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
|
> # test.parsnip.R: test the parsnip package with earth and other models
> # Stephen Milborrow Sep 2020 Petaluma
>
> source("test.prolog.R")
> options(warn=1) # print warnings as they occur
> library(earth)
Loading required package: Formula
Loading required package: plotmo
Loading required package: plotrix
> cat("loading parsnip libraries\n") # these libraries take several seconds to load
loading parsnip libraries
> library(tidymodels, quietly=TRUE, verbose=FALSE)
── Attaching packages ────────────────────────────────────── tidymodels 1.4.1 ──
✔ broom 1.0.11 ✔ recipes 1.3.1
✔ dials 1.4.2 ✔ rsample 1.3.1
✔ dplyr 1.1.4 ✔ tailor 0.1.0
✔ ggplot2 4.0.1 ✔ tidyr 1.3.2
✔ infer 1.1.0 ✔ tune 2.0.1
✔ modeldata 1.5.1 ✔ workflows 1.3.0
✔ parsnip 1.4.0 ✔ workflowsets 1.1.1
✔ purrr 1.2.0 ✔ yardstick 1.3.2
── Conflicts ───────────────────────────────────────── tidymodels_conflicts() ──
✖ purrr::discard() masks scales::discard()
✖ dplyr::filter() masks stats::filter()
✖ dplyr::lag() masks stats::lag()
✖ recipes::step() masks stats::step()
> library(timetk)
> library(lubridate)
Attaching package: 'lubridate'
The following objects are masked from 'package:base':
date, intersect, setdiff, union
> cat("loaded parsnip libraries\n")
loaded parsnip libraries
> cat("parsnip version:", as.character(packageVersion("parsnip")[[1]]), "\n")
parsnip version: 1.4.0
>
> vdata <- data.frame(
+ resp = 1:23,
+ bool = c(F, F, F, F, F, T, T, T, T, T, T, T, T, F, F, T, T, T, T, T, T, T, T),
+ ord = ordered(c("ORD1", "ORD1", "ORD1",
+ "ORD1", "ORD1", "ORD1",
+ "ORD1", "ORD3", "ORD1",
+ "ORD2", "ORD2", "ORD2", "ORD2",
+ "ORD2", "ORD2", "ORD2",
+ "ORD3", "ORD3", "ORD3",
+ "ORD2", "ORD2", "ORD2", "ORD2"),
+ levels=c("ORD1", "ORD3", "ORD2")),
+ fac = as.factor(c("FAC1", "FAC1", "FAC1",
+ "FAC2", "FAC2", "FAC2",
+ "FAC3", "FAC1", "FAC1",
+ "FAC1", "FAC2", "FAC2", "FAC2",
+ "FAC2", "FAC2", "FAC2",
+ "FAC3", "FAC3", "FAC3",
+ "FAC1", "FAC3", "FAC3", "FAC3")),
+ str = c("STR1", "STR1", "STR1", # WILL BE TREATED LIKE A FACTOR
+ "STR1", "STR1", "STR1",
+ "STR2", "STR2", "STR2",
+ "STR3", "STR3", "STR2", "STR3",
+ "STR2", "STR3", "STR2",
+ "STR3", "STR3", "STR3",
+ "STR3", "STR3", "STR3", "STR3"),
+ num = c(1, 9, 2, 3, 14, 5, 6, 4, 5, 6.5, 3, 6, 5,
+ 3, 4, 5, 6, 4, 5, 16.5, 3, 16, 15),
+ sqrt_num = sqrt(
+ c(1, 9, 2, 3, 14, 5, 6, 4, 5, 6.5, 3, 6, 5,
+ 3, 4, 5, 6, 4, 5, 16.5, 3, 16, 15)),
+ int = c(1L, 1L, 3L, 3L, 4L, 4L, 3L, 5L, 3L, 6L, 7L, 8L, 10L,
+ 13L, 14L, 3L, 13L, 5L, 13L, 16L, 17L, 18L, 11L),
+ date = as.Date(
+ c("2018-08-01", "2018-08-02", "2018-08-03",
+ "2018-08-04", "2018-08-05", "2018-08-06",
+ "2018-08-07", "2018-08-08", "2018-08-08",
+ "2018-08-10", "2018-08-10", "2018-08-11", "2018-08-11",
+ "2018-08-11", "2018-08-12", "2018-08-13",
+ "2018-08-10", "2018-08-15", "2018-08-17",
+ "2018-08-04", "2018-08-19", "2018-08-03", "2018-08-18")),
+ date_num = as.numeric(as.Date(
+ c("2018-08-01", "2018-08-02", "2018-08-03",
+ "2018-08-04", "2018-08-05", "2018-08-06",
+ "2018-08-07", "2018-08-08", "2018-08-08",
+ "2018-08-10", "2018-08-10", "2018-08-11", "2018-08-11",
+ "2018-08-11", "2018-08-12", "2018-08-13",
+ "2018-08-10", "2018-08-15", "2018-08-17",
+ "2018-08-04", "2018-08-19", "2018-08-03", "2018-08-18"))))
>
> set.seed(2020)
> splits <- initial_time_split(vdata, prop=.9)
>
> #--- lm ----------------------------------------------------------------------
>
> lm1 <- lm(resp~num+fac:int+date+ord+str, data=training(splits))
> cat("lm1:\n")
lm1:
> print(summary(lm1))
Call:
lm(formula = resp ~ num + fac:int + date + ord + str, data = training(splits))
Residuals:
Min 1Q Median 3Q Max
-3.9119 -0.6559 -0.0438 0.7549 3.1946
Coefficients:
Estimate Std. Error t value Pr(>|t|)
(Intercept) -1.396e+04 5.559e+03 -2.511 0.0309 *
num 1.818e-01 1.883e-01 0.966 0.3571
date 7.867e-01 3.132e-01 2.512 0.0308 *
ord.L 1.254e+00 2.009e+00 0.624 0.5465
ord.Q 3.783e-01 1.910e+00 0.198 0.8470
strSTR2 5.801e-01 2.381e+00 0.244 0.8124
strSTR3 3.341e-01 3.136e+00 0.107 0.9173
facFAC1:int 6.908e-01 3.066e-01 2.253 0.0479 *
facFAC2:int 2.891e-01 2.116e-01 1.366 0.2018
facFAC3:int 5.818e-01 2.621e-01 2.220 0.0507 .
