# covsPredDistant <- colMeansAndModes(covsData)
### initialise with first row...
covsPredDistant <- covsData[1,,drop=FALSE]
### don't include the depths in finding medoid...
idCovs <- which(names(covsData) == 'dIMidPts')
if(length(idCovs) > 0){ covsPredDistant[1,idCovs] <- NA }else{}
if(length(idCovs) < ncol(covsData)){ covsPredDistant[1,-idCovs] <- medoid(covsData[,-idCovs,drop=FALSE]) }else{}
nxPred <- dim(xPred)[[1]]
# return(list('xTmp' = rbind(xPred , xPredDistant) , 'dIMap' = dIPred , 'covsMap' = rbind(covsPred , covsPredDistant) , 'lmmFit' = lmmFit))
### call the predict function...
tmp <- predictIAK3D(xMap = rbind(xPred , xPredDistant) , dIMap = dIPred ,
covsMap = rbind(covsPred , covsPredDistant) , siteIDMap = siteID4Predict ,
lmmFit = lmmFit , rqrBTfmdPreds = rqrBTfmdPreds , constrainX4Pred = FALSE)
zPred <- tmp$zMap[,1:nxPred]
vPred <- tmp$vMap[,1:nxPred]
pi90UPred <- tmp$pi90UMap[,1:nxPred]
pi90LPred <- tmp$pi90LMap[,1:nxPred]
zPredDistant <- tmp$zMap[,nxPred+1]
vPredDistant <- tmp$vMap[,nxPred+1]
### to back transform data for mapping...
if(lnTfmdData & rqrBTfmdPreds){
###(were inputted and used all the way through on the log scale,
### but for plot put back to real scale)
zData <- exp(zData)
}else{}
#################################################
### make a pdf with the distant profile prediction (page 1) and all data profiles (6 per page thereafter)...
### note these are not validation predictions, they are predicted at the data profiles given the data for the same profiles
### also note, they are a bi-product of the method (ie you can use the method to predict at profiles where we have data),
### not to be confused with the spline-then-krige type approach where similar plots may be produced in the first step of analysis
#################################################
print('Making the profile plots...')
tmp <- plotProfilesIAK3D(namePlot = namePlot , xData = xData , dIData = dIData , zData = zData ,
xPred = xPred , dIPred = dIPred , zPred = zPred , pi90LPred = pi90LPred , pi90UPred = pi90UPred , zPredDistant = zPredDistant , profNames = profNamesPlot)
}else{
zPredDistant <- vPredDistant <- zPred <- vPred <- pi90LPred <- pi90UPred <- xPred <- dIPred <- NA
}
#####################################################
### and return...
#####################################################
return(list('parsFit' = parsFit , 'lmmFit' = lmmFit ,
'zPredDistant' = zPredDistant , 'vPredDistant' = vPredDistant ,
'zPred' = zPred , 'vPred' = vPred , 'pi90LPred' = pi90LPred , 'pi90UPred' = pi90UPred ,
'xPred' = xPred , 'dIPred' = dIPred))
}
###################################################################
### the nll function that is to be minimized...
### nCores arg is ignored in this fn!
###################################################################
nllIAK3D <- function(pars , zData , XData , vXU , iU , modelx , nud ,
sdfdType_cd1 , sdfdType_cxd0 , sdfdType_cxd1 , cmeOpt , prodSum , setupMats = NULL , parBnds ,
useReml , lnTfmdData , rtnAll = F , forCompLik = FALSE , attachBigMats = FALSE , nCores = 8){
### the final parameters for the ssre (sire-specific random effect) bit
### put in for modelling within-field variation using data from field to update regression parameters.
# if(!is.null(setupMats$listStructs4ssre)){
# ipars_iak3d <- seq(1 , length(pars) - length(setupMats$listStructs4ssre))
# ipars_ssre <- seq(length(pars) - length(setupMats$listStructs4ssre) + 1 , length(pars))
# }else{
# ipars_iak3d <- seq(1,length(pars))
# ipars_ssre <- c()
# }
### default values for return...
badnll <- 9E99
cxdhat <- NA
WiAW <- NA
lndetA <- NA
p <- ncol(XData)
n <- length(zData)
if(exists("printnllTime") && printnllTime){
ptm <- proc.time()
}else{}
if(exists("parsTrace4Optim") && (!forCompLik)){
if(is.null(ncol(parsTrace4Optim)) || (ncol(parsTrace4Optim) == length(pars))){
if(length(parsTrace4Optim) > 0){
parsTmp <- rbind(parsTrace4Optim , pars)
nllTmp <- c(nllTrace4Optim , NA) # for this trace, use NA for failed evaluation.
