kernlab (version 0.9-25)

onlearn: Kernel Online Learning algorithms

Description

Online Kernel-based Learning algorithms for classification, novelty detection, and regression.

Usage

# S4 method for onlearn
onlearn(obj, x, y = NULL, nu = 0.2, lambda = 1e-04)

Arguments

obj

obj an object of class onlearn created by the initialization function inlearn containing the kernel to be used during learning and the parameters of the learned model

x

vector or matrix containing the data. Factors have to be numerically coded. If x is a matrix the code is run internally one sample at the time.

y

the class label in case of classification. Only binary classification is supported and class labels have to be -1 or +1.

nu

the parameter similarly to the nu parameter in SVM bounds the training error.

lambda

the learning rate

Value

The function returns an S4 object of class onlearn containing the model parameters and the last fitted value which can be retrieved by the accessor method fit. The value returned in the classification and novelty detection problem is the decision function value phi. The accessor methods alpha returns the model parameters.

Details

The online algorithms are based on a simple stochastic gradient descent method in feature space. The state of the algorithm is stored in an object of class onlearn and has to be passed to the function at each iteration.

References

Kivinen J. Smola A.J. Williamson R.C. Online Learning with Kernels IEEE Transactions on Signal Processing vol. 52, Issue 8, 2004 http://users.cecs.anu.edu.au/~williams/papers/P172.pdf

See Also

inlearn

Examples

# NOT RUN {
## create toy data set
x <- rbind(matrix(rnorm(100),,2),matrix(rnorm(100)+3,,2))
y <- matrix(c(rep(1,50),rep(-1,50)),,1)

## initialize onlearn object
on <- inlearn(2,kernel="rbfdot",kpar=list(sigma=0.2),
              type="classification")

ind <- sample(1:100,100)
## learn one data point at the time
for(i in ind)
on <- onlearn(on,x[i,],y[i],nu=0.03,lambda=0.1)

## or learn all the data 
on <- onlearn(on,x[ind,],y[ind],nu=0.03,lambda=0.1)

sign(predict(on,x))
# }