---
Signif. codes: 0 '***' 0.001 '**' 0.01 '*' 0.05 '.' 0.1 ' ' 1
Residual standard error: 2.341 on 10 degrees of freedom
Multiple R-squared: 0.9176, Adjusted R-squared: 0.8435
F-statistic: 12.38 on 9 and 10 DF, p-value: 0.0002531
> set.seed(2020)
> lmpar <- linear_reg(mode = "regression") %>%
+ set_engine("lm") %>%
+ fit(resp~num+fac:int+date+ord+str, data = training(splits))
> stopifnot(identical(lm1$coeff, lmpar$fit$coeff))
>
> predict.lm1 <- predict(lm1, testing(splits))
> predict.lmpar <- lmpar %>% predict(testing(splits))
> stopifnot(all(predict.lm1 == predict.lmpar))
>
> par(mfrow = c(3, 3), mar = c(3, 3, 3, 1), mgp = c(1.5, 0.5, 0))
> plotmo(lm1, do.par=2, SHOWCALL=TRUE)
plotmo grid: num fac int date ord str
5 FAC2 5 2018-08-09 ORD1 STR3
> plotres(lm1, which=c(3,1), do.par=FALSE)
> plotmo(lmpar, do.par=2, SHOWCALL=TRUE)
plotmo grid: num fac int date ord str
5 FAC2 5 2018-08-09 ORD1 STR3
> plotres(lmpar, which=c(3,1), do.par=FALSE)
> plotmo(lmpar$fit, do.par=2, SHOWCALL=TRUE)
plotmo grid: num fac int date ord str
5 FAC2 5 2018-08-09 ORD1 STR3
> plotres(lmpar$fit, which=c(3,1), do.par=FALSE)
> par(org.par)
>
> lmpar.sqrtnum <- linear_reg(mode = "regression") %>%
+ set_engine("lm") %>%
+ fit(resp~sqrt(num), data = training(splits))
> #$$ TODO
> # expect.err(try(plotmo(lmpar.sqrtnum)),
> # "cannot get the original model predictors")
>
> #--- earth -------------------------------------------------------------------
>
> # note that sqrt(num) is ok, unlike parsnip models for lm and rpart
> earth1 <- earth(resp~sqrt(num)+int+ord:bool+fac+str+date, degree=2,
+ data=training(splits), pmethod="none")
> cat("earth1:\n")
earth1:
> print(summary(earth1))
Call: earth(formula=resp~sqrt(num)+int+ord:bool+fac+str+date,
data=training(splits), pmethod="none", degree=2)
coefficients
(Intercept) 7.86702
ordORD2:boolTRUE -0.81733
h(5-int) 0.46965
h(int-5) 2587.95933
h(17751-date) -1.23206
h(date-17751) 1.48020
h(int-5) * facFAC2 -0.35097
h(int-5) * date -0.14573
Selected 8 of 8 terms, and 4 of 13 predictors (pmethod="none")
Termination condition: GRSq -Inf at 8 terms
Importance: int, date, facFAC2, ordORD2:boolTRUE, sqrt(num)-unused, ...
Number of terms at each degree of interaction: 1 5 2
GCV 19.29495 RSS 2.170681 GRSq 0.47628 RSq 0.9967358
> set.seed(2020)
> earthpar <- mars(mode = "regression", prune_method="none", prod_degree=2) %>%
+ set_engine("earth") %>%
+ fit(resp~sqrt(num)+int+ord:bool+fac+str+date, data = training(splits))
> cat("earthpar:\n")
earthpar:
> print(earthpar)
parsnip model object
Selected 8 of 8 terms, and 4 of 13 predictors (pmethod="none")
Termination condition: GRSq -Inf at 8 terms
Importance: int, date, facFAC2, ordORD2:boolTRUE, sqrt(num)-unused, ...
