SemiSupervised (version 1.0)

SemiSupervised.control: Control Parameters for the S4-generic generic functions derived from virtual class SemiSupervised

Description

Controls various aspects of fitting any ‘SemiSupervised’ object.

Usage

SemiSupervised.control(normalize=TRUE,stability=NULL,k=NULL,nok=FALSE,
dissimilar=TRUE,l.eps=1e-5,l.thresh=25L,h.thresh=1e-5,U.as.anchor.thresh=600L,
U.as.anchor=TRUE,sig.est=TRUE,sig.frac=0.5,iter.max=1000L,
anchor.seed=100,sfac=5L,cn=4L,LAE.thresh=100L,LAE.eps=1e-4,
cv.fold=3L,cv.seed=100L,cv.cl=TRUE,cv.type="scv",cv.adjust=0.001)

Arguments

normalize

flags whether or not the normalized or combinatorial graph operator should be used. This is not used by ‘jtharm’. Further, it has no effect on anchor graph based approaches.

stability

stabilization parameter for necessary inverses. A NULL value allows it to be set internally (recommended) but can be set manually.

k

in fitting a distance graph with the ‘formula’ as y~. this will set the default k-NN graph parameter. It is not used otherwise. In the case of an anchor graph the k parameter is the number of anchors used by k-means.

nok

flags the y~. ‘formula’ call to either fit or not fit a k-NN parameter which overrides the k argument.

dissimilar

flags whether the graph is similar or dissimilar. This is not necessary in the ‘formula’ call since a call to sG in the formula automatically flags the graph as similarity or a call to dG to dissimilar. This is necessary in all non-formula calls with similarity graphs.

sfac

the ‘s’ parameter for the LAE method.

cn

the ‘cn’ parameter for the LAE method.

LAE.thresh

thresh hold for LAE algorithm.

LAE.eps

convergence tolerance for LAE algorithm.

iter.max

maximum number of iterations for kmeans.

anchor.seed

sets the seed for the kmeans algorithm.

sig.est

use an internal estimation scheme to estimate the parameter for an rbf kernel.

sig.frac

when ‘sig.est’ is true, the fraction of training data necessary for the computaton is used.

l.thresh

max iteration parameter for the underlying logistic regression algorithm when fitting classification.

l.eps

threshold for stopping the underlying logistic regression algorithm when fitting classification.

U.as.anchor

if n>U.as.anchor.thresh then the anchor points are fitted as the unlabeled cases to speed up the approach. This only works in the formula call where the graph is unspecified, i.e., y~.

U.as.anchor.thresh

threshold for determining when the unlabeled are to be anchors.

h.thresh

minimum allowable choice for h in the grid used by CV.

cv.fold

the number of folds used for K-fold CV.

cv.seed

the seed to generate the folds for K-fold CV.

cv.cl

forces in classification at least two distinct responses in each fold and should be set to TRUE.

cv.type

use "scv" or "cv". The "scv" is faster and should be used.

cv.adjust

forces an inverse stabilization in "scv".