Detail: Hybrid (Conditional Gaussian) Instantiated Model

This page describes Hybrid (conditional Gaussian) instantiated models in the Instantiated Model box. These are mixed discrete/continuous conditional Gaussian (CG) models fitted to data, starting from a Hybrid parametric model.

Hybrid Instantiated Model

Hybrid Instantiated Model

A Hybrid instantiated model contains:

  • A graph over discrete and continuous variables with typed nodes.

  • For discrete variables:

    • Estimated probabilities for (P(X \mid \text{Parents}(X))).

  • For continuous variables:

    • For each configuration of discrete parents, estimated linear-Gaussian regression parameters (coefficients and variances) conditional on parents.

How Hybrid instantiated models are created

  1. In the Parametric Model box, create a Hybrid (conditional Gaussian) model, making sure that variable types (discrete/continuous) match the data.

  2. In the Estimator box, select:

    • The Hybrid parametric model, and

    • A mixed dataset from the Data box.

  3. Choose a Hybrid/CG estimator (when available) and run it.

  4. Save or send the fitted result to the Instantiated Model box.

Instantiated Model box layout (Hybrid)

When you select a Hybrid instantiated model, the main panel typically shows:

  • For discrete variables:

    • Estimated CPTs for their conditional distributions.

  • For continuous variables:

    • Estimated regression coefficients and error variances, often broken down by discrete parent configuration.

  • Optional likelihood- or score-based summaries for the overall model.

Because Hybrid models combine discrete and continuous pieces, the instantiated view often looks like a mix of the Bayes and SEM views.

Typical uses

Hybrid instantiated models are useful when you want to:

  • Simulate realistic mixed data from a fitted CG model in the Simulation box.

  • Compare mixed-model search algorithms against a known generative Hybrid model using the Compare box.

  • Inspect how continuous variables behave under different discrete parent configurations.

Tips

  • Watch sample sizes for each discrete parent configuration; small cell counts can lead to unstable continuous-parameter estimates.

  • Confirm that variable types and coding (especially for discrete variables) are consistent between the data, graph, and parametric model.