Installing PyTables when you’re not root

By Koen van de Sande.

Warning

contents of this recipe recipe may be outdated.

This guide describes how to install PyTables and its dependencies on Linux or other *nix systems when your user account is not root. Installing the HDF5 [1] shared libraries and Python extensions NumArray and NumPy requires some non-trivial steps to work. We describe all steps needed. They only assumption is that you have Python 2.3 or higher and a C/C++ compiler (gcc) installed.

Installing HDF5

  • First go to or make a temporary folder where we can download and compile software. We’ll assume you’re in this temporary folder in the rest of this section.

  • Download hdf5-1.6.5.tar.gz from ftp://ftp.hdfgroup.org/HDF5/current16/src/:

    wget ftp://ftp.hdfgroup.org/HDF5/current16/src/hdf5-1.6.5.tar.gz
    
  • Extract the archive to the current folder:

    tar xzvf hdf5-1.6.5.tar.gz
    
  • Go to the extracted HDF5 folder:

    cd hdf5-1.6.5
    
  • Run the configure script:

    ./configure
    
  • Run make:

    make install
    
  • We’ve now compiled HDF5 [1] into the hdf5 folder inside the source tree. We’ll need to move this to its final location. For this guide, we’ll make a software folder inside your home directory to store installed libraries:

    mkdir ~/software
    
  • Move the files to the right location:

    mv hdf5 ~/software/
    

Installing NumArray

  • From the NumArray SourceForge page download NumArray 1.5.2 to our temporary folder.

  • Extract the archive:

    tar xzvf numarray-1.5.2.tar.gz
    
  • Go to the NumArray folder:

    cd numarray-1.5.2
    
  • Build and install the Python module into our software folder (it will actually end up in ~/software/lib/python:

    python setup.py install --home=~/software
    

    We will also need to copy the header files of NumArray so PyTables can use them later on for compilation. Skipping this step will lead to compilation errors for PyTables.

  • Go into the header file folder:

    cd include
    
  • Copy the header files. We’ll put them together with the HDF5 [1] header files:

    cp -r numarray ~/software/hdf5/include/
    

Installing NumPy (optional)

It is not required to install NumPy; PyTables will work with just NumArray installed. However, I do recommend that you install NumPy as well, because PyTables can optionally use it.

  • From the NumPy SourceForge page download NumPy 1.0 (at time of writing) to our temporary folder.

  • Extract the archive:

    tar xzvf numpy-1.0.tar.gz
    
  • Go to the NumPy folder:

    cd numpy-1.0
    
  • Build and install the Python module into our software folder:

    python setup.py install --home=~/software
    

Python wrapper script

We’ve installed all dependencies of PyTables. We need to create a wrapper script for Python to let PyTables actually find all these dependencies. Had we installed them as root, they’d be trivial to find, but now we need to help a bit.

  • Create a script with the following contents (I’ve called this script p on my machine):

    #!/bin/bash
    export PYTHONPATH=~/software/lib/python
    export HDF5_DIR=~/software/hdf5
    export LD_LIBRARY_PATH=~/software/lib/python/tables:~/software/hdf5/lib
    python $*
    
  • Make the script executable:

    chmod 755 p
    
  • Place the script somewhere on your path (for example, inside a folder called bin inside your home dir, which is normally added to the path automatically). If you do not add this script to your path, you’ll have to replace p in scripts below by the full path (and name of) your script, e.g. ~/pytablespython.sh if you called it pytablespython.sh and put it in your home dir.

  • Test your Python wrapper script:

    p
    
  • It should now start Python. And you should be able to import numarray (and optionally numpy) without errors:

    Python 2.3.4 (#1, Feb  2 2005, 12:11:53)
    [GCC 3.4.2 20041017 (Red Hat 3.4.2-6.fc3)] on linux2
    Type "help", "copyright", "credits" or "license" for more information.
    >>> import numarray
    >>> import numpy
    >>>
    

Note

you could do this differently by defining these environment settings somewhere in your startup scripts, but this wrapper script approach is cleaner.

Installing PyTables

  • From the SourceForge page download PyTables 1.3.3 (at time of writing) to our temporary folder.

  • Extract the archive:

    tar xzvf pytables-1.3.3.tar.gz
    
  • Go to the PyTables folder:

    cd pytables-1.3.3
    
  • Install PyTables using our wrapper script:

    p setup.py install --home=~/software
    
  • If you get the following error then you are not using the wrapper script properly!

    .. ERROR:: Can't find a local numarray Python installation.
       Please, read carefully the ``README`` file and remember that
       PyTables needs the numarray package to compile and run.}}}
    

Running Python with PyTables support

  • Use your Python wrapper script to start Python:

    p
    
  • You can now import tables without errors:

    Python 2.3.4 (#1, Feb  2 2005, 12:11:53)
    [GCC 3.4.2 20041017 (Red Hat 3.4.2-6.fc3)] on linux2
    Type "help", "copyright", "credits" or "license" for more information.
    >>> import tables
    >>> tables.__version__
    '1.3.3'
    >>>
    

Concluding remarks

  • It is safe to remove the temporary folder we have used in this guide, there are no dependencies on it.
  • This guide was written for and tested with HDF5 1.6.5, PyTables 1.3.3 and NumArray 1.5.2.

Enjoy working with PyTables!

Koen


[1](1, 2, 3) http://www.hdfgroup.org/HDF5