Open HeuristicLab-Trac-Bot opened 9 years ago
r12957: Worked on expert system
- using newly created ERT helper class
r13475: updated from trunk, worked on okb connection for downloading knowledge base
- fixed bugs in iterated sampling view
- worked on download runs from okb feature
r13551: Added characteristic calculator for qap and adapted expert system view
- added two types of problem instance mappings: PCA and MDS
r13568: added code to auto-sync stopping criterion with embedded algorithm in iterated algorithms
r13569: fixed bugs in iterated algorithm, fixed bug in expert system view
- Added stripped-down version of FLA branch
- Added appropriate calculators
- Fixed detecting maximization in RLD view
- Changed ICharacteristicCalculator interface
- Updated OKB Problem view
r13706: Added pre-existing solution creator for permutation and respective interface
- Renamed remaining files from ExpertSystem to KnowledgeCenter
- Added ability to scatter plot to display a regression line
- Allowed to execute multiple instances at once and displaying either only final result or tracking result
- Split runs in seeded runs and instance runs
r13743: adding relational SOM projection for solution network visualization
r13745: added line annotations to the network visu to highlight close solutions
r13750: Added SOM projection for problem instances, fixed a bug (learningRadius was not used)
- improved mapping of problem instances by normalization (z-score) and handling of missing values (median)
- enabled to select the characteristics that should be included in the mapping
- improved speed of OKB download by calling service methods in parallel
- worked on problem instance mapping
- started working on improved suggestion algorithm
r13797: worked on testing recommendation algorithms through x-validation
- changed expected runtime calculation
- now outputs positive infinity instead of nan when no run was successful
- now filters outliers in successful and unsuccessful runs by using two standard deviations of the mean of successful runs as lower bound
- this change allows having unsuccessful runs in the database with low evaluations / runtime (e.g. due to being aborted early or from an experiment where the max budget was lower)
- worked on recommendation algorithms
- implemented several performance measures (absolute error, absolute log error, ndcp, kendall's tau) to evaluate the ranking
r13804: removed algorithm instances that do not appear in any run
r13878: added standardization of features for recommendation and using log10 of the expected runtime for clustering
r13920: added repetitions to LocalAnalysis and computing average
Issue migrated from trac ticket # 2457
milestone: HeuristicLab 3.3.x Backlog | component: Optimization | priority: medium
2015-08-11 09:43:37: @abeham created the issue