}else{
parsTmp <- matrix(pars , nrow = 1)
nllTmp <- NA # for this trace, use NA for failed evaluation.
}
assign("parsTrace4Optim" , parsTmp , envir = .GlobalEnv)
assign("nllTrace4Optim" , nllTmp , envir = .GlobalEnv)
}else{}
}else{}
parsBTfmd <- readParsIAK3D(pars = pars , modelx = modelx , nud = nud ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , prodSum = prodSum , parBnds = parBnds , lnTfmdData = lnTfmdData)
if(any(is.na(c(parsBTfmd$sdfdPars_cd1,parsBTfmd$sdfdPars_cxd0,parsBTfmd$sdfdPars_cxd1))) ||
any(is.nan(c(parsBTfmd$sdfdPars_cd1,parsBTfmd$sdfdPars_cxd0,parsBTfmd$sdfdPars_cxd1))) ||
any(is.infinite(c(parsBTfmd$sdfdPars_cd1,parsBTfmd$sdfdPars_cxd0,parsBTfmd$sdfdPars_cxd1)))){
parsOK <- FALSE
A <- NA
sigma2Vec <- NA
p <- ncol(XData)
if(is.null(p)){ p <- 1 }else{}
}else{
tmp <- setCIAK3D(parsBTfmd = parsBTfmd , modelx = modelx ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 ,
cmeOpt = cmeOpt , setupMats = setupMats)
A <- tmp$C
sigma2Vec <- tmp$sigma2Vec
remove(tmp) ;
parsOK <- T
if(max(is.na(A)) == 1){ parsOK <- F }else{}
if(is.infinite(parsBTfmd$cx0) | is.infinite(parsBTfmd$cx1) | is.infinite(parsBTfmd$cd1) | is.infinite(parsBTfmd$cxd0) | is.infinite(parsBTfmd$cxd1)){ parsOK <- F }else{}
}
### add on the ssre bit
### all lmer parameters are on log scale and relative to cxd0hat
# ### [ie final cov mtx vals are cxdhat * exp(pars4lmer)]
# if(!is.null(setupMats$listStructs4ssre)){
# for(ip in 1:length(ipars_ssre)){
# A <- A + setupMats$listStructs4ssre[[ip]] * exp(pars[ipars_ssre[ip]])
# }
# # A <- A + do.call('+' , mapply('*' , setupMats$listStructs4ssre , exp(pars[ipars_ssre]) , SIMPLIFY = FALSE))
# }else{}
if(parsOK){
n <- length(zData)
XData <- as.matrix(XData , nrow = n)
p <- ncol(XData)
W <- cbind(XData,zData)
### just to make sure numerical errors have not made it non-symmetric...
if(!is(A , 'dspMatrix')){ A <- forceSymmetric(A) }else{}
if(lnTfmdData){
tmp <- nrUpdatesIAK3DlnN(zData = zData , XData = XData , vXU = vXU , iU = iU , C = A , sigma2Vec = sigma2Vec)
nll <- tmp$nll
betahat <- tmp$betahat
if(rtnAll){
# vbetahat <- solve(tmp$fim)
### or, alternatively,
vXUTmp <- matrix(0 , n * p , p)
vXUTmp[kronecker(seq(0 , (n - 1) * p , p) , matrix(1 , length(iU) , 1)) + rep(iU , n) , iU] <- vXU
tmp <- gradHessIAK3DlnN(zData = zData , XData = XData , vXU = vXUTmp , iU = seq(p) , betaU = betahat , C = A , lndetC = 0 , TK = tmp$iC , iXKiCXKXKiC = c() , sigma2Vec = sigma2Vec)
vbetahat <- solve(tmp$fim)
}else{
vbetahat <- NA
}
cxdhat <- 1
C <- A
}else{
iAW <- lndetANDinvCb(A , W)
lndetA <- iAW$lndetC
iAW <- iAW$invCb
if(is.na(lndetA) | is.infinite(lndetA)){
print('Inversion of A failed!')
