Groupings

In the types of visualizations that NetPanorama was intended to help create, it is common to aggregate data based on common elements.

For example, we may have a list of links, and want to aggregate this into a list of lists, each containing links with the same source node and same target node.

Limitation

In Vega, this sort of grouping would be achieved by using a transform of type aggregation, with an aggregation operation of values. We chose to deviate from this behaviour: this means that the aggregation transforms consistently aggregate multiple values into a single number, and for our intended uses groupings are sufficiently important to be reified as top-levle objects.

Example

Suppose that the value of edges is:

[
    {source: 1, target: 2, id: 1},
    {source: 1, target: 2, id: 2},
    {source: 1, target: 3, id: 3},
    {source: 1, target: 4, id: 4},
    {source: 1, target: 5, id: 5},

    {source: 2, target: 2, id: 6},
    {source: 2, target: 2, id: 7},
    {source: 3, target: 3, id: 8},
    {source: 3, target: 4, id: 9},
    {source: 4, target: 5, id: 10},
]

We can group by a single attribute - the grouping

{
  "groupings": {
    "name": "grouped_links",
    "data": "links",
    "common": ["source"]
  }
}

would give

          
  [
    [
        {"source": 1, "target": 2, "id": 1},
        {"source": 1, "target": 2, "id": 2},
        {"source": 1, "target": 3, "id": 3 },
        {"source": 1, "target": 4, "id": 4},
        {"source": 1, "target": 5, "id": 5}
    ],
     [
        {"source": 2, "target": 2, "id": 6},
        {"source": 2, "target": 2, "id": 7}
    ],
     [
        {"source": 3, "target": 3, "id": 8},
        {"source": 3, "target": 4, "id": 9}
    ],
     [{"source": 4, "target": 5, "id": 10}]
]

Alternatively, we could group by two attributes - the grouping

{
  "groupings": {
    "name": "grouped_links",
    "data": "links",
    "common": ["source", "target"]
  }
}

would give

[
  [{ "source": 1, "target": 2, "id": 1 }, { "source": 1, "target": 2, "id": 2 }],
  [{ "source": 1, "target": 3, "id": 3 }],
  [{ "source": 1, "target": 4, "id": 4 }],
  [{ "source": 1, "target": 5, "id": 5 }],
  [{ "source": 2, "target": 2, "id": 6 }, { "source": 2, "target": 2, "id": 7 }],
  [{ "source": 3, "target": 3, "id": 8 }],
  [{ "source": 3, "target": 4, "id": 9 }],
  [{ "source": 4, "target": 5, "id": 10 }]
]

Additional attributes

Each computed group get the following properties:

• key: the common value of the group
• data: the data elements that have the common value
• represents: nodes | links. Type of network element that each group represents
• representsId: nodes | links. Id of the nodes of links represented by each group. This is useful to specify interactions.

Grouping networks

It is also possible to construct a grouping from a network rather than a table, by setting the network attribute of the specification. In this case, each entry in the resulting group array will include nodes and links arrays. Links whose source and target nodes are in different groups are saved separately in the <GROUP_NAME>_INTER_GROUP_LINKS attribute of the state.