Implement automated knot refinement based on the Bayesian Information Criterion (BIC). For example, trial knots can be inserted between existing knots (one interval at a time), and their positions are optimized. A knot is permanently added only if the model BIC improves by a user-defined margin. Conversely, a simplification phase can be employed in which radius-ratio knots are removed one by one to determine whether their absence significantly affects the model likelihood.
Implement automated knot refinement based on the Bayesian Information Criterion (BIC). For example, trial knots can be inserted between existing knots (one interval at a time), and their positions are optimized. A knot is permanently added only if the model BIC improves by a user-defined margin. Conversely, a simplification phase can be employed in which radius-ratio knots are removed one by one to determine whether their absence significantly affects the model likelihood.