liquidSVM (version 1.2.2)

getSolution: Retrieve the solution of an SVM

Description

Gives the solution of an SVM that has been trained and selected in an ad-hoc list.

Usage

getSolution(model, task = 1, cell = 1, fold = 1)

Arguments

model

the model

task

the task between 1 and number of tasks

cell

the cell between 1 and number of cells

fold

the fold between 1 and number of folds

Value

a list with three entries: the offset of the solution (not yet implemented), the indices of the support vectors in the training data set, and the coefficients of the support vectors

Details

liquidSVM splits all problems into tasks (e.g. for multiclass classification or if using multiple weights), then each task is split into cells (maybe only a global one), and every cell then is trained in one or more folds to yiele a solution. Hence these coordinates have to be specified.

Examples

# NOT RUN {
# simple example: regression of sinus curve
x <- seq(0,1,by=.01)
y <- sin(x*10)
a <- lapply(1:5, function(i)getSolution(model <- lsSVM(x,y,d=1), 1,1,i))
plot(x,y,type='l',ylim=c(-5,5));
for(i in 1:5) lines(coeff~samples, data=a[[i]],col=i)

# a more typical example
banana <- liquidData('banana-mc')
model <- mcSVM(Y~.,banana$train,d=1)
# task 4 is predicting 2 vs 3, there is only cell 1 here
solution <- getSolution(model,task=4,cell=1,fold=1)
supportvecs <- solution$samples
# we are considering task 4 and hence only show labels 2 and 3:
bananaSub <- banana$train[banana$train$Y %in% c(2,3),]
plot(bananaSub[,-1],col=bananaSub$Y)
points(supportvecs,pch='x',cex=2)
# }