In sunburstR: Sunburst 'Htmlwidget'

devtools::install_github("cpsievert/XML2R")

devtools::install_github("cpsievert/pitchRx")

devtools::install_github("timelyportfolio/sunburstR")

install.packages("dplyr")

install.packages("tidyr")

install.packages("stringr")

install.packages("rvest")

library(sunburstR) library(pitchRx) library(dplyr)

get all data from 2016-08-25

dat <- scrape(start = "2016-08-25", end = "2016-08-25")

Make the Data sunburst-able

# use runner data to get idea of action with a runner on base
#  please note this will not be all action from a game
#  but I think it is an easier dataset to understand
action <- dat$runner %>%
  group_by(event_num) %>%
  filter(row_number() == 1) %>%
  ungroup() %>%
  group_by(gameday_link, inning, inning_side) %>%
  summarize(event = paste(c(event),collapse="-"))

sequences <- action %>%
  ungroup() %>%
  group_by(event) %>%
  summarize(count = n())

# sorry this is messy, but get data in a form
#  so sunburst can build hierarchy
#  which means we will sort in descending order of depth
# note: this will eventually improve
sequences$depth <- unlist(lapply(strsplit(sequences$event,"-"),length))

Create a Sunburst

sb <- sequences %>%
  arrange(desc(depth), event) %>%
  sunburst()
sb

Use Sunburst Events

In this commit, we added some basic event dispatch to sunburstR that can be used in Shiny and non-Shiny contexts. We will use it to display a link to the games that fit the hovered paths.

# use sunburst event handling to provide games for hovered sequence
library(htmltools)

sb$x$tasks <- list(htmlwidgets::JS(
'
function(){
debugger;
  var chart = this.instance.chart;
  chart.on("mouseover",mouseovered);
}
'
))

sb$height = 400
sb$width = 600

tagList(
  sb,
  tags$div(id="games", style="margin-top:100px"),
  tags$script(HTML(
    sprintf(
'
var action = %s;

function mouseovered(d){
var thiz = this;
var games = action.filter(function(evt){
  return evt.event === thiz.join("-");
});

var div = document.getElementById("games");
div.innerHTML = games.map(function(game){
  return [
    "<a href=\'http://gd2.mlb.com/components/game/mlb/year_2016/month_08/day_25/",
    game.gameday_link,
    "\'>",
    game.gameday_link,
    "</a><br/>"
  ].join("");
}).join("\\n");
}
',
      jsonlite::toJSON(action, auto_unbox=TRUE, dataframe="row")
    )
  ))
)

Sequence of Pitches

Dan Malter wrote a fantastic post Using Markov Chains to Predict Pitches. Let's visualize his pitch data for Jake Arrieta.

Scrape the Data

# pitch sequence data from Markov Chain
#   http://danmalter.github.io/r/2016/03/28/Markov-chains.html
library(sunburstR)
library(rvest)
library(stringr)
library(tidyr)
library(dplyr)

# get table from post to avoid having to run all the code
ht <- read_html("http://danmalter.github.io/r/2016/03/28/Markov-chains.html")
# get pitch type as proportion of total pitches
ht_pitch_arrieta <- html_table(
  xml_find_first(ht,'//*[@id="jake-arrieta---overall-pitch-proportions"]/table')
)
# get markov table for pitch and following pitch
ht_tbl_arrieta <- html_text(
  xml_find_all(ht,'//*[@id="jake-arrieta---multi-class-markov-chain"]/pre/code')[2]
)
ht_tbl_arrieta <- str_replace_all(
  ht_tbl_arrieta,
  c("4-seam FB"="4seamFB", "Int. Ball"="IntBall")
)
tbl <- read.table(
  textConnection(ht_tbl_arrieta),
  skip=2,
  header=FALSE,
  stringsAsFactors=FALSE
)
colnames(tbl) <- c("pitch",tbl[,1])

# multiple by pct of total
tbl[,-1] <- tbl[,-1] * t(ht_pitch_arrieta)

tbl_long <- gather(tbl, key=pitch2, value=value, -pitch)

Draw the Sunburst

tbl_long %>%
  mutate(path = paste(pitch,pitch2,sep="-")) %>%
  select(path, value) %>%
  sunburst()

Thanks

Thanks so much to Kerry Rodden who provided the original sunburst on which this is based.

Thanks Mike Bostock for d3.js (please give us a way to pay you).

Thanks Tim Holman for the great github-corners.

Thanks skeleton.

Thanks Carson Sievert for pitchRx and all your work on plotly.

Thanks Dan Malter for the great markov pitch post.