search

heal-research / HeuristicLab

HeuristicLab - An environment for heuristic and evolutionary optimization
https://dev.heuristiclab.com
GNU General Public License v3.0
36 stars 16 forks source link

Provide a well-configured sample for symbolic regression with GP with strict offspring selection #2802

Closed HeuristicLab-Trac-Bot closed 7 years ago

HeuristicLab-Trac-Bot commented 7 years ago

Issue migrated from trac ticket # 2802

milestone: HeuristicLab 3.3.15 | component: Problems.DataAnalysis.Symbolic.Regression | priority: medium | resolution: done

2017-07-11 08:29:57: @gkronber created the issue

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-12 20:56:09: @mkommend changed status from new to accepted

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-12 20:56:09: @mkommend set owner to @mkommend

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-12 20:56:09: @mkommend changed type from defect to feature request

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-13 16:36:12: @mkommend changed status from accepted to reviewing

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-13 16:36:12: @mkommend changed owner from @mkommend to @gkronber

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-13 16:36:12: @mkommend commented

r15235: Implemented new sample with strict OS for sym reg (Paige-I) with constant optimization and a simplified grammar.

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-14 16:20:01: @NimZwei commented

Tested the Sample with gkronberger on the builder and compared to my local machine

  • unit test results were the same when executed on x86 (via MStest.exe without testsettings)
  • unit test results were different when the test was executed on x64 (Builder.testsettings specify hostprocessplatform=MSIL, meaning the test is executed on x64)
  • unit test sample executed in HL (x64) yielded different results on builder and local machine
  • when using another evaluator (R²) results were the same on builder and local machine

We suspect that the alglib's nonlinear least squares fitting within the constant opt. evaluator somehow behaves differently on both machines.

  • .net framework on which HL was executed was the same
  • builder runs on windows server 2012, local machine on windows 10
  • builder has an Intel Xeon processor, local machine Intel Core i7

We currently suspect that either the operating system (less likely) or the processor type differences causes the unit tests to behave slightly different (also for other unittests). However, more tests must be done to be certain.

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-15 09:34:52: @gkronber commented

r15246: changed result vales which are checked in the unit test (for x64) based on the results produced on the builder.

This should fix the unit test on the builder. (TODO: check)

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-15 09:45:33: @gkronber commented

r15247: hide parameters for elites, solution creator and relative number of samples and updated sample

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-16 09:43:32: @gkronber commented

r15257: merged r15235,r15246,r15247 from trunk to stable

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-16 10:04:57: @gkronber changed status from reviewing to readytorelease

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-16 10:05:02: @gkronber changed status from readytorelease to closed

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-16 10:05:02: @gkronber set resolution to done

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-19 12:40:29: @gkronber commented

r15271: fixed a typo in the sample description

HeuristicLab-Trac-Bot commented 7 years ago

2017-07-19 12:42:08: @gkronber commented

r15272: merged r15271 from trunk to stable