Ticket #1255 (accepted feature request)
OffspringSelection criterion for symbolic regression with size constraints
| Reported by: | mkommend | Owned by: | cfischer |
|---|---|---|---|
| Priority: | medium | Milestone: | HeuristicLab 3.3.x Backlog |
| Component: | ZZZ OBSOLETE: Problems.DataAnalysis.Regression | Version: | branch |
| Keywords: | Cc: | maffenze |
Description
Implement a new OS criterion that takes the symbolic expression complexity into account. The complexity results as a linear combination of the expression size, expression depth and the variables used in the expression.
Attachments
Change History
comment:1 Changed 3 years ago by cfischer
- Owner changed from mkommend to cfischer
- Status changed from new to assigned
comment:3 in reply to: ↑ description Changed 3 years ago by cfischer
The various complexity measures are calculated and scaled as follows:
- expression size = treeSize / maxTreeSize
- expression depth = treeHeight / maxTreeHeight
- used variables = count distinct usedVariables / count inputVariables
- quality: mean squared error = mse / variance(target)
- quality: R² = 1 - R²
4 weights are used for a linear combination of these 4 complexity measures.
Changed 2 years ago by cfischer
-
attachment
OS Criterion range+bound.hl
added
Current version of OS Criterion logical with GenerationRanges and Lower/UpperBound for SymbolicRegression and Classification
Changed 2 years ago by cfischer
-
attachment
OS Criterion net40.hl
added
OS Criterion logical (added assembly reference to System.Core due to change to .NET 4; added trainings analyzer; added solution complexity for results)
