forked from snoopryan123/fourth_down
-
Notifications
You must be signed in to change notification settings - Fork 0
/
d0_data2h_add_ep_wp.R
1422 lines (1209 loc) · 66 KB
/
d0_data2h_add_ep_wp.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
### purpose of this file: helper functions for `0_data2_acquisition.R`
################################################################################
# Author: Sebastian Carl
# Purpose: Functions to add ep(a) and wp(a) variables
# Code Style Guide: styler::tidyverse_style()
################################################################################
# For now those functions are only a call to nflscrapR but we may want to
# use other models and epa wpa definitions. This is the place where this
# could happen
add_ep <- function(pbp) {
out <- pbp %>% add_ep_variables()
user_message("added ep variables", "done")
return(out)
}
add_air_yac_ep <- function(pbp) {
if (nrow(pbp %>% dplyr::filter(!is.na(.data$air_yards))) == 0) {
out <- pbp %>%
dplyr::mutate(
air_epa = NA_real_,
yac_epa = NA_real_,
comp_air_epa = NA_real_,
comp_yac_epa = NA_real_,
home_team_comp_air_epa = NA_real_,
away_team_comp_air_epa = NA_real_,
home_team_comp_yac_epa = NA_real_,
away_team_comp_yac_epa = NA_real_,
total_home_comp_air_epa = NA_real_,
total_away_comp_air_epa = NA_real_,
total_home_comp_yac_epa = NA_real_,
total_away_comp_yac_epa = NA_real_,
home_team_raw_air_epa = NA_real_,
away_team_raw_air_epa = NA_real_,
home_team_raw_yac_epa = NA_real_,
away_team_raw_yac_epa = NA_real_,
total_home_raw_air_epa = NA_real_,
total_away_raw_air_epa = NA_real_,
total_home_raw_yac_epa = NA_real_,
total_away_raw_yac_epa = NA_real_
)
user_message("No non-NA air_yards detected. air_yac_ep variables set to NA", "info")
} else {
out <- pbp %>% add_air_yac_ep_variables()
user_message("added air_yac_ep variables", "done")
}
return(out)
}
add_wp <- function(pbp) {
out <- pbp %>% add_wp_variables()
user_message("added wp variables", "done")
return(out)
}
add_air_yac_wp <- function(pbp) {
if (nrow(pbp %>% dplyr::filter(!is.na(.data$air_yards))) == 0) {
out <- pbp %>%
dplyr::mutate(
air_wpa = NA_real_,
yac_wpa = NA_real_,
comp_air_wpa = NA_real_,
comp_yac_wpa = NA_real_,
home_team_comp_air_wpa = NA_real_,
away_team_comp_air_wpa = NA_real_,
home_team_comp_yac_wpa = NA_real_,
away_team_comp_yac_wpa = NA_real_,
total_home_comp_air_wpa = NA_real_,
total_away_comp_air_wpa = NA_real_,
total_home_comp_yac_wpa = NA_real_,
total_away_comp_yac_wpa = NA_real_,
home_team_raw_air_wpa = NA_real_,
away_team_raw_air_wpa = NA_real_,
home_team_raw_yac_wpa = NA_real_,
away_team_raw_yac_wpa = NA_real_,
total_home_raw_air_wpa = NA_real_,
total_away_raw_air_wpa = NA_real_,
total_home_raw_yac_wpa = NA_real_,
total_away_raw_yac_wpa = NA_real_
)
user_message("No non-NA air_yards detected. air_yac_wp variables set to NA", "info")
} else {
out <- pbp %>% add_air_yac_wp_variables()
user_message("added air_yac_wp variables", "done")
}
return(out)
}
#get predictions for a set of pbp data
#for predict stage
get_preds <- function(pbp) {
preds <- as.data.frame(
matrix(stats::predict(fastrmodels::ep_model, as.matrix(pbp %>% ep_model_select())), ncol=7, byrow=TRUE)
)
colnames(preds) <- c("Touchdown","Opp_Touchdown","Field_Goal","Opp_Field_Goal",
"Safety","Opp_Safety","No_Score")
return(preds)
}
#get predictions for a set of pbp data
#for predict stage
get_preds_wp <- function(pbp) {
preds <- stats::predict(fastrmodels::wp_model, as.matrix(pbp %>% wp_model_select()))
return(preds)
}
#get predictions for a set of pbp data
#for predict stage
get_preds_wp_spread <- function(pbp) {
preds <- stats::predict(fastrmodels::wp_model_spread, as.matrix(pbp %>% wp_spread_model_select()))
return(preds)
}
#get the columns needed for ep predictions
#making sure they're in the right order
ep_model_select <- function(pbp) {
pbp <- pbp %>%
dplyr::select(
"half_seconds_remaining",
"yardline_100",
"home",
"retractable",
"dome",
"outdoors",
"ydstogo",
"era0", "era1", "era2", "era3", "era4",
"down1", "down2", "down3", "down4",
"posteam_timeouts_remaining",
"defteam_timeouts_remaining",
)
return(pbp)
}
#get the columns needed for wp predictions
#making sure they're in the right order
wp_model_select <- function(pbp) {
pbp <- pbp %>%
dplyr::select(
"receive_2h_ko",
"home",
"half_seconds_remaining",
"game_seconds_remaining",
"Diff_Time_Ratio",