Number of terms at each degree of interaction: 1 5 2
GCV 19.29495 RSS 2.170681 GRSq 0.47628 RSq 0.9967358
> cat("summary(earthpar$fit)\n")
summary(earthpar$fit)
> print(summary(earthpar$fit))
Call: earth(formula=resp~sqrt(num)+int+ord:bool+fac+str+date, data=data,
pmethod=~"none", keepxy=TRUE, degree=~2)
coefficients
(Intercept) 7.86702
ordORD2:boolTRUE -0.81733
h(5-int) 0.46965
h(int-5) 2587.95933
h(17751-date) -1.23206
h(date-17751) 1.48020
h(int-5) * facFAC2 -0.35097
h(int-5) * date -0.14573
Selected 8 of 8 terms, and 4 of 13 predictors (pmethod="none")
Termination condition: GRSq -Inf at 8 terms
Importance: int, date, facFAC2, ordORD2:boolTRUE, sqrt(num)-unused, ...
Number of terms at each degree of interaction: 1 5 2
GCV 19.29495 RSS 2.170681 GRSq 0.47628 RSq 0.9967358
> stopifnot(identical(earth1$coeff, earthpar$fit$coeff))
>
> predict.earth1 <- predict(earth1, testing(splits))
> predict.earthpar <- earthpar %>% predict(testing(splits))
> stopifnot(all(predict.earth1 == predict.earthpar))
>
> par(mfrow = c(3, 3), mar = c(3, 3, 3, 1), mgp = c(1.5, 0.5, 0))
> plotmo(earth1, do.par=2, pt.col=3, SHOWCALL=TRUE)
plotmo grid: num int ord bool fac str date
5 5 ORD1 TRUE FAC2 STR3 2018-08-09
> set.seed(2020)
> plotres(earth1, which=c(1,3), do.par=FALSE, pt.col=3, legend.pos="topleft")
> par(org.par)
>
> par(mfrow = c(3, 3), mar = c(3, 3, 3, 1), mgp = c(1.5, 0.5, 0))
> plotmo(earthpar, do.par=2, pt.col=3, SHOWCALL=TRUE)
plotmo grid: num int ord bool fac str date
5 5 ORD1 TRUE FAC2 STR3 2018-08-09
> set.seed(2020)
> plotres(earthpar, which=c(1,3), do.par=FALSE, pt.col=3, legend.pos="topleft")
> par(org.par)
>
> par(mfrow = c(3, 3), mar = c(3, 3, 3, 1), mgp = c(1.5, 0.5, 0))
> plotmo(earthpar$fit, do.par=2, pt.col=3, SHOWCALL=TRUE)
plotmo grid: num int ord bool fac str date
5 5 ORD1 TRUE FAC2 STR3 2018-08-09
> set.seed(2020)
> plotres(earthpar$fit, which=c(1,3), do.par=FALSE, pt.col=3, legend.pos="topleft")
> par(org.par)
>
> #--- rpart -------------------------------------------------------------------
>
> library(rpart)
Attaching package: 'rpart'
The following object is masked from 'package:dials':
prune
> library(rpart.plot)
> rpart1 <- rpart(resp~num+fac+int+date+ord+str, data=training(splits),
+ control=rpart.control(minsplit=1, cp=.0001))
> cat("\nrpart.rules(rpart1)\n")
rpart.rules(rpart1)
> print(rpart.rules(rpart1))
resp
1 when ord is ORD1 & date < 17748 & num < 5.0 & int < 2
2 when ord is ORD1 & date < 17748 & num >= 5.0 & int < 2
3 when ord is ORD1 & date < 17748 & num < 2.5 & int >= 2
4 when ord is ORD1 & date < 17748 & num >= 2.5 & int >= 2
5 when ord is ORD1 & date >= 17748 & num >= 10.0 & fac is FAC2 or FAC3
6 when ord is ORD1 & date >= 17748 & num < 5.5 & fac is FAC2 or FAC3
7 when ord is ORD1 & date >= 17748 & num is 5.5 to 10.0 & fac is FAC2 or FAC3
9 when ord is ORD1 & date >= 17748 & fac is FAC1
14 when ord is ORD3 or ORD2
>
> set.seed(2020)
> # TODO note need of model=TRUE below (needed only for further processing with e.g. plotmo)
> rpartpar <- decision_tree(mode = "regression", min_n=1, cost_complexity=.0001) %>%
+ set_engine("rpart", model=TRUE) %>%
+ fit(resp~num+fac+int+date+ord+str, data = training(splits))
> cat("\nrpart.rules(rpartpar$fit)\n")
rpart.rules(rpartpar$fit)
> print(rpart.rules(rpartpar$fit))
resp
1 when ord is ORD1 & date < 17748 & num < 5.0 & int < 2
