Open HeuristicLab-Trac-Bot opened 4 years ago
AI-Feynman.png
(93.0 KiB)r17638 Added first part of feynman equations as instances
- Added rest of part I equations
- Set Training/Test Partitions to 10^5
r17642 Added field and constructor to define sample size in instances
- Updated instances (removed 2*pi term from equations)
- Added instances to InstanceProvider
- Added possibility to add noise to the feynman instances
- Sorted instances by name
r17649 Added noise instances for feynman bonus instances
- r17670 Fix order of instances and add information on reference publication and description.
- r17671 Removed "Feynman" prefix from all instances
- r17672 rename file for reodering
- r17673 reorder instances
- r17674 small changes while reviewing
- r17675 reordering part1
- r17676 reorder part2
- r17677 update class names to match file names, split problem instances into 'small' and 'large' instances with separate providers
- r17678 changed title for noise part
r17683 (not migrated): deleted branch
The problem instances provider should not be the default selected one. Instead CSV should be the default. Right now AI is ordered first.
Discovered problem instance providers are ordered by
Name
in ProblemView line 70:problemInstanceProviderComboBox.DataSource # ProblemInstanceProviders.OrderBy(x> x.Name).ToList();
r17804: start
Name
of CVS instance provider with a space to make sure it is ordered first.
- Use the same noise levels and calculation as in our experiments for the IEEE TeC paper.
- Reorder instances by name first and noise level second.
- Remove number of samples from the name.
r17966: Changed Feynman problem instances to new normal distributed RNG.
r17967: Add GlobalSuppression file to HL.Problems.DataAnalysis.Instances to prevent persistence analyzer messages about missing StorableTypeAttributes.
r17969: Correct project file for HL.Problems.Instances.DataAnalysis.
- Added target without noise to noisy instances
- Moved noise calculation to descriptor
Check and adapt (if necessary) noise ratios of Feynman instances so that these are equivalent to the noise ratios of the Friedman problems.
If the different calculation has been introduced on purpose, we should define what we understand under noise ratio and how to calculate it.
r18032 noise generation method to ValueGenerator; use same method for generating noise in friedman and feynman instances
Noise generation introduced in r18032 does not work for noise ratio = 1 (which means no signal left)
Issue migrated from trac ticket # 3075
milestone: HeuristicLab 3.3.17 | component: Problems.DataAnalysis.Symbolic.Regression | priority: medium
2020-06-18 09:13:47: @gkronber created the issue