Details about base shogun framework

Base_Shogun.md

machine

• CMulticlassMachine > CBaseMulticlassMachine > [3]
• CLinearLatentMachine > CLinearMachine > [3]
• gp/CDualVariationalGaussianLikelihood > [4]
• gp/CNumericalVGLikelihood > [4]
• gp/CSingleFITCInference > CSingleSparseInference > CSparseInference > [2]
• gp/CKLLowerTriangularInference > CKLInference > [2]
• gp/CLaplaceInference > [2]
• gp/CMeanFunction > CSGObject

evaluation

• CSplittingStrategy > CSGObject
• CEvaluationResult > CSGObject
• CMachineEvaluation > CSGObject
• CClusteringEvaluation > [10]
• CBinaryClassEvaluation > [10]
• CCrossValidationOutput > CSGObject

distance

• CRealDistance > CDenseDistance > [5]
• CStringDistance > [5]
• CSparseDistance > [5]

classifier

• CMKL > CSVM > CKernelMachine > CMachine > CSGObject
• vw/cache/CVwCacheWriter > CSGObject
• vw/cache/CVwCacheReader > CSGObject
• vw/cache/CVwLearner > CSGObject

mathematics

• linalg/linsolver/CIterativeShiftedLinearFamilySolver > CIterativeLinearSolver > CLinearSolver > CSGObject
• CMatrixOperator > CLinearOperator > CSGObject
• linalg/eigsolver/CEigenSolver > CSGObject
• linalg/ratapprox/opfunc/COperatorFunction > CSGObject
• linalg/ratapprox/tracesampler/CTraceSampler > CSGObject
• linalg/ratapprox/opfunc/CRationalApproximation > COperatorFunction > CSGObject
• CFunction > CSGObject

distributions

• CDiscreteDistribution > CDistribution > CSGObject
• CEMBase > CSGObject
• classical/CProbabilityDistribution > CSGObject

loss

• CLossFunction > CSGObject

ui

• CSGInterface > CSGObject

io

• CFile > CSGObject # Has commented out pure virtual func

structure

• CPlifBase > CSGObject
• CStructuredModel > CSGObject
• CMAPInference > CSGObject
• CStateModel > CSGObject

optimization

• FirstOrderBoundConstraintsCostFunction > FirstOrderCostFunction
• FirstOrderSAGCostFunction > FirstOrderStochasticCostFunction > FirstOrderCostFunction
• FirstOrderStochasticMinimizer > FirstOrderMinimizer > Minimizer
• liblinear/CTron > CSGObject
• LearningRate
• MappingFunction
• DescendCorrection
• SparsePenalty > ProximalPenalty > Penalty
• DescendUpdater

preprocessor

• CSparsePreprocessor > [8]
• CDependenceMaximization > CFeatureSelection > [8]
• CDensePreprocessor > [8]
• CStringPreprocessor > [8]
• CKernelDependenceMaximization > CDependenceMaximization > CFeatureSelection > [8]

lib

• SGReferencedData
• CTokenizer > CSGObject
• computation/aggregator/CStoreVectorAggregator > CJobResultAggregator > CSGObject
• computation/engine/CIndependentComputationEngine > CSGObject
• computation/job/CIndependentJob > CSGObject

multiclass

• CECOCSimpleDecoder > CECOCDecoder > CSGObject
• CMulticlassStrategy > CSGObject
• CConditionalProbabilityTree > [9]
• CNbodyTree > [9]
• CVwConditionalProbabilityTree > [9]
• CECOCEncoder > CSGObject
• CECOCDecoder > CSGObject
• CRejectionStrategy > CSGObject

latent

• CLatentModel > CSGObject

kernel

• CExponentialARDKernel > CDotKernel > [6]
• CSparseKernel > [6]
• CStringKernel > [6]
• /normalizer/CKernelNormalizer > CSGObject

features

• CStreamingDotFeatures > CStreamingFeatures > [7]
• CRandomKitchenSinksDotFeatures > CDotFeatures > [7]
• CAttributeFeatures > [7]

statistics

• CMMDKernelSelection > CKernelSelection > CSGObject
• CKernelTwoSampleTest > CTwoSampleTest > CHypothesisTest > CSGObject
• CKernelIndependenceTest > CIndependenceTest > CHypothesisTest > CSBObject

transfer/multitask

• CMultitaskKernelMklNormalizer > CKernelNormalizer > CSGObject
• CTaskRelation > CSGObject

modelselection

• CModelSelection > CSGObject

labels

• CDenseLabels > CLabels > CSGObject

converter

• CEmbeddingConverter > [11]
• CICAConverter > [11]

ensemble

• CCombinationRule > CSGObject

[1]: CVariationalLikelihood > CLikelihoodModel > CSGObject

[2]: CInference > CDifferentiableFunction > CSGObject

[3]: CMachine > CSGObject

[4]: CVariationalGaussianLikelihood > [1]

[5]: CDistance > CSGObject

[6]: CKernel > CSGObject

[7]: CFeatures > CSGObject

[8]: CPreprocessor > CSGObject

[9]: CTreeMachine > CBaseMulticlassMachine > [3]

[10]: CEvaluation > CSGObject

[11]: CConverter > CSGObject

It is definitely:

• Kernels and Distances
• Features, DotFeatures, DenseFeatures, SparseFeatures
• Machine, LinearMachine, MulticlassMachine
• Preprocessor, Converter

Let's see other suggestions

I would more do this as a class diagram

Starting from ML paradigms:

• supervised learning
• unsupervised learning
• statistical testing
• feature extraction
• preprocessing
• model selection
• optimization
• different feature representations
• etc

To classes of these algorithms

• density estimation
• classification
• regression
• two-sample testing

and subclasses sucha s

• kernel machine
• linear model
• densities with gradients
• continuous /discrete

To more concrete algorithms

This needs to be far more concrete than above btw!!
We need signatures and a very concrete list that contains at least 80% of all shogun classes