R/winProbDL.R

In yorkr: Analyze Cricket Performances Based on Data from Cricsheet

##########################################################################################
# Designed and developed by Tinniam V Ganesh
# Date : 25 Dec 2022
# Function: winProbabilityDL
# This function computes  the ball by ball win probability using Deep Learning Keras
#
###########################################################################################
#' @title
#' Plot the  win probability using Deep Learning model
#'
#' @description
#' This function  plots the  win probability of the teams in a T20 match
#'
#' @usage
#' winProbabilityDL(match,t1,t2,plot=1)
#'
#' @param match
#' The dataframe of the match
#'
#' @param t1
#' The 1st team of the match
#'
#' @param t2
#' the 2nd team in the match
#'
#' @param plot
#' Plot=1 (static), Plot=2(interactive)
#'
#' @return none
#'
#' @references
#' \url{https://cricsheet.org/}\cr
#' \url{https://gigadom.in/}\cr
#' \url{https://github.com/tvganesh/yorkrData/}
#'
#' @author
#' Tinniam V Ganesh
#' @note
#' Maintainer: Tinniam V Ganesh \email{tvganesh.85@gmail.com}
#'
#' @examples
#' \dontrun{
#' #Get the match details
#' a <- getMatchDetails("England","Pakistan","2006-09-05",dir="../temp")
#'
#' # Plot tne match worm plot
#' winProbabilityDL(a,'England',"Pakistan")
#' }
#' @seealso
#' \code{\link{getBatsmanDetails}}\cr
#' \code{\link{getBowlerWicketDetails}}\cr
#' \code{\link{batsmanDismissals}}\cr
#' \code{\link{getTeamBattingDetails}}\cr
#'
#' @export
#'
winProbabilityDL <- function(match,t1,t2,plot=1){

  team=ball=totalRuns=wicketPlayerOut=ballsRemaining=runs=numWickets=runsMomentum=perfIndex=isWinner=NULL
  predict=ml_model=winProbability=ggplotly=runs=runRate=batsman=bowler=NULL
  batsmanIdx=bowlerIdx=NULL
  if (match$winner[1] == "NA") {
    print("Match no result ************************")
    return()
  }
  team1Size=0
  requiredRuns=0

  # Read batsman, bowler vectors
  batsmanMap=readRDS("batsmanMap.rds")
  bowlerMap=readRDS("bowlerMap.rds")

  teams=unique(match$team)
  teamA=teams[1]

  # Filter the performance of team1
  a <-filter(match,team==teamA)

  #Balls in team 1's innings
  ballsIn1stInnings= dim(a)[1]


  b <- select(a,batsman, bowler,ball,totalRuns,wicketPlayerOut,team1,team2,date)
  c <-mutate(b,ball=gsub("1st\\.","",ball))

  # Compute the total runs scored by team
  d <- mutate(c,runs=cumsum(totalRuns))
  # Check if team1 won or lost the match

  if(match$winner[1]== teamA){
    d$isWinner=1
  } else{
    d$isWinner=0
  }
  #Get the ball num
  d$ballNum = seq.int(nrow(d))

  # Compute the balls remaining for the team
  d$ballsRemaining = ballsIn1stInnings - d$ballNum +1

  # Wickets lost by team
  d$wicketNum = d$wicketPlayerOut != "nobody"
  d=d %>% mutate(numWickets=cumsum(d$wicketNum==TRUE))

  #Performance index is based on run rate (runs scored/ ball number) with wickets in hand
  d$perfIndex = (d$runs/d$ballNum) * (11 - d$numWickets)

  # Compute run rate
  d$runRate = (d$runs/d$ballNum)
  d$runsMomentum = (11 - d$numWickets)/d$ballsRemaining

  df8 = select(d, batsman,bowler,ballNum, ballsRemaining, runs, runRate,numWickets,runsMomentum,perfIndex, isWinner)
  df9=left_join(df8,batsmanMap)
  df9=left_join(df9,bowlerMap)

  dfa = select(df9, batsmanIdx,bowlerIdx,ballNum,ballsRemaining,runs,runRate,numWickets,
               runsMomentum,perfIndex, isWinner)
  print(dim(dfa))


  #############################################################################################
  ######## Team 2
  # Compute for Team 2

  # Required runs is the team made by team 1 + 1
  requiredRuns=d[dim(d)[1],]$runs +1
  teamB=teams[2]


  # Filter the performance of team1
  a1 <-filter(match,team==teamB)

  #Balls in team 1's innings
  ballsIn2ndInnings= dim(a1)[1] + 1


  b1 <- select(a1,batsman,bowler,ball,totalRuns,wicketPlayerOut,team1,team2,date)
  c1 <-mutate(b1,ball=gsub("2nd\\.","",ball))

  # Compute total Runs
  d1 <- mutate(c1,runs=cumsum(totalRuns))
  # Check of team2 is winner
  if(match$winner[1]== teamB){
    d1$isWinner=1
  } else{
    d1$isWinner=0
  }

  # Compute ball number
  d1$ballNum= ballsIn1stInnings + seq.int(nrow(d1))

  # Compute remaining balls in 2nd innings
  d1$ballsRemaining=  ballsIn2ndInnings -  seq.int(nrow(d1))


  # Compute wickets remaining
  d1$wicketNum = d1$wicketPlayerOut != "nobody"
  d1=d1 %>% mutate(numWickets=cumsum(d1$wicketNum==TRUE))

  ballNum=d1$ballNum - ballsIn1stInnings
  #Performance index is based on run rate (runs scored/ ball number) with wickets in hand
  d1$perfIndex = (d1$runs/ballNum) * (11 - d1$numWickets)