print(paste0('nll = ' , badnll , '; nmPars = ' , paste(round(pars , digits = 3) , collapse = ', ')))
nll <- badnll
cxdhat <- NA
if(forCompLik){
return(list('pars' = pars , 'parsBTfmd' = parsBTfmd , 'sigma2Vec' = sigma2Vec , 'WiAW' = matrix(NA , p+1 , p+1) , 'lndetA' = NA))
}else{}
if(rtnAll){
### -999 * A returned because it is not the covariance matrix yet, as cxhat was not computed. Still allows A to be retrieved if required though.
if(attachBigMats){
return(list('nll' = nll , 'pars' = pars , 'parsBTfmd' = parsBTfmd , 'betahat' = NA , 'vbetahat' = NA , 'cxdhat' = NA ,
'C' = -999 * A , 'sigma2Vec' = sigma2Vec , 'XData' = XData , 'vXU' = vXU , 'iU' = iU))
}else{
return(list('nll' = nll , 'pars' = pars , 'parsBTfmd' = parsBTfmd , 'betahat' = NA , 'vbetahat' = NA , 'cxdhat' = NA ,
'sigma2Vec' = sigma2Vec , 'XData' = XData , 'vXU' = vXU , 'iU' = iU))
}
}else{
return(nll)
}
}else{}
iAW <- matrix(iAW , ncol = ncol(W))
WiAW <- t(W) %*% iAW
if(!is(WiAW , 'dspMatrix')){ WiAW <- forceSymmetric(WiAW) }else{}
if(forCompLik){
if(attachBigMats){
return(list('pars' = pars , 'parsBTfmd' = parsBTfmd , 'sigma2Vec' = sigma2Vec , 'WiAW' = WiAW , 'lndetA' = lndetA , 'A' = A , 'iAW' = iAW))
}else{
return(list('pars' = pars , 'parsBTfmd' = parsBTfmd , 'sigma2Vec' = sigma2Vec , 'WiAW' = WiAW , 'lndetA' = lndetA , 'iAW' = iAW))
}
}else{}
XiAX <- WiAW[1:p , 1:p , drop = FALSE]
XiAz <- WiAW[1:p , p+1 , drop = FALSE]
ziAz <- as.numeric(WiAW[p+1 , p+1])
betahat <- lndetANDinvCb(XiAX , XiAz)
lndetXiAX <- betahat$lndetC
betahat <- betahat$invCb
if(is.na(lndetXiAX) | is.infinite(lndetXiAX)){
print('Inversion of t(XData) iA XData failed!')
print(paste0('nll = ' , badnll , '; nmPars = ' , paste(round(pars , digits = 3) , collapse = ', ')))
nll <- badnll
if(rtnAll){
### -999 * A returned because it is not the covariance matrix yet, as cxhat was not computed. Still allows A to be retrieved if required though.
if(attachBigMats){
return(list('nll' = nll , 'pars' = pars , 'parsBTfmd' = parsBTfmd , 'betahat' = NA , 'vbetahat' = NA , 'cxdhat' = NA ,
'C' = -999 * A , 'sigma2Vec' = sigma2Vec , 'XData' = XData , 'vXU' = vXU , 'iU' = iU))
}else{
return(list('nll' = nll , 'pars' = pars , 'parsBTfmd' = parsBTfmd , 'betahat' = NA , 'vbetahat' = NA , 'cxdhat' = NA ,
'sigma2Vec' = sigma2Vec , 'XData' = XData , 'vXU' = vXU , 'iU' = iU))
}
}else{
return(nll)
}
}else{}
betahat <- matrix(betahat , ncol = 1)
resiAres <- as.numeric(ziAz - 2 * t(betahat) %*% XiAz + t(betahat) %*% XiAX %*% betahat)
if(useReml){
cxdhat <- resiAres / (n - p)
