Procedures

Full name Microsoft.VisualBasic.MachineLearning.SVM.Procedures Assembly Microsoft.VisualBasic.MachineLearning Members 21

01 Syntax

Microsoft.VisualBasic.MachineLearning.SVM.Procedures

02 Methods

Name	Overloads	Summary
setRandomSeed	1	Set the random seed value which is used by the LibSVM procedure.
sigmoid_train	1	Platt's binary SVM Probablistic Output: an improvement from Lin et al.
multiclass_probability	1	Method 2 from the multiclass_prob paper by Wu, Lin, and Weng
svm_binary_svc_probability	1	Cross-validation decision values for probability estimates
svm_svr_probability	1	Return parameter of a Laplace distribution
svm_group_classes	1	group training data of the same class
oneClassSvm	1	regression or one-class-svm
multipleClassification	1	classification
svm_train	1	Train a support vector machine model from the given problem and parameters, this is the entry point of the whole training procedure.
svm_cross_validation	1	Stratified cross validation
svm_get_svm_type	1	Gets the svm type of the given model.
svm_get_nr_class	1	Gets the number of the classes of the given model.
svm_get_labels	1	Copy the class label values of the given model into the target array.
svm_get_sv_indices	1	Copy the indices of the support vectors into the target array.
svm_get_nr_sv	1	Gets the number of the support vectors of the given model.
svm_get_svr_probability	1	Gets the sigma value of the laplace distribution which was estimated for the support vector regression model.
svm_predict_values	1	Evaluate the decision function values of the given input sample.
svm_predict	1	Predict the label or the target value of a single input sample.
svm_predict_probability	1	Evaluate the class probability distribution of the given input sample.
svm_check_parameter	1	Validate the training parameters against the given problem before the training procedure starts.
svm_check_probability_model	1	Check whether the given model contains the information which is required by the probability inference.

03 Members

method setRandomSeed #

setRandomSeed(Int32)

Set the random seed value which is used by the LibSVM procedure.

Parameters

Name	Type	Description
`seed`	`Int32`	The random seed value.

method sigmoid_train #

sigmoid_train(Int32, Double(), ColorClass(), Double())

Platt's binary SVM Probablistic Output: an improvement from Lin et al.

Parameters

Name	Type	Description
`l`	`Int32`	The number of the decision values.
`dec_values`	`Double()`	The decision value of each training sample.
`labels`	`ColorClass()`	The class label of each training sample.
`probAB`	`Double()`	The output array which will receive the two parameters `A` and `B` of the sigmoid function, its length should be 2 or more.

method multiclass_probability #

multiclass_probability(Int32, Double[0:,0:], Double())

Method 2 from the multiclass_prob paper by Wu, Lin, and Weng

Parameters

Name	Type	Description
`k`	`Int32`	The number of the classes.
`r`	`Double[0:,0:]`	The pairwise probability matrix, in which the element `(i, j)` is the probability of the class `i` against the class `j`.
`p`	`Double()`	The output array which will receive the probability value of each class.

method svm_binary_svc_probability #

svm_binary_svc_probability(Problem, Parameter, Double, Double, Double())

Cross-validation decision values for probability estimates

Parameters

Name	Type	Description
`prob`	Problem	The training data of the binary sub-problem.
`param`	Parameter	The training parameters.
`Cp`	`Double`	The C value of the positive side samples.
`Cn`	`Double`	The C value of the negative side samples.
`probAB`	`Double()`	The output array which will receive the two parameters `A` and `B` of the sigmoid function, its length should be 2 or more.

method svm_svr_probability #

svm_svr_probability(Problem, Parameter)

Return parameter of a Laplace distribution

Parameters

Name	Type	Description
`prob`	Problem	The training data of the regression problem.
`param`	Parameter	The training parameters.

Returns

The sigma value of the estimated Laplace distribution; a large value is returned when the regression result is not reliable.

method svm_group_classes #

svm_group_classes(Problem, Int32, Int32(), Int32(), Int32(), Int32())

group training data of the same class

Parameters

Name	Type	Description
`prob`	Problem	The source training data which will be grouped.
`nr_class_ret`	`Int32`	The output number of the classes.
`label_ret`	`Int32()`	label name
`start_ret`	`Int32()`	begin of each class
`count_ret`	`Int32()`	#data of classes
`perm`	`Int32()`	indices to the original data, perm, length l, must be allocated before calling this subroutine

method oneClassSvm #

oneClassSvm(Model, Problem, Parameter)

regression or one-class-svm

Parameters

Name	Type	Description
`model`	Model	The model object which will be filled with the training result.
`prob`	Problem	The training data of the one-class or the regression problem.
`param`	Parameter	The training parameters.

method multipleClassification #

multipleClassification(Model, Problem, Parameter)

classification

Parameters

Name	Type	Description
`model`	Model	The model object which will be filled with the training result.
`prob`	Problem	The training data of the multi-class classification problem.
`param`	Parameter	The training parameters.

method svm_train #

svm_train(Problem, Parameter)

Train a support vector machine model from the given problem and parameters, this is the entry point of the whole training procedure.

