[R] plotting issue with timestamps

[Gabor Grothendieck]

On 8/17/05, Dhiren DSouza <dhiren22 at hotmail.com> wrote:
> I have a dataset with transactions and a timestamp at which they occoured
> during a day.  The time stamp is in the format YYYY/MM/DD hh:mm:ss.  I would
> like to plot a timeseries of the transactions to see if there is a
> particular time in the day when there is a spike in transactions.  Ofcourse
> the YYYY/MM/DD can be dropped since I am monitoring activity for the day and
> the actual date is unimportant.
> 
> Can anyone give me some direction on this.  I could possibly build a
> frequency table but not sure how to plot against the timestamp in the format
> above.  Any help would be appreciated.


Lets say your times are in a character vector tt.  (If the dates
are there too use substring to remove them.) Then convert them
to times class (of library chron) which will represent them
internally as a fraction of a day where the day starts at 0 and 
goes to 1 so that 0.5 is noon.  Since we want a more meaningful
axis do not use the automatic plot axis (xaxt = "n") and also
restrict the xaxis to the internval 0:1.  So that we can get
all the hours to fix reduce the size of the axis labels to 70%.
Then redraw it yourself with the hours labelled and add some 
light grey vertical lines if you like.

library(chron)
tt <- c("10:11:12", "09:10:11", "01:02:03")
tt.times <- times(tt)
plot(density(tt.times), xaxt = "n", xlim = 0:1, xlab = "Hour")
axis(1, 0:24/24, 0:24, cex.axis = 0.7)
abline(v = 0:24/24, col = "lightgrey")

See RNews 4/1 Help Desk article for more.