TSVMSSLR {SSLR}

General Interface for TSVM (Transductive SVM classifier using the convex concave procedure) model

Description

model from RSSL package Transductive SVM using the CCCP algorithm as proposed by Collobert et al. (2006) implemented in R using the quadprog package. The implementation does not handle large datasets very well, but can be useful for smaller datasets and visualization purposes. C is the cost associated with labeled objects, while Cstar is the cost for the unlabeled objects. s control the loss function used for the unlabeled objects: it controls the size of the plateau for the symmetric ramp loss function. The balancing constraint makes sure the label assignments of the unlabeled objects are similar to the prior on the classes that was observed on the labeled data.

Usage

TSVMSSLR(
  C = 1,
  Cstar = 0.1,
  kernel = kernlab::vanilladot(),
  balancing_constraint = TRUE,
  s = 0,
  x_center = TRUE,
  scale = FALSE,
  eps = 1e-09,
  max_iter = 20,
  verbose = FALSE
)

Arguments

C

numeric; Cost parameter of the SVM

Cstar

numeric; Cost parameter of the unlabeled objects

kernel

kernlab::kernel to use

balancing_constraint

logical; Whether a balancing constraint should be enfored that causes the fraction of objects assigned to each label in the unlabeled data to be similar to the label fraction in the labeled data.

s

numeric; parameter controlling the loss function of the unlabeled objects (generally values between -1 and 0)

x_center

logical; Should the features be centered?

scale

If TRUE, apply a z-transform to all observations in X and X_u before running the regression

eps

numeric; Stopping criterion for the maximinimization

max_iter

integer; Maximum number of iterations

verbose

logical; print debugging messages, only works for vanilladot() kernel (default: FALSE)

References

Collobert, R. et al., 2006. Large scale transductive SVMs. Journal of Machine Learning Research, 7, pp.1687-1712.

Examples

library(tidyverse)
library(caret)
library(tidymodels)
library(SSLR)

data(breast)

set.seed(1)
train.index <- createDataPartition(breast$Class, p = .7, list = FALSE)
train <- breast[ train.index,]
test  <- breast[-train.index,]

cls <- which(colnames(breast) == "Class")

#% LABELED
labeled.index <- createDataPartition(breast$Class, p = .2, list = FALSE)
train[-labeled.index,cls] <- NA

library(kernlab)
m <- TSVMSSLR(kernel = kernlab::vanilladot()) %>% fit(Class ~ ., data = train)


#Accesing model from RSSL
model <- m$model