Example methodology description

Both examples consider the following combinations of hyperparameters used for Bayesian network learning:

• K2 metric;
• K2 metric with gaussian mixtures (GMM);
• K2 metric with GMM and logit nodes;
• K2 with initial structure.

All the examples are executed using cross-validation.

Geological data example

Data Description

The data set contains 9 variables with 442 samples. The target variable for prediction in the following example is 'Depth'. The variable is also used to visually evaluate sampling quality via distribution plot.

Sampling and corresponding network structure

K2 metric sampling example

Sampling with K2 + GMM example

Sampling with K2 + GMM + logit nodes example

K2 with initial structure sampling

Social data example

Data Description

The second example is similar to the previous one, but carried out on different data set. Social data set consists of 30000 anonymous bank records with 9 variables each, bayesian networks were learnt on a sample with 2000 records. The target variable is 'mean_tr' which is mean transaction of client.

Sampling

K2 metric sampling example

Sampling with K2 + GMM example

Sampling with K2 + GMM + logit nodes example

K2 with initial structure sampling

Prediction MSE table for both examples