Installing pySIML

Prerequisites

• pySIML requires a working Python installation, since it is a collection of Python bindings. It has been successfully tested on Python 2.4 and 2.5 on both Linux and Mac OS X. It also requires that the development header files for the Python interpreter used be installed on the machine (e.g., on Ubuntu Linux, package python-dev must be installed, not just python).
• The NumPy package, and its headers, must also be installed. pySIML makes extensive use of NumPy to store input and output data for the SIML algorithm.
• An OpenMP-capable compiler is required to take advantage of multiple CPUs (parallel computations over multiple rows of a Tanimoto matrix in CPULingo).
• PyCUDA version 0.94 or greater is required for NVIDIA GPU support using GPULingo. Versions 0.93 and previous will not work properly! Note that at the time of this writing, 0.93 is the most recent release version. If this is still true, then you must retrieve a copy of the PyCUDA source code from the source repository (following the directions given on the PyCUDA homepage).
• PyOpenCL is required for NVIDIA/AMD GPU support using OCLLingo.

Setup Procedure

pySIML is distributed as a source tarball using a mostly-standard Python distutils-based setup procedure. After untarring the package, most people should be able to run:

python setup.py build
sudo python setup.py install

In some cases, the setup script will not be able to detect one or more settings properly, in which case, the configure option can be used:

python setup.py configure <options>

The following options are available:

• --enable-openmp: Force pySIML to be built with OpenMP support.
• --disable-openmp: Force pySIML to be built without OpenMP support.
• --numpy-include=<dir>: Indicates that the C headers for numpy can be found in <dir>. Note that if, for example, arrayobject.h is in /usr/include/python2.5/numpy/arrayobject.h, <dir> should be specified as /usr/include/python2.5, NOT /usr/include/python2.5/numpy.
• --python-include=<dir>: Indicates that the C headers for Python (e.g., Python.h) can be found in <dir>.