"score_differential",
"down",
"ydstogo",
"yardline_100",
"posteam_timeouts_remaining",
"defteam_timeouts_remaining"
)
return(pbp)
}
#get the columns needed for wp predictions
#making sure they're in the right order
wp_spread_model_select <- function(pbp) {
pbp <- pbp %>%
dplyr::select(
"receive_2h_ko",
"spread_time",
"home",
"half_seconds_remaining",
"game_seconds_remaining",
"Diff_Time_Ratio",
"score_differential",
"down",
"ydstogo",
"yardline_100",
"posteam_timeouts_remaining",
"defteam_timeouts_remaining"
)
return(pbp)
}
prepare_wp_data <- function(pbp) {
pbp <- pbp %>%
dplyr::group_by(.data$game_id) %>%
dplyr::mutate(
receive_2h_ko = dplyr::if_else(.data$qtr <= 2 & .data$posteam == dplyr::first(stats::na.omit(.data$defteam)), 1, 0)
) %>%
dplyr::ungroup() %>%
dplyr::mutate(
posteam_spread = dplyr::if_else(.data$home == 1, .data$spread_line, -1 * .data$spread_line),
elapsed_share = (3600 - .data$game_seconds_remaining) / 3600,
spread_time = .data$posteam_spread * exp(-4 * .data$elapsed_share),
Diff_Time_Ratio = .data$score_differential / (exp(-4 * .data$elapsed_share))
)
return(pbp)
}
#add ep variables
#All of these are heavily borrowed from nflscrapR (Maksim Horowitz, Ronald Yurko, and Samuel Ventura)
add_ep_variables <- function(pbp_data) {
#testing
#pbp_data <- g
#this function is below
base_ep_preds <- get_preds(pbp_data)
# ----------------------------------------------------------------------------
# ---- special case: deal with FG attempts
# Now make another dataset that to get the EP probabilities from a missed FG:
missed_fg_data <- pbp_data
# Subtract 5.065401 from TimeSecs:
missed_fg_data$half_seconds_remaining <- missed_fg_data$half_seconds_remaining - 5.065401
# Correct the yrdline100:
missed_fg_data$yardline_100 <- 100 - (missed_fg_data$yardline_100 + 8)
# Now first down:
missed_fg_data$down1 <- rep(1,nrow(pbp_data))
missed_fg_data$down2 <- rep(0,nrow(pbp_data))
missed_fg_data$down3 <- rep(0,nrow(pbp_data))
missed_fg_data$down4 <- rep(0,nrow(pbp_data))
# 10 ydstogo:
missed_fg_data$ydstogo <- rep(10,nrow(pbp_data))
# Get the new predicted probabilites:
if (nrow(missed_fg_data) > 1) {
missed_fg_ep_preds <- get_preds(missed_fg_data)
} else{
missed_fg_ep_preds <- get_preds(missed_fg_data)
}
# Find the rows where TimeSecs_Remaining became 0 or negative and make all the probs equal to 0:
end_game_i <- which(missed_fg_data$half_seconds_remaining <= 0)
missed_fg_ep_preds[end_game_i,] <- rep(0,ncol(missed_fg_ep_preds))
# if the half ends, no one scored
missed_fg_ep_preds[end_game_i, "No_Score"] <- 1
# Get the probability of making the field goal:
make_fg_prob <- as.numeric(mgcv::predict.bam(fastrmodels::fg_model, newdata = pbp_data, type="response"))
# Multiply each value of the missed_fg_ep_preds by the 1 - make_fg_prob
missed_fg_ep_preds <- missed_fg_ep_preds * (1 - make_fg_prob)
# Find the FG attempts:
fg_attempt_i <- which(pbp_data$play_type == "field_goal")
# Now update the probabilities for the FG attempts (also includes Opp_Field_Goal probability from missed_fg_ep_preds)
base_ep_preds[fg_attempt_i, "Field_Goal"] <- make_fg_prob[fg_attempt_i] + missed_fg_ep_preds[fg_attempt_i,"Opp_Field_Goal"]
# Update the other columns based on the opposite possession:
base_ep_preds[fg_attempt_i, "Touchdown"] <- missed_fg_ep_preds[fg_attempt_i,"Opp_Touchdown"]
base_ep_preds[fg_attempt_i, "Opp_Field_Goal"] <- missed_fg_ep_preds[fg_attempt_i,"Field_Goal"]
base_ep_preds[fg_attempt_i, "Opp_Touchdown"] <- missed_fg_ep_preds[fg_attempt_i,"Touchdown"]
base_ep_preds[fg_attempt_i, "Safety"] <- missed_fg_ep_preds[fg_attempt_i,"Opp_Safety"]
base_ep_preds[fg_attempt_i, "Opp_Safety"] <- missed_fg_ep_preds[fg_attempt_i,"Safety"]
base_ep_preds[fg_attempt_i, "No_Score"] <- missed_fg_ep_preds[fg_attempt_i,"No_Score"]
# ----------------------------------------------------------------------------------
# ---- special case: deal with kickoffs
# Calculate the EP for receiving a touchback (from the point of view for recieving team)
# and update the columns for Kickoff plays:
kickoff_data <- pbp_data
# Change the yard line to be 80 for 2009-2015 and 75 otherwise
# (accounting for the fact that Jan 2016 is in the 2015 season:
kickoff_data$yardline_100 <- with(kickoff_data,
ifelse(season < 2016,
80, 75))
# Now first down:
kickoff_data$down1 <- rep(1,nrow(pbp_data))
kickoff_data$down2 <- rep(0,nrow(pbp_data))
