[Rd] HDF5 compound data types and h5r/hdf5 R packages

Brian G. Peterson brian at braverock.com
Thu Dec 1 11:41:39 CET 2011


On Thu, 2011-12-01 at 19:22 +1300, Jeremy Reeve wrote:
> Hi there,
> 
> 
> I have a Java process that writes HDF5 files with the following
> approximate structure:
> 
> group "xxx" {
> 
>     group "yyy" {
> 
>         dataset {}
>         dataset {}
> 
>     }
> 
>     group "zzz" {
> 
>         dataset {}
>         dataset {}
> 
>     }
> 
> }
> 
> where dataset is a rank one dataspace having a compound datatype defined as:
> 
> H5T_UNIX_TIME, float, float, float, float
> 
> I have tried R packages h5r and hdf5 in an attempt to read the file
> but examining the source of h5r and reading documentation for hdf5
> here:  http://xweb.geos.ed.ac.uk/~hcp/Rhdf5.html  leads me to believe
> that compound datatypes are not supported by these packages.  My guess
> is that mapping arbitrary type definitions in HDF5 to available types
> in R might be somewhat tricky.  Incidentally, h5dump has trouble
> displaying the data in the dataset but I think this is to do with the
> time 'field':
> 
>       DATASPACE  SIMPLE { ( 33 ) / ( H5S_UNLIMITED ) }
>       DATA {
>          h5dump error: unable to print data
>       }
> 
> Since I am new to both HDF5 and R I wonder if there are better
> approaches to storing the information I have that will allow me to use
> either h5r or hdf5 packages unmodified.  I expect I can contribute
> changes to either of the packages that will allow me to do what I
> describe but fall short of general compound datatype support.
> 
> Any comments or advice gratefully received.
> 
> Regards...Jeremy

You should contact the maintainers of the two hdf5 packages you
reference.  Hopefully one of them will be open to assisting you,
advising of your immediate problem, and collaborating on extension of
the package. 

If that doesn't work, perhaps the R Bioconductor list would be a better
place for the discussion, since I believe that HDF5 files are mostly
used in the geological and biological sciences.

Regards,

   - Brian

-- 
Brian G. Peterson
http://braverock.com/brian/
Ph: 773-459-4973
IM: bgpbraverock



More information about the R-devel mailing list