svmlin: svmlin implementation by Sindhwani & Keerthi (2006)

In RSSL: Implementations of Semi-Supervised Learning Approaches for Classification

svmlin implementation by Sindhwani & Keerthi (2006)

Description

R interface to the svmlin code by Vikas Sindhwani and S. Sathiya Keerthi for fast linear transductive SVMs.

Usage

svmlin(X, y, X_u = NULL, algorithm = 1, lambda = 1, lambda_u = 1,
  max_switch = 10000, pos_frac = 0.5, Cp = 1, Cn = 1,
  verbose = FALSE, intercept = TRUE, scale = FALSE, x_center = FALSE)

Arguments

X	Matrix or sparseMatrix containing the labeled feature vectors, without intercept
y	factor containing class assignments
X_u	Matrix or sparseMatrix containing the unlabeled feature vectors, without intercept
algorithm	integer; Algorithm choice, see details (default:1)
lambda	double; Regularization parameter lambda (default 1)
lambda_u	double; Regularization parameter lambda_u (default 1)
max_switch	integer; Maximum number of switches in TSVM (default 10000)
pos_frac	double; Positive class fraction of unlabeled data (default 0.5)
Cp	double; Relative cost for positive examples (only available with algorithm 1)
Cn	double; Relative cost for positive examples (only available with algorithm 1)
verbose	logical; Controls the verbosity of the output
intercept	logical; Whether an intercept should be included
scale	logical; Should the features be normalized? (default: FALSE)
x_center	logical; Should the features be centered?

Details

The codes to select the algorithm are the following: 0. Regularized Least Squares Classification 1. SVM (L2-SVM-MFN) 2. Multi-switch Transductive SVM (using L2-SVM-MFN) 3. Deterministic Annealing Semi-supervised SVM (using L2-SVM-MFN).

References

Vikas Sindhwani and S. Sathiya Keerthi. Large Scale Semi-supervised Linear SVMs. Proceedings of ACM SIGIR, 2006 @references V. Sindhwani and S. Sathiya Keerthi. Newton Methods for Fast Solution of Semi-supervised Linear SVMs. Book Chapter in Large Scale Kernel Machines, MIT Press, 2006

See Also

Other RSSL classifiers: EMLeastSquaresClassifier, EMLinearDiscriminantClassifier, GRFClassifier, ICLeastSquaresClassifier, ICLinearDiscriminantClassifier, KernelLeastSquaresClassifier, LaplacianKernelLeastSquaresClassifier(), LaplacianSVM, LeastSquaresClassifier, LinearDiscriminantClassifier, LinearSVM, LinearTSVM(), LogisticLossClassifier, LogisticRegression, MCLinearDiscriminantClassifier, MCNearestMeanClassifier, MCPLDA, MajorityClassClassifier, NearestMeanClassifier, QuadraticDiscriminantClassifier, S4VM, SVM, SelfLearning, TSVM, USMLeastSquaresClassifier, WellSVM

Examples

data(svmlin_example)
t_svmlin_1 <- svmlin(svmlin_example$X_train[1:50,],
                 svmlin_example$y_train,X_u=NULL, lambda = 0.001)
t_svmlin_2 <- svmlin(svmlin_example$X_train[1:50,],
                       svmlin_example$y_train,
                       X_u=svmlin_example$X_train[-c(1:50),], 
                       lambda = 10,lambda_u=100,algorithm = 2)
                       
# Calculate Accuracy
mean(predict(t_svmlin_1,svmlin_example$X_test)==svmlin_example$y_test)
mean(predict(t_svmlin_2,svmlin_example$X_test)==svmlin_example$y_test)

data(testdata)

g_svm <- SVM(testdata$X,testdata$y)
g_sup <- svmlin(testdata$X,testdata$y,testdata$X_u,algorithm = 3)
g_semi <- svmlin(testdata$X,testdata$y,testdata$X_u,algorithm = 2)

mean(predict(g_svm,testdata$X_test)==testdata$y_test)
mean(predict(g_sup,testdata$X_test)==testdata$y_test)
mean(predict(g_semi,testdata$X_test)==testdata$y_test)

RSSL documentation built on March 31, 2023, 7:27 p.m.