ISSUE 381 · April 5, 2022
In this week's issue, Roni Kobrosly continues his "Invited Topics" series on causal inference with a selection of core methods and tools. Roni is the head of data science at a Health Tech company in D.C. and he's also the creator of the open-source python package causal-curve, which offers a collection of tools to perform causal inference analysis.
In case you missed it, Part 1 of the series is in Issue 378 of Data Elixir, starting about halfway down.
Organizations
The “0/1/Done” Strategy for Data Science
To consistently achieve operational excellence, the data science team at News UK Technology uses a strategy they call "0/1/Done"... 0-day Handovers, 1-day Prototyping, and projects are declared "Done" when Completely Done. Here's how they work backwards from these goals to identify what's needed in terms of team, process, tooling & governance.
News UK Technology | Marios Perrakis
Your Data Initiatives Can’t Just Be for Data Scientists
Without buy-in from your organization’s rank and file, even the cleverest model will sit idle and “data-driven decision-making” will just go around in circles. Organizations need to start seeing regular people as part of their data strategy. When starting a new data project, start with two questions: 1) Who will this effect? And 2) How can we get them involved as soon as possible?
Harvard Business Review | Thomas C. Redman
Faster data analysis with PostgreSQL & TimescaleDB
Excel, R, and Python are great tools for data analysis, but may not be the best for every job. Fast-track your work with larger datasets of time-series data directly in your database, using PostgreSQL and TimescaleDB. Learn how.
// sponsored
Tutorials, Projects & Opinions
What does it mean to give someone what they want?
It's a mistake to think that the choices we make are the same as what we value. Or even what we need. This post surveys recent work in recommender systems and shows why recommenders often fail. Ultimately, Choices ≠ Preferences ≠ Welfare. Includes practical ideas for doing better.
Luke Thorburn, Jonathan Stray, Priyanjana Bengani
How to I read research papers
Great set of slides that walk through strategies and tools to use for reading papers — especially long technical papers. The tools outlined here will also help you keep track of key points so you don't forget them.
Learning Theory Alliance | Aaditya Ramdas
The Value of Descriptive Analytics
Companies invest a lot into analytics but are those investments valuable? In this paper, Ron Berman and Ayelet Israeli studied 1500 online retailers and found that retailers who used a descriptive dashboard did, indeed, increase their revenues — and by a substantial amount. The paper explores how.
Ron Berman, Ayelet Israeli
Text summarization with Hugging Face Transformers
This two-part tutorial is a practical guide to setting up and assessing the quality of text summarization models for your domain. It's intended for readers who have used Transformer models before, but are at the start of their text summarization journey and want to go deeper.
AWS ML Blog | Heiko Hotz
Good-Bye Digital Natives. Hello AI Natives.
TikTok understands what AI Natives want. Do you? AI is fundamentally changing the rules of business & creating a new consumer class. Learn about AI Natives & how you can win their loyalty. Download a free copy of Prolego’s ‘AI Natives Among Us’
research report.
// sponsored
Invited Topics: Causal Inference
Part 2: Core methods and tools
Causal inference can be a frustrating discipline to crack. It can be very theory-heavy, there isn't a clear, singular software library that nearly all practitioners use (e.g. `scikit-learn` for machine learning users), and oftentimes the same approach can go by different names depending on the company, field, or research team. The links below should give you a very solid first pass at the landscape of methods out there.
Part 1: Understanding the Fundamentals
Part 2: Causal inference: Selecting algorithms
In this multi-part series, Jane Huang and colleagues at Microsoft outline the critical effect measures to know in causal inference, the data assumptions necessary to employ causal inference in the first place, a comparison of the many open-source packages to carry out the job, and lay out a nice categorization of the algorithms that one can employ.
Data Science at Microsoft
Implementation of G-Computation on simulated data
In this excellent tutorial paper, Snowden and colleagues describe one of the more straight-forward modeling approaches for causal inference. This approachable paper has a FAQ and comes with code if the reader is interested in seeing how the authors arrived at their numbers. It's an approach that performs well against
others in simulation studies too.
Jonathan M. Snowden, Sherri Rose, Kathleen M. Mortimer
Trustworthy Online Controlled Experiments:
A Practical Guide to A/B Testing
This gem of a book was written by the current industry giants in A/B testing and evaluating treatment effects. Understanding the methods and metrics of A/B testing will give you a fantastic intuition for various causal inference approaches. Chapter One is available online as a free PDF on the book's website.
Ron Kohavi, Diane Tang, Ya Xu
Outlier
Visualization experiments with Svelte.js and Three.js
There's no code here and it's not a tutorial. But wow! These are beautifully designed, immersive data visualization experiments.
cybernetic
 |