kickoff_data$down3 <- rep(0,nrow(pbp_data))
kickoff_data$down4 <- rep(0,nrow(pbp_data))
# 10 ydstogo:
kickoff_data$ydstogo <- rep(10,nrow(pbp_data))
# Get the new predicted probabilites:
kickoff_preds <- get_preds(kickoff_data)
# Find the kickoffs:
kickoff_i <- which(pbp_data$play_type == "kickoff" | pbp_data$kickoff_attempt == 1)
# Now update the probabilities:
base_ep_preds[kickoff_i, "Field_Goal"] <- kickoff_preds[kickoff_i, "Field_Goal"]
base_ep_preds[kickoff_i, "Touchdown"] <- kickoff_preds[kickoff_i, "Touchdown"]
base_ep_preds[kickoff_i, "Opp_Field_Goal"] <- kickoff_preds[kickoff_i, "Opp_Field_Goal"]
base_ep_preds[kickoff_i, "Opp_Touchdown"] <- kickoff_preds[kickoff_i, "Opp_Touchdown"]
base_ep_preds[kickoff_i, "Safety"] <- kickoff_preds[kickoff_i, "Safety"]
base_ep_preds[kickoff_i, "Opp_Safety"] <- kickoff_preds[kickoff_i, "Opp_Safety"]
base_ep_preds[kickoff_i, "No_Score"] <- kickoff_preds[kickoff_i, "No_Score"]
# ----------------------------------------------------------------------------------
# Insert probabilities of 0 for everything but No_Score for QB Kneels that
# occur on the possession team's side of the field:
# Find these QB Kneels:
qb_kneels_i <- which(pbp_data$play_type == "qb_kneel" & pbp_data$yardline_100 > 50)
# Now update the probabilities:
base_ep_preds[qb_kneels_i, "Field_Goal"] <- 0
base_ep_preds[qb_kneels_i, "Touchdown"] <- 0
base_ep_preds[qb_kneels_i, "Opp_Field_Goal"] <- 0
base_ep_preds[qb_kneels_i, "Opp_Touchdown"] <- 0
base_ep_preds[qb_kneels_i, "Safety"] <- 0
base_ep_preds[qb_kneels_i, "Opp_Safety"] <- 0
base_ep_preds[qb_kneels_i, "No_Score"] <- 1
# ----------------------------------------------------------------------------------
# Create two new columns, ExPoint_Prob and TwoPoint_Prob, for the PAT events:
base_ep_preds$ExPoint_Prob <- 0
base_ep_preds$TwoPoint_Prob <- 0
# Find the indices for these types of plays:
extrapoint_i <- which((pbp_data$play_type == "extra_point" | pbp_data$play_type_nfl == "XP_KICK") &
(is.na(pbp_data$play_type_nfl) | pbp_data$play_type_nfl != "PAT2"))
twopoint_i <- which(pbp_data$two_point_attempt == 1)
#new: special case for PAT or kickoff with penalty
#for inserting NAs
st_penalty_i_1 <- which(
# pat: prior play was TD or PAT and next play is PAT and this play isn't a td
(pbp_data$touchdown == 0 & (dplyr::lag(pbp_data$touchdown == 1) | dplyr::lag(pbp_data$play_type_nfl == "XP_KICK")) &
(dplyr::lead(pbp_data$two_point_attempt)==1 | dplyr::lead(pbp_data$extra_point_attempt)==1 | dplyr::lead(pbp_data$play_type_nfl) == "XP_KICK")) |
#kickoff: prior play was PAT and next play is kickoff
((dplyr::lag(pbp_data$two_point_attempt)==1 | dplyr::lag(pbp_data$extra_point_attempt)==1) & dplyr::lead(pbp_data$kickoff_attempt == 1))
)
st_penalty_i_2 <- which(
is.na(dplyr::lead(pbp_data$down)) &
# has a key term in desc
(((stringr::str_detect(pbp_data$desc, 'Kick formation') & is.na(pbp_data$down) & pbp_data$play_type == 'no_play') |
(stringr::str_detect(pbp_data$desc, 'Pass formation') & is.na(pbp_data$down) & pbp_data$play_type == 'no_play') |
(stringr::str_detect(pbp_data$desc, 'kicks onside') & is.na(pbp_data$down) & pbp_data$play_type == 'no_play') |
(stringr::str_detect(pbp_data$desc, 'Offside on Free Kick') & is.na(pbp_data$down) & pbp_data$play_type == 'no_play') |
(stringr::str_detect(pbp_data$desc, 'TWO-POINT CONVERSION')) &
# down is NA and play type no play and next play isn't a kickoff
is.na(pbp_data$down) & pbp_data$play_type == 'no_play' & dplyr::lead(pbp_data$kickoff_attempt) == 0))
)
# Assign the make_fg_probs of the extra-point PATs:
base_ep_preds$ExPoint_Prob[extrapoint_i] <- make_fg_prob[extrapoint_i]
# Assign the TwoPoint_Prob with the historical success rate:
base_ep_preds$TwoPoint_Prob[twopoint_i] <- 0.4735
# ----------------------------------------------------------------------------------
# Insert NAs for timeouts and end of play rows:
missing_i <- which(
(pbp_data$timeout == 1 &
pbp_data$play_type == "no_play" &
!stringr::str_detect(pbp_data$desc, ' pass ') &
!stringr::str_detect(pbp_data$desc, ' sacked ') &
!stringr::str_detect(pbp_data$desc, ' scramble ') &
!stringr::str_detect(pbp_data$desc, ' punts ') &
!stringr::str_detect(pbp_data$desc, ' up the middle ') &
!stringr::str_detect(pbp_data$desc, ' left end ') &
!stringr::str_detect(pbp_data$desc, ' left guard ') &
!stringr::str_detect(pbp_data$desc, ' left tackle ') &
