This is the same bug as that reported by Wolfram Fischer a few days back. I'm working on it, and hopefully it would be fixed by the next release of lattice (sometime next week). Deepayan
On Friday 13 December 2002 08:57 am, Luis Torgo wrote:
I've come across the following error when using free scales with bwplot (I use a small example data set just to illustrate the problem):
d <- data.frame(
x=c(34.4, 12.4, NA, 65.3, NA, 12.0, 45.0, 645.0, 644.0,323.0),
fac1=c('a','a','b','a','b','a','a','c','c','c'),
fac2=c('v2','v2','v1','v2','v2','v2','v1','v2','v1','v2')
)
# ok, although "x" has only NA values for fac1=='b'
bwplot(fac2 ~ x | fac1,data=d)
# not ok, if I try to use different scales in the X axis
bwplot(fac2 ~ x | fac1,data=d,scales=list(x="free"))
Error in pretty(x[is.finite(x)], ...) : x must be numeric # I can go around the error with
bwplot(fac2 ~ x | fac1,data=d[!is.na(d$x),],scales=list(x="free"))
The problem is that one of the panels has no data because 'x' has only NA values for that particular factor combination. This causes no problem with uniform scales, but generates that error when trying to guess the best scale for each panel. I think that it should be easy to add some test to the code obtaining the scales for each panel, to avoid the always unpleasant criptographic error messages ;-) Thanks, Luis Torgo