2 when ord is ORD1 & date < 17748 & num >= 5.0 & int < 2
3 when ord is ORD1 & date < 17748 & num < 2.5 & int >= 2
4 when ord is ORD1 & date < 17748 & num >= 2.5 & int >= 2
5 when ord is ORD1 & date >= 17748 & num >= 10.0 & fac is FAC2 or FAC3
6 when ord is ORD1 & date >= 17748 & num < 5.5 & fac is FAC2 or FAC3
7 when ord is ORD1 & date >= 17748 & num is 5.5 to 10.0 & fac is FAC2 or FAC3
9 when ord is ORD1 & date >= 17748 & fac is FAC1
14 when ord is ORD3 or ORD2
>
> predict.rpart1 <- predict(rpart1, testing(splits))
> predict.rpartpar <- rpartpar %>% predict(testing(splits))
> stopifnot(all(predict.rpart1 == predict.rpartpar))
>
> par(mfrow = c(3, 3), mar = c(3, 3, 3, 1), mgp = c(1.5, 0.5, 0))
> plotmo(rpart1, do.par=2, SHOWCALL=TRUE, trace=0)
plotmo grid: num fac int date ord str
5 FAC2 5 2018-08-09 ORD1 STR3
> plotres(rpart1, which=c(3,1), do.par=FALSE)
> plotmo(rpartpar, do.par=2, SHOWCALL=TRUE, trace=0)
plotmo grid: num fac int date ord str
5 FAC2 5 2018-08-09 ORD1 STR3
> plotres(rpartpar, which=c(3,1), do.par=FALSE)
> plotmo(rpartpar$fit, do.par=2, SHOWCALL=TRUE)
plotmo grid: num fac int date ord str
5 FAC2 5 2018-08-09 ORD1 STR3
> plotres(rpartpar$fit, which=c(3,1), do.par=FALSE)
> par(org.par)
>
> # TODO note that this differs from the above rpart model in that we don't use model=TRUE
> rpartpar.nosavemodel <- decision_tree(mode = "regression", min_n=1, cost_complexity=.0001) %>%
+ set_engine("rpart") %>%
+ fit(resp~num+fac+int+date+str, data = training(splits))
>
> cat("\nrpart.rules(rpartpar.nosavemodel$fit)\n")
rpart.rules(rpartpar.nosavemodel$fit)
> options(warn=2)
> expect.err(try(rpart.rules(rpartpar.nosavemodel$fit)),
+ "Cannot retrieve the data used to build the model")
Error : (converted from warning) Cannot retrieve the data used to build the model (so cannot determine roundint and is.binary for the variables).
To silence this warning:
Call rpart.rules with roundint=FALSE,
or rebuild the rpart model with model=TRUE.
Got expected error from try(rpart.rules(rpartpar.nosavemodel$fit))
> options(warn=1)
> expect.err(try(plotmo(rpartpar.nosavemodel)),
+ "Cannot plot parsnip rpart model: need model=TRUE in call to rpart")
Error : Cannot plot parsnip rpart model: need model=TRUE in call to rpart
Do it like this: set_engine("rpart", model=TRUE)
Got expected error from try(plotmo(rpartpar.nosavemodel))
>
> rpart.sqrtnum <- decision_tree(mode = "regression", min_n=1, cost_complexity=.0001) %>%
+ set_engine("rpart", model=TRUE) %>%
+ fit(resp~sqrt(num)+fac+int+date+ord+str, data = training(splits))
> cat("\nrpart.rules(rpart.sqrtnum$fit)\n")
rpart.rules(rpart.sqrtnum$fit)
> print(rpart.rules(rpart.sqrtnum$fit)) # ok
resp
1 when ord is ORD1 & date < 17748 & sqrt(num) < 2.0 & int < 2
2 when ord is ORD1 & date < 17748 & sqrt(num) >= 2.0 & int < 2
3 when ord is ORD1 & date < 17748 & sqrt(num) < 1.6 & int >= 2
4 when ord is ORD1 & date < 17748 & sqrt(num) >= 1.6 & int >= 2
5 when ord is ORD1 & date >= 17748 & sqrt(num) >= 3.1 & fac is FAC2 or FAC3
6 when ord is ORD1 & date >= 17748 & sqrt(num) < 2.3 & fac is FAC2 or FAC3
7 when ord is ORD1 & date >= 17748 & sqrt(num) is 2.3 to 3.1 & fac is FAC2 or FAC3
9 when ord is ORD1 & date >= 17748 & fac is FAC1
14 when ord is ORD3 or ORD2
> #$$ TODO
> # expect.err(try(plotmo(rpart.sqrtnum)),
> # "cannot get the original model predictors")
>
> #-----------------------------------------------------------------------------------
