0

A newbie question. (ADDED NEW INFO)

I have a set of time stamped data that were collected randomly. I like to create a matrix of plots, but I could not create using either scatterplot or xyplot & time objects.

my data

dataset$Time  #POSIX time objects (no set sampling period)
              #i built POSIX time objects by dataset$Time<-strptime(tt, "%H:%M:%OS")
              #the origial string was formated like this 12:12:12.234 (HH:MM:SS:msec)
dataset$d1, dataset$d2 #integers
dataset$d3 #factor with 10 levels

.

i can do these

plot( dataset$Time, dataset$d1)
scatterplot(dataset$d1 ~ dataset$d2 | dataset$d3, data=dataset)
xyplot(dataset$d1 ~ dataset$d2 | dataset$d3, data=dataset)

However, i can NOT do these (POSIX time object in x-axis)

scatterplot(dataset$d1 ~ dataset$Time | dataset$d3, data=dataset)
xyplot(dataset$d1 ~ dataset$Time | dataset$d3, data=dataset)

(NEW INFO)

Error in structure(.Internal(as.POSIXct(x, tz)), class = c("POSIXt", "POSIXct"), : invalid 'x' argument.

(NEW INFO) but this works (POSIX time object in y-axis) 

xyplot(dataset$Time ~ dataset$d1 | dataset$d3, data=dataset)

related, but different question is hexbin. When time objects are added to hexbin, the plot from the hexbin does not show correct time format on the units.

bin<-hexbin(dataset$Time, dataset$d1) 
plot(bin))

What should I do?

Thanks for looking into it!!

user186477
  • 1,039
  • 3
  • 11
  • 13

3 Answers3

2

Apparently R has some problem dealing with addiction of POSIX times... it gives me the error:

Error in +.POSIXt(x[floor(d)], x[ceiling(d)]) : binary '+' is not defined for "POSIXt" objects

just cast them to numeric and it should work. You can hide the corresponding axis and redraw it later with the correct dates

E.g.:

scatterplot(dataset$d1 ~ as.numeric(dataset$Time) | dataset$d3, data=dataset, xaxt="n")
axis(1, at=as.numeric(dataset$Time), labels=dataset$Time, cex.axis=0.5)
nico
  • 50,859
  • 17
  • 87
  • 112
  • Hi nico! as.numeric() and as.POSIXct() let me use xyplot but NOT with scatterplot(). have you tried with scatterplot() from car package? – user186477 Jun 23 '10 at 00:51
  • any alternative way to represent time from time stamps? – user186477 Jun 23 '10 at 01:33
  • 1
    yes, I've tried with scatterplot, as in the example I wrote in my answer... may be an error in how you create/import the data? can you maybe provide a sample of your data? – nico Jun 23 '10 at 06:15
2

For me it just works, so you probably has a time vector in a bad format. What do you get when you call class(dataset$Time)? It should contain "POSIXct" for this to work.

On the other hand, you don't need to put dataset$ in formula if you supply data=dataset.

mbq
  • 18,510
  • 6
  • 49
  • 72
  • mbq, you're right. with help from nico below, i was able to use xyplot by casting either by as.numeric() or by as.POSIXct(). However, scatterplot was a no-go so far. Thank you!! – user186477 Jun 23 '10 at 00:49
2

Here is a working example. You may have gotten your data formats wrong -- one needs to be careful about the precise data formats.

First, a simple data frame:

R> X <- data.frame(pt=Sys.Date()+0:4, x1=100+cumsum(rnorm(5)),
+                                     x2=90+cumsum(rt(5,4)))
R> X
          pt     x1    x2
1 2010-06-22  98.73 90.33
2 2010-06-23  99.43 89.56
3 2010-06-24  98.85 86.95
4 2010-06-25  99.08 88.52
5 2010-06-26 100.30 94.08
R>

This is so-called wide form which lattice does not use. You need to transform it into long format. I use stack() here, you can also use cast() and melt() from the reshape package:

R> Y <- data.frame(pt=rep(X$pt,2), stack(X, select=c(x1,x2)))
R> Y
           pt values ind
1  2010-06-22  98.73  x1
2  2010-06-23  99.43  x1
3  2010-06-24  98.85  x1
4  2010-06-25  99.08  x1
5  2010-06-26 100.30  x1
6  2010-06-22  90.33  x2
7  2010-06-23  89.56  x2
8  2010-06-24  86.95  x2
9  2010-06-25  88.52  x2
10 2010-06-26  94.08  x2
R>

Now the xyplot call is simply:

R> xyplot(values ~ pt | ind, data=Y, panel=panel.lines)

and you can of course use much more complicated conditioning expressions.

Dirk Eddelbuettel
  • 360,940
  • 56
  • 644
  • 725
  • Hi Dirk! myabe, it is because of the difference in localization, but my time object is shown as this "2010-06-21 17:45:42" What are x1, x2, and ind? Thank you! – user186477 Jun 22 '10 at 22:19
  • 1
    I *created the example data on the spot* which is why you get different time values. The two columns are cumulative sums of random number from the normal and t distributions. – Dirk Eddelbuettel Jun 22 '10 at 23:54
  • the problem is not the data format, but time format. it seems that R has bit of issues with POSIXct vs POSIXlt. Thanks! – user186477 Jun 23 '10 at 00:17
  • 1
    That's a FAQ. You want `POSIXct` (== a single double) and not `POSIXlt` (== a list of nine elements) in your `data.frame` -- so call `as.POSIXct()` on the columns. – Dirk Eddelbuettel Jun 23 '10 at 00:38
  • hexbinplot() giving me errors with POSIXct "Error in '/.difftime'(diff(ybonds), diff(.ylim) : second argument of / cannot be a "difftime" object" ----- scatterplot() giving me "Error in '+.Date'(x[floor(d)], x[ceiling(d)]) : binary + is not defined for Date objects --- any suggestion to represent time in a format i would not get these errors? Thank you!! – user186477 Jun 23 '10 at 01:38
  • Nobody claimed `hexbin()` would with DateTime classes, that is only you assuming it does. I showed you how to use `xyplot()` with DateTime data, this does in no way imply `hexbin()` would. So once again, if it doesn't work, transform your data. `as.numeric()` on a `POSIXct` gives you a nice and clean floating point that `hexbin()` will have no trouble with. – Dirk Eddelbuettel Jun 23 '10 at 01:50