Charting time series as calendar heat maps in R

Last month we showcased the JSM Data Expo, where the winning entry was a visualization of airline delays represented as a color-coded calendar. That graphic was created in SAS, but now thanks to reader Paul Bleicher, we can show you how to create the same graphic in R. 

Paul Bleicher, MD PhD is Chief Medical Officer at Humedica, a next-generation clinical informatics company that provides novel business intelligence solutions to the health care and life science industries. Paul is leading a team that is using R extensively for a wide variety of predictive analytics and data visualization applications with medical record data. Paul has been kind enough to share his R code that takes a sequence of numeric values indexed by date, and represents them as a calendar with the days filled with colors representing the values. It's easier to explain by example: let's download Microsoft's stock price from 2006 to date from Yahoo, and plot it using Paul's calendarHeat function:

stock <- "MSFT"

start.date <- "2006-01-12"

end.date <- Sys.Date()

quote <- paste("http://ichart.finance.yahoo.com/table.csv?s=",

                stock,

                "&a=", substr(start.date,6,7),

                "&b=", substr(start.date, 9, 10),

                "&c=", substr(start.date, 1,4), 

                "&d=", substr(end.date,6,7),

                "&e=", substr(end.date, 9, 10),

                "&f=", substr(end.date, 1,4),

                "&g=d&ignore=.csv", sep="")             

stock.data <- read.csv(quote, as.is=TRUE)

calendarHeat(stock.data$Date, stock.data$Adj.Close, varname="MSFT Adjusted Close")

Pretty cool (click the calendar to see the details better). We used financial data here because it's easier to access than the airline data, but it's actually a pretty interesting way of looking at a financial time series. Weekend and holiday effects are a bit more obvious, and it's a bit like being able to see the daily, weekly, monthly and yearly closes all at once (by scanning your eye over the calendar in different directions).

The calendarHeat function takes two primary arguments: the first, dates, is a vector of dates. You can provide these in POSIXlt format or as characters as in the example above (where they were converted using the default "%Y-%m-%d" value of the optional date.form argument). The second is a vector of numeric values (here, stock closing prices) which by default are converted into a scale of 99 shades from red to green. You can control the number of shades with the ncolors option, and the options colors="r2b" and colors="w2b" give you the alternative colorschemes red-to-blue and white-to-blue, respectively.

The source code for the calendarHeat function is attached in the link below, licensed as open-source under GPL v2. Share and enjoy. And if someone downloads the airline data and reproduces this chart, let us know!

Update Dec 3: With thanks to Jon Egil Strand in the comments, the calendarHeat.R source file has been updated to fix an off-by-one error.

Paul Bleicher: calendarHeat.R (source code for calendarHeat function)

Comments

[Jay]

Nice thanks! It's amazing how giving the R community is. Thanks for the blog.

[Hadley]

Here's a quick rendition in ggplot2:

stock.data <- transform(stock.data,
  week = as.POSIXlt(Date)$yday %/% 7 + 1,
  wday = as.POSIXlt(Date)$wday,
  year = as.POSIXlt(Date)$year + 1900)

library(ggplot2)

ggplot(stock.data, aes(week, wday, fill = Adj.Close)) + 

  geom_tile(colour = "white") + 

  scale_fill_gradientn(colours = c("#D61818","#FFAE63","#FFFFBD","#B5E384")) + 

  facet_wrap(~ year, ncol = 1)

[Hadley]

But maybe a better display of the same data is :

qplot(week, Adj.Close, data = stock.data, colour = factor(wday), geom = "line") + 
  facet_wrap(~ year, ncol = 1)

[Jay]

Your package allows such concise definition of graphics Hadley. This really demonstrates the power of your paradigm for ggplot yet again.

[Andrew Wooster]

Neat! I just used it to chart my iPhone app sales over time:
http://blog.planetaryscale.com/2009/11/04/iphone-app-sales-heatmap/

[Sean]

Would be interesting to see the % daily change (instead of closing price) in this format.

[Jon Egil Strand]

A bug in merge used in original sourceode shifts data one day. Instead use this:

# Merge moves data by one day, avoid
caldat <- data.frame(date.seq = seq(min.date, max.date, by="days"), value = NA)
dates <- as.Date(dates)
caldat$value[match(dates, caldat$date.seq)] <- values

[Neil]

Very nice. Even better - hook this into a web application, make the heat map a clickable image map taking you to data for that date.