!stringr::str_detect(pbp_data$desc, ' right end ') &
!stringr::str_detect(pbp_data$desc, ' right guard ') &
!stringr::str_detect(pbp_data$desc, ' right tackle ')
) |
is.na(pbp_data$play_type))
# Now update the probabilities for missing and PATs:
base_ep_preds$Field_Goal[c(missing_i, extrapoint_i, twopoint_i, st_penalty_i_1, st_penalty_i_2)] <- 0
base_ep_preds$Touchdown[c(missing_i, extrapoint_i, twopoint_i, st_penalty_i_1, st_penalty_i_2)] <- 0
base_ep_preds$Opp_Field_Goal[c(missing_i, extrapoint_i, twopoint_i, st_penalty_i_1, st_penalty_i_2)] <- 0
base_ep_preds$Opp_Touchdown[c(missing_i, extrapoint_i, twopoint_i, st_penalty_i_1, st_penalty_i_2)] <- 0
base_ep_preds$Safety[c(missing_i, extrapoint_i, twopoint_i, st_penalty_i_1, st_penalty_i_2)] <- 0
base_ep_preds$Opp_Safety[c(missing_i, extrapoint_i, twopoint_i, st_penalty_i_1, st_penalty_i_2)] <- 0
base_ep_preds$No_Score[c(missing_i, extrapoint_i, twopoint_i, st_penalty_i_1, st_penalty_i_2)] <- 0
# Rename the events to all have _Prob at the end of them:
base_ep_preds <- dplyr::rename(base_ep_preds,
Field_Goal_Prob = "Field_Goal",
Touchdown_Prob = "Touchdown",
Opp_Field_Goal_Prob = "Opp_Field_Goal",
Opp_Touchdown_Prob = "Opp_Touchdown",
Safety_Prob = "Safety",
Opp_Safety_Prob = "Opp_Safety",
No_Score_Prob = "No_Score")
# Join them together:
pbp_data <- cbind(pbp_data, base_ep_preds)
# Calculate the ExpPts:
pbp_data_ep <- dplyr::mutate(pbp_data,
ExpPts = (0*.data$No_Score_Prob) + (-3 * .data$Opp_Field_Goal_Prob) +
(-2 * .data$Opp_Safety_Prob) +
(-7 * .data$Opp_Touchdown_Prob) + (3 * .data$Field_Goal_Prob) +
(2 * .data$Safety_Prob) + (7 * .data$Touchdown_Prob) +
(1 * .data$ExPoint_Prob) + (2 * .data$TwoPoint_Prob))
#just going to set these to NA bc we have no way of calculating EPA for them
if (length(st_penalty_i_1) > 0) {
pbp_data_ep$ExpPts[st_penalty_i_1] <- NA_real_
}
if (length(st_penalty_i_2) > 0) {
pbp_data_ep$ExpPts[st_penalty_i_2] <- NA_real_
}
pbp_data_ep$ExpPts[missing_i] <- NA_real_
#################################################################
# Calculate EPA:
### Adding Expected Points Added (EPA) column
# Create multiple types of EPA columns
# for each of the possible cases,
# grouping by GameID (will then just use
# an ifelse statement to decide which one
# to use as the final EPA):
pbp_data_ep %>%
dplyr::group_by(.data$game_id) %>%
dplyr::mutate(# Now conditionally assign the EPA, first for possession team
# touchdowns:
ep = .data$ExpPts,
tmp_posteam = .data$posteam
) %>%
tidyr::fill(
.data$ep, .direction = "up"
) %>%
tidyr::fill(
.data$tmp_posteam, .direction = "up"
) %>%
dplyr::mutate(
# get epa for non-scoring plays
home_ep = dplyr::if_else(.data$tmp_posteam == .data$home_team, .data$ep, - .data$ep),
home_epa = dplyr::lead(.data$home_ep) - .data$home_ep,
epa = dplyr::if_else(.data$tmp_posteam == .data$home_team, .data$home_epa, -.data$home_epa),
# td
epa = dplyr::if_else(!is.na(.data$td_team),
dplyr::if_else(.data$td_team == .data$posteam,
7 - .data$ep, -7 - .data$ep),
.data$epa),
# Offense field goal:
epa = dplyr::if_else(is.na(.data$td_team) & .data$field_goal_made == 1,
3 - .data$ep, .data$epa, missing = .data$epa),
# Offense extra-point:
epa = dplyr::if_else(is.na(.data$td_team) & .data$field_goal_made == 0 &
.data$extra_point_good == 1,
1 - .data$ep, .data$epa, missing = .data$epa),
# Offense two-point conversion:
epa = dplyr::if_else(is.na(.data$td_team) & .data$field_goal_made == 0 &
.data$extra_point_good == 0 &
(.data$two_point_rush_good == 1 |
.data$two_point_pass_good == 1 |
.data$two_point_pass_reception_good == 1),
2 - .data$ep, .data$epa, missing = .data$epa),
# Failed PAT (both 1 and 2):
epa = dplyr::if_else(is.na(.data$td_team) & .data$field_goal_made == 0 &
.data$extra_point_good == 0 &
((.data$extra_point_failed == 1 |
.data$extra_point_blocked == 1 |
.data$extra_point_aborted == 1) |
(.data$two_point_rush_failed == 1 |
.data$two_point_pass_failed == 1 |
.data$two_point_pass_reception_failed == 1)),
0 - .data$ep, .data$epa, missing = .data$epa),
# Opponent scores defensive 2 point:
epa = dplyr::if_else(
.data$defensive_two_point_conv == 1, -2 - .data$ep, .data$epa, missing = .data$epa
),
# Safety:
epa = dplyr::case_when(
!is.na(.data$safety_team) & .data$safety_team == .data$posteam ~ 2 - .data$ep,
!is.na(.data$safety_team) & .data$safety_team == .data$defteam ~ -2 - .data$ep,
TRUE ~ .data$epa
)
) %>%
