The Big Picture of AVA v2.0

[arranzeyuwang]

AVA is the core automated visual analytics framework of AntV. For historical reasons, there're some complex code structures and redundant functional implementations on the current AVA v1.0, which made it unfriendly to both developers and users. Meanwhile, the overall structure and big picture were not clear enough, which was not conducive for users to understand.

We're delighted to announce that the refactor of AVA (i.e., AVA v2.0, see branch v2) is in progress now. We'd like to share our plan and how it is going now.

Pipeline

The following figure illustrates the overall structure of AVA 2.0.

We redesigned to divide the auto VA layer into Empiric-Driven and Insight-Driven paths, using Empirical Rules and Insight Types as the benchmarks for recommendation respectively.

Empiric-Driven approach aims to empirically output expressive charts based on experienced visualization rules.

Insight-Driven approach aims to exploratory discover charts that demonstrate potential data attributes.

Bases

1. antv-spec defines a declarative grammar that supports various technology stacks of AntV.

2. CKB is a chart knowledge base library that offers knowledge base entries for chart wikis in a JSON format.

Main process

1. DataWizard is AVA's data pre-processing lib.

2. ChartAdvisor is an empiric-driven chart recommendation lib that also contains a chart lint sub-component named Linter. Besides, Rules gives the employed empirical visual consensus.

3. LiteInsight is the insight exploration lib of AVA.

4. SmartBoard aims to solve chart dashboarding based on AVA recommendations.

Plugins

1. smart-color provides excellent color generation, optimization, and other computation options

2. vis-steg can conceal information within visualization images via visual steganography methods.

3. NarrativeTextVis tends to combine the benefits of both text visual analytics and storytelling.