1 of 43

Putting the Fun in Functional Data:

A tidy pipeline to identify routes in NFL tracking data

Dani Chu (@chuurveg) Quantitative Analyst - Statistics, NHL Seattle

1

2 of 43

Collaborators

Lucas Wu

PhD Student

Simon Fraser University

yifan_wu@sfu.ca

@lucaswu123

James Thomson

MSc Student

Simon Fraser University

james_thomson_2@sfu.ca

@stat_jamest

Dani Chu

Quantitative Analyst - Statistics

NHL Seattle

danic.stats@gmail.com

@chuurveg

Matthew Reyers

MSc Student

Simon Fraser University

matthew_reyers@sfu.ca

@Stats_By_Matt

2

3 of 43

A little about me

3

4 of 43

NFL Big Data Bowl

4

5 of 43

The Data

NFL player tracking data from the first 6 weeks of the 2017 regular season

10 observations per second
(x, y) location of each player and the ball

Game, play, player level information
One example file still available at github.com/nfl-football-ops

5

6 of 43

Example - KC vs TEN, 2020 AFC Championship

6

Credit: NFL Next Gen Stats

7 of 43

Route Identification

7

8 of 43

Receivers with most routes per team

8

9 of 43

Route Clustering

9

10 of 43

Cluster 20 - Post

10

11 of 43

Route Development

11

12 of 43

What I hope you learn

12

Tidy data when its functional data!
purrr::map()
purrr::safely()

13 of 43

Our Data

13

14 of 43

The Tracking Data

1 tracking data file per game (91)
Linked to other files by

game_id
play_id
nfl_id (player)

14

15 of 43

Read in our Data

15

tracking_data

# A tibble: 316,025 x 14

time x y s dis dir event nflId displayName jerseyNumber team frame.id gameId playId

1 2017-09-08 00:41:59 41.6 16.5 3.91 0.41 78.9 NA 2495340 Anthony Sherman 42 away 1 2017090700 44

2 2017-09-08 00:41:59 42.0 16.6 4.28 0.4 79.2 NA 2495340 Anthony Sherman 42 away 2 2017090700 44

3 2017-09-08 00:41:59 42.4 16.7 4.66 0.47 79.5 NA 2495340 Anthony Sherman 42 away 3 2017090700 44

4 2017-09-08 00:41:59 42.8 16.8 5.04 0.46 79.8 NA 2495340 Anthony Sherman 42 away 4 2017090700 44

5 2017-09-08 00:41:59 43.4 16.9 5.39 0.51 80.1 kickoff 2495340 Anthony Sherman 42 away 5 2017090700 44

6 2017-09-08 00:41:59 43.9 17.0 5.6 0.52 80.6 NA 2495340 Anthony Sherman 42 away 6 2017090700 44

7 2017-09-08 00:41:59 44.4 17.1 5.78 0.580 81.0 NA 2495340 Anthony Sherman 42 away 7 2017090700 44

8 2017-09-08 00:41:59 45.1 17.2 5.97 0.63 81.1 NA 2495340 Anthony Sherman 42 away 8 2017090700 44

9 2017-09-08 00:41:59 45.7 17.3 6.18 0.64 81.2 NA 2495340 Anthony Sherman 42 away 9 2017090700 44

# ... with 316,016 more rows

tracking_data <- readr::read_csv(file_name, col_types = cols())

16 of 43

Cut! Cut! Cut!

16

17 of 43

Columns First!

17

> # A tibble: 316,025 x 8

nflId gameId playId x y frame.id team event

1 2495340 2017090700 44 41.6 16.5 1 away NA

2 2495340 2017090700 44 42.0 16.6 2 away NA

3 2495340 2017090700 44 42.4 16.7 3 away NA

4 2495340 2017090700 44 42.8 16.8 4 away NA

5 2495340 2017090700 44 43.4 16.9 5 away kickoff

6 2495340 2017090700 44 43.9 17.0 6 away NA

7 2495340 2017090700 44 44.4 17.1 7 away NA

# ... with 316,018 more rows

tracking_data %>%

dplyr::select(nflId, gameId, playId, x, y, frame.id, team, event)

18 of 43

And the Rows!

