Scaling Altair visualizations with VegaFusion
summary
Claims that VegaFusion automatically moves Vega/Altair data transformations from the browser into the Python kernel so charts scale Altair charts to millions of rows. It provides a MIME renderer that pre-evaluates transforms, a widget renderer that keeps a live kernel–browser link for selections, and a transform_data function that returns a pandas DataFrame for downstream use. Implementation uses Vega/Vega-Lite, Rust + WebAssembly, Apache DataFusion, Apache Arrow and protobuf, with demos showing smooth cross-filtering on 1M flights and 10M taxi rides.
AI-written summary · about
chapters (18)
- 0:00Altair and VegaFusion intro
- 1:03Why Altair for exploration
- 3:52Declarative vs imperative APIs
- 5:37Composability in Altair
- 7:01Altair data transforms
- 8:45Interactivity without Python
- 10:11Altair limitations
- 14:04VegaFusion fixes limits
- 15:52VegaFusion MIME renderer
- 16:33Extracting transform data
- 17:35VegaFusion widget renderer
- 18:38Cross-filtering demo
- 19:46Taxi data scaling demo
- 20:52VegaFusion with Vega/Lite
- 21:55VegaFusion tech stack
- 22:39VegaFusion with Polars/DuckDB
- 24:03VegaFusion for chart editors
- 24:46Resources and closing
related talks
29mBuilding customer-visible data science dashboards with Altair / Vega / Vue
42mInteractive Data Analysis: Visualization and Beyond
42mAltair Declarative, Statistical Visualization for Python
28mHigh Performance Data Visualization for the Web
46mExplorations in Interactive Visual Analytics
29mHigh Performance Data Visualization for the Web
43mThe Future of Data Visualization on the Web
39mUse of AI in Modern Dataviz