> # Test fix for github bug report https://github.com/tidymodels/parsnip/issues/341
> # (fixed Sep 2020)
>
> cat("===m750a first example===\n")
===m750a first example===
> set.seed(2020)
> m750a <- m4_monthly %>%
+ filter(id == "M750") %>%
+ select(-id)
> print(m750a) # a tibble
# A tibble: 306 × 2
date value
<date> <dbl>
1 1990-01-01 6370
2 1990-02-01 6430
3 1990-03-01 6520
4 1990-04-01 6580
5 1990-05-01 6620
6 1990-06-01 6690
7 1990-07-01 6000
8 1990-08-01 5450
9 1990-09-01 6480
10 1990-10-01 6820
# ℹ 296 more rows
> set.seed(2020)
> splits_a <- initial_time_split(m750a, prop = 0.9)
> earth_m750a <- earth(log(value) ~ as.numeric(date) + month(date, label = TRUE), data = training(splits_a), degree=2)
> print(summary(earth_m750a))
Call: earth(formula=log(value)~as.numeric(date)+month(date,label=TRUE),
data=training(splits_a), degree=2)
coefficients
(Intercept) 1.000000e+01
h(as.numeric(date)-7639) 0.000000e+00
h(as.numeric(date)-9100) 0.000000e+00
h(12022-as.numeric(date)) 0.000000e+00
h(as.numeric(date)-13483) 0.000000e+00
h(as.numeric(date)-14579) 0.000000e+00
h(0.370142-month(date, label = TRUE)^7) 0.000000e+00
h(month(date, label = TRUE)^7-0.370142) 0.000000e+00
h(month(date, label = TRUE)^10-0.491049) -3.077492e+12
h(as.numeric(date)-9100) * h(-0.254544-month(date, label = TRUE)^8) 0.000000e+00
h(as.numeric(date)-13483) * h(month(date, label = TRUE)^11- -0.392904) 0.000000e+00
h(as.numeric(date)-13483) * h(-0.392904-month(date, label = TRUE)^11) 0.000000e+00
h(0.491049-month(date, label = TRUE)^10) * h(month(date, label = TRUE)^11-0.065484) 0.000000e+00
h(0.491049-month(date, label = TRUE)^10) * h(0.065484-month(date, label = TRUE)^11) 0.000000e+00
Selected 14 of 17 terms, and 5 of 12 predictors
Termination condition: RSq changed by less than 0.001 at 17 terms
Importance: as.numeric(date), month(date, label = TRUE)^10, ...
Number of terms at each degree of interaction: 1 8 5
GCV 0.0004725457 RSS 0.1002179 GRSq 0.9834104 RSq 0.9871125
> set.seed(2020)
> model_m750a <- mars(mode = "regression", prod_degree=2) %>%
+ set_engine("earth") %>%
+ fit(log(value) ~ as.numeric(date) + month(date, label = TRUE), data = training(splits_a))
> print(summary(model_m750a$fit))
Call: earth(formula=log(value)~as.numeric(date)+month(date,label=TRUE),
data=data, keepxy=TRUE, degree=~2)
coefficients
(Intercept) 1.000000e+01
h(as.numeric(date)-7639) 0.000000e+00
h(as.numeric(date)-9100) 0.000000e+00
h(12022-as.numeric(date)) 0.000000e+00
h(as.numeric(date)-13483) 0.000000e+00
h(as.numeric(date)-14579) 0.000000e+00
h(0.370142-month(date, label = TRUE)^7) 0.000000e+00
h(month(date, label = TRUE)^7-0.370142) 0.000000e+00
h(month(date, label = TRUE)^10-0.491049) -3.077492e+12
h(as.numeric(date)-9100) * h(-0.254544-month(date, label = TRUE)^8) 0.000000e+00
h(as.numeric(date)-13483) * h(month(date, label = TRUE)^11- -0.392904) 0.000000e+00
h(as.numeric(date)-13483) * h(-0.392904-month(date, label = TRUE)^11) 0.000000e+00
h(0.491049-month(date, label = TRUE)^10) * h(month(date, label = TRUE)^11-0.065484) 0.000000e+00
h(0.491049-month(date, label = TRUE)^10) * h(0.065484-month(date, label = TRUE)^11) 0.000000e+00
Selected 14 of 17 terms, and 5 of 12 predictors
Termination condition: RSq changed by less than 0.001 at 17 terms
Importance: as.numeric(date), month(date, label = TRUE)^10, ...
