Time series visualization for start, end, duration in R

Question

I have the following data:

> Data
          Date    Start       End
1   2011-11-15 12:01:27 12:30:15 
2   2011-11-16 12:01:25 12:32:15 
3   2011-11-17 12:01:02 12:39:12 
4   2011-11-19 12:01:12 12:30:18

to which I've also appended a Duration column

Data[,4] <- as.numeric(difftime(Data$End,Data$Start))
names(Data)[4] <- "Duration"

I have in my head a visualization for Start,End that looks kind of like a stock candlestick or OHLC chart, where the x value is the Date, and y is End - Start.

End is at the top with a rectangle descending down to Start---the height of the rectangle changes over time with the Duration. That is, each Date has a different rectangle height determined by the difference between Start and End.

The x axis, here, goes from 2011-11-15 to 2011-11-19. The y axis goes from 12:00:00 to 12:40:00.

Do any ggplot wizards see an easy way to do this? Since both Start and End are changing over time, would I have to use geom_ribbon or geom_polygon rather than geom_bar or geom_area?

It would be extra cool if the color of the bar can change to red on days when the value of Duration is greater than 2 standard deviations!

csgillespie · Accepted Answer · 2012-04-26T11:58:41.373

I use a similar structure to nico (thanks!):

date = c("2011-11-15", "2011-11-16", "2011-11-17", "2011-11-19")
start = c("12:01:27", "12:01:25", "12:01:02", "12:01:12")
end = c("12:30:15", "12:32:15", "12:39:12", "12:30:18")

Next, we put it in a data frame that contains the corners of the rectangles:

##I've made the rectangles 2 hours wide
df = data.frame(date = as.POSIXct(date),
         ystart = as.POSIXct(start, format="%H:%M:%S"), 
         yend = as.POSIXct(end, format="%H:%M:%S"),
         xstart=as.POSIXct(paste(date, "12:00:00"), format="%Y-%m-%d %H:%M:%S"),
         xend = as.POSIXct(paste(date, "14:00:00"), format="%Y-%m-%d %H:%M:%S"))

Then we just use geom_rect:

ggplot() + geom_rect(data=df, aes(ymin=ystart, ymax=yend,
                           xmin=xend, xmax=xstart))

If you want to make some of them red based on a condition, just create an additional column on your data frame:

##Your condition is something to do with the sd
df$isRed = c(TRUE, FALSE)

Then add two ggplot layers:

ggplot() + geom_rect(data=subset(df, !isRed), aes(ymin=ystart, ymax=yend,
                           xmin=xend, xmax=xstart)) +
           geom_rect(data=subset(df, isRed), aes(ymin=ystart, ymax=yend,
                           xmin=xend, xmax=xstart), colour="red")

Example graph

enter image description here

It's debatable whether this is any easier, but you could specify the `colour` parameter as an aesthetic and manually adjust the scale, like so: `ggplot(df, aes(date, ymin = y.from, ymax = y.to, colour = isRed)) + geom_linerange() + scale_colour_manual(values = c("TRUE" = "red", "FALSE" = "black"), guide = "none")`. With two colours, adding separate geoms is arguably just as easy, but perhaps not as scale-able. — Chase, Apr 25 '12 at 15:00
Sorry to be confusing. Nico got what I was looking for aesthetically. This is close, and I like the ggplot, but can geom_linerange produce wider bars than just lines? — Mittenchops, Apr 25 '12 at 17:38

score 3 · Answer 2 · answered Apr 25 '12 at 07:12

I don't use ggplot, but I can give you a base R solution

# Generate the data
date <- c("2011-11-15", "2011-11-16", "2011-11-17", "2011-11-19")
start <- c("12:01:27", "12:01:25", "12:01:02", "12:01:12")
end <- c("12:30:15", "12:32:15", "12:39:12", "12:30:18")

# Put everything in a data frame and convert to POSIXct objects
# The times will be all converted to today's date
# but this will not influence the plot
df <- data.frame(date = as.POSIXct(date),
                 start = as.POSIXct(start, format="%H:%M:%S"), 
                 end = as.POSIXct(end, format="%H:%M:%S"))

# Get the working range for the axes in order to make them nicer (see below)
x.from <- as.POSIXct(min(date))
x.to <- as.POSIXct(max(date))
y.from <- as.POSIXct(min(start), format="%H:%M:%S")
y.to <- as.POSIXct(max(end), format="%H:%M:%S")

# Create an empty plot, as rect will not create a new one
# We put no axes on the plot
plot(0, "n", xaxt="n", yaxt="n", ylab="", xlab="Day", 
     ylim=c(from, to), xlim=range(df$date))

# Now draw the rectangles (I made them 2 hours-wide)
rect(df$date-3600, df$start, df$date+3600, df$end, col="black")

days <- seq(x.from, x.to, 24*3600)
times <- seq(y.from, y.to, 300) # 5 min (=300 s) axis ticks
# Finally add the axes
axis(1, at=days, labels=strftime(days, "%d/%m"))
axis(2, at=times, labels=strftime(times, "%H:%M"), las=1)

The result:

candlestick-like plot

sorry, don't have time now to add the colour bit, will try to add it tonight (unless some other more elegant solution comes up) — nico, Apr 25 '12 at 07:13
This is pretty close. I'm looking for wider bars (that almost touch), but once I added the 2 years or so of data points I have, this will definitely be an interesting chart. Thanks for the direction! — Mittenchops, Apr 25 '12 at 17:31
@Mittenchops: just change the 3600 to something higher in the `rect` call to have wider bars ;) — nico, Apr 25 '12 at 17:46
You're amazing, @nico, thanks. I need to understand the mechanics better here (and how I'll have to change it to work on a time series that expands over multiple years, and to change the colors for extremes), but this helped me make a fantastic graph. Thanks! — Mittenchops, Apr 25 '12 at 18:50
@Mittenchops: R "thinks" in seconds so each tick on the x axis is 1 day, if your bars are 7200 seconds wide it means they will cover 2 hours, so 1/12 of the space between 2 ticks. — nico, Apr 25 '12 at 18:59

Time series visualization for start, end, duration in R

2 Answers2

Linked