Target Node and Target State

Context

In BayesiaLab, Supervised Learning includes algorithms that learn Bayesian networks with a structure focused on characterizing a Target Node.
In other words, the Target Node is the equivalent of a dependent variable in traditional modeling approaches.

The Target Node is visually distinguished from a normal node by its “bullseye” styling.
Additionally, the monitor corresponding to the Target Node features a green background.

Loading SVG...

In Unsupervised Learning, having a Target Node in the network has no influence on the learning algorithm.

There are several ways to define a node as a Target Node:

You can define a node as a Target Node by right-clicking on it and selecting Set as Target Node from the Node Context Menu. Selecting the same item from the menu reverts the node to its original role.
Holding T and double-clicking on a node turns it into a Target Node. Holding T and double-clicking on a Target Node makes it a normal node again.

You can define a node as a Target Node by right-clicking on it and selecting Set as Target Node from the Node Context Menu. You then choose the Target State from the drop-down list.
Holding T and double-clicking on a node turns it into a Target Node.
You can also define the target through the Monitor:
- Bring up the Monitor of the desired node in the Monitor Panel.
- Then, while holding T, double-click into the bar chart area of that Monitor.
- This sets the Target Node and the Target State at the same time.

If there is already a Target Node in the network, assigning a new Target Node automatically reverts the previous Target Node to its default status.

At any given time, there can only be one Target Node in a network.

Whenever you have a Target Node, there is automatically a Target State.
By default, the first state of a node is set as the Target State of a Target Node.
The Target State is marked with a bullseye icon on the monitor of the Target Node.
While a Target State is present whenever there is a Target Node, it does not affect the learning algorithms in any way.
As such, you can disregard the Target State for all modeling purposes.
In fact, the Target State is neither visible nor accessible in Modeling Mode F4.
However, for some analysis purposes in the Validation Mode F5, it can be useful to specify a Target State.
For instance, a Target State can be used to evaluate a model for predicting a specific state.
In many applications, only one state is relevant. For example, a Target Node may represent multiple virus types (Dengue, Zika, Rhinovirus, Influenza, and Coronavirus), but only the model’s performance for diagnosing the Target State (Coronavirus) is critical.

Loading SVG...

For structural learning, it matters which node is the Target Node, but the Target State does not.

Given that it is irrelevant for modeling purposes, the Target State is only visible and modifiable in Validation Mode F5.
In Validation Mode, there are multiple ways to set the Target State to the state of interest:

Option 1:
- Right-click on the Target Node and select Set as Target Node from the Node Context Menu again.
- This brings up a Target Selection window, from which you can choose the Target State.
Option 2:
- Bring up the Monitor corresponding to the Target Node.
- Then, while holding T, double-click in the bar chart area of the desired state.

Option 1:
- Right-click on the desired node and select Set as Target Node from the Node Context Menu.
- In the dialog, check the Target Node box and select the desired Target State from the drop-down list.
Option 2:
- Bring up the Monitor of the desired node in the Monitor Panel.
- Then, while holding T, double-click in the bar chart area of the desired state.
- This sets the Target Node and the Target State at the same time.