Rediscretize Continuous Nodes

Context

Rediscretize Continuous Nodes is an option when setting up a Structural Coefficient Analysis. When enabled, it performs automatic rediscretization at each iteration (each Structural Coefficient value) for nodes that were originally discretized with an automatic discretization algorithm. Nodes that were discretized manually are not included in the rediscretization; such nodes keep their original intervals across all iterations. Rediscretize Continuous Nodes uses the originally selected discretization algorithm and the originally selected number of intervals.

Tree-Based Discretization Algorithms

Rediscretize Continuous Nodes is particularly useful in the context of Supervised Learning when used in conjunction with Tree-based discretization algorithms. Tree-based discretization algorithms utilize the Minimum Description Length Score, which depends directly on the Structural Coefficient. As a result, the discretization thresholds are optimized for each value of the Structural Coefficient. This approach can help you fine-tune a model by trading off fit and complexity. Note that the Target Node in a network is never rediscretized automatically in this context.

Stochastic Discretization Algorithms

If a Fixed Seed is set for the Random Number Generator under Menus > Window > Preferences > General > Random Number Generator, the following stochastic discretization algorithms will always produce identical discretizations each time they run:

• R2-GenOpt,
• K-Means,
• Supervised Multivariate,
• Unsupervised Multivariate.

If a Fixed Seed is not set, Rediscretize Continuous Nodes will use the stochastic nature of the above discretization algorithms and produce stochastic variations in the discretization results.