18

# A tibble: 32,282 x 9

nflId gameId playId x y frame.id team event PositionAbbr

1 497240 2017090700 68 83.9 34.8 1 home NA TE

2 497240 2017090700 68 83.9 34.8 2 home NA TE

3 497240 2017090700 68 83.9 34.8 3 home NA TE

# ... with 32,271 more rows

... %>%

# keep only the passing plays

dplyr::inner_join(pass_playIds, by = c("gameId", "playId"))

# keep only the tracking data for route runners

dplyr::inner_join(route_runners_pos_id_key, by = c("nflId"))

19 of 43

Tidy Data

19

20 of 43

Tidy Data

20

Credit: R for Data Science

21 of 43

21

22 of 43

No!

22

23 of 43

Curves are the Observations!

23

24 of 43

Tidy with tidyr

24

# A tibble: 394 x 6

nflId gameId playId team PositionAbbr tracking_data

1 497240 2017090700 68 home TE [87 x 4]

2 2530515 2017090700 68 home WR [87 x 4]

3 2533046 2017090700 68 home TE [87 x 4]

4 2539265 2017090700 68 home RB [87 x 4]

5 2543498 2017090700 68 home WR [87 x 4]

6 2649 2017090700 94 home WR [89 x 4]

7 497240 2017090700 94 home TE [89 x 4]

# ... with 387 more rows

... %>%

# nest the x, y data tidyness

tidyr::nest(tracking_data = c(x, y, frame.id, event))

25 of 43

Now do it for all the files!

25

26 of 43

Map to All Files

26

routes_data <-

# list of all the files

tracking_files %>%

# read in each file and create a data frame of routes

purrr::map_dfr(read_routes_from_csv)

27 of 43

Routes Data Frame

27

> routes_data

# A tibble: 34,698 x 9

nflId gameId playId team quarter PositionAbbr tracking_data direction_left line_of_scrimmage

1 497240 2017090700 68 home 1 TE <tibble [87 x 5]> home 83.7

2 2530515 2017090700 68 home 1 WR <tibble [87 x 5]> home 83.7

3 2533046 2017090700 68 home 1 TE <tibble [87 x 5]> home 83.7

4 2539265 2017090700 68 home 1 RB <tibble [87 x 5]> home 83.7

5 2543498 2017090700 68 home 1 WR <tibble [87 x 5]> home 83.7

6 2649 2017090700 94 home 1 WR <tibble [89 x 5]> home 83.4

7 497240 2017090700 94 home 1 TE <tibble [89 x 5]> home 83.4

8 2530515 2017090700 94 home 1 WR <tibble [89 x 5]> home 83.4

9 2539265 2017090700 94 home 1 RB <tibble [89 x 5]> home 83.4

# ... with 34,689 more rows

28 of 43

purrr::map()

28

29 of 43

Function Development then Map!

29

routes_data <-

routes_data %>%

dplyr::mutate(tracking_data = purrr::map(tracking_data, cut_plays),

tracking_data = purrr::pmap(list(tracking_data, team, direction_left, line_of_scrimmage),

flip_length(..1, ..2, ..3, ..4)),

tracking_data = purrr::map(tracking_data, flip_width))

30 of 43

Finally Cluster!

30

cluster_trajectory_data(routes_data$tracking_data)

31 of 43

Tips for development

31

32 of 43

Work with One “Observation”

32

# A tibble: 87 x 12

nflId gameId playId team quarter PositionAbbr x y frame.id event direction_left line_of_scrimmage

1 497240 2017090700 68 home 1 TE 83.9 34.8 1 NA home 83.7

2 497240 2017090700 68 home 1 TE 83.9 34.8 2 NA home 83.7

3 497240 2017090700 68 home 1 TE 83.9 34.8 3 NA home 83.7

4 497240 2017090700 68 home 1 TE 83.9 34.8 4 NA home 83.7

5 497240 2017090700 68 home 1 TE 83.9 34.8 5 NA home 83.7

6 497240 2017090700 68 home 1 TE 83.9 34.8 6 NA home 83.7

# ... with 81 more rows

routes_data %>%

dplyr::slice(1) %>%

tidyr::unnest(cols = tracking_data)

