3 Takeaways from RStudio Conference 2020

In late January, Brad Kossmann and I were  fortunate enough to represent our Data Services’ Data Science team, and attend the annual RStudio conference in San Francisco. While there were many things worthy of receiving a gold sticker, the following are the three things that had me leaving the conference EXTRA excited to continue my work in Data Science.

Organizationally Leveraging Data Science Teams

From the production track at the conference, I was enamored with a common pattern that organizations are using to form their data science teams. These organizations are horizontally integrating their data science teams, resulting in more agile teams that contain data scientists, engineers, and DevOps specialists. This type of team structure creates an immense amount of team agency because they have the ability to tackle the end to end project lifecycle: Exploration/Ideation, Infrastructure/Data Pipeline, Modeling, Deployment, & Maintenance – without having to reach out to other parts of the organization for support.

RStudio Connect

There were numerous examples of RStudio Connect serving as a primary component for data science infrastructure within organizations. Its ability to support Data Scientists wielding either R or Python was impressive. However, what stood out most to me was how easy it was to deploy data science artifacts to the platform. Whether that was APIs (Via Plumber), Datasets (Via Pins),  or scheduling and delivering dynamically constructed Rmarkdown reports. Ultimately, my takeaway was that RStudio Connect looks to be the real deal and provides a unified platform for data scientists to collaborate and deploy their work.

Tidymodels Ecosystem is Rapidly Improving

It feels like it was just yesterday that Max Khun (@topepos) was hired at RStudio and the Tidymodels ecosystem was announced, but it’s actually been just over 3 years now!

Max Khun and I at RStudio Conference 2020I’m happy I put his session on my top 3 most anticipated list, because it highlighted the strides the ecosystem has made over the last year. During his session he walked us through using Parnsip as the primary model library in unison with recipes (preprocessing), and tune for hyperparameter tuning. I was particularly impressed with the level of control and nuance available for hyperparameter via tune. It looks easily approachable for beginners in ML, especially with its “sensible defaults” for different model parameters. Overall, Max and team are making incredible strides to creating a unified ecosystem for modeling with tidy data design principles.

See how our team is leveraging Data Science every day here!