Data transforms in non-group marks about vega HOT 1 CLOSED

Transforms can indeed be applied to non-group mark types. However, the semantics of the transforms applied are critical. Clearly we need better documentation of input and output data for many of the transforms.

For example, given input data like this:

[{x:"a", y:1}, {x:"b", y:2}, {x:"c", y:3}]

The stats operator {type:"stats", value:"data.y"} will produce a data set with a single element, like so:

[{mean:2, min:1, max:3, ...}]

Note that no reference to the original data is maintained, hence no data.x property exists. In short, some transforms add new properties to data. Other transforms (such as stats) may produce completely new derived data elements.

Moving data from group marks to leaf-level marks also has major implications. Data at the group level is recursively divided among group instances. For example, the data [[1],[2],[3]] will result in 3 groups with "from" data of [1], [2] and [3], respectively. It is these sub-arrays (or values arrays, for faceted data) that are then passed to child marks. If you instead reference the input data at the level of (say) a rect mark, you will be attempting to visualize data that is still an array of arrays.

from vega.