R devel - Nov 2003 - wishlist item: changing origin of plot (PR#5045)

[This is an edited version of an email that I sent to Paul Murrell.
He was in favour of the idea (although he noted the obstacles that: it
could make par() longer;  somebody has to implement it) and
suggested I submit this as a wishlist item.]

Do you think it would be worth adding options to plot commands to
reverse axes?

Here is a simple example:
 
> x <- 1:12
> plot(x)
  
Whereas if I want the x-axis reversed, I need to know the range of
the x data points:
 
>  plot(x, xlim=c(12,1))
 
So, maybe something like:
 
plot(x, x.rev=TRUE)
 
could be implemented.  X.REV could default to FALSE, and then 
plot.default could have something like:
 
 xlim <- if (is.null(xlim)) {
 	if (x.rev)
 		rev( range(xy$x[is.finite(xy$x)]))
 	else 
 	        range(xy$x[is.finite(xy$x)])
 }
 
So, it would reverse the xlimits only if xlim was not specified.
Likewise, code could be added for ylim.
 
The alternative that Paul suggested was to add an option "origin" to
par(), rather than flags x.rev and y.rev.  Presumably then origin
could take values something like "top-left", "bottom-left",
"top-right", "bottom-right".  (However, this will get longer
for
specifying all corners for 3-d plots.)

Stephen Eglen

On Wed, 12 Nov 2003 10:24:28 +0100 (CET), you wrote:
>Do you think it would be worth adding options to plot commands to
>reverse axes?
>
>Here is a simple example:
> 
>> x <- 1:12
>> plot(x)
>  
>Whereas if I want the x-axis reversed, I need to know the range of
>the x data points:
> 
>>  plot(x, xlim=c(12,1))
This seems like a fairly obscure need, and it can be done by
calculating the range of x at plot time, e.g.

 plot(x, y, xlim=rev(range(x)))

or

 plot(y, xlim=c(length(y),1))

I'd be happier with including one of those in the example code rather
than with adding another couple of parameters to par.

Duncan Murdoch

>>>>> "Paul" == Paul Murrell
<p.murrell@auckland.ac.nz>
>>>>>     on Fri, 14 Nov 2003 09:21:12 +1300 writes:
    Paul> Hi
    Paul> Duncan Murdoch wrote:
    >> On Wed, 12 Nov 2003 10:24:28 +0100 (CET), you wrote:
    >> 
    >> 
    >>> Do you think it would be worth adding options to plot commands
to
    >>> reverse axes?
    >>> 
    >>> Here is a simple example:
    >>> 
    >>> 
    >>>> x <- 1:12
    >>>> plot(x)
    >>> 
    >>> 
    >>> Whereas if I want the x-axis reversed, I need to know the range
of
    >>> the x data points:
    >>> 
    >>> 
    >>>> plot(x, xlim=c(12,1))
    >>> 
    >> 
    >> This seems like a fairly obscure need, and it can be done by
    >> calculating the range of x at plot time, e.g.
    >> 
    >> plot(x, y, xlim=rev(range(x)))
    >> 
    >> or
    >> 
    >> plot(y, xlim=c(length(y),1))


    Paul> That assumes that you know how the axis range is going
    Paul> to be calculated.  If a plotting function (barplot
    Paul> springs to mind) calculates the axis range internally
    Paul> then the user has a hard task to try and emulate that
    Paul> behaviour (in order to rev() it).  OTOH, the same
    Paul> point raises the potential problem that LOTS of
    Paul> plotting functions would possibly have to be updated
    Paul> because they assume that the x-scale goes
    Paul> left-to-right (here interaction.plot springs to mind
    Paul> because it allocates space within the plot for a
    Paul> legend).

    Paul> Paul

    >> I'd be happier with including one of those in the example code
rather
    >> than with adding another couple of parameters to par.
    >> 
    >> Duncan Murdoch

StephenE> So, maybe something like:
StephenE>  
StephenE> plot(x, x.rev=TRUE)
StephenE>  
StephenE> could be implemented.  X.REV could default to FALSE, and then 
StephenE> plot.default could have something like:
StephenE>  
StephenE>  xlim <- if (is.null(xlim)) {
StephenE>  	if (x.rev)
StephenE>  		rev( range(xy$x[is.finite(xy$x)]))
StephenE>  	else 
StephenE>  	        range(xy$x[is.finite(xy$x)])
StephenE>  }
StephenE>  
StephenE> So, it would reverse the xlimits only if xlim was not specified.
StephenE> Likewise, code could be added for ylim.

A much simpler alternative (and more in line with Stephen's
original whish ?!) is to only make this a new argument
for plot.default();  no par()'s at all.

This would make the argument work correctly for quite a few plot
methods that pass their "..." to plot.default(), but might not
be clean enough, since some things won't work {as e.g. Paul has mentioned}.

Martin