Opened 13 months ago

Last modified 6 weeks ago

#2635 accepted defect

Predict if a child should be rejected during OSGA offspring selection

Reported by: bburlacu Owned by: bburlacu
Priority: medium Milestone: HeuristicLab 3.3.16
Component: Algorithms.OffspringSelectionGeneticAlgorithm Version: branch
Keywords: Cc:

Description

The idea is to try to predict if a child should be rejected by evaluating its fitness on a smaller partition of the training data. It would then make sense to save computation time when the child is already very weak compared to its parents and has no chance of being selected.

A new plugin should be implemented containing a specialized evaluator together with an analyzer.

Change History (16)

comment:1 Changed 13 months ago by bburlacu

  • Owner set to bburlacu
  • Status changed from new to accepted

r14072: Initial implementation.

comment:2 Changed 13 months ago by bburlacu

r14084: Forgot to add solution and project files.

comment:3 Changed 13 months ago by bburlacu

r14104: Improve performance and accuracy of evaluator and analyzer.

comment:4 Changed 13 months ago by bburlacu

r14184: Simplified evaluator code.

comment:5 Changed 13 months ago by bburlacu

r14231: Add evaluator option to aggregate statistics during evaluation (slower). Improve analyzer.

comment:6 Changed 11 months ago by bburlacu

r14279: Implement an alternative way of assessing which offspring should be rejected early: the child is evaluated on part of the training data and the remaining rows are assumed to be perfectly correlated with the target. If the offspring selection criteria is not fulfilled, we can reject early. Otherwise the child is evaluated on an additional slice of the training and so on.

comment:7 Changed 11 months ago by bburlacu

r14280: Small refactor.

comment:8 Changed 11 months ago by bburlacu

r14281: Small cosmetic refactor.

comment:9 Changed 11 months ago by bburlacu

r14301: Use linear scaling in the evaluator to bring the target in the range of the estimated (work in progress)

comment:10 Changed 11 months ago by mkommend

r14302: Rewrote parts of the OSGA evaluator.

comment:11 Changed 9 months ago by bburlacu

r14428: Add analyzer for counting the AdjustedEvaluatedSolutions (according to the actual number of evaluated rows). Add option to preserve compatibility with the standard evaluator. Optimize performance.

comment:12 Changed 7 months ago by bburlacu

r14584: Improve reporting of statistics about rejected individuals.

comment:13 Changed 7 months ago by bburlacu

r14585: Limit estimated values to range by default.

comment:14 Changed 7 months ago by bburlacu

r14608: Simplify code.

comment:15 Changed 7 months ago by bburlacu

r14609: Small improvement.

comment:16 Changed 6 weeks ago by mkommend

  • Milestone changed from HeuristicLab 3.3.15 to HeuristicLab 3.3.16
Note: See TracTickets for help on using tickets.