Support vector machines (SVMs) and related kernel-based learning algorithms are a well-known class of machine learning algorithms, for non-parametric classification and regression. liquidSVM is an implementation of SVMs whose key features are: fully integrated hyper-parameter selection, extreme speed on both small and large data sets, full flexibility for experts, and inclusion of a variety of different learning scenarios: multi-class classification, ROC, and Neyman-Pearson learning, and least-squares, quantile, and expectile regression.
liquidSVM is a package written in C++ that
provides SVM-type solvers for various classification and regression tasks.
Because of a fully integrated hyper-parameter selection, very carefully implemented solvers,
multi-threading and GPU support,
and several built-in data decomposition strategies it provides unprecedented speed
for small training sizes as well as for data sets of tens of millions of samples.
You can use it e.g. for multi-class classification, least squares (kernel) regression, or even quantile regression, etc.:
install.packages("liquidSVM")library(liquidSVM) model <- mcSVM(Species ~ ., iris)predict(model, iris) model <- lsSVM(Height ~ ., trees)y <- predict(model, trees) model <- svmQuantileRegression(Height ~ ., trees)y <- test(model, trees)If you install build the package to be used on several machines please use the following:
install.packages("liquidSVM", configure.args="generic")For details please look at the vignettes demo and documentation.
Also check the help ?liquidSVM and ?svm.
For the command-line version and other bindings go to (http://www.isa.uni-stuttgart.de/software/).