Generate a Predefined Class

Context

• With the Generate a Predefined Class function, you can automatically generate a very specific type of Classes.
• Based on a qualitative analysis of the graph, this function generates one Class for each "layer" or "level" in the network.
  • Root Nodes, i.e., nodes without parent nodes, represent the reference level and belong to the Class [Depth_0].
  • Following the direction of the arcs from parent nodes one level down, the child nodes form Class [Depth_1].
  • The last level is comprised of the Leaf Nodes, i.e., nodes that do not have any child nodes (descendants). They form Class [Depth_n], where n indicates the hierarchical separation from the root level.

Example

• The following simple causal Bayesian network represents the dependencies of engine components in an automobile.
• Here, the Classes corresponding to Depth are already defined and arranged hierarchically to illustrate the purpose of this function.

Note: Our use of spatial references such as "down", "below", "depth", "downward" reflects the convention that parent nodes are typically positioned vertically above their child nodes in a network graph.

Workflow Animation

• The above layout, however, represents the endpoint. The following animation now shows the workflow associated with producing this hierarchical representation.
• The starting point is a typical model that represents individual causal relationships in a problem domain, in our case, an internal combustion engine.
• The initial layout may reflect how domain experts have directly encoded their knowledge by drawing arcs. Anyone who is somewhat familiar with motor vehicles should be able to recognize and qualitatively validate each relationship in this model.
• However, despite the validity of the model, a "big picture" is not immediately obvious.
• In this context, generating Classes for Depth will be remarkably useful. In addition to defining the Classes, we apply colors to them and manually rearrange them to achieve an intuitive layout.

The animation shows the following steps:

1. Select Graph Panel Context Menu > Edit Classes.
2. Click Generate a Predefined Class.
3. Highlight all generated Classes, i.e., [Depth_0] through [Depth_n].
4. Click Associate Colors.
5. Select Associate Default Colors with Classes.
6. Manually rearrange the color-coded Classes hierarchically (not shown).
7. Right-Click on the Classes icon to display individual Classes.

This hierarchical visualization of relationships can help with human understanding of the problem domain.