# Now rename each of the expected points columns to match the style of
# the updated code:
dplyr::rename(
no_score_prob = "No_Score_Prob",
opp_fg_prob = "Opp_Field_Goal_Prob",
opp_safety_prob = "Opp_Safety_Prob",
opp_td_prob = "Opp_Touchdown_Prob",
fg_prob = "Field_Goal_Prob",
safety_prob = "Safety_Prob",
td_prob = "Touchdown_Prob",
extra_point_prob = "ExPoint_Prob",
two_point_conversion_prob = "TwoPoint_Prob"
) %>%
# Create columns with cumulative epa totals for both teams:
dplyr::mutate(
# Change epa for plays occurring at end of half with no scoring
# plays to be just the difference between 0 and starting ep:
epa = dplyr::if_else(((.data$qtr == 2 &
(dplyr::lead(.data$qtr) == 3 |
dplyr::lead(.data$desc) == "END QUARTER 2")) |
(.data$qtr == 4 &
(dplyr::lead(.data$qtr) == 5 |
dplyr::lead(.data$desc) == "END QUARTER 4" |
dplyr::lead(.data$desc) == "END GAME"))) &
.data$sp == 0 &
!is.na(.data$play_type),
0 - .data$ep, .data$epa),
# last play of OT
epa = dplyr::if_else(.data$qtr > 4 & dplyr::lead(.data$desc) == "END GAME" & .data$sp == 0,
0 - .data$ep,
.data$epa),
epa = dplyr::if_else(.data$desc == "END QUARTER 2", NA_real_, .data$epa),
epa = dplyr::if_else(.data$desc == "GAME", NA_real_, .data$epa),
ep = dplyr::if_else(.data$desc == "END QUARTER 2", NA_real_, .data$ep),
ep = dplyr::if_else(.data$desc == "GAME", NA_real_, .data$ep),
home_team_epa = dplyr::if_else(.data$posteam == .data$home_team,
.data$epa, -.data$epa),
away_team_epa = dplyr::if_else(.data$posteam == .data$away_team,
.data$epa, -.data$epa),
home_team_epa = dplyr::if_else(is.na(.data$home_team_epa),
0, .data$home_team_epa),
away_team_epa = dplyr::if_else(is.na(.data$away_team_epa),
0, .data$away_team_epa),
total_home_epa = cumsum(.data$home_team_epa),
total_away_epa = cumsum(.data$away_team_epa),
# Same thing but separating passing and rushing:
home_team_rush_epa = dplyr::if_else(.data$play_type == "run",
.data$home_team_epa, 0),
away_team_rush_epa = dplyr::if_else(.data$play_type == "run",
.data$away_team_epa, 0),
home_team_rush_epa = dplyr::if_else(is.na(.data$home_team_rush_epa),
0, .data$home_team_rush_epa),
away_team_rush_epa = dplyr::if_else(is.na(.data$away_team_rush_epa),
0, .data$away_team_rush_epa),
total_home_rush_epa = cumsum(.data$home_team_rush_epa),
total_away_rush_epa = cumsum(.data$away_team_rush_epa),
home_team_pass_epa = dplyr::if_else(.data$play_type == "pass",
.data$home_team_epa, 0),
away_team_pass_epa = dplyr::if_else(.data$play_type == "pass",
.data$away_team_epa, 0),
home_team_pass_epa = dplyr::if_else(is.na(.data$home_team_pass_epa),
0, .data$home_team_pass_epa),
away_team_pass_epa = dplyr::if_else(is.na(.data$away_team_pass_epa),
0, .data$away_team_pass_epa),
total_home_pass_epa = cumsum(.data$home_team_pass_epa),
total_away_pass_epa = cumsum(.data$away_team_pass_epa)) %>%
dplyr::ungroup() %>%
return()
}
#################################################################
# Calculate WP and WPA:
add_wp_variables <- function(pbp_data) {
#testing only
# pbp_data <- g
# Initialize the df to store predicted win probability
OffWinProb <- rep(NA_real_, nrow(pbp_data))
OffWinProb_spread <- rep(NA_real_, nrow(pbp_data))
pbp_data <- pbp_data %>%
prepare_wp_data()
# First check if there's any overtime plays:
if (any(pbp_data$qtr > 4)){
# Find the rows that are overtime:
overtime_i <- which(pbp_data$qtr > 4)
# Separate the dataset into regular_df and overtime_df:
overtime_df <- pbp_data[overtime_i,]
# Separate routine for overtime:
# Create a column that is just the first drive of overtime repeated:
overtime_df$First_Drive <- rep(min(overtime_df$drive,
na.rm = TRUE),
nrow(overtime_df))
# Calculate the difference in drive number
overtime_df <- dplyr::mutate(overtime_df,
Drive_Diff = .data$drive - .data$First_Drive)
# Create an indicator column that means the posteam is losing by 3 and
# its the second drive of overtime:
overtime_df$One_FG_Game <- ifelse(overtime_df$score_differential == -3 &
overtime_df$Drive_Diff == 1, 1, 0)
# Now create a copy of the dataset to then make the EP predictions for when
# a field goal is scored and its not sudden death:
overtime_df_ko <- overtime_df
overtime_df_ko$yrdline100 <- with(overtime_df_ko,
ifelse(game_year < 2016 |
(game_year == 2016 & game_month < 4),
80, 75))
# Now first down:
overtime_df_ko$down1 <- rep(1,nrow(overtime_df_ko))
overtime_df_ko$down2 <- rep(0,nrow(overtime_df_ko))
