This post is a continuation of the basis of hacking out an UpSet plot to analyse matchsheet data. In the previous post, I touched upon the basics of data loading, cleaning, wrangling and finally visualisation with ggplot using a dummy dataset. In here, real data is at work.
This article is the first of a series with supporting codes that will act as a step-by-step guide to automating the entire process of loading matchsheet data for any team and visualising it, part by part. Presenting matchsheet data is easy, but building a system to automate the process of loading, processing and presenting matchsheet data with one click makes the life of a performance analyst much easier.
The tasks in this part are made extremely easy by functions. We write two functions which will handle all aspects of connecting, loading, transforming and visualising.
Jason Zivkovic's worldfootballR package is of great help here in that it allows us to pull data from fbref.com with a few helper functions. Even though ISL data isn't as extensive and clean as big leagues' data on fbref.com, matchsheet data is fairly consistent; and since it's available to scrap and laod via worldfootballR, there's no point looking for other data sources.
To use the worldfootballR package, first install it (instructions are here) and then load the library with library(worldfootballR) command.
The first function we will write is the teamsheet_data() function. This function takes in an integer season_end_year (like 2021 for 2020-21 season) and a string team_name (like "Odisha" for Odisha FC) as arguments, and returns a dataframe containing the starting lineup details of every match of the given team's given season.
teamsheet_data <- function(season_end_year,team_name){
library(worldfootballR)
url<-as.data.frame(get_match_urls(country = "IND",gender = "M",season_end_year = season_end_year,tier = "1st"))
names(url)<-"urls"
url<-url%>%filter(str_detect(urls,"Indian-Super-League"))
url<-url%>%filter(str_detect(urls,team_name))
lineup<-get_match_lineups(url$urls)
#lineup dataframe is ready
lineup.team<-lineup%>%
filter(str_detect(Team,team_name),Starting=="Pitch")%>%
arrange(Matchday)
#opponent & away formation dataframe
lineup.meta<-lineup%>%
group_by(Matchday)%>%
dplyr::summarise(home=unique(Team)[1],
away=unique(Team)[2],
home_tactic=unique(Formation)[1],
away_tactic=ifelse(length(unique(Formation))==2L,
unique(Formation)[2],unique(Formation)[1]))
lineup.meta<-lineup.meta%>%
mutate(vs.team=ifelse(home==team_name,away,home),
vs.tactics=ifelse(home==team_name,away_tactic,home_tactic),
match.num=c(1:length(unique(lineup.meta$Matchday))))
lineup.allinfo<-merge(lineup.team,lineup.meta[,c("Matchday","vs.team","vs.tactics","match.num")],by="Matchday")
return(lineup.allinfo)
}
The second funnction start.changes() uses the first function teamsheet_data() in a nested capacity. It generates a dataframe in the format of the dummy dataframe in this post so that we can plot the UpSet plot depicting the number of starting XI changes per game. The start.changes() function returns the resultant plot like this:
The code chunk:
start.changes <- function(season_end_year,team_name){
#use teamsheet_data() function to gather lineup info of particular team
team_name_lineup<-teamsheet_data(season_end_year,team_name)
#initialize an empty list to store lineups
empty_list<-vector(mode = "list",length = length(unique(team_name_lineup$match.num)))
#populate the empty list
for (i in 1:length(unique(team_name_lineup$match.num))) {
empty_list[i]=list(unique(team_name_lineup$Player_Name[which(team_name_lineup$match.num==i)]))
}
lineup_list<-as_tibble(empty_list,.name_repair = "unique")
#make changes in the column names of the starting XI tibble (as column names are "...1", "...2",...
#& we need better names)
names(lineup_list)<-as.character(unique(team_name_lineup$match.num))
#create dataframe for visualising changes data
#we create a "changes.df" dataframe with 3 columns - match, vs, changes
#populate the match column
changes.df<-data.frame(match=as.numeric(names(lineup_list)))
#initialize empty vs & changes columns
changes.df$vs<-character(nrow(changes.df))
changes.df$changes<-integer(nrow(changes.df))
#iterate through the vs column to populate it
# for (i in 1:nrow(changes.df)) {
# changes.df$vs[i]=unique(lineup_list$vs.team[which(lineup_list$match.num==i)])
# }
#iterate through the changes column to populate it
for (j in 2:(nrow(changes.df))) {
changes.df$changes[j]=11-(length(intersect(as_vector(lineup_list[,j-1]),as_vector(lineup_list[,j]))))
}
#we can now plot the changes upset plot using ggplot
changes.plot<-ggplot()+
scale_x_continuous(breaks = seq(1,max(changes.df$match),1),position = "bottom")+
scale_y_continuous(breaks = seq(1,max(changes.df$changes),1))+
geom_segment(data = changes.df,
aes(x = match,y = 0,xend = match,yend = changes,colour = changes),
size = 1,
alpha = 0.7,
show.legend = F)+
geom_point(data = changes.df,
aes(x = match,y = changes),
size=2.5)+
geom_point(data = changes.df,
aes(x = match,y = 0),
size = 2.5)+
labs(title = "Starting XI Changes",
subtitle = paste0(team_name_lineup$vs.team[str_detect(team_name_lineup$vs.team,team_name)][1]," \nISL ",season_end_year-1,"-",season_end_year))+
xlab("Match no.")+
ylab("No. of changes")+
scale_color_gradient(low = "blue",high = "red")+
theme(
text = element_text(family = "Courier"),
plot.background = element_rect(fill="#F6FCF8"),
panel.background = element_rect(fill = "#F6FCF8"),
panel.grid = element_blank(),
title = element_text(hjust = 0.5),
plot.title = element_text(hjust = 0,face = "bold"),
plot.subtitle = element_text(hjust = 0,size = 10)
)
return(changes.plot)
}
Now, we can generate UpSet plot of any ISL team from any season by keying in the season_end_year and the team_name.
Like this...
start.changes(season_end_year = 2020,team_name = "Kerala")
... which gives us this plot (everything else happens under the hood):
Fin.
NOTE: Functions will run for ISL seasons starting from 2017-18. Pre-2017 ISL seasons have different formatting in the fbref.com data and we need to write different functions to extract and visualise data from those seasons.