R/plot.oCPD.R

In onlineCPD: Detect Changepoints in Multivariate Time Series

plot.oCPD <-
function(x,lines=TRUE,title="",leg.name="Variable",cleanCP=TRUE,buffer=10, ...) {
  value <- variable <- val <- Var1 <- Var2 <- num <- NULL

  T <- dim(x$data)[1]
  R1 <- x$R[1:T,1:T]

  if(is.null(x$time))
  {
    time <- 1:(T)
    melted <- melt(data.frame(cbind(time,data=x$data)),id.vars="time")
    a <- ggplot(data=melted,aes(x=time,y=value,color=variable))+geom_point()+labs(x="",title=title) +
      scale_color_discrete(name=leg.name,label=colnames(x$data))
    meltedR1 <- melt(R1)
  } else
  {
    time <- x$time
    melted <- melt(data.frame(cbind(time=time,data=x$data)),id.vars="time")
    melted$time <- as.POSIXct(melted$time,origin="1970-01-01 00:00:00")
    a <- ggplot(data=melted,aes(x=time,y=value,color=variable))+geom_point()+labs(x="",title=title)+
      scale_color_discrete(name=leg.name)
    colnames(R1) <- time
    meltedR1 <- melt(R1)
    meltedR1$Var2 <- as.POSIXlt(meltedR1$Var2,origin="1970-01-01")
  }
  
  if(lines){
    ymin <- ggplot_build(a)$panel$ranges[[1]]$y.range[1]
    ymax <- ggplot_build(a)$panel$ranges[[1]]$y.range[2]
    if(cleanCP) changes <- findCP(x,buffer)
    else        changes <- x$changes
    for(k in changes[-1]){
      a$layers <- c(geom_segment(data=data.frame(val=k),aes(x=time[val],xend=time[val],y=ymin,yend=ymax,alpha=0.02),color="red",show.legend=FALSE),a$layers)
    }
  }
  
  b <- ggplot(data=meltedR1,aes(Var2,Var1,fill=value))+labs(x="data",y="runlength")+
    geom_raster() +scale_fill_gradient(low="#FFFFFF", high="#000000",na.value="grey92",name="Probability",trans=log_trans(exp(1)),limits=c(1e-5,1),
                                       breaks=c(1,0.01,0.0001))+
    geom_line(data=data.frame(num=append(time,time[T]),max=x$max[1:(T+1)]),aes(num,max),alpha=0.5,color="red",inherit.aes=FALSE) + theme_grey()
  grid.arrange(a,b)
}