[R] floating point in single precision

[Sebastian Gibb]

Am Sonntag, 27. Juni 2010, 08:39:10 schrieb Prof Brian Ripley:
> On Sun, 27 Jun 2010, Sebastian Gibb wrote:
> > Hello,
> > 
> > is there a possibilty in R to convert numbers (double precision, 64bit)
> > into single precision ones (32bit).
> > I need that for compatibility reasons. Until now I call a C application
> > which casts a double to a float.
> > 
> > float precision32(double value) {
> > 
> >  return (float)value;
> > 
> > }
> > 
> > But I want to use a R only method. What can I do?
> 
> ?as.single
> ?writeBin
Thanks a lot for the hints.
I use the following code now:

### function double2singlePrecision
##  wrapper function for .changePrecision(x, size=4)
##
double2singlePrecision <- function(x) {
  stopifnot(is.double(x));
  return(.changePrecision(x, size=4));
}

### function .changePrecision
##  converts double values to double values in a given precision
##  (only correctly working for cut a higher precision to a lower one; e.g.
##  IEEE 754 double precision to IEEE 754 single precision)
##
.changePrecision <- function(x, size) {
  # create a raw object to avoid direct file access
  virtualCon <- raw();
  # write binary data to raw object and change (mostly cut) precision to size
  # size==4 # 32bit, single precision
  # size==8 # 64bit, double precision
  virtualCon <- writeBin(object=x, con=virtualCon, size=size);
  # re-read data
  x <- readBin(con=virtualCon, what=double(), size=size, n=length(x));
  return(x);
}

Kind regards,

Sebastian