It's hard to evaluate when the projects are so new. It was a similar problem with Crossfilter/Datavore/Data.js/Miso Dataset when they all came out.
If I had to pick, I'd go with Vega because I think it has the most potential to develop a community around it. Especially as a bridge between Python/R and D3.js through JSON. Think iPython notebooks, R Shiny, etc.
They are all libraries for managing datasets with fast or convenient filtering. Useful for linking visualizations together through a common data model.
None have become as popular as d3.js though, so it's still hard to compare them.
The interaction model is something that very few charting libraries have. Also we try to make Polychart.js very easy to get started in, so that creating a simple line or bar chart would take minimal amount of code. Vega is amazing as well for how flexible it is.
https://github.com/trifacta/vega