overtime_df_ko$down3 <- rep(0,nrow(overtime_df_ko))
overtime_df_ko$down4 <- rep(0,nrow(overtime_df_ko))
# 10 ydstogo:
overtime_df_ko$ydstogo <- rep(10,nrow(overtime_df_ko))
# Get the predictions from the EP model and calculate the necessary probability:
overtime_df_ko_preds <- get_preds(overtime_df_ko)
overtime_df_ko_preds <- dplyr::mutate(overtime_df_ko_preds,
Win_Back = .data$No_Score + .data$Opp_Field_Goal + .data$Opp_Safety + .data$Opp_Touchdown)
# Calculate the two possible win probability types, Sudden Death and one Field Goal:
overtime_df$Sudden_Death_WP <- overtime_df$fg_prob + overtime_df$td_prob + overtime_df$safety_prob
overtime_df$One_FG_WP <- overtime_df$td_prob + (overtime_df$fg_prob * overtime_df_ko_preds$Win_Back)
# Decide which win probability to use:
OffWinProb[overtime_i] <- ifelse(overtime_df$game_year >= 2012 & (overtime_df$Drive_Diff == 0 | (overtime_df$Drive_Diff == 1 & overtime_df$One_FG_Game == 1)),
overtime_df$One_FG_WP, overtime_df$Sudden_Death_WP)
OffWinProb_spread[overtime_i] <- OffWinProb[overtime_i]
}
#regulation plays
regular_i <- which(pbp_data$qtr <= 4)
# df of just the regulation plays:
regular_df <- pbp_data[regular_i,]
# do predictions for the regular df
OffWinProb[regular_i] <- get_preds_wp(regular_df)
OffWinProb_spread[regular_i] <- get_preds_wp_spread(regular_df)
## set to NA WP for plays down is missing
# for kickoffs and PATs, these will get overwritten by the fixes after this
down_na <- which(is.na(pbp_data$down))
OffWinProb[down_na] <- NA_real_
OffWinProb_spread[down_na] <- NA_real_
## start PAT fix
make_pat_prob <- as.numeric(
mgcv::predict.bam(
fastrmodels::fg_model,
newdata = pbp_data %>%
mutate(
yardline_100 = ifelse(.data$season >= 2015, 15, 3)
), type="response")
)
# plays with 1 point PAT attempts
pat_i <- which(
(pbp_data$kickoff_attempt == 0 &
!(stringr::str_detect(pbp_data$desc, 'Onside Kick')) &
(stringr::str_detect(pbp_data$desc, 'Kick formation')) &
is.na(pbp_data$down)) |
# or has PAT indicators
stringr::str_detect(pbp_data$desc, 'extra point') |
!is.na(pbp_data$extra_point_result)
)
# plays with 2 point PAT attempts
two_pt_i <- which(
(pbp_data$kickoff_attempt == 0 &
!(stringr::str_detect(pbp_data$desc, 'Onside Kick')) &
(stringr::str_detect(pbp_data$desc, 'Pass formation')) &
is.na(pbp_data$down)) |
# or has PAT indicators
stringr::str_detect(pbp_data$desc, 'TWO-POINT CONVERSION ATTEMPT') |
!is.na(pbp_data$two_point_conv_result)
)
# some rare 2 point PAT attempts have duplicated matches in 1 point PAT attempts
# so we remove them in the next line
pat_i <- pat_i[!pat_i %in% two_pt_i]
# make df of post-PAT plays
pat_data <- pbp_data %>%
dplyr::mutate(
# swap timeouts
to_pos = .data$posteam_timeouts_remaining,
to_def = .data$defteam_timeouts_remaining,
posteam_timeouts_remaining = .data$to_def,
defteam_timeouts_remaining = .data$to_pos,
# swap score
score_differential = -.data$score_differential,
# 1st and 10
down = 1,
ydstogo = 10,
# flip receive_2h_ko var
receive_2h_ko = case_when(
.data$qtr <= 2 & .data$receive_2h_ko == 0 ~ 1,
.data$qtr <= 2 & .data$receive_2h_ko == 1 ~ 0,
TRUE ~ .data$receive_2h_ko
),
# switch posteam
posteam = if_else(.data$home_team == .data$posteam, .data$away_team, .data$home_team),
yardline_100 = 75
) %>%
dplyr::mutate(
home = case_when(
.data$home == 0 ~ 1,
.data$home == 1 ~ 0
),
posteam_spread = dplyr::if_else(.data$home == 1, .data$spread_line, -1 * .data$spread_line),
elapsed_share = (3600 - .data$game_seconds_remaining) / 3600,
spread_time = .data$posteam_spread * exp(-4 * .data$elapsed_share)
)
## start with spread version
# get pat if 0, 1, or 2
pat_0 <- get_preds_wp_spread(pat_data %>% add_esdtr())
pat_1 <- get_preds_wp_spread(pat_data %>% dplyr::mutate(score_differential = .data$score_differential - 1) %>% add_esdtr())
pat_2 <- get_preds_wp_spread(pat_data %>% dplyr::mutate(score_differential = .data$score_differential - 2) %>% add_esdtr())
# Using nflscrapR version of 2pt make prob on 2nd line here
pat_go_for_1 <- 1 - (make_pat_prob * pat_1 + (1 - make_pat_prob) * pat_0)
pat_go_for_2 <- 1 - (0.4735 * pat_2 + (1 - 0.4735) * pat_0)
OffWinProb_spread[two_pt_i] <- pat_go_for_2[two_pt_i]
OffWinProb_spread[pat_i] <- pat_go_for_1[pat_i]
## repeat for non-spread version
# get pat if 0, 1, or 2
pat_0 <- get_preds_wp(pat_data %>% add_esdtr())
