Opened 12 years ago
Closed 12 years ago
#1976 closed feature request (done)
Implement ConstantOptimizationAnalyzer to perform constant optimization after a generation is completed
| Reported by: | mkommend | Owned by: | gkronber |
|---|---|---|---|
| Priority: | high | Milestone: | HeuristicLab 3.3.8 |
| Component: | Problems.DataAnalysis.Symbolic | Version: | 3.3.8 |
| Keywords: | Cc: |
Description
Change History (18)
comment:1 Changed 12 years ago by mkommend
- Status changed from new to accepted
comment:2 Changed 12 years ago by mkommend
comment:3 Changed 12 years ago by mkommend
I had a more detailed look at the constant optimization evaluator and I don't think we should skip the optimization of variables with a weight of 1.0, just because this could have been set by the simplifier.
comment:4 Changed 12 years ago by mkommend
r8828: Allowed constant optimization for variable with a weight of 1.0 and made the constant optimization thread-safe (access to evaluated node results).
comment:5 Changed 12 years ago by mkommend
r8844: Added constant optimization analyzer.
comment:6 Changed 12 years ago by mkommend
r8845: Corrected access modifiers in cloning and storable ctor of ConstantOptimizationAnalyzer.
comment:7 Changed 12 years ago by mkommend
r8849: Added code to clean up the state of the ConstantOptimizationAnalyzer.
comment:8 Changed 12 years ago by mkommend
r8893: Added median value to the datatable containing the constant optimization improvments.
comment:10 Changed 12 years ago by mkommend
r8938: Removed obsolete code in SymbolicRegressionConstantOptimizationEvaluator.
comment:11 Changed 12 years ago by mkommend
- Owner changed from mkommend to gkronber
- Status changed from accepted to reviewing
r8976: Minor code clean up in constant optimization analyzer.
comment:14 follow-up: ↓ 17 Changed 12 years ago by gkronber
SymbolicRegressionConstantOptimizationEvaluator line 134: shouldn't we call const opt evaluator here? In the exception handlers it would be better to return originalQuality?
comment:15 Changed 12 years ago by gkronber
- Owner changed from gkronber to mkommend
comment:16 Changed 12 years ago by mkommend
r8984: Returned original quality in case of an exception in the SymbolicRegressionConstantOptimizationEvaluator.
comment:17 in reply to: ↑ 14 Changed 12 years ago by mkommend
Replying to gkronber:
SymbolicRegressionConstantOptimizationEvaluator line 134: shouldn't we call const opt evaluator here?
In the evaluate method (line 134) the pearson R² evaluator is used on purpose instead of the const opt evaluator, because this method is used to get the quality of evolved models on different partitions of the dataset (e.g., best validation model).
comment:18 Changed 12 years ago by mkommend
- Owner changed from mkommend to gkronber
comment:19 Changed 12 years ago by gkronber
r8985: comment about call to R² evaluator in const-opt evaluator
comment:20 Changed 12 years ago by gkronber
- Status changed from reviewing to readytorelease
comment:21 Changed 12 years ago by swagner
- Resolution set to done
- Status changed from readytorelease to closed
- Version changed from 3.3.7 to 3.3.8
r8823: Updated constant optimization evaluator to allow optimization of subtraction and added a parameter stating if constants should be updated in the tree.