Opened 5 years ago
Last modified 5 years ago
#2993 new enhancement
User-defined Evaluators for Symbolic Data Analysis Problems
| Reported by: | bburlacu | Owned by: | bburlacu |
|---|---|---|---|
| Priority: | medium | Milestone: | HeuristicLab 3.3.x Backlog |
| Component: | Problems.DataAnalysis.Symbolic | Version: | branch |
| Keywords: | Cc: |
Description
The idea behind a user-defined evaluator is to provide a more flexible way of combining different objective measures in data analysis problems.
For example, multi-objective regression problems currently support combinations of two objectives (R2 + length, R2 + complexity, R2 + diversity, etc.), where each combination is implemented as a separate evaluator class. This becomes cumbersome or even infeasible when more/different combinations of objectives are desired.
A user-defined evaluator maintains a collection of basic evaluators (R2, length, etc.) which it then executes sequentially, returning either an aggregate value (ie., weighted sum) in the single-objective case or an array of quality values in the multi-objective case.
More complex evaluation could then be easily achieved by nesting user-defined evaluators.
r16687: Initial implementation of user-defined evaluators as separate plugin.