ISSUE 357 · October 12, 2021
In the News
State of AI Report 2021
This slide deck is one of the most comprehensive resources you'll find about the current state of AI. Explores a variety of topics including AI research, talent supply, commercial applications, and predictions for the coming year. It's dense but it's a very visual deck that's easy to follow.
Nathan Benaich and Ian Hogarth
Covid response hampered by population data glitches
Great Twitter thread about bad Covid data and how it's costing lives around the world. It turns out that many countries don't even know how many residents they have. This thread breaks down exactly how that's a problem and how to think about the data you see in the news, reports, papers, etc.
Twitter | John Burn-Murdoch
Sponsored Link
Distilled Insights from Data Leaders
The past year has been marked by a great acceleration. As a result, becoming data-driven is a priority for every organization. In this white paper, you will find distilled insights in data transformation from enterprise CDOs to data startup CEOs shared on the most useful episodes of DataCamp’s DataFramed podcast. Download now.
Tutorials, Projects & Opinions
Interpreting A/B test results: false positives and statistical significance
Nice introduction to false positives and statistical significance with simple examples along the way to build up intuition. This is the 3rd post of a multi-part series on how Netflix uses A/B tests to inform decisions.
Netflix Technology Blog
Choosing the right KPIs to evaluate your models
This first of a two-part series introduces a framework for choosing the "right" KPIs to evaluate your ML models. In Part 2, Amir Dolev explores a related question: how do you know if your model is actually correct?
Riskified | Amir Dolev
Pull Request Flow with usethis
If you use git and GitHub to collaborate on R packages or projects, then the "pr_" helper functions in the usethis package are super helpful. This flowchart/cheatsheet is a great reference that shows how.
R Markdown Source | Garrick Aden‑Buie
Cracking the Mystery of Deep Learning
To help them explain the shocking success of deep neural networks, researchers are turning to older but better-understood models of ML.
Quanta Magazine | Anil Ananthaswamy
Activity Schema: a faster, simpler data model
An activity schema enables faster querying, easier maintenance, and simpler governance. Create models with no dependencies or foreign keys, and query any combination of them to build all necessary tables for BI and analysis. Narrator is the only platform for building, maintaining, and querying an activity schema.
// sponsored
Resources
Bayesian Optimization Book
This text on Bayesian optimization aims to provide a self-contained and comprehensive introduction to Bayesian optimization, starting from scratch and carefully developing all the key ideas along the way. The intended audience is graduate students and researchers in machine learning, statistics, and related fields.
Roman Garnett
Data Visualization
Map Projection Playground
There are other places on the web that let you play with map projections but this new tool offers a wide selection of customization options, the code is easily edited, and the output can be exported as svg files for further editing in Illustrator or Inkscape.
Observable | Florian Ledermann
Robservable
This package allows Observable notebooks to be used as htmlwidgets in R. It doesn't embed an entire notebook but rather, it lets you choose which cells to display, update cell values from R, and add observers to get cell values back into a Shiny application.
Julien Barnier, Kenton Russell
 |