Many wp variables with different values when running different number of games

[mrcaseb]

library(tidyverse)
library(arsenal)
library(nflfastR)

ids <- fast_scraper_schedules(2020) %>% head(10) %>% pull(game_id)
pbp <- fast_scraper(ids) %>% decode_player_ids()
#> ✓ Download finished. Adding variables...
#> ✓ added game variables
#> ✓ added nflscrapR variables
#> ✓ added ep variables
#> ✓ added air_yac_ep variables
#> ✓ added wp variables
#> ✓ added air_yac_wp variables
#> ✓ added cp and cpoe
#> ✓ added fixed drive variables
#> ✓ added series variables
#> ✓ Procedure completed.
#> ✓ Decoding of player ids completed
data_repo <- load_pbp(2020) %>% 
  filter(game_id %in% ids) %>% 
  select_at(vars(names(pbp)))

s <- summary(comparedf(data_repo, pbp))
s$diffs.byvar.table %>% filter(n>0)
#>                      var.x                   var.y    n NAs
#> 1                       wp                      wp   53   0
#> 2                   def_wp                  def_wp   53   0
#> 3                  home_wp                 home_wp   53   0
#> 4                  away_wp                 away_wp   53   0
#> 5                      wpa                     wpa  104   0
#> 6                vegas_wpa               vegas_wpa  108   1
#> 7           vegas_home_wpa          vegas_home_wpa  108   1
#> 8             home_wp_post            home_wp_post   53   0
#> 9             away_wp_post            away_wp_post   53   0
#> 10                vegas_wp                vegas_wp   57   1
#> 11           vegas_home_wp           vegas_home_wp   53   0
#> 12     total_home_rush_wpa     total_home_rush_wpa 1330   0
#> 13     total_away_rush_wpa     total_away_rush_wpa 1330   0
#> 14     total_home_pass_wpa     total_home_pass_wpa 1199   0
#> 15     total_away_pass_wpa     total_away_pass_wpa 1199   0
#> 16                 air_wpa                 air_wpa   18   0
#> 17                 yac_wpa                 yac_wpa   12   0
#> 18            comp_air_wpa            comp_air_wpa   18   0
#> 19            comp_yac_wpa            comp_yac_wpa   12   0
#> 20 total_home_comp_air_wpa total_home_comp_air_wpa 1103   0
#> 21 total_away_comp_air_wpa total_away_comp_air_wpa 1103   0
#> 22 total_home_comp_yac_wpa total_home_comp_yac_wpa  751   0
#> 23 total_away_comp_yac_wpa total_away_comp_yac_wpa  751   0
#> 24  total_home_raw_air_wpa  total_home_raw_air_wpa 1103   0
#> 25  total_away_raw_air_wpa  total_away_raw_air_wpa 1103   0
#> 26  total_home_raw_yac_wpa  total_home_raw_yac_wpa  751   0
#> 27  total_away_raw_yac_wpa  total_away_raw_yac_wpa  751   0
d <- diffs(s)
d %>% head(30)
#>    var.x var.y ..row.names..     values.x     values.y row.x row.y
#> 1     wp    wp            94 0.792873.... 0.793337....    94    94
#> 2     wp    wp           107 0.910384.... 0.910609....   107   107
#> 3     wp    wp           139 0.097398.... 0.097866....   139   139
#> 4     wp    wp           169 0.213206.... 0.213757....   169   169
#> 5     wp    wp           180 0.718435.... 0.719560....   180   180
#> 6     wp    wp           212 0.752501.... 0.752887....   212   212
#> 7     wp    wp           230 0.285728.... 0.285877....   230   230
#> 8     wp    wp           263 0.847960.... 0.848386....   263   263
#> 9     wp    wp           283 0.929228.... 0.929174....   283   283
#> 10    wp    wp           309 0.992384.... 0.992403....   309   309
#> 11    wp    wp           392 0.682618.... 0.683063....   392   392
#> 12    wp    wp           440 0.730897.... 0.732254....   440   440
#> 13    wp    wp           446 0.866797.... 0.867168....   446   446
#> 14    wp    wp           503 0.963208.... 0.963421....   503   503
#> 15    wp    wp           523 0.997528.... 0.997546....   523   523
#> 16    wp    wp           566 0.613061.... 0.613371....   566   566
#> 17    wp    wp           580 0.556122.... 0.557336....   580   580
#> 18    wp    wp           604 0.763171.... 0.763544....   604   604
#> 19    wp    wp           635 0.934121.... 0.934154....   635   635
#> 20    wp    wp           668 0.997358.... 0.997368....   668   668
#> 21    wp    wp           694 0.004272.... 0.004209....   694   694
#> 22    wp    wp           720 0.612394.... 0.612816....   720   720
#> 23    wp    wp           756 0.556122.... 0.557336....   756   756
#> 24    wp    wp           769 0.642990.... 0.643484....   769   769
#> 25    wp    wp           783 0.557104.... 0.558263....   783   783
#> 26    wp    wp           850 0.727532.... 0.728479....   850   850
#> 27    wp    wp           895 0.529801.... 0.530521....   895   895
#> 28    wp    wp           907 0.654755.... 0.655375....   907   907
#> 29    wp    wp           945 0.598712.... 0.599226....   945   945
#> 30    wp    wp           987 0.850935.... 0.851262....   987   987