Parameters

Name	Type	Description
`prob`	Problem	The training data.
`param`	Parameter	The training parameters.

Returns

A trained Model object.

method svm_cross_validation #

svm_cross_validation(Problem, Parameter, Int32, SVMPrediction())

Stratified cross validation

Parameters

Name	Type	Description
`prob`	Problem	The training data which will be cross validated.
`param`	Parameter	The training parameters.
`nr_fold`	`Int32`	The number of the folds of the cross validation.
`target`	SVMPrediction()	The output array which will receive the prediction result of each sample.

method svm_get_svm_type #

svm_get_svm_type(Model)

Gets the svm type of the given model.

Parameters

Name	Type	Description
`model`	Model	The target model.

Returns

A SvmType value.

method svm_get_nr_class #

svm_get_nr_class(Model)

Gets the number of the classes of the given model.

Parameters

Name	Type	Description
`model`	Model	The target model.

Returns

An Int32 value.

method svm_get_labels #

svm_get_labels(Model, Int32())

Copy the class label values of the given model into the target array.

Parameters

Name	Type	Description
`model`	Model	The target model.
`label`	`Int32()`	The output array which will receive the class labels, its length should be equals to Model.numberOfClasses.

method svm_get_sv_indices #

svm_get_sv_indices(Model, Int32())

Copy the indices of the support vectors into the target array.

Parameters

Name	Type	Description
`model`	Model	The target model.
`indices`	`Int32()`	The output array which will receive the support vector indices, its length should be equals to Model.supportVectorCount.

method svm_get_nr_sv #

svm_get_nr_sv(Model)

Gets the number of the support vectors of the given model.

Parameters

Name	Type	Description
`model`	Model	The target model.

Returns

An Int32 value.

method svm_get_svr_probability #

svm_get_svr_probability(Model)

Gets the sigma value of the laplace distribution which was estimated for the support vector regression model.

Parameters

Name	Type	Description
`model`	Model	The target model.

Returns

The sigma value; a message will be written to the error console and 0 will be returned when the target model does not contain the information for the SVR probability inference.

method svm_predict_values #

svm_predict_values(Model, Node(), Double())

Evaluate the decision function values of the given input sample.

Parameters

Name	Type	Description
`model`	Model	The target model.
`x`	Node()	The sparse feature vector of the input sample.
`dec_values`	`Double()`	The output array which will receive the pairwise decision values; for the one-class and the regression models its length should be 1, otherwise it should be `nr_class * (nr_class - 1) / 2`.

Returns

The SVMPrediction result of the given sample.

method svm_predict #

svm_predict(Model, Node())

Predict the label or the target value of a single input sample.

Parameters

Name	Type	Description
`model`	Model	The target model.
`x`	Node()	The sparse feature vector of the input sample.

Returns

兼容分类以及打分这两种工作模式

method svm_predict_probability #

svm_predict_probability(Model, Node(), Double())

Evaluate the class probability distribution of the given input sample.

Parameters

Name	Type	Description
`model`	Model	The target model, it should be trained with the probability estimation enabled.
`x`	Node()	The sparse feature vector of the input sample.
`prob_estimates`	`Double()`	The output array which will receive the probability value of each class, its length should be equals to Model.numberOfClasses.

Returns

The SVMPrediction result of the given sample; when the model does not provide the pairwise probability, this function falls back to the Procedures.svm_predict() function.

method svm_check_parameter #

svm_check_parameter(Problem, Parameter)

Validate the training parameters against the given problem before the training procedure starts.

Parameters

Name	Type	Description
`prob`	Problem	The training data.
`param`	Parameter	The training parameters that will be validated.

Returns

An error message which describes the first invalid parameter, or Nothing when all of the parameters are valid.

method svm_check_probability_model #

svm_check_probability_model(Model)

Check whether the given model contains the information which is required by the probability inference.

Parameters

Name	Type	Description
`model`	Model	The target model.

Returns

1 when the probability information is available, otherwise 0.