Prof Brian D Ripley
Mon, 15 Nov 1999 09:00:06 +0000 (GMT)

On Mon, 15 Nov 1999, Martin Maechler wrote:

> >>>>> On Sun, 14 Nov 1999 23:10:26 +0000 (GMT), Prof Brian D Ripley
> >>>>> <> said:
>     BDR> On Sat, 13 Nov 1999, Jonathan Rougier wrote:
>     >> May I raise the issue of the behaviour of image()?  In image()
>     >> dim(z) is supposed to be c(length(x), length(y)).  This suggests
>     >> that the user is supplying the midpoints of the rectangles.  For all
>     >> but the outer rectangles this is true.  The outer rectangles,
>     >> however, are only displayed in half width.  So it is not possible to
>     >> show correctly an image plot of equally-spaced abscissa.  I would
>     >> hazard a guess that these kinds of plots make up the bulk of calls
>     >> to image() either directly or after a simple transformation (eg
>     >> starting with boundary values rather than midpoints).
>     >> 
>     >> Can I suggest that the code for image() be modified to show full
>     >> rectangles for all values on the basis that x and y are midpoints?
>     >> According to my code for R (which is a bit out of date!) the minimum
>     >> change would require 4 lines to be modified in the function do_image
>     >> of plot3d.c.  The alternative, which I think would be more
>     >> transparent, would be to call the function do_image with boundary
>     >> values having previously made the appropriate calculations in
>     >> image() according to whether midpoints or boundary values are
>     >> supplied.  Something like this happens in S+ (although not very
>     >> transparently!).
>     BDR> I needed this in hurry this weekend for some fMRI plots. I've
>     BDR> altered image so that if z is m x n, x can be of length m or m+1
>     BDR> and y of length n or n+1.  If x is of length m it supplies the
>     BDR> midpoints, if of length m+1 it supplies the boundaries of the
>     BDR> cells. (All S compatible.)  The internal call is made with the
>     BDR> boundary values.  The outer cells are plotted in full width.
> Excellent, thank you!
>     BDR> The one change is if m=1 or n=1. I reckon that is an error, but S
>     BDR> and R previously filled the whole width. Does anyone want that
>     BDR> behaviour?
> yes, I do.
> - If I remember properly, I even spent an hour or so to make this work in R:
>  When I was porting the wavethresh package,
>  one of the multiresolution plots didn't work in R, first, because  image()
>  didn't work for 1x1 images.
> - Also, for the matter of completeness and consistency, I think this just
>   should work.
>     BDR>   At present those cases are an error.
OK, I know how to do it and will implement it soon (but probably not

Brian D. Ripley,        
Professor of Applied Statistics,
University of Oxford,             Tel:  +44 1865 272861 (self)
1 South Parks Road,                     +44 1865 272860 (secr)
Oxford OX1 3TG, UK                Fax:  +44 1865 272595

r-devel mailing list -- Read
Send "info", "help", or "[un]subscribe"
(in the "body", not the subject !)  To: