Automatic Bayesian Network Generator

Overview

The Automatic Bayesian Network Generator is a feature of Hellixia, BayesiaLab’s Large Language Model-powered layer. Its purpose is to produce constraint-free Bayesian Networks.

“Constraint-free” means that the generated network imposes no restrictions on variable type or the number of states a variable can have. This flexibility is important because in many domains you don’t know in advance whether a concept should be treated as Boolean, categorical, ordinal, or continuous. The generator handles all these cases automatically:

Nodes can be Boolean (true/false)
Categorical (multiple unordered states)
Ordinal (ordered categories)
Continuous (automatically discretized into bins)

The Conditional Probability Tables (CPTs) in these networks are plain probability tables, in which each cell is defined directly, without using built-in functions like Noisy-OR, logistic formulas, or other parametric forms. The generator therefore produces fully general CPTs that can capture any arbitrary conditional distribution.

The generator builds both the graphical structure (nodes and arcs) and the probability parameters entirely from the model’s knowledge, without requiring you to manually define variable types or states beforehand. It is one of several ways to create a Bayesian Network in Hellixia; others include conversational modeling (through the Conversational Network Modeler) or using the Causal Relationship Finder on a set of user-supplied nodes.

Workflow

To start the Automatic Bayesian Network Generator, select Menus > Hellixia > Automatic Bayesian Network Generator.

BayesiaLab window with the Hellixia menu open and Automatic Bayesian Network Generator highlighted

Click to Zoom

A window opens up and prompts you for the topic of the Bayesian network to be modeled: What would you like to model today?

The topic you provide can be as simple as a single concept, e.g., “Coral Reef Health”, or an entire paragraph to describe the domain of interest.

Automatic Bayesian Network Generator dialog with a Great Barrier Reef modeling prompt entered

Click to Zoom

Checking Advanced Mode extends the window and provides you with an additional Settings panel.

Hellixia Automatic Bayesian Network Generator dialog in Advanced Mode with a Great Barrier Reef modeling prompt, gpt-5.5 completion engine, and Marine Ecosystems general context

Click to Zoom

With the Completion Engine dropdown menu, you can specify which LLM you want to employ for creating the Bayesian network model.

The optional Knowledge File allows you to introduce very specific resources on the particular topic to be studied.

Under General Context you can also provide helpful guidance to the model development, especially if the topic you provide is ambiguous. For example, “Java” could refer to coffee, the Indonesian island, or the programming language.

Click OK to start generating the model.

BayesiaLab status bar with a progress indicator and stop button showing the Generating Model message

Upon completion, you will see a Bayesian network representing the domain you described in the previous step.

Bayesian network generated by Hellixia for Great Barrier Reef ecosystem health, with colored nodes and labeled arcs converging on an Integrated Great Barrier Reef Health target node

Click to Zoom

Zooming in and out allows you to explore the generated network in more detail.

Zoomed detail of the reef ecosystem network showing arc comment boxes explaining how Inshore water quality stress influences Coral bleaching severity and Coral disease prevalence

Click to Zoom

By clicking the Arc Comments icon in the Toolbar, you can turn the comment boxes on the arcs on and off.

Furthermore, you can show the Node Comments by highlighting the nodes of interest and then selecting Node Contextual Menu > Properties > Rendering Properties > Show Comment.

Node context menu cascade through Properties and Rendering Properties with Show Comment highlighted, on the Inshore water quality stress node

Click to Zoom

To confirm, check Show Comment in the pop-up window.

Rendering Properties dialog with the Show Comment checkbox selected

Now, the Bayesian network features Node Comments, providing descriptions of the nodes.

Reef ecosystem nodes displaying their comment definitions inside each box, including Inshore water quality stress, Coral recruitment success, Coral bleaching severity, and Coral disease prevalence

Click to Zoom

Beyond creating nodes and arcs in Bayesian networks, the generator groups the nodes into Classes. You can bring up the Class Editor from Menus > Edit > Edit Classes, which reveals how classes were applied to the nodes. Especially in large and complex networks, classes can help organize and manage the nodes.

In the screenshot below, we see the class $\mathit{Biodiversity}$ and the nodes that are part of this class.

Class Editor listing the generated node classes alongside the Edit Class Biodiversity dialog assigning reef nodes to the Biodiversity class

Click to Zoom

It is important to point out that the generated Bayesian network is fully estimated, i.e., all Conditional Probability Tables are filled with values, as shown in the Node Editor below. You can open the Node Editor by selecting Node Contextual Menu > Edit.