  #Compute required runs
  d1$requiredRuns = requiredRuns - d1$runs
  d1$runRate = (d1$requiredRuns/d1$ballsRemaining)

  d1$runsMomentum = (11 - d1$numWickets)/d1$ballsRemaining


  # Rename required runs as runs
  df10 = select(d1,batsman,bowler,ballNum,ballsRemaining, requiredRuns,runRate,numWickets,runsMomentum,perfIndex, isWinner)
  names(df10) =c("batsman","bowler","ballNum","ballsRemaining","runs","runRate","numWickets","runsMomentum","perfIndex","isWinner")
  print(dim(df10))

  df11=left_join(df10,batsmanMap)
  df11=left_join(df11,bowlerMap)
  df2=rbind(df9,df11)



  dfb = select(df11, batsmanIdx,bowlerIdx,ballNum,ballsRemaining,runs,runRate,numWickets,
               runsMomentum,perfIndex, isWinner)
  print(dim(dfb))

  # load the model
  m=predict(dl_model,dfa,type = "prob")


  m1=m*100
  m2=matrix(m1)


  n=predict(dl_model,dfb,type="prob")

  n1=n*100
  n2=matrix(n1)

  m3= 100-n2
  n3=100-m2

  team1=rbind(m2,m3)
  team2=rbind(n3,n2)

  team11=as.data.frame(cbind(df2$ballNum,team1))
  names(team11) = c("ballNum","winProbability")
  team22=as.data.frame(cbind(df2$ballNum,team2))
  names(team22) = c("ballNum","winProbability")


  # Add labels to chart team 1
  #Mark when players were dismissed
  k <- cbind(b,m1)
  k$ballNum = seq.int(nrow(k))
  k1= filter(k,wicketPlayerOut !=  "nobody")
  k2 = select(k1,ballNum,m1,wicketPlayerOut)
  #print(k2)

  # Mark when batsman started
  batsmen = unique(k$batsman)
  p =  data.frame(matrix(nrow = 0, ncol = dim(k[2])))
  for(bman in batsmen){
      l <-k  %>% filter(batsman == bman)
      n=l[1,]
      p=rbind(p,n)
  }
  p1 = select(p,ballNum,m1,batsman)
  #print(p1)

  # Add labels to team 2
  #Mark when players were dismissed
  r <- cbind(b1,n1)
  r$ballNum = seq.int(nrow(r))
  r1= filter(r,wicketPlayerOut !=  "nobody")
  r2 = select(r1,ballNum,n1,wicketPlayerOut)

  # Mark when batsman started
  batsmen = unique(r$batsman)
  s =  data.frame(matrix(nrow = 0, ncol = dim(k[2])))
  for(bman1 in batsmen){
      t1 <-r  %>% filter(batsman == bman1)
      t2=t1[1,]
      s=rbind(s,t2)
  }
  s1 = select(s,ballNum,n1,batsman)

  # Plot both lines
  if(plot ==1){ #ggplot
    df3 = as.data.frame(cbind(d$ballNum,m1))
    names(df3) <- c("ballNum","winProbability")
    df4 = as.data.frame(cbind(d1$ballNum,n1))
    names(df4) <- c("ballNum","winProbability")
    maxBallNum = max(df3$ballNum)
    df4$ballNum = df4$ballNum - maxBallNum
    g <- ggplot() +
    geom_line(data = df3, aes(x = ballNum, y = winProbability, color = teamA)) +
    geom_line(data = df4, aes(x = ballNum, y = winProbability, color = teamB))+
    geom_point(data=k2, aes(x=ballNum, y=m1,color="blue"),shape=15) +
    geom_text(data=k2, aes(x=ballNum,y=m1,label=wicketPlayerOut,color="blue"),nudge_x =0.5,nudge_y = 0.5)+
    geom_point(data=p1, aes(x=ballNum, y=m1,colour="red"),shape=16) +
    geom_text(data=p1, aes(x=ballNum,y=m1,label=batsman,colour="red"),nudge_x =0.5,nudge_y = 0.5) +
    geom_point(data=r2, aes(x=ballNum, y=n1,colour="black"),shape=15) +
    geom_text(data=r2, aes(x=ballNum,y=n1,label=wicketPlayerOut,colour="black"),nudge_x =0.5,nudge_y = 0.5) +
    geom_point(data=s1, aes(x=ballNum, y=n1,colour="grey"),shape=16) +
    geom_text(data=s1, aes(x=ballNum,y=n1,label=batsman,colour="grey"),nudge_x =0.5,nudge_y = 0.5)+
    geom_vline(xintercept=36, linetype="dashed", color = "red") +
    geom_vline(xintercept=96, linetype="dashed", color = "red") +
    ggtitle("Ball-by-ball Deep Learning Win Probability (Overlapping)")


    ggplotly(g)

  }else { #ggplotly
    g <- ggplot() +
      geom_line(data = team11, aes(x = ballNum, y = winProbability, color = teamA)) +
      geom_line(data = team22, aes(x = ballNum, y = winProbability, color = teamB))+
      ggtitle("Ball-by-ball Deep Learning Win Probability (Side-by-side)")


    ggplotly(g)

  }
}