pat_1 <- get_preds_wp(pat_data %>% dplyr::mutate(score_differential = .data$score_differential - 1) %>% add_esdtr())
pat_2 <- get_preds_wp(pat_data %>% dplyr::mutate(score_differential = .data$score_differential - 2) %>% add_esdtr())
# Using nflscrapR version of 2pt make prob on 2nd line here
pat_go_for_1 <- 1 - (make_pat_prob * pat_1 + (1 - make_pat_prob) * pat_0)
pat_go_for_2 <- 1 - (0.4735 * pat_2 + (1 - 0.4735) * pat_0)
OffWinProb[two_pt_i] <- pat_go_for_2[two_pt_i]
OffWinProb[pat_i] <- pat_go_for_1[pat_i]
## end PAT fix
## now we need to fix WP on kickoffs, which will be WP associated with touchback
kickoff_data <- pbp_data
# Change the yard line to be 80 for 2009-2015 and 75 otherwise
kickoff_data$yardline_100 <- with(kickoff_data,
ifelse(season < 2016,
80, 75))
# Now first down:
kickoff_data$down <- rep(1,nrow(pbp_data))
kickoff_data$down1 <- rep(1,nrow(pbp_data))
kickoff_data$down2 <- rep(0,nrow(pbp_data))
kickoff_data$down3 <- rep(0,nrow(pbp_data))
kickoff_data$down4 <- rep(0,nrow(pbp_data))
# 10 ydstogo:
kickoff_data$ydstogo <- rep(10,nrow(pbp_data))
# Get the new predicted probabilites:
kickoff_preds <- get_preds_wp(kickoff_data)
kickoff_preds_spread <- get_preds_wp_spread(kickoff_data)
# Find the kickoffs in regulation:
kickoff_i <- which((pbp_data$play_type == "kickoff" | pbp_data$kickoff_attempt == 1) & pbp_data$qtr <= 4)
# Now update the probabilities:
OffWinProb[kickoff_i] <- kickoff_preds[kickoff_i]
OffWinProb_spread[kickoff_i] <- kickoff_preds_spread[kickoff_i]
## end fix for kickoffs
# Now create the win probability columns and return:
pbp_data <- pbp_data %>%
dplyr::mutate(
wp = OffWinProb,
vegas_wp = OffWinProb_spread,
# for figuring out posteam on NA posteam lines
tmp_posteam = .data$posteam
) %>%
tidyr::fill(
.data$wp, .direction = "up"
) %>%
tidyr::fill(
.data$vegas_wp, .direction = "up"
) %>%
tidyr::fill(
.data$tmp_posteam, .direction = "up"
) %>%
dplyr::group_by(.data$game_id) %>%
dplyr::mutate(
#add columns for home WP
home_wp = dplyr::if_else(.data$tmp_posteam == .data$home_team, .data$wp, 1 - .data$wp),
vegas_home_wp = dplyr::if_else(.data$tmp_posteam == .data$home_team, .data$vegas_wp, 1 - .data$vegas_wp),
# convenience to mark end of game
end_game = ifelse(
stringr::str_detect(tolower(.data$desc), "(end of game)|(end game)"),
1, 0
),
# convenience for marking home win prob on last line
final_value = dplyr::case_when(
.data$home_score > .data$away_score ~ 1,
.data$away_score > .data$home_score ~ 0,
.data$home_score == .data$away_score ~ .5
),
#make 1 or 0 the final win prob
vegas_home_wp = dplyr::if_else(
.data$end_game == 1,
.data$final_value,
.data$vegas_home_wp
),
# can we make this and the above into a function? feels like a lot of repitition
home_wp = dplyr::if_else(
.data$end_game == 1,
.data$final_value,
.data$home_wp
),
away_wp = 1 - .data$home_wp,
# make wp of posteam on last line NA because there's no posteam
vegas_wp = dplyr::if_else(
.data$end_game == 1,
NA_real_,
.data$vegas_wp
),
wp = dplyr::if_else(
.data$end_game == 1,
NA_real_,
.data$wp
),
def_wp = 1 - .data$wp,
# make wpa
vegas_home_wpa = dplyr::lead(.data$vegas_home_wp) - .data$vegas_home_wp,
vegas_wpa = dplyr::if_else(.data$tmp_posteam == .data$home_team, .data$vegas_home_wpa, -.data$vegas_home_wpa),
vegas_wpa = dplyr::if_else(
stringr::str_detect(tolower(.data$desc), "( kneels )|(end of game)|(end game)"), NA_real_, .data$vegas_wpa
),
# home wpa isn't saved but needed for next line
home_wpa = dplyr::lead(.data$home_wp) - .data$home_wp,
wpa = dplyr::if_else(.data$tmp_posteam == .data$home_team, .data$home_wpa, -.data$home_wpa),
wpa = dplyr::if_else(
stringr::str_detect(tolower(.data$desc), "( kneels )|(end of game)|(end game)"), NA_real_, .data$wpa
)
) %>%
dplyr::ungroup()
# Home and Away post:
pbp_data$home_wp_post <- ifelse(pbp_data$posteam == pbp_data$home_team,
pbp_data$home_wp + pbp_data$wpa,
pbp_data$home_wp - pbp_data$wpa)
pbp_data$away_wp_post <- ifelse(pbp_data$posteam == pbp_data$away_team,
pbp_data$away_wp + pbp_data$wpa,
pbp_data$away_wp - pbp_data$wpa)
# If next thing is end of game, and post score differential is tied because it's
# overtime then make both the home_wp_post and away_wp_post equal to 0:
pbp_data <- pbp_data %>%
dplyr::mutate(home_wp_post = dplyr::if_else(.data$qtr == 5 &
stringr::str_detect(tolower(dplyr::lead(.data$desc)),
"(end of game)|(end game)") &
.data$score_differential_post == 0,
0, .data$home_wp_post),
