twinsvm

twinsvm brings twin support vector machines to R with a standard C-SVC SVM baseline, one-vs-one multiclass classification, and visualization-first helpers for small to moderate problems.

Install locally

install.packages("twinsvm_0.0.1.tar.gz", repos = NULL, type = "source")
library(twinsvm)

Quick start

set.seed(1)
dat <- gen_moons(100, noise = 0.12)

fit <- tsvm(dat$x, dat$y, kernel = "rbf", gamma = 2)
pred <- predict(fit, dat$x)
mean(pred == dat$y)

plot(fit)

Multiclass

tsvm() and svms() support three or more classes with one-vs-one majority voting. Ties are resolved by the first factor level.

set.seed(4)
x3 <- rbind(
  matrix(rnorm(30, -2, 0.25), ncol = 2),
  cbind(rnorm(15, 2, 0.25), rnorm(15, -2, 0.25)),
  matrix(rnorm(30, 2, 0.25), ncol = 2)
)
y3 <- factor(rep(c("alpha", "beta", "gamma"), each = 15))

multi <- tsvm(x3, y3, kernel = "linear")
predict(multi, x3[1:5, , drop = FALSE])
predict(multi, x3[1:5, , drop = FALSE], type = "votes")
confusion(multi, x3, y3)

Visualization

Show the RBF lift as a static pseudo-3D ggplot:

circles <- gen_circles(100, noise = 0.04)
lift_plot(circles$x, circles$y, gamma = 1)

Compare least-squares twin SVM, original QP twin SVM, and the standard SVM baseline side by side:

compare_methods(dat$x, dat$y, gamma = 1, c1 = 0.2, c2 = 0.2, cost = 1)

Compare against the standard SVM baseline:

base <- svms(dat$x, dat$y, kernel = "rbf", cost = 1, gamma = 2)
mean(predict(base, dat$x) == dat$y)

Tune the twin SVM:

cv <- cv_tsvm(
  dat$x, dat$y,
  c1_grid = c(0.1, 1),
  c2_grid = c(0.1, 1),
  gamma_grid = c(1, 2),
  kernel = "rbf",
  k = 3
)
cv$best_params
plot(cv)

Validation note

For the standard SVM baseline, tests compare predictions against e1071, the LIBSVM-backed R package. For twin SVM there is no existing R implementation to compare against, so tests check the defining plane-distance behavior, nonlinear kernel improvement, and agreement between least-squares and original QP twin-SVM variants. Kernel twin-SVM models invert an (n + 1) matrix and are intended for small to moderate data sets.