Thrifty Machine Learning

August 01, 2020

We live with an abundance of ML resources; from open source tools, to GPU workstations, to cloud-hosted autoML. What’s more, the lines between AI research and everyday ML have blurred; you can recreate a state-of-the-art model from arxiv papers at home. But can you afford to? In this post, we explore ways to recession-proof your ML process without sacrificing on accuracy, explainability, or value.

Why UX will Get Worse Before it Gets Better

July 19, 2020

The things we make are not user-friendly by accident; we have to make them that way. And that’s hard. For most of the time people have been making things, we’ve been mainly concerned with making them user-friendly for just us. And, even if the last few decades of app development have brought more focus to the importance of empathy (especially if it helps you capture market share), the last few years have forced app developers to acknowledge the continued pervasiveness of white privilege, gender privilege, and class privilege in UI/UX. Unfortunately, as our efforts to become more intentional and empathetic as technologists continue in the coming years, there is another problem we have barely considered that will become a major threat to compassionate, egalitarian, and even subversive app development. Spoiler: it’s our cloud infrastructure.

Subplots with Yellowbrick

June 08, 2020

In this quick post, we’ll see how to use Yellowbrick together with Matplotlib’s subplots feature to create a visual story about the modeling process:

The Music and Mayhem of Machine Translation

April 26, 2020

Conversational context is more important than the precision of the generalized translation model. Leveraging conversation history requires a model that can “remember” what the speakers have been talking about, making it an ideal task for sequence-to-sequence models. But what about conversations that traverse two (or more) languages?

Designing Democratic APIs

December 18, 2019

When we set out to create the open source machine learning diagnostics library Yellowbrick, we were faced with tough decisions: will our users mostly be experienced ML practitioners or beginners? Should we prioritize ease of use or ease of contribution? How best to coordinate between the existing APIs of our two main dependencies, scikit-learn and Matplotlib? Here we walk through our decisions, including some of our biggest challenges, successes, and lessons learned along the way.