33 of 43

Use purrr::safely()

33

routes_data %>%

dplyr::sample_n(5) %>%

dplyr::mutate(tracking_data = purrr::map(tracking_data, cut_plays_rand_err))

Error in cutt_plays(data) : could not find function "cutt_plays"

34 of 43

Use purrr::safely()

34

routes_data %>%

dplyr::sample_n(5) %>%

dplyr::mutate(tracking_data = purrr::map(tracking_data,

purrr::safely(cut_plays_rand_err,

otherwise = NA)),

35 of 43

Use purrr::safely()

35

routes_data %>%

dplyr::sample_n(5) %>%

dplyr::mutate(tracking_data = purrr::map(tracking_data,

purrr::safely(cut_plays_rand_err,

otherwise = NA)),

error = purrr::map(tracking_data, ~ purrr::pluck(., "error")),

tracking_data = purrr::map(tracking_data, ~ purrr::pluck(., "result")))

36 of 43

Use purrr::safely()

36

# A tibble: 5 x 10

nflId gameId playId team quarter PositionAbbr tracking_data direction_left line_of_scrimmage error

1 2558069 2017091000 2314 away 3 WR <lgl [1]> home 48.1 <smplErrr>

2 2543488 2017100100 1405 home 2 WR <tibble [41 x 4]> away 57 <NULL>

3 71265 2017092413 595 away 1 TE <lgl [1]> home 34.7 <smplErrr>

4 2555224 2017091012 3213 home 4 RB <lgl [1]> away 67.1 <smplErrr>

5 2555415 2017100101 3945 home 4 TE <tibble [32 x 4]> away 77.2 <NULL>

37 of 43

Use purrr::safely()

37

routes_data %>%

dplyr::sample_n(5) %>%

dplyr::mutate(tracking_data = purrr::map(tracking_data,

purrr::safely(cut_plays_rand_err,

otherwise = NA)),

error = purrr::map(tracking_data, ~ purrr::pluck(., "error")),

tracking_data = purrr::map(tracking_data, ~ purrr::pluck(., "result")),

type_error = purrr::map_chr(error, typeof)) %>%

dplyr::filter(type_error != "NULL")

38 of 43

Use purrr::safely()

38

# A tibble: 3 x 10

nflId gameId playId team quarter PositionAbbr tracking_data direction_left line_of_scrimmage error

1 2558069 2017091000 2314 away 3 WR <lgl [1]> home 48.1 <smplErrr>

2 71265 2017092413 595 away 1 TE <lgl [1]> home 34.7 <smplErrr>

3 2555224 2017091012 3213 home 4 RB <lgl [1]> away 67.1 <smplErrr>

39 of 43

Use purrr::safely()

39

routes_data %>%

dplyr::sample_n(5) %>%

dplyr::mutate(tracking_data = purrr::map(tracking_data,

purrr::safely(cut_plays_rand_err,

otherwise = NA)),

error = purrr::map(tracking_data, ~ purrr::pluck(., "error")),

tracking_data = purrr::map(tracking_data, ~ purrr::pluck(., "result")),

type_error = purrr::map_chr(error, typeof)) %>%

dplyr::filter(type_error != "NULL") %>%

dplyr::pull(error)

40 of 43

Use purrr::safely()

40

[[1]]

[[2]]

[[3]]

41 of 43

Takeaways!

41

Tidy data when its functional data!
purrr::map()
purrr::safely()

42 of 43

Thank You!

Any Questions?

Paper in Journal of Quantitative Analysis in Sports

Slides on twitter @chuurveg or on my website danichusfu.github.io

Dani Chu

dani c.stats @gmail .com

@chuurveg

Matthew Reyers

matthew_reyers@sfu.ca

@Stats_By_Matt

James Thomson

james_thomson_2@sfu.ca

@stat_jamest

Lucas Wu

yifan_wu@sfu.ca

@lucaswu123

Email:

Twitter:

Email:

Twitter:

42

43 of 43

43