PyData 2016 Madrid

Recently I have attended to the PyData 2016 conference held in Madrid. It has been a two-days single-tracked conference covering aspects of Python and data science. I will try to share some links and comments of that have been more relevant to me.

Christine Doig (@ch_doing) opened the conference with a keynote entitled The Hitchhiker’s Guide to Data Science. Christine works at Continuum, one of the companies leading the data science revolution in the Python ecosystem. In her talk, she introduced the new products and services that Continuum is delivering. Many interesting things, such as:

  • Math Kernel Library optimizations for supporting vectorized math routines hardware-accelerated by the Intel Math Kernel Library, now shipped in the Anaconda distribution

  • Anaconda navigator, a GUI for managing the packages, environments and channels

  • Conda-forge: A community-curated collection of conda recipes and distributions

  • Support for R language as a first-class citizen in Anaconda and Jupyter

  • A Jupyter Notebook extension named nbpresent for creating slideshows based on your notebooks

  • A Bokeh server to create beautiful visualizations

  • Support for Numba for enabling JIT compilation of code and Dask for distributed computing using the NumPy and Pandas interface

Juan Luis Cano (@astrojuanlu) talked about packaging software with Conda. Very interesting talk. I have really liked the idea of using Continuous Integration in data science, which I think that is not (unfortunately) a mainstream practice.

Manuel Garrido (@manugarri) explained A primer on recommendation systems. The talk is great if you want to get hands-on with a simple yet functional first approximation to a recommendation system, and to grasp the concepts behind them.

Tomás Gómez Álvarez-Arenas spoke about inverse problems and how to solve them using Python. Maybe you didn’t know that you can reproduce the results of the LIGO findings using a Jupyter Notebook and the data publicly released. Whoa!

Jaime Fernández gave a comprehensive talk on The future of NumPy indexing. Fancy indexing, orthogonal indexing, it all makes sense when Jaime explain them, including the three indexers (orthogonal, vectorized and legacy) that will eventually be incorporated to NumPy codebase.

Claudia Guirao (@claudiaguirao) spoke about Whoosh: a fast pure-Python search engine library. This talk shows that it is important to find the best tool for the job, and sometimes a simple search engine can be a better fit than a commercial-grade but complex one.

The team behind AeroPython presented Remove before flight: Analysing flight safety data with Python. Really cool Pandas magic to do data massaging with aerial accidents data.

In the section of the lightning talks:

Thanks for reading!