ISSUE 394 · July 5, 2022

Insight

We Take Our Units of Measurement for Granted

In a perfect world, you choose the unit of measurement that is optimal for the question you’re trying to answer in some analysis. But practically, more often than not, this isn’t the case. Take Randy Au’s advice and think twice next time you’re deciding on a unit of measurement.
Counting Stuff | Randy Au

Staff Data Scientist: Comments on Will Larson's Staff Engineer Book

There’s not a ton of writing out there about what it’s like filling a “staff-plus” data science role. What does it entail? What does it take to succeed? Harlan Harris answers both of these questions relating to Will Larson’s excellent book, Staff Engineer: Leadership beyond the management track.
Harlan D. Harris

Tutorials, Projects & Opinions

Causal Forecasting at Lyft

I really enjoyed this zoomed-out look at how Lyft thinks about causal forecasting. The post proposes thinking about each team as “a function which optimizes their metrics given a set of inputs” and through this lens, takes you on a tour of what that looks like for them.
Lyft Engineering | DJ Rich

A Beginner's Introduction to Mixed Effects Models

This wakthrough is a masterclass from Megan Hall on mixed effects models. If you’re interested in the statistics behind these models or how to implement them in R, check it out.
Meghan Hall

Geo-based A/B Testing at Instacart

This post from the Catalog Team at Instacart goes deep into how their new geo-based A/B testing system works, along with the “Difference-in-Difference” analysis that it enables. The explanations here are clear and easy to follow.
Tech at Instacart | Xiaoding Krause

Introducing ShopifyQL: Our New Commerce Data Querying Language

Shopify built out an intriguing new commerce-focused querying language called ShopifyQL. The idea here is to build abstractions on common SQL patterns for commerce so that anyone can easily access their data. Something tells me we’ll be seeing more of this across more use cases.
Shopify Data Science | Ranko Cupovic

Explore, develop, and grow with world-class Data Scientists

Want to join a trusted community of DS leaders from companies like Amazon, Stripe, Notion, and Asana? Join On Deck Data Science, a continuous community for ambitious Data Science leaders who want to maximize their impact and accelerate their careers alongside a highly-curated network of peers. Learn More and Apply Here.
// sponsored

Resources

Pen and Paper Exercises in Machine Learning

Sometimes, the best tools for thinking are just a pen and piece of paper. If you agree and want to put your machine learning knowledge to the test, you’re in luck. This paper has a bunch of exercises ranging from linear algebra to Markov models to variational inference.
arXiv | Michael U. Gutmann

Lightweight (Bayesian) Marketing Mix Modeling

Google open-sourced a Prophet-like library that allows users to easily train Bayesian Marketing Mix Models. If you’ve ever been posed with the question of how effective your marketing channels are, then MMMs could be of use. For more background, this talk on the subject from a few years back is a good one.
GitHub | Google