Stream Suitable Online Support Vector Machines

Soft-margin support vector machines (SVMs) are a common class of classification models. The training of SVMs usually requires that the data be available all at once in a single batch, however the Stochastic majorization-minimization (SMM) algorithm framework allows for the training of SVMs on streamed data instead Nguyen, Jones & McLachlan(2018). This package utilizes the SMM framework to provide functions for training SVMs with hinge loss, squared-hinge loss, and logistic loss.


CRAN_Status_Badge Travis-CI BuildStatus DOI

The goal of SSOSVM is to use R to allow batch and online training of soft-margin support vector machines (SVMs). The training of SVMs usually requires that the data be available all at once in a single batch, however the Stochastic majorization-minimization (SMM) algorithm framework allows for the training of SVMs on streamed data instead http://doi.org/10.1007/s42081-018-0001-y. This package utilizes the SMM framework to provide functions for training SVMs with hinge loss, squared-hinge loss, and logistic loss, functions.

Installation

You can install SSOSVM from github with:

# install.packages("devtools")
devtools::install_github("andrewthomasjones/SSOSVM")

Example

Here is a very simple example using simulated data:

#setup
library(SSOSVM)
library(ggplot2)
 
#simulations
sims <- generateSim(100, DELTA=3)
 
#fit using various loss functions
sq1<-SVMFit(sims$YMAT,"square")
h1<-SVMFit(sims$YMAT,"hinge")
l1<-SVMFit(sims$YMAT,"logistic")
 
#plot results
plot<-ggplot(data.frame(sims$YMAT), aes(colour=factor(YY), x=V2, y=V3))
plot<-plot+geom_point()+theme_bw()+xlab("X")+ylab("Y")+guides(colour=FALSE)
plot<-plot+geom_abline(intercept=sq1$THETA[1],
                       slope=sq1$THETA[2]/sq1$THETA[3], colour="blue")
plot<-plot+geom_abline(intercept=h1$THETA[1],
                       slope=h1$THETA[2]/h1$THETA[3], colour="green")
plot<-plot+geom_abline(intercept=l1$THETA[1],
                       slope=l1$THETA[2]/l1$THETA[3], colour="red")
plot

Animated figures

Here is an animated example to demostrate the online nature of of the SSOSVM method:

library(ggplot2)
library(gganimate)
 
#set up
sims <- generateSim(10^2, DELTA=1.5)
 
#fit using various loss functions
sq1<-SVMFit(sims$YMAT,"square", returnAll = TRUE)
h1<-SVMFit(sims$YMAT,"hinge", returnAll = TRUE)
l1<-SVMFit(sims$YMAT,"logistic", returnAll = TRUE)
 
#dataframe
data<-data.frame(sample=1:10^2, 
                 sims$YMAT,
                 logistic_int=l1$THETA_list[,1],
                 square_int=sq1$THETA_list[,1],
                 hinge_int=h1$THETA_list[,1],
                 logistic_sl=l1$THETA_list[,2]/l1$THETA_list[,3],
                 square_sl=sq1$THETA_list[,2]/sq1$THETA_list[,3],
                 hinge_sl=h1$THETA_list[,2]/h1$THETA_list[,3])  
 
#base plot
plot<-ggplot(data, aes(colour=factor(YY), x=V2, y=V3))+ 
  geom_point(size=2)+theme_bw()+xlab("X")+ylab("Y")+
  guides(colour=FALSE)+geom_abline(size=1.6,alpha=.5, aes(intercept=square_int, slope=square_sl))
 
#animate
example <- plot + transition_time(sample)+
  labs(title =  "Sample: {frame_time}")+
  shadow_mark(alpha = 1, size = 1, exclude_layer = 2)
 
#save animation
anim_save("./inst/example.gif", example, fps=2.5)

News

SSOSVM 0.2.1

First published version

Reference manual

It appears you don't have a PDF plugin for this browser. You can click here to download the reference manual.

install.packages("SSOSVM")

0.2.1 by Andrew Thomas Jones, 14 days ago


Browse source code at https://github.com/cran/SSOSVM


Authors: Andrew Thomas Jones , Hien Duy Nguyen , Geoffrey J. McLachlan


Documentation:   PDF Manual  


GPL-3 license


Imports Rcpp, mvtnorm, MASS

Suggests testthat, knitr, rmarkdown, ggplot2, gganimate, gifski

Linking to Rcpp, RcppArmadillo


See at CRAN