Python in Scientific Computing

Thursday July 6, 2006

In recent years there has been increased interest in using the scripting language Python for scientific programming. As well as Python packages such as SciPy (Scientific Python) there are several applications where Python is used to create a more user friendly and application specific environment for accessing highly tuned code such as linear solvers. The user can pose their problem in a natural way using Python module extensions and the interaction with sophisticated numerical packages or visualization is handled by the Python bindings.

A one day workshop on the use of Python in scientific computing was held on July 6, 2006 at James Cook University, Townsville, following the CTAC'06 conference.

There is to be no formal publication of proceedings of this workshop but presentations will be recorded informally at this site.

The organising committee includes:

• Markus Hegland, ANU
• Steve Roberts, ANU
• Lutz Gross, ESSCC, UQ
• Ole Nielsen, Geoscience Australia

The program for the day is as follows:

• Introduction to Python
  
  Slides and examples for the introductory sessions can be seen here.
  
  9-9:30: Introductory talk by Ole Nielsen: The Python language
  
  9:30-10:30: Hands on self-paced tutorial including Numeric and Matplotlib.
  
  10:30-10:45 buffer or coffee
  
  10:45-11:15: Presentation: Summary and advanced topics:
  • Generators
  • Callable objects
  • Weak typing
  
  
  11:15-12:00: Stephen Roberts to talk about parallelisation
  Python and MPI
  Pypar is an efficient but easy-to-use module that allows programs/scripts written in the Python programming language to run in parallel on multiple processors and communicate using message passing. Pypar provides bindings to an important subset of the message passing interface standard MPI.
  Details on PyPar are here.
  
  
• Applications : escript and Finley, Laurent Bourgouin ESSCC University of Qld.
  
  The modeling library escript has been developed as a module extension of the scripting language Python to facilitate the rapid development of 3D parallel simulations on the Altix 3700. The finite element kernel library, Finley, has been specifically designed for solving large scale problems on ccNUMA architectures and has been incorporated as a differential equation solver into escript. In the escript programming model, Python scripts orchestrate numerical algorithms that are parallelised behind the scenes in escript module calls, without low level technical threading implementation by the user. In particular, the Finley module extends the escript functionality of data array abstraction with functions that assemble a system of linear equations from a linear boundary value problem defined over a mesh.
  
  
• Applications:Solving the Master Equations in Python, Markus Hegland ANU

  Slides from the talk are here.

  The investigation of biochemical reactions with low copy numbers but many different species poses a collection of computational problems. I will discuss some of these problems and some solutions using Python. We chose Python as a vehicle as with a low initial investment it provides returns right from the start. Several basic concepts like object oriented programming, list processing and binding in Fortran or C code will be illustrated.

  I will start with the implementation of a sparse grid function space for modeling probability distributions on a discrete but infinite and high dimensional state space. Besides the infinity of the state space one also has to overcome the curse of dimensionality -- with sparse grids of course. Then I will talk about how we bind in the Expokit code by Roger Sidje for the solution of the master equations which are large linear ordinary differential equations. Finally I will cover an example of the application and implementation of a simple Python matrix class using a Python wrapper (Lineal) around PetSC which was created using pyrex and pyxelator by Simon Burton.
  
  

• Applications: PyMCT and PyCPL: Refactoring the Community Climate System Model Using Python
  Authors: Michael Tobis, Michael Steder, Raymond T. Pierrehumbert (Department of Geoscience, U. Chicago), Robert L. Jacob (Argonne National Laboratory), J. Walter Larson (speaker; ANU Supercomputer Facility, The Australian National University), and Everest T. Ong (Department of Atmospheric and Oceanic Sciences, University of Wisconsin)

  Slides of the talk are here.

  Abstract:
  Multiphysics models such as coupled climate models are typically built from numerous separately-developed codes that are integrated into a single mutually interacting system. The couplings between these codes pose their own software architecture challenges, and object oriented programming (OOP) is one tool for conquering this complexity. Python is the language of choice for fast prototyping and assembly of codes from multiple components. Python is superior to languages such as C and Fortran in its support of OOP. We have pursued this goal for version 3.0 of the Community Climate System Model (CCSM), a coupled climate model with a large international academic user base. CCSM uses a hub-and-spokes architecture, with its subsystem models (atmosphere, ocean, sea-ice and land surface existing as 'spokes' that communicate via a central hub (the flux coupler). CCSM's coupler, CPL6 is a fairly flexible, modular, and extensible hub, and a dramatic improvement in these aspects over its predecessor CPL5 found in CCSM 2.0. Still, CPL6 is not sufficiently flexible nor configurable for many types of climate studies that might be pursued using CCSM. A Python implementation of CPL6, with better exploitation of object-oriented programming techniques can meet these requirements. In this presentation we report on a proof-of-concept Python implementation of CPL6--pyCPL--and describe the technical challenges surmounted in its realization. We will discuss a Python API for CCSM's coupling middleware, the Model Coupling Toolkit (MCT), pyMCT, outline how pyMCT was used to build pyCPL, and subsequently how pyCPL has enabled a prototype Python-assembled CCSM.