A Canonical Representation of Data-Linear Visualization Algorithms
This work provides a foundational framework for visualization researchers and practitioners, though it is incremental in building upon existing dataflow concepts.
The paper tackles the problem of unifying diverse visualization algorithms by introducing a canonical model called linear-state dataflows, which enables declarative construction and easy mixing of visual mappings for data-linear visualizations.
We introduce linear-state dataflows, a canonical model for a large set of visualization algorithms that we call data-linear visualizations. Our model defines a fixed dataflow architecture: partitioning and subpartitioning of input data, ordering, graphic primitives, and graphic attributes generation. Local variables and accumulators are specific concepts that extend the expressiveness of the dataflow to support features of visualization algorithms that require state handling. We first show the flexibility of our model: it enables the declarative construction of many common algorithms with just a few mappings. Furthermore, the model enables easy mixing of visual mappings, such as creating treemaps of histograms and 2D plots, plots of histograms... Finally, we introduce our model in a more formal way and present some of its important properties. We have implemented this model in a visualization framework built around the concept of linear-state dataflows.