rankSVMcompare: an R package which implements Support Vector Machines for ranking and comparing. https://arxiv.org/abs/1401.8008

Installation

install.packages("quadmod", repos="http://r-forge.r-project.org")
devtools::install_github("tdhock/rankSVMcompare")

Usage examples

The main function for fitting the Support Vector Machine ranking/comparison model is softCompareQP.

library(rankSVMcompare)
example(softCompareQP)

The softCompareQP function returns a list which represents the trained model. If fit is the trained model object, then

• fit$rank is a ranking function r(x) that takes an object x and returns a real-valued rank score – bigger is better.
• fit$predict is a comparison function c(x1, x2) that takes two objects x1 and x2, and returns
  • 1 if x2 is significantly better,
  • -1 if x1 is significantly better,
  • 0 if there is no significant difference between x1 and x2.
• in our model, an absolute rank difference greater than 1 is considered a significant difference.
• the model is trained on labeled pairs (x1,x2,y) where x1 and x2 are p-dimensional feature vectors, and y is a label (-1, 0, 1).