— Insight —
At the recent Rev event, Chris Wiggins, the Chief Data Scientist at The New York Times, described how his team re-frames real-world questions as machine learning tasks and executes on those tasks to deploy working solutions. This is a great talk with a lot of practical insights that extends beyond the newsroom. This post provides distilled highlights of the talk, a transcript, and a video of the session.
Is your organization still “data-driven” or has it become “AI-driven?” It's not just semantics and this new essay by Eric Colson is sure to spur debate about the value of human contributions in business decision-making.
— How-to —
What would have happened in the 2016 election had every eligible voter actually turned out? This post by G. Elliott Morris walks through how The Economist’s data team answered this surprisingly difficult question.
This tutorial explores unsupervised anomaly detection for univariate and multivariate data. Covers a variety of detection strategies with python code snippets and screenshots.
— Resources —
Nice collection of mostly pure NumPy implementations of machine learning models. These are bare-bones implementations and aren't optimized to be efficient. They're optimized to be clear and useful for understanding how they work.
fast.ai's latest course is a code-first introduction to NLP. The course covers a variety of traditional NLP topics, recent neural network approaches, and current ethical issues, such as bias and disinformation. The course includes a library of Jupyter Notebooks, video lectures, and an online forum. This is a free course with modest pre-requisites.
— Data Viz —
Coming up next in Lisa Charlotte Rost's Data Vis Book Club is an online discussion of “Observe, Collect, Draw!” by Giorgia Lupi and Stefanie Posavec. The format is a live online discussion that will happen on August 20th and the authors are planning to participate! Check out the announcement for details.