}else{
cxdhat <- resiAres / n
}
### update the values in parsBTfmd so that it makes sense when returned to user.
parsBTfmd$cx0 <- cxdhat * parsBTfmd$cx0
parsBTfmd$cx1 <- cxdhat * parsBTfmd$cx1
parsBTfmd$cd1 <- cxdhat * parsBTfmd$cd1
parsBTfmd$cxd0 <- cxdhat * parsBTfmd$cxd0
parsBTfmd$cxd1 <- cxdhat * parsBTfmd$cxd1
if(cmeOpt == 1){ parsBTfmd$cme <- cxdhat * parsBTfmd$cme }else{}
if(!is.null(parsBTfmd$cssre)){ parsBTfmd$cssre <- cxdhat * parsBTfmd$cssre }else{}
lndetC <- lndetA + n * log(cxdhat)
lndetXiCX <- lndetXiAX - p * log(cxdhat)
resiCres <- resiAres / cxdhat
if(rtnAll){
vbetahat <- cxdhat * solve(XiAX)
C <- cxdhat * A
sigma2Vec <- cxdhat * sigma2Vec
}else{}
if(useReml){
nll <- 0.5 * ((n - p) * log(2 * pi) + lndetC + lndetXiCX + resiCres)
}else{
nll <- 0.5 * (n * log(2 * pi) + lndetC + resiCres)
}
nll <- as.numeric(nll)
}
}else{
nll <- badnll
if(forCompLik){
return(list('pars' = pars , 'parsBTfmd' = parsBTfmd , 'sigma2Vec' = sigma2Vec , 'WiAW' = matrix(NA , p+1 , p+1) , 'lndetA' = NA))
}else{}
}
if(is.na(nll) | is.nan(nll) | is.infinite(nll)){ nll <- badnll }else{}
if(!lnTfmdData){
txtOut <- paste0('nll = ' , round(nll , digits = 3) ,
'; nmPars = ' , paste(round(pars , digits = 3) , collapse = ', ') ,
', cxdhat = ' , round(cxdhat , digits = 4))
}else{
txtOut <- paste0('nll = ' , round(nll , digits = 3) ,
'; nmPars = ' , paste(round(pars , digits = 3) , collapse = ', '))
}
if(verboseOptim){
print(txtOut)
}else{}
if(exists("printnllTime") && printnllTime){
print('time for nll evaluation was:')
print(proc.time() - ptm)
}else{}
if(exists("parsTrace4Optim") && (length(nllTrace4Optim) > 0) && (!forCompLik)){
nllTmp <- nllTrace4Optim # update the final value, now fn has been successful.
if(nll != badnll){
nllTmp[length(nllTmp)] <- nll
assign("nllTrace4Optim" , nllTmp , envir = .GlobalEnv)
}else{}
}else{}
if(rtnAll){
### added extra info for quicker prediction, 28/2/19...
if(!lnTfmdData){
muhat <- XData %*% betahat
}else{
muhat <- setmuIAK3D(XData = XData , vXU = vXU , iU = iU , beta = betahat , diagC = diag(C) , sigma2Vec = sigma2Vec)
}
if(attachBigMats){
### 30-08-2021 - some numerical issues with chol2inv(chol(C)) for teh edgeroi eg???
# iC <- chol2inv(chol(C))
iC <- solve(C)
iCXRes <- matrix(iC %*% cbind(XData , zData - muhat) , ncol = p+1)
return(list('nll' = nll , 'pars' = pars , 'parsBTfmd' = parsBTfmd , 'betahat' = betahat , 'vbetahat' = vbetahat , 'cxdhat' = cxdhat ,
'C' = C , 'sigma2Vec' = sigma2Vec , 'XData' = XData , 'vXU' = vXU , 'iU' = iU , 'iCX' = iCXRes[,1:p,drop=FALSE] , 'iCz_muhat' = iCXRes[,p+1,drop=FALSE] , 'iC' = iC))
}else{
# return(list('nll' = nll , 'pars' = pars , 'parsBTfmd' = parsBTfmd , 'betahat' = betahat , 'vbetahat' = vbetahat , 'cxdhat' = cxdhat ,
