Algorithm: Compares Multiclass SVM, K nearest neighbor and Multiclass...

In gargjhanvi/SSLLabelGuide: Classify what data points should be labelled(for training) based on budget constraints and very small or no labelled data for semi supervised learning, predict the labels of unlabelled data and determine the confidence of labelling them

Description

Compares Multiclass SVM, K nearest neighbor and Multiclass Logistic regression on the labelled data points using five fold cross validation

Usage

Algorithm(Xtilde, Y, K = NULL)

Arguments

Xtilde	n * p centered and scaled data points
Y	A numeric n vector of labels of Xtilde
K	Number of classes, ( If K is NULL, it calculates K as maximum entry in Y)

Value

A list of three elements a list of three elements:

error_svm	returns five fold cross-validation error when Multiclass svm is used
error_knn	returns five fold cross-validation error when K-nearest neighbor is used with K = round(sqrt(N)) where N is the number of training data points
error_logistic	returns five fold cross-validation error when Multiclass logistic regression is used with default numIter = 50, eta = 0.1, lambda = 1

Examples

X <- rbind(matrix(rnorm(10,0,1),5,2),matrix(rnorm(10,1,2),5,2), matrix(rnorm(10,4,3),5,2))
fold_ids <- sample(1:15)
X <- X[fold_ids, ]
Xtilde <- standardizeX(X)$tilde
Y <- c(1,1,1,1,1,2,2,2,2,2,3,3,3,3,3)
Y <- Y[fold_ids]
out <- Algorithm(Xtilde,Y,3)

gargjhanvi/SSLLabelGuide documentation built on Dec. 20, 2021, 9:48 a.m.