Proposals:Refactoring Statistics Framework 2007 New Statistics Framework
From KitwarePublic
Jump to navigationJump to search
Class Manifesto of New Statistics Framework
Summary Table
The classes that integrate the new statistics framework are categorized in the following table
| Conceptual Class | Number |
|---|---|
| Traits | 1 |
| Data Objects | 4 |
| Filters | 11 |
| Total | 16 |
List of Classes per Category
Traits
- MeasurementVectorTraits
Data Objects
- Sample
- ListSample
- Histogram
- Subsample
Filters
- SampleToHistogramFilter
- MeanFilter
- WeightedMeanFilter
- CovarianceFilter
- WeightedCovarianceFilter
- HistogramToTextureFeaturesFilter
- ImageToListSampleFilter
- ScalarImageToCooccurrenceMatrixFilter
- SampleToSubsampleFilter
- SampleClassifierFilter
- NeighborhoodSubsampler
Classifiers (Suggested Design)
Elements
- MembershipFunctionBase
- DistanceToCentroidMembershipFunction (plugs in a DistanceMetric)
- DistanceMetrics
- Euclidean
- Mahalanobis
- 1_1
Filters
- Sample, Array of Membership Functions --> MembershipSample(sample,labels) == SampleClassifierFilter
- Sample, Array of Membership Functions --> GoodnessOfFitComponent (sample,weights) == SampleGoodnessOfFitFilter
Class Diagrams
Traits
Data Objects
Filters
Classifiers (Suggested Design)
Distance notation
- Manhattan (L1) = sum of absolute values
- Euclidean = square root of ( sum of squares )
- Euclidean Squared (L2) = sum of squares
- Mahalanobis = square root of ( V . M . VT )
API
- DistanceToCentroidMembershipFunction
- SetDistanceMetric( const DistanceMetric * ) (new)
- const GetDistanceMetric() (new)
- Evaluate( Measurement vector ) (already there)
- SetCentroid( ) (already there)