# 'sigma2Vec' = sigma2Vec , 'XData' = XData , 'vXU' = vXU , 'iU' = iU))
### change to have iAX and iAz, 02/09/2021...
return(list('nll' = nll , 'pars' = pars , 'parsBTfmd' = parsBTfmd , 'betahat' = betahat , 'vbetahat' = vbetahat , 'cxdhat' = cxdhat ,
'sigma2Vec' = sigma2Vec , 'XData' = XData , 'vXU' = vXU , 'iU' = iU ,
'iAX' = iAW[,1:p,drop=FALSE] ,'iAz' = iAW[,p+1,drop=FALSE]))
}
}else{
return(nll)
}
}
#######################################################
### use parsBTfmd to make cov mtx C for IAK3D
### updated from the above function, 28/2/20-5/3/20, to use dspMatrix class for more efficient memory.
#######################################################
setCIAK3D <- function(parsBTfmd , modelx ,
sdfdType_cd1 , sdfdType_cxd0 , sdfdType_cxd1 , cmeOpt , setupMats){
### check sdfdTypes - if any > 0, use the approx version...
if(max(c(sdfdType_cd1 , sdfdType_cxd0 , sdfdType_cxd1)) > 0){
return(setCApproxIAK3D(parsBTfmd = parsBTfmd , modelx = modelx ,
sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1 , cmeOpt = cmeOpt , setupMats = setupMats))
}else{}
### if setupMats doesn't include 'Kx', it should include 'xData' and 'dIData', so that proper setupMats can be made now...
if(is.null(setupMats$Kx)){
stop('I am not sure this option to call setCIAK3D without a prior call to setupIAK3D is needed!')
# if(max(c(sdfdType_cd1 , sdfdType_cxd0 , sdfdType_cxd1)) > 0){ stop('Not ready to run setCIAK3D without passing in setupMats yet, because this requires passing in sdfdKnots - update the function!') }else{}
# setupMats <- setupIAK3D(xData = setupMats$xData , dIData = as.data.frame(setupMats$dIData) , nDscPts = 0 ,
# sdfdType_cd1 = sdfdType_cd1 , sdfdType_cxd0 = sdfdType_cxd0 , sdfdType_cxd1 = sdfdType_cxd1)
}else{}
n <- nrow(setupMats$Kx)
C <- sparseMatrix(i=seq(n),j=seq(n),x=parsBTfmd$cme , symmetric = TRUE) # initiate with just meas err on diag. , dims = as.integer(c(n,n))
### adding Cx0...
C <- C + parsBTfmd$cx0 * forceSymmetric(setupMats$Kx %*% t(setupMats$Kx))
if ((modelx == 'matern') | (modelx == 'spherical') | (modelx == 'wendland')){
phix1 <- spatialCovIAK3D(setupMats$Dx , c(1 , parsBTfmd$ax , parsBTfmd$nux) , covModel = modelx)
}else if(modelx == 'nugget'){
phix1 <- 0
}else{
stop('Not ready!')
}
### if any components are stationary, function only neds to be run once...
iStat <- which(c(sdfdType_cd1 , sdfdType_cxd0 , sdfdType_cxd1) == 0)
if(length(iStat) > 0){
tmp <- iaCovMatern(dIData = setupMats$dIU , ad = parsBTfmd$ad , nud = parsBTfmd$nud , sdfdPars = c(1 , 0 , 1) , sdfdType = 0 , abcd = setupMats$dIUabcd , iUElements = setupMats$dIUiUElements)
phidStat <- tmp$avCovMtx
avVardStat <- tmp$avVarVec
remove(tmp) ;
}else{}
if(sdfdType_cd1 == 0){
phid_cd1 <- phidStat
avVard_cd1 <- avVardStat
}else if(sdfdType_cd1 == -9){
phid_cd1 <- 0
avVard_cd1 <- 0