Number of terms at each degree of interaction: 1 8 5
GCV 0.0004725457 RSS 0.1002179 GRSq 0.9834104 RSq 0.9871125
> stopifnot(identical(earth_m750a$coeff, model_m750a$fit$coeff))
> predict_earth_m750a <- predict(earth_m750a, newdata=testing(splits_a)[1:3,])
> predict_m750a <- model_m750a %>% predict(testing(splits_a)[1:3,])
> stopifnot(max(c(9.238049628, 9.240535151, 9.232361834) - predict_m750a) < 1e-8)
> stopifnot(max(predict_earth_m750a - predict_m750a) < 1e-20)
>
> par(mfrow = c(2, 2), mar = c(3, 3, 3, 1), mgp = c(1.5, 0.5, 0))
> set.seed(2020)
> plotmo(model_m750a, trace=2, do.par=FALSE, pt.col="green", main="model_m750a", SHOWCALL=TRUE)
plotmo trace 2: plotmo(object=model_m750a, pt.col="green", do.par=FALSE,
trace=2, main="model_m750a", SHOWCALL=TRUE)
--get.model.env for object with class _earth
plotmo parsnip model: will plot model_m750a$fit, not 'model_m750a' itself
object call is earth(formula=log(value)~as.numeric(date)+month(date,
label=TRUE), data=data, keepxy=TRUE, degree=~2)
using the environment saved in $terms of the earth model: env(data, weights)
--plotmo_prolog for _earth object 'model_m750a'
--plotmo_x for earth object
get.object.x:
object$x is NULL (and it has no colnames)
object call is earth(formula=log(value)~as.numeric(date)+month(date, label=TRUE), data=data,...
get.x.from.model.frame:
formula(object) is log(value) ~ as.numeric(date) + month(date, label = TRUE)
naked formula is log(value) ~ date
formula is valid, now looking for data for the model.frame
object$model is NULL (and it has no colnames)
object$data is usable and has column names date value
na.action(object) is "na.fail"
stats::model.frame(log(value) ~ date, data=object$data, na.action="na.fail")
x=model.frame[,-1] is usable and has column name date
plotmo_x returned[275,1]:
date
1 1990-01-01
2 1990-02-01
3 1990-03-01
... 1990-04-01
275 2012-11-01
----Metadata: plotmo_predict with nresponse=NULL and newdata=NULL
calling predict.earth with NULL newdata
stats::predict(earth.object, NULL, type="response")
predict returned[275,1]:
log(value)
1 8.779940
2 8.777069
3 8.795003
... 8.799953
275 9.244442
predict after processing with nresponse=NULL is [275,1]:
log(value)
1 8.779940
2 8.777069
3 8.795003
... 8.799953
275 9.244442
----Metadata: plotmo_fitted with nresponse=NULL
stats::fitted(object=earth.object)
fitted(object) returned[275,1]:
log(value)
1 8.779940
2 8.777069
3 8.795003
... 8.799953
275 9.244442
fitted(object) after processing with nresponse=NULL is [275,1]:
log(value)
1 8.779940
2 8.777069
3 8.795003
... 8.799953
275 9.244442
----Metadata: plotmo_y with nresponse=NULL
--plotmo_y with nresponse=NULL for earth object
get.object.y:
object$y is usable and has column name log(value)
plotmo_y returned[275,1]:
log(value)
1 8.759355
2 8.768730
3 8.782630
... 8.791790
275 9.271435
plotmo_y after processing with nresponse=NULL is [275,1]:
log(value)
1 8.759355
2 8.768730
3 8.782630
... 8.791790
275 9.271435
converted nresponse=NA to nresponse=1
nresponse=1 (was NA) ncol(fitted) 1 ncol(predict) 1 ncol(y) 1
----Metadata: plotmo_y with nresponse=1
--plotmo_y with nresponse=1 for earth object
get.object.y:
object$y is usable and has column name log(value)
got model response from object$y
plotmo_y returned[275,1]:
log(value)
1 8.759355
2 8.768730
3 8.782630
... 8.791790
275 9.271435
plotmo_y after processing with nresponse=1 is [275,1]:
log(value)
1 8.759355
2 8.768730
3 8.782630
... 8.791790
275 9.271435
got response name "log(value)" from yhat
resp.levs is NULL
----Metadata: done
number of x values: date 275
----plotmo_singles for earth object
singles: 1 date
----plotmo_pairs for earth object
no pairs
----Figuring out ylim
--get.ylim.by.dummy.plots
--plot.degree1(draw.plot=FALSE)
degree1 plot1 (pmethod "plotmo") variable date
newdata[50,1]:
date
1 1990-01-01
2 1990-06-20
3 1990-12-07
... 1991-05-26
50 2012-11-01
stats::predict(earth.object, data.frame[50,1], type="response")
predict returned[50,1]:
log(value)
1 8.779940
2 8.797283
3 8.835027
... 8.843453
50 9.244442
predict after processing with nresponse=1 is [50,1]:
log(value)
1 8.779940
2 8.797283
3 8.835027
... 8.843453
50 9.244442
--done get.ylim.by.dummy.plots