^{Created on 2021-02-13 by the reprex package (v1.0.0)}

[mrcaseb]

For comparison now same thing but for the first 30 games instead of the first 10

library(tidyverse)
library(arsenal)
library(nflfastR)

ids <- fast_scraper_schedules(2020) %>% head(30) %>% pull(game_id)
pbp <- fast_scraper(ids) %>% decode_player_ids()
#> ✓ Download finished. Adding variables...
#> ✓ added game variables
#> ✓ added nflscrapR variables
#> ✓ added ep variables
#> ✓ added air_yac_ep variables
#> ✓ added wp variables
#> ✓ added air_yac_wp variables
#> ✓ added cp and cpoe
#> ✓ added fixed drive variables
#> ✓ added series variables
#> ✓ Procedure completed.
#> ✓ Decoding of player ids completed
data_repo <- load_pbp(2020) %>% 
  filter(game_id %in% ids) %>% 
  select_at(vars(names(pbp)))

s <- summary(comparedf(data_repo, pbp))
s$diffs.byvar.table %>% filter(n>0)
#>            var.x          var.y n NAs
#> 1           desc           desc 1   0
#> 2      vegas_wpa      vegas_wpa 2   1
#> 3 vegas_home_wpa vegas_home_wpa 2   1
#> 4       vegas_wp       vegas_wp 2   1
d <- diffs(s)
d %>% head(30)
#>            var.x          var.y ..row.names..     values.x     values.y row.x
#> 1           desc           desc          2563 (15:00) .... (15:00) ....  2563
#> 2      vegas_wpa      vegas_wpa          4955 0.362151.... 0.350065....  4955
#> 3      vegas_wpa      vegas_wpa          5488 0.267549....           NA  5488
#> 4 vegas_home_wpa vegas_home_wpa          4955 -0.36215.... -0.35006....  4955
#> 5 vegas_home_wpa vegas_home_wpa          5488 -0.26754....           NA  5488
#> 6       vegas_wp       vegas_wp          4955 0.637848.... 0.649934....  4955
#> 7       vegas_wp       vegas_wp          5488 0.267549....           NA  5488
#>   row.y
#> 1  2563
#> 2  4955
#> 3  5488
#> 4  4955
#> 5  5488
#> 6  4955
#> 7  5488

^{Created on 2021-02-13 by the reprex package (v1.0.0)}

[guga31bb]

Looked for plays with different wp and LOL I think there's a pattern

[mrcaseb]

Looked for plays with different wp and LOL I think there's a pattern

Haha that's obvious! So if the 12th row is "END GAME" something goes wrong,,,

[guga31bb]

Have to leave computer soon but to fix the END GAME line I think we just need something like this for vegas_wp as well

[guga31bb]

The win prob issue on PAT issue is different and I have no idea what's causing it but it also is a very small difference

[mrcaseb]

Since number of different values gets smaller if the number of games increases I think the PATs somehow interfere with each other

[guga31bb]

I added a group_by to #185 where there should have been one before but there are still differences. But these are at like the 4th decimal point on PATs and I'm not sure if I care to track it down if that's all it is

[mrcaseb]

Dropping some code here in case I want to investigate this at some point. For now we don't care as the wp values differ less than 0.3 percentage points

library(tidyverse)
library(arsenal)
library(nflfastR)
progressr::handlers(global = TRUE)
dat <- load_pbp(2020)
ids <- fast_scraper_schedules(2020) %>% slice(1:10) %>% pull(game_id)
pbp <- fast_scraper(ids) %>% decode_player_ids() %>% filter(desc != "END GAME")
data_repo <- dat %>% 
  filter(game_id %in% ids) %>% 
  select_at(vars(names(pbp))) %>% 
  filter(desc != "END GAME")

s <- summary(comparedf(data_repo, pbp))
# s$diffs.byvar.table %>% filter(n>0)
d <- diffs(s)
# d %>% head(30)
big <- d %>% 
  mutate(
    data_repo = furrr::future_map_dbl(values.x, ~ ifelse(is.numeric(.x), .x, 0)),
    small_sample = furrr::future_map_dbl(values.y, ~ ifelse(is.numeric(.x), .x, 0)),
    diff_abs = abs(data_repo - small_sample),
    diff = diff_abs %>% scales::percent(accuracy = 0.01)
  ) %>% 
  # filter(diff_abs > 0.1/100) %>%
  select(var = var.x, row = ..row.names.., data_repo:diff) %>% 
  filter(var != "desc") %>% 
  arrange(desc(diff_abs))

big 

pbp %>% slice(big$row) %>% select(game_id, play_id, desc)
# pbp %>% slice(499) %>% select(game_id, play_id, desc, penalty) %>% view()

unique(pbp$game_id[big$row])

[mrcaseb]

To do

• make all wpa variables NA for the END GAME line
• make all wp variable 0, 0.5 or 1 for the END GAME line