— Insight —
Mathematical ideas are some of the most transformative and beautiful in history. So why do they get so little attention?
In the right hands, metadata is amazing.
— Tools and Techniques —
This case study by Michael Betancourt shows how to build and evaluate probabilistic models in Bayesian inference. This is a great read.
Camelot is a new Python library that makes it easy to extract tables from PDF files. There are already tools that extract data from PDFs but Camelot gives you more control for those times when things don't go perfectly. Check out the docs for info. They're awesome.
This Tips and Tricks post has been growing for a few months on the fast.ai blog and there are a lot of good ideas and resources here.
Brandon Rohrer's four-part series on optimization is a nice introduction that's easy to follow with lots of visuals and short videos.
— Datasets —
Some of the most useful datasets for studying and training models are being made available by fast.ai in a single place, using standard formats, on reliable and fast infrastructure.
— Data Viz —
ArviZ (pronounced "AR-vees") is a Python package for exploratory analysis of Bayesian models. Includes functions for posterior analysis, sample diagnostics, model checking, and comparison. A Quickstart Guide and "Cookbook" of examples make it easy to get started.
Sports analytics fans will love this new site by Nicholas Canova and Bryant O'Brien. Big League Graphs is where sports analytics and data visualization intersect. It's mostly NBA content right now but hockey and baseball visualizations are also being developed and the site will eventually include a variety of flagship visualizations for each league.