Last week, I had the pleasure of speaking at SciPy 2017 in Austin about my work on modeling trends in music. Thanks to the SciPy committee and Enthought for hosting a wonderful conference with excellent speakers, and to everyone who postponed happy hour just a bit longer to join me and ask some great questions.
Here's my talk. I hope you enjoy the memes and bad puns as much as I did!
I especially enjoyed these other talks from the conference:
- Andrew Therriault's discussion of sustainable machine learning models;
- Jake VanderPlas's candid and thorough comparison of distributed systems used for image analysis;
- Jim Crist's extremely useful advice on debugging and profiling Dask;
- Gaël Varoquaux's thoughts on scientific software development (worth a listen, since he's an expert on the subject); and
- Peter Skipper's introduction to propensity matching as a tool for observational studies.
Like Andrew, I'm also very excited to check out Andy Mueller and Alexandre Gram's tutorial on scikit-learn. The comprehensiveness of the library's documentation allows it to double as a text for an introductory machine learning course, and the only possible improvement at this point is a detailed walkthrough of those docs by two core committers.
I've also bookmarked Jim Crist's tutorial on Dask. The Python community seems to be consolidating around Dask as the "native Python" solution for distributed work. While I only have experience working on HDFS-based teams, it's always important to understand other solutions.
In addition to these quality, hand-curated video selections, you can find the rest of the talks here.