Python has a large collection of tools for scientific computing. However, finding the right pieces and assembling them into a fast and scale-able app can be a daunting task. This talk will explore common requirements of scientific apps and how to fulfill those from the Python ecosystem. It will also provide a blueprint for building applications using tools like PyQt, PyQwt, NumPy, and HDF5.
Python is a common tool for serious scientists of all disciplines. However, many people outside the scientific and Python communities still fail to see Python's potential in the world of scientific computing. This talk will provide you with the background for starting to prototype your own scientific applications and showing Python doubters the potential power of the ecosystem.
This talk will give a few solid reasons for considering Python for a high performance scientific application. For example, many novices don't know that Python plays very nicely with many accepted 'fast' languages such as C and Fortran. Both of these languages are considered cornerstones of the scientific computing world and by interfacing with them Python can instantly tap into fast, robust libraries.
In fact, the SciPy package is built directly on top of some very robust Fortran and C libraries. So not all of your number crunching needs to be in Python. You can use Python to prototype quickly and build infrastructure while still utilizing these faster languages for some of the heavy duty work.
Another benefit of using Python is easy access to lots of open source GUI and plotting libraries such as PyQt, matplotlib, and PyQwt. These projects allow you to easily visualize your data so you can spent more time reasoning about the cutting-edge data instead of plotting it. In addition, all of these projects are cross-platform so your users can be comfortable in their own computing environments.
However, just knowing some of the tools available only gets you so far. So this talk will also explain how to use these tools to piece together a typical MVC (Model-View-Controller) application. So, you'll see how you can store massive amounts of data on disk with the fast HDF5 binary format, manipulate that data easily in memory with the NumPy, and then use the PyQwt plotting tools to quickly display nice customizable graphs.
The presentation requires landslide
to view. This can be installed via pip and the included requirements.txt
file (pip install -r requirements.txt).
The code samples require the following libraries. However, there is no provided requirements file because some of these require C-extentions and installation can be somewhat involved. Please see the associated websites for information about installing on your platform.
Please review the Requirements section of this document for what packages to install.
You can run the main sample application by doing the following:
cdinto code_samples/python controller.py
There is also a profiling script for Pandas that can be executed with the following:
cdinto code_samples/python profiling_pandas_data_import.py
You can view a live version of the presentation here.