At Sandtable we use Python throughout our stack. We’re huge fans.
Our Data Scientists use a full scientific Python stack for exploratory data analysis, machine learning and prototyping agent-based models. At the other end, our platform for running large-scale experiments is written mainly in Python.
We really love Python’s flexibility and agility.
So when the opportunity arose to attend PyData 2014 here in London, we were very excited. It was also the first PyData outside of the US.
The conference was a packed three day event from Friday to Sunday (21-23 of February), and held at L39 at One Canada Square, Canary Wharf. As you can imagine, views from the 39th floor were spectacular! Cue photo:
The theme of the conference was Python and (Big) Data, with speakers talking on a range of topics, including machine learning, high-performance Python, and data visualisation.
The first day, Friday, was for tutorials: more hands-on sessions than the weekend’s. We particularly enjoyed the opening talk by Yves Hilpisch on Python and Financial Analytics, and the presentation by Bryan Van de Ven on Bokeh, which shows great promise.
On the second day, we thought Ian Ozvald did a great job elucidating on the high performance python landscape. Lessons learnt: profile (of course!), and Cython still good for the win. However, stay agile and be careful of technical debt. Also, some interesting, emerging projects to checkout: Numba, Shedskin, and Pythran. We’re looking forward to the book coming out: High Performance Python. Ian’s own write-up of the conference can be found here.
Later in the day, and something quite different, we particularly enjoyed one-eyed artist Eric Drass’ thoughtful and entertaining presentation on the mix of art and technology. Check out his topical piece: ‘Who watches the watchers?’ — there’s more than meets the eye, promise.
On Sunday, we found Gael Varoquaux’s keynote compelling. He mused on building a cutting edge data processing environment with limited resources. Gael reminded us that software development isn’t just about tools, it’s a social process. And, again, we received a warning about technical debt; we must plan for changes. Also, we found it fascinating hearing about the vision (reality?) of scikit-learn: ‘Machine learning without learning the machinery.’
Later we were mesmerised by James Powell’s whizz-bang tour of Python generators. A Python feature we’ll be sure to harness more in future.
In the afternoon, Bart Baddeley gave a very accessible introduction to similarity and clustering using scikit-learn.
Lots of great talks! We really enjoyed the event. Thanks to all those who organised it!