Open Sourcing deck.gl 4.0: Uber Engineering’s Framework for Advanced Data Visualization

Data is mission-critical to the advancement of any technology, but it loses value if we are unable to understand and visualize it.

In 2016, we acted on this philosophy by open sourcing deck.gl, a WebGL-powered framework specifically designed for exploring and visualizing data sets at scale. deck.gl lets us extract both historical and real-time insights from large, complex data sets, allowing us to think in 3D.

Today, we open sourced our latest version of the framework, deck.gl 4.0, empowering anyone, anywhere to rapidly visualize and explore data sets. Our goals for 4.0 were threefold:

deck.gl 4.0 brings a new set of powerful layers to our catalog, now capable of covering more use cases from across the mapping world. Check out a few of the dozen new layers and examples of how they can be used:

GeoJSON layer: Able to render any type of GeoJSON data set, this layer is designed to process and depict multiple types of geometry features defined in the GeoJSON format including points, lines, multi-lines, polygons, and multipolygons.  

Below, we rendered Vancouver’s property value data using the extrusion feature in the GeoJSON layer. Height encodes property value and color is percentage growth for that value.

Grid and Hexagon layers: These layers encode the aggregation results of location data by square or hexagon in the color or height of their cells, providing a 2D heatmap or 3D heightmap visualization of the data.

In the model below, we used data from data.gov.uk to depict the number of personal injuries based on location in the UK since 1979. The model demonstrates variable granularity for grouping data, variable radius for hexagons, and a simple way to select percentiles from the overall distribution.

For a full list of layers and examples, please refer to the layer catalog and our gallery.

Some internal needs at Uber related to machine learning and network visualization have generated new use cases for deck.gl. To meet these requests, we worked on decoupling the mercator projection viewport from deck.gl so that a user can take in any type of viewport, including standard 3D perspective and orthographic viewports.

3D surface layer: This layer can be used for rendering things like Partial Dependence Plots for machine learning models, highlighting the correlations between two variables and their impact on a prediction. The 3-D surface layer depicted below shows how to use deck.gl to render 3D surface plots, which makes use of our built-in picking mechanism to query the value for the surface at any arbitrary point.

Point-cloud layer: deck.gl 4.0 also incorporates a point-cloud layer for 3D point-cloud data visualization to get accurate models of reality.

In addition to adding new layers and functionalities, deck.gl 4.0 makes the framework more accessible to new and experienced users alike.

Documentation has been extensively improved and reorganized in response to user feedback. In particular, every layer now has an interactive layer browser allowing the user to play with all the properties of the layer while reading the docs. This makes it much easier to understand how the provided deck.gl layers work, as well as determine whether or not they work for your use cases.

deck.gl also provides multiple stand-alone examples, with minimal configuration files (package.json, webpack.config.js, etc.). These examples are easy to copy directly to new files, which makes visualizing your data quicker than before.

With its new set of powerful visualization layers, re-architected codebase, and user-friendly, interactive documentation and examples, we hope that deck.gl 4.0 can help your team, too! To stay informed on the latest updates and get visualizing, check out the deck.gl website.

Nicolas Garcia Belmonte leads Uber Engineering’s data visualization team.