On 4/3/06, David E. Konerding <dekonerding@lbl.gov> wrote:
Hello,
Recently I have been converting most of my scientific data to the HDF5 format. HDF5 is a powerful, high performance data format library. I found an absolutely wonderful open source code, PyTables, which provides an HDF5 interface. It adds enormous value to HDF5 by making it much easier to use, and exposing a lot of the complex functionality that HDF5 supports (such as variable-length records).
I would really appreciate it if you guys would consider adding PyTables. I would definitely convert a lot of my existing extensions and processing tools to use PyTables- for example, when I analyze trajectories to find hydrogen bond data and use that to render long movies, PyTables makes it a lot easier (I already used the Chimera headers to build my own copy of PyTables on Linux, but it would be a lot more painful to do on Windows).
Actually windows is pretty much binary compatible, so just download the tables-1.3.win32-py2.4.exe and copy the module into chimera's site-packages folder. I have done this with numpy and matplotlib. You'll have to also get the hdf5 just as if you were normally installing on windows. Info found here, http://www.pytables.org/docs/manual/x457.html .
PyTables depends on numarray (although it also provides Numeric Python support). I'd like to see numarray in Chimera as well (although I could live with just Numeric support)
PyTables also supports numpy now, which is the new agreed pyarray package. I have been using numpy in chimera with no problems. The one hitch on OSX is that numpy and Numeric can't be imported in the same script (complaints about multiarray.so). It's not a problem on other platforms. I believe I made an informal request on the dev list to migrate completely from Numeric to numpy, but this probably doesn't make sense until the numpy 1.0 release. Actually the OSX issue brings up another feature request. I have a plugin that creates a surface as seen in, http://www.cgl.ucsf.edu/chimera/docs/ProgrammersGuide/Reference/surface.html . Could this approach please support float32 numpy arrays as well? I have to do a numpy->Numeric hack in windows/linux, but I can't do this in OSX due to the import issue. Thanks, Charlie