ylim c(8.603, 9.289) clip TRUE
--plot.degree1(draw.plot=TRUE)
graphics::plot.default(x=Date:1990-01-01 1990-06-20 1990-12...,
y=c(8.78,8.797,8...), type="n", main="model_m750a",
xlab="", ylab="", xaxt="s", yaxt="s",
xlim=Date:1990-01-01 2012-11-01, ylim=c(8.603,9.289))
> set.seed(2020)
> plotmo(model_m750a$fit, trace=1, do.par=FALSE, pt.col="green", main="model_m750a$fit", SHOWCALL=TRUE)
stats::predict(earth.object, NULL, type="response")
stats::fitted(object=earth.object)
got model response from object$y
> set.seed(2020)
> plotmo(earth_m750a, trace=1, do.par=FALSE, pt.col="green", main="earth_m750a", SHOWCALL=TRUE)
stats::predict(earth.object, NULL, type="response")
stats::fitted(object=earth.object)
got model response from model.frame(log(value) ~ as.numeric(date) + month...,
data=call$data, na.action="na.fail")
> par(org.par)
>
> cat("===m750a second example===\n")
===m750a second example===
> set.seed(2020)
> m750b <- m4_monthly %>%
+ filter(id == "M750") %>%
+ select(-id) %>%
+ rename(date2 = date)
> print(m750b) # tibble
# A tibble: 306 × 2
date2 value
<date> <dbl>
1 1990-01-01 6370
2 1990-02-01 6430
3 1990-03-01 6520
4 1990-04-01 6580
5 1990-05-01 6620
6 1990-06-01 6690
7 1990-07-01 6000
8 1990-08-01 5450
9 1990-09-01 6480
10 1990-10-01 6820
# ℹ 296 more rows
> set.seed(2020)
> splits_b <- initial_time_split(m750b, prop = 0.9)
> set.seed(2020)
> model_m750b <- mars(mode = "regression") %>%
+ set_engine("earth") %>%
+ fit(log(value) ~ as.numeric(date2) + month(date2, label = TRUE), data = training(splits_b))
> # new data that only contains the feature "date" as a predictor
> future_data <- m750b %>% future_frame(date2, .length_out = "3 years")
> print(future_data) # a tibble with a single column of class "Date"
# A tibble: 36 × 1
date2
<date>
1 2015-07-01
2 2015-08-01
3 2015-09-01
4 2015-10-01
5 2015-11-01
6 2015-12-01
7 2016-01-01
8 2016-02-01
9 2016-03-01
10 2016-04-01
# ℹ 26 more rows
> stopifnot(class(future_data[,1,drop=TRUE]) == "Date")
> predict_m750a <- model_m750b %>% predict(new_data = future_data)
>
> par(mfrow = c(2, 2), mar = c(3, 3, 3, 1), mgp = c(1.5, 0.5, 0))
> set.seed(2020)
> plotmo(model_m750b, trace=2, do.par=FALSE, pt.col="green", main="model_m750b", SHOWCALL=TRUE)
plotmo trace 2: plotmo(object=model_m750b, pt.col="green", do.par=FALSE,
trace=2, main="model_m750b", SHOWCALL=TRUE)
--get.model.env for object with class _earth
plotmo parsnip model: will plot model_m750b$fit, not 'model_m750b' itself
object call is earth(formula=log(value)~as.numeric(date2)+month(date2,
label=TRUE), data=data, keepxy=TRUE)
using the environment saved in $terms of the earth model: env(data, weights)
--plotmo_prolog for _earth object 'model_m750b'
--plotmo_x for earth object
get.object.x:
object$x is NULL (and it has no colnames)
object call is earth(formula=log(value)~as.numeric(date2)+month(date2, label=TRUE), data=dat...
get.x.from.model.frame:
formula(object) is log(value) ~ as.numeric(date2) + month(date2, label = TRUE)
naked formula is log(value) ~ date2
formula is valid, now looking for data for the model.frame
object$model is NULL (and it has no colnames)
object$data is usable and has column names date2 value
na.action(object) is "na.fail"
stats::model.frame(log(value) ~ date2, data=object$data, na.action="na.fail")
x=model.frame[,-1] is usable and has column name date2
plotmo_x returned[275,1]:
date2
1 1990-01-01
2 1990-02-01
3 1990-03-01
... 1990-04-01
275 2012-11-01
----Metadata: plotmo_predict with nresponse=NULL and newdata=NULL
calling predict.earth with NULL newdata
stats::predict(earth.object, NULL, type="response")
predict returned[275,1]:
log(value)
1 8.773349
2 8.779320
3 8.797022
... 8.803553
275 9.243245
predict after processing with nresponse=NULL is [275,1]:
log(value)
1 8.773349
2 8.779320
3 8.797022
... 8.803553
275 9.243245
----Metadata: plotmo_fitted with nresponse=NULL
stats::fitted(object=earth.object)
fitted(object) returned[275,1]:
log(value)
1 8.773349
2 8.779320
3 8.797022
... 8.803553
275 9.243245
fitted(object) after processing with nresponse=NULL is [275,1]:
log(value)
1 8.773349
2 8.779320
3 8.797022
... 8.803553
275 9.243245
----Metadata: plotmo_y with nresponse=NULL