away_wp_post = dplyr::if_else(.data$qtr == 5 &
stringr::str_detect(tolower(dplyr::lead(.data$desc)),
"(end of game)|(end game)") &
.data$score_differential_post == 0,
0, .data$away_wp_post))
# For plays with playtype of End of Game, use the previous play's WP_post columns
# as the pre and post, since those are already set to be 1 and 0:
pbp_data$home_wp_post <- with(pbp_data,
ifelse(stringr::str_detect(tolower(desc),
"(end of game)|(end game)"), dplyr::lag(home_wp_post),
ifelse(dplyr::lag(play_type) == "no_play" & play_type == "no_play", dplyr::lag(home_wp_post),home_wp_post)))
pbp_data$away_wp_post <- with(pbp_data,
ifelse(stringr::str_detect(tolower(desc),
"(end of game)|(end game)"), dplyr::lag(away_wp_post),
ifelse(dplyr::lag(play_type) == "no_play" & play_type == "no_play", dplyr::lag(away_wp_post),away_wp_post)))
# Now drop the unnecessary columns, rename variables back, and return:
pbp_data %>%
dplyr::group_by(.data$game_id) %>%
dplyr::mutate(
# Generate columns to keep track of cumulative rushing and
# passing WPA values:
home_team_wpa = dplyr::if_else(.data$posteam == .data$home_team,
.data$wpa, -.data$wpa),
away_team_wpa = dplyr::if_else(.data$posteam == .data$away_team,
.data$wpa, -.data$wpa),
home_team_wpa = dplyr::if_else(is.na(.data$home_team_wpa),
0, .data$home_team_wpa),
away_team_wpa = dplyr::if_else(is.na(.data$away_team_wpa),
0, .data$away_team_wpa),
# Same thing but separating passing and rushing:
home_team_rush_wpa = dplyr::if_else(.data$play_type == "run",
.data$home_team_wpa, 0),
away_team_rush_wpa = dplyr::if_else(.data$play_type == "run",
.data$away_team_wpa, 0),
home_team_rush_wpa = dplyr::if_else(is.na(.data$home_team_rush_wpa),
0, .data$home_team_rush_wpa),
away_team_rush_wpa = dplyr::if_else(is.na(.data$away_team_rush_wpa),
0, .data$away_team_rush_wpa),
total_home_rush_wpa = cumsum(.data$home_team_rush_wpa),
total_away_rush_wpa = cumsum(.data$away_team_rush_wpa),
home_team_pass_wpa = dplyr::if_else(.data$play_type == "pass",
.data$home_team_wpa, 0),
away_team_pass_wpa = dplyr::if_else(.data$play_type == "pass",
.data$away_team_wpa, 0),
home_team_pass_wpa = dplyr::if_else(is.na(.data$home_team_pass_wpa),
0, .data$home_team_pass_wpa),
away_team_pass_wpa = dplyr::if_else(is.na(.data$away_team_pass_wpa),
0, .data$away_team_pass_wpa),
total_home_pass_wpa = cumsum(.data$home_team_pass_wpa),
total_away_pass_wpa = cumsum(.data$away_team_pass_wpa)) %>%
dplyr::ungroup() %>%
return()
}
# helper function to get expected score diff to time ratio
# needed after flipping teams in WP for getting PAT WP
add_esdtr <- function(data) {
data %>%
dplyr::mutate(
Diff_Time_Ratio = .data$score_differential / (exp(-4 * .data$elapsed_share))
) %>%
return()
}
#################################################################
# air and YAC EP:
# as with the rest, heavily borrowed from nflscrapR:
# https://github.com/maksimhorowitz/nflscrapR/blob/master/R/add_ep_wp_variables.R
add_air_yac_ep_variables <- function(pbp_data) {
#testing
#pbp_data <- g
# Final all pass attempts that are not sacks:
pass_plays_i <- which(!is.na(pbp_data$air_yards) & pbp_data$play_type == 'pass')
pass_pbp_data <- pbp_data[pass_plays_i,]
# Using the air_yards need to update the following:
# - yrdline100
# - TimeSecs_Remaining
# - ydstogo
# - down
# - timeouts
# Get everything set up for calculation
pass_pbp_data <- pass_pbp_data %>%
dplyr::mutate(
posteam_timeouts_pre = .data$posteam_timeouts_remaining,
defeam_timeouts_pre = .data$defteam_timeouts_remaining
) %>%
# Rename the old columns to update for calculating the EP from the air:
dplyr::rename(old_yrdline100 = .data$yardline_100,
old_ydstogo = .data$ydstogo,
old_TimeSecs_Remaining = .data$half_seconds_remaining,
old_down = .data$down) %>%
dplyr::mutate(Turnover_Ind = dplyr::if_else(.data$old_down == 4 & .data$air_yards < .data$old_ydstogo,
1, 0),
yardline_100 = dplyr::if_else(.data$Turnover_Ind == 0,
.data$old_yrdline100 - .data$air_yards,
100 - (.data$old_yrdline100 - .data$air_yards)),
ydstogo = dplyr::if_else(.data$air_yards >= .data$old_ydstogo |
.data$Turnover_Ind == 1,
10, .data$old_ydstogo - .data$air_yards),
down = dplyr::if_else(.data$air_yards >= .data$old_ydstogo |
.data$Turnover_Ind == 1,
1, as.numeric(.data$old_down) + 1),
half_seconds_remaining = .data$old_TimeSecs_Remaining - 5.704673,
down1 = dplyr::if_else(.data$down == 1, 1, 0),