Sample code to demonstrate some ways of making circular time-of-day plots in R (i.e. polar plots with 24 major hourly units)

demo_circular_day_plots.R

##
## Sample code to demonstrate circular time plots in R
## [email protected]
## Dec 22, 2012

# inspired by 
# http://stackoverflow.com/questions/2076370/most-underused-data-visualization

library(lubridate)
library(circular)
library(ggplot2)   # use at least 0.9.3 for theme_minimal()

## generate random data in POSIX date-time format

N=500
events <- as.POSIXct("2011-01-01", tz="GMT") + 
              days(floor(365*runif(N))) + 
              hours(floor(24*rnorm(N))) +  # using rnorm here
              minutes(floor(60*runif(N))) +
              seconds(floor(60*runif(N)))

# extract hour with lubridate function
hour_of_event <- hour(events)

# make a dataframe
eventdata <- data.frame(datetime=events, eventhour=hour_of_event)

# determine if event is in business hours
eventdata$Workday<-eventdata$eventhour%in%seq(9,17)

## first method of plotting is from ggplot2 

ggplot(eventdata,aes(x=eventhour,fill=Workday))+
       geom_histogram(breaks=seq(0,24),width = 2,colour="grey")+ 
       coord_polar(start=0)+
       theme_minimal()+
       scale_fill_brewer()+
       ylab("Count")+
       ggtitle("Events by Time of day")+
       scale_x_continuous("", limits=c(0,24),
                      breaks=seq(0,24),
                      labels=seq(0,24))


## two remaining examples use package::circular
## first is plot.circular(), second is roseplot()

# make circular class from package circular
eventdata$eventhour <- circular(hour_of_event%%24, # convert to 24 hrs
      units="hours", template="clock24")

# plot events on a circle
plot.circular(eventdata$eventhour, stack=TRUE, shrink=2, cex=0.7,col="red")

# estimate density, class is still circular not density
bw <- bw.nrd0(eventdata$eventhour) # may not be best method: experiment
dens <- density.circular(eventdata$eventhour, bw=bw)  #bw must be given

# plot.density.circular is not useful for large numbers of events
# or we have a bad bandwidth -- need to determine better method
plot(dens,plot.type="circle", join=TRUE, shrink=1.3, cex=0.5, col="blue")

# plot.density on type = "line" is better, but bw still questionable
plot(dens, plot.type="line", join=TRUE, cex=0.5, col="blue")

# plot a rose diagram, default looks bad
rose.diag(eventdata$eventhour, bin=24, col="blue", 
    main="Events by Hour (sqrt scale)")

# change prop argument, now looks better
rose.diag(eventdata$eventhour, bin=24, col="blue", 
    main="Events by Hour (sqrt scale)", prop=3)

# linear scale will emphasize outliers, need to change prop more
rose.diag(eventdata$eventhour, bin=24, col="blue", 
    main="Events by Hour (linear scale)", prop=10,
    radii.scale = "linear")

# plot sqrt and linear side by side
oldpar <- par()
par(mfcol=c(1, 2))
rose.diag(eventdata$eventhour, bin=24, col="blue", cex = 0.5, 
    main="Sqrt scale", prop=3)
rose.diag(eventdata$eventhour, bin=24, col="blue", cex = 0.5,
    main="Linear scale", prop=10,
    radii.scale = "linear")
par(oldpar)

# redundantly add points to surface 
# need to adjust parameters like shrink, cex, and prop
rp <- rose.diag(eventdata$eventhour, bin=24, col="blue", 
    main="Events by Hour (linear scale)", prop=12, 
    radii.scale = "linear", shrink=2, cex=0.5)
points(eventdata$eventhour, plot.info=rp, col="purple",stack=TRUE)


## in addition to what is in circular
## library(psych) has helpful circular statistics. 
## circadian.mean(): circular mean 
## circadian.cor(): circular (phasic) correlations 
## circadian.linear.cor(): correlation between linear variables and circular variables  
## cosinor(): fit phase angle for measures taken with a fixed period (e.g., 24 hours)