--plotmo_y with nresponse=NULL for earth object
get.object.y:
object$y is usable and has column name log(value)
plotmo_y returned[275,1]:
log(value)
1 8.759355
2 8.768730
3 8.782630
... 8.791790
275 9.271435
plotmo_y after processing with nresponse=NULL is [275,1]:
log(value)
1 8.759355
2 8.768730
3 8.782630
... 8.791790
275 9.271435
converted nresponse=NA to nresponse=1
nresponse=1 (was NA) ncol(fitted) 1 ncol(predict) 1 ncol(y) 1
----Metadata: plotmo_y with nresponse=1
--plotmo_y with nresponse=1 for earth object
get.object.y:
object$y is usable and has column name log(value)
got model response from object$y
plotmo_y returned[275,1]:
log(value)
1 8.759355
2 8.768730
3 8.782630
... 8.791790
275 9.271435
plotmo_y after processing with nresponse=1 is [275,1]:
log(value)
1 8.759355
2 8.768730
3 8.782630
... 8.791790
275 9.271435
got response name "log(value)" from yhat
resp.levs is NULL
----Metadata: done
number of x values: date2 275
----plotmo_singles for earth object
singles: 1 date2
----plotmo_pairs for earth object
no pairs
----Figuring out ylim
--get.ylim.by.dummy.plots
--plot.degree1(draw.plot=FALSE)
degree1 plot1 (pmethod "plotmo") variable date2
newdata[50,1]:
date2
1 1990-01-01
2 1990-06-20
3 1990-12-07
... 1991-05-26
50 2012-11-01
stats::predict(earth.object, data.frame[50,1], type="response")
predict returned[50,1]:
log(value)
1 8.773349
2 8.797894
3 8.831375
... 8.848941
50 9.243245
predict after processing with nresponse=1 is [50,1]:
log(value)
1 8.773349
2 8.797894
3 8.831375
... 8.848941
50 9.243245
--done get.ylim.by.dummy.plots
ylim c(8.603, 9.289) clip TRUE
--plot.degree1(draw.plot=TRUE)
graphics::plot.default(x=Date:1990-01-01 1990-06-20 1990-12...,
y=c(8.773,8.798,8...), type="n", main="model_m750b",
xlab="", ylab="", xaxt="s", yaxt="s",
xlim=Date:1990-01-01 2012-11-01, ylim=c(8.603,9.289))
> set.seed(2020)
> plotmo(model_m750b$fit, trace=1, do.par=FALSE, pt.col="green", main="model_m750b$fit", SHOWCALL=TRUE)
stats::predict(earth.object, NULL, type="response")
stats::fitted(object=earth.object)
got model response from object$y
> par(org.par)
>
> #-----------------------------------------------------------------------------------
> # multiple response earth model
>
> data(etitanic)
>
> etit <- etitanic
> etit$survived <- factor(ifelse(etitanic$survived == 1, "yes", "no"),
+ levels = c("yes", "no"))
> etit$notsurvived <- factor(ifelse(etitanic$survived == 0, "notsurvived", "survived"),
+ levels = c("notsurvived", "survived"))
> set.seed(2020)
> earth_tworesp <- earth(survived + notsurvived ~ ., data=etit, degree=2)
> print(summary(earth_tworesp))
Call: earth(formula=survived+notsurvived~., data=etit, degree=2)
survived notsurvived
(Intercept) 0.03829050 0.96170950
pclass3rd 0.81545352 -0.81545352
sexmale 0.57003496 -0.57003496
h(age-32) 0.00471938 -0.00471938
pclass2nd * sexmale 0.26568920 -0.26568920
pclass3rd * sexmale -0.19310203 0.19310203
pclass3rd * h(4-sibsp) -0.10222181 0.10222181
sexmale * h(16-age) -0.04505232 0.04505232
Selected 8 of 17 terms, and 5 of 6 predictors
Termination condition: Reached nk 21
Importance: sexmale, pclass3rd, pclass2nd, age, sibsp, parch-unused
Number of terms at each degree of interaction: 1 3 4
GCV RSS GRSq RSq
survived 0.1404529 141.7629 0.4197106 0.4389834
notsurvived 0.1404529 141.7629 0.4197106 0.4389834
All 0.2809057 283.5258 0.4197106 0.4389834
>
> # TODO following commented out because parsnip (version 0.1.5) says "'+' not meaningful for factors"
> # set.seed(2020)
> # mars_tworesp <- mars(mode = "regression", prod_degree=2) %>%
> # set_engine("earth") %>%
> # fit(survived + notsurvived~., data=etit)
> # print(summary(mars_tworesp))
> # print(summary(mars_tworesp$fit))
> #
> # stopifnot(identical(earth_tworesp$coeff, mars_tworesp$fit$coeff))
> #
> # predict.earth_tworesp <- predict(earth_tworesp, etit[3:6,])
> # predict.mars_tworesp <- mars_tworesp %>% predict(etit[3:6,])
> # stopifnot(all(predict.earth_tworesp == predict.mars_tworesp))
> #
> # plotmo(earth_tworesp, trace=0, nresponse=1, SHOWCALL=TRUE)
> # plotmo(mars_tworesp, trace=0, nresponse=1, SHOWCALL=TRUE)
> # plotmo(mars_tworesp, trace=0, nresponse=2, SHOWCALL=TRUE)
>
> source("test.epilog.R")
|
