5. Conditional Gaussian BIC Score
5.1. Summary
The Conditional Gaussian BIC Score is a BIC-type score for conditional Gaussian (CG) models with mixed continuous and discrete variables. It evaluates a DAG or CG structure by combining CG log-likelihood with a penalty on the number of parameters.
5.2. When to use
Data are a mix of continuous and discrete variables.
You assume a CG model: given the discrete variables, continuous variables are multivariate normal with means and covariances that may depend on discrete configurations.
You are using score-based or hybrid algorithms that support CG models.
5.3. Model class
Conditional Gaussian Bayesian networks with discrete parents and linear- Gaussian continuous components.
5.4. Score form (conceptual)
As with other BIC scores:
BIC = 2 * logL − k * ln(N)
where logL is the CG log-likelihood and k is the number of free parameters
(conditional means, covariances, and discrete probabilities).
5.5. Parameters
Parameter (camelCase) |
Description |
|---|---|
|
Double ≥ 0.0. The penalty multiplier “c” in the modified BIC-type criterion (for example, a score of the form 2·log-likelihood − c·k·log(N), where k is the number of free parameters and N is the sample size). Larger values impose a stronger complexity penalty and yield sparser graphs; smaller values allow denser graphs. Default is 2.0. |
|
Double ≥ 0.0. Structure prior coefficient controlling a binomial-style prior on the number of parents per node. When 0.0 (default), the score uses essentially a flat structure prior. Larger values encode a stronger preference for a particular expected parent count and can bias the search toward graphs with that typical in-degree. |
|
Boolean. If |
|
Integer ≥ 2. Number of categories used when discretizing continuous variables in the backup discretization step. Default is 3. Larger values give a finer discretization but increase the size of the conditional tables and the number of parameters. |
|
Integer ≥ 2. Minimum required sample size per configuration (cell) in the conditional Gaussian model. If the per-cell sample size is too small, the exact CG calculations become unstable, and the score may fall back to the discretization strategy. Default is 4. |
5.6. Strengths
Properly accounts for mixed data types without discretizing continuous variables.
Integrates naturally with CG independence tests and CG learning algorithms.
5.7. Limitations
Requires enough data per discrete configuration.
Assumes linear-Gaussian behavior for continuous components.
5.8. References
Andrews, B., Ramsey, J., & Cooper, G. F. (2018). Scoring Bayesian networks of mixed variables. International Journal of Data Science and Analytics, 6(1), 3–18.