ISSUE 385  ·   May 3, 2022

Insight

Making Data Actionable

Great data products should be easily actionable. It's not enough that users may be "data-driven." Where are they going? Why are they going there? Are there other routes? This is a short, practical post for helping your users get what they really need from your data.
Eric Weber

σ-Driven Project Management: When is the Optimal Time to Give Up?

σ-Driven Project Management: When is the Optimal Time to Give Up?
We all know that projects (often) take longer than we estimate. This is especially true for more research-oriented projects, as is often the case in data science. Erik Bernhardsson has developed a toy log-normal model for these estimates, and expands on it here by defining the “blowup factor” of a given project.
Erik Bernhardsson

The 3 Mental Models Model for Data Visualization

I hadn’t thought of how data visualization could benefit from mental models before, but thankfully Enrico Bertini has. In short, understanding how visualization works means better results. There are plenty of heuristics in here that might make you think twice next time you plot() data for an audience.
Enrico Bertini

Sponsored Link

Introducing HARI - The First Annotation And Dataset Analytics Platform

Achieve significant cost and time savings by improving your machine learning models with HARI, a semi-automated user interface for data curation, annotation, and model analysis of vision-based datasets.

Tutorials, Projects & Opinions

Comparing quantiles at scale in online A/B-testing

This post from Spotify’s engineering blog covers their latest approach to more effective experimentation: Poisson bootstrap difference-in-quantiles confidence intervals. How does the algorithm work and more importantly for their case, how do you scale it?
Spotify R&D

tidypandas

tidypandas is a new tidyverse-inspired python package that uses a tidy data frame and lets you use tidy verbs such as select, filter, summarize, etc. Follow the links at the top for a tutorial and tour of features.
GitHub | Srikanth

Graphing the Blogsphere

This is a cool idea for a project resulting in a fascinating interactive network visualization that you can browse around as you please. Furthermore, Jacob shares the process of collecting and processing the data, along with the code on Github.
Jacob Wood

How is Felix Today?

What if you put your entire life into a database? What types of data questions would you want to dig into? If your interest is piqued, Felix’s dashboard and write-up are well-worth exploring. Oh and the code is on Github. 
Felix Krause

The New Standard in Dataflow Automation

Join our community and see why 20,000 other data engineers, scientists and analysts use Prefect to confidently build, run and monitor robust pipelines at scale.
// sponsored

Machine Learning

How to use big AI models (Like GPT-3) in your startup

Many problems we face don’t require the likes of AI and deep learning, but when one does, how should you practically put these models into production? Fear not: there’s lots of solid advice in this post around the opportunities, limitations, and risks of going down the rabbit hole.
a16z future | Elliot Turner

Machine Learning Has a Validity Problem

Ben Recht dives into distribution shift in this introduction. Distribution shift describes whenever the data on which a machine learning algorithm is deployed differs from the data it was trained on. Turns out, this happens far more than we think.
Ben Recht

Career

Transitioning From Academia to Industry

The academia to industry path is a popular one for data scientists and is often cited as quite the jump. Danielle Navarro outlines her experience here with lots of practical advice for anyone thinking about a similar transition.
Danielle Navarro

How did you like this issue of Data Elixir?

👎    1     2     3     4     5   👍

Was this email forwarded to you?  Sign up here >>