Notes taken during the lightning talks session EuroScipy 2010 conference in Paris. Most talks were about presenting a new scientific python module or software package. This was also an opportunity for the authors to make a call for contributions.
glumpy
A lightweight visualization library based on PyOpenGL and numpy. Does not attempt to replace Mayavi and serves a different purpose. A nice way to have fast animations for figures rendered with matplotlib.
joblib
A lightweight pipelining library. Syntatic sugar over the multiprocessing module
NPK 2.0
A nuclear magnetic resonance processing kernel. The current version uses a Java GUI with some Jython . It does not currently use numpy/scipy, but there is an ongoing rewrite in that direction.
fermhub
Multiple notebooks for SAGE.
peewit
An agnostic machine learning framework. Rather than focusing on the actual computations, peewit aims to do the housekeeping of running the experiments.
scikits.learn
A machine learning library, implements general-purpose ML algorithms. It has efficient bindings to libSVM and LibLinear. Many examples on the site.
From the audience : > How does it compare to > Orange?
Orange is a GUI over many algorithms. It also has many dependencies. scikits.learn aims to be an easy to install, easy to use API.
There are many machine learning libraries currently being developed. There should be an effort to unify everything.
Author does not disagree.
RST/Sphinx based thesis authoring tool-chain
This is more a wish than an actual implementation. LaTeX has many great features and beautiful typography, for text and equations, but is too complicated. Any self-respecting scientist won't use Word to publish her work.
reStructuredText and Sphinx fit nicely in the middle of this spectrum. However, as a tool originally developed for making documentation and manuals, it currently lacks some basic functionality to author scientific work. The most critical one being proper support for citations.
An RST-based presentation tool would also be nice. An audience member pointed to the rst2beamer package.
Personal note : It's not RST, but the markdown-based showoff presentation system is pretty enjoyable.
uncertainties
A python module to handle calculations on numbers with uncertainties. Uses error propagation theory.
upy
Another python module to handle error propagation and uncertainty in computations. Uses Gaussian error propagation.
SNPedia
An effort to centralize informations about SNPs ("snips").
For every SNP, SNPedia links to related research papers. The database is used by the Promethease reporting tool, which can import data data from genetic testing services like 23andme.
The slides are still available as a Google Docs presentation.
Interfacing with instruments for experimental physics
This is call for feedback on how python could help replacing Labview, to interface with measurements and control hardware used in experimental physics. Vendors rarely ships a python API. And a GUI tool like Labview does not shine to run batch experiments, where a script would be great.
Questions :
Why and when did you leave matlab and labview?
The speaker did not leave matlab. Only the "nightmare" (sic) that is Labview.
If you get a binary vendor DLL, why not use ctypes ?
ctypes works well with DLLs written in C, not so much when they are written in C++. In that case, you need to write a C wrapper first.
matplotlib and HTML5/Canvas
As of version 1.0, the popular python plotting library matplotlib got a
HTML5 Canvas backend. Really useful for SAGE. A nice step to make anything with a modern browser a decent platform for scientific work. The 3D plot demo wasn't really convincing though.