[R] RFC: default background on lattice plots
deepayan.sarkar at gmail.com
Tue Mar 14 23:59:52 CET 2006
On 3/13/06, Prof Brian Ripley <ripley at stats.ox.ac.uk> wrote:
> On Mon, 13 Mar 2006, Martin Maechler wrote:
> >>>>>> "JohnF" == John Fox <jfox at mcmaster.ca>
> >>>>>> on Sat, 11 Mar 2006 13:29:34 -0500 writes:
> > JohnF> Dear Deepayan, As you say, it's currently very easy
> > JohnF> to change settings (which is what I do routinely),
> > JohnF> but since you asked, I much prefer the settings in
> > JohnF> canonical.theme("pdf") and therefore would prefer
> > JohnF> that the windows() device use these settings as a
> > JohnF> default (independent of the printing issue).
> > JohnF> Regards, and thanks for the lattice package,
> > indeed, thanks a lot, Deepayan!
> > I think it would make much sense to use the *same* canonical.theme
> > for all interactive default devices.
> > This may be important in teaching, packages, user-written
> > functions which often are all meant to be used interactively; I
> > think it would be painful if students in my class looked at
> > quite differently colored pictures depending on if they are
> > using MacOS X, Linux or Windows.
> > Hence, if you change the setting for windows(), I think you
> > should also do so for x11() and quartz().
> That would be an even stronger argument if those devices all rendered RGB
> colours the same, but MacOS is running a different default interpretation,
> AFAIK (and many PC displays are still way off sRGB, hence the gamma
> argument of some devices but unfortunately not quartz()). Beyond that,
> the visual effect is dependent both on the brilliance setting of the
> display and ambient lighting levels, and most people have screens set far
> too bright (and work in too harshly lit environments) to achieve optimal
> I can see two defensible positions.
> 1) Default themes are chosen for each device with a common 'perceptual
> intent' (a technical phrase). I believe that was the aim of Trellis, but
> one not achieved on many displays (I remember a muddy brown background on
> an old Sun display Bill Venables had).
> 2) The same theme is chosen for all devices, and the user is expected to
> establish proper viewing conditions. That would mean using the same
> default theme for _all_ devices.
> I think 2) has to be the way forward, as nowadays there is no reason to
> suppose that postscript() or pdf() output will be printed rather than say
> included into a lecture presentation (and if so whether that will be
> printed or 'beamed').
I completely agree (personally, I would prefer to retain black and
white defaults for postscript, but that's only because most of our
printers are still B&W). The important question is then, what should
this default be?
The current default (the one with a grey background) obviously has
issues, but at least some thought was put into it (see, e.g.
http://cm.bell-labs.com/stat/doc/trellis.tour.col.ps). The other
available themes are
(a) the current PDF default, which is essentially the screen defaults
with background changed to white, followed by some of the lighter
symbol colors made darker according to my whims at the time.
(b) col.whitebg(), which was meant to be a proof of concept, and
should not be considered seriously.
There has been no systematic study of how these settings affect
perception. I know of some work on the optimal choice of colors (e.g.
but they mostly apply to fill colors, not symbol/line colors.
Personally, I'm happy enough with the PDF defaults (except maybe the
third superpose.symbol color), but I'm sure they can be improved.
> (Ross Ihaka did once mention an intent to include colour management into
> R, so that R RGB colours were rendered as accurately as possible in sRGB.
> That would remove the MacOS anomaly.)
> Brian D. Ripley, ripley at stats.ox.ac.uk
> Professor of Applied Statistics, http://www.stats.ox.ac.uk/~ripley/
> University of Oxford, Tel: +44 1865 272861 (self)
> 1 South Parks Road, +44 1865 272866 (PA)
> Oxford OX1 3TG, UK Fax: +44 1865 272595
More information about the R-help