Package {twinsvm}

twinsvm: Twin Support Vector Machines

Description

Twin support vector machine classifiers with a standard C-SVC baseline.

Author(s)

Maintainer: Shamika Tissera nimeshshamika@gmail.com

Authors:

• Shamika Tissera nimeshshamika@gmail.com

References

Jayadeva, Khemchandani, R., and Chandra, S. (2007). Twin support vector machines for pattern classification. IEEE Transactions on Pattern Analysis and Machine Intelligence, 29(5), 905-910.

Kumar, M. A. and Gopal, M. (2009). Least squares twin support vector machines for pattern classification. Expert Systems with Applications, 36(4), 7535-7543.

Extract Twin-SVM Coefficients

Description

Extract Twin-SVM Coefficients

Usage

## S3 method for class 'tsvm'
coef(object, ...)

Arguments

Value

A list with the two plane coefficients.

Examples

dat <- gen_moons(30)
fit <- tsvm(dat$x, dat$y)
coef(fit)

Compare Twin-SVM and Standard SVM Boundaries

Description

Fits least-squares twin SVM, original QP twin SVM, and the standard C-SVC SVM baseline on the same two-dimensional data, then plots the three decision boundaries side by side.

Usage

compare_methods(x, y, kernel = "rbf", gamma = 0.5, c1 = 1, c2 = 1, cost = 1)

Arguments

Value

A faceted ggplot object.

See Also

Other visualization: kernel_lift(), lift_plot(), lift_plotly()

Examples

set.seed(30)
dat <- gen_moons(50, noise = 0.1)
compare_methods(dat$x, dat$y, gamma = 1, c1 = 0.2, c2 = 0.2, cost = 1)

Confusion Matrix and Accuracy

Description

Predicts on x and returns a confusion matrix plus overall accuracy. This works for both binary and one-vs-one multiclass tsvm and svms fits.

Usage

confusion(object, x, y)

Arguments

Value

A list with table, a table(truth, predicted) confusion matrix, and accuracy, the overall classification accuracy.

See Also

Other multiclass: predict.svms_multiclass(), predict.tsvm_multiclass(), print.svms_multiclass(), print.tsvm_multiclass()

Examples

set.seed(50)
x <- rbind(
  matrix(rnorm(20, -2, 0.2), ncol = 2),
  matrix(rnorm(20, 0, 0.2), ncol = 2),
  matrix(rnorm(20, 2, 0.2), ncol = 2)
)
y <- factor(rep(c("a", "b", "c"), each = 10))
fit <- tsvm(x, y, kernel = "linear")
confusion(fit, x, y)

Cross-Validate Twin-SVM Hyperparameters

Description

Runs k-fold cross-validation over c1, c2, and optionally gamma.

Usage

cv_tsvm(x, y, c1_grid, c2_grid, gamma_grid = NULL, k = 5, ...)

Arguments

Value

A cv_tsvm object with best_params and results.

Examples

set.seed(10)
dat <- gen_moons(40, noise = 0.1)
cv <- cv_tsvm(dat$x, dat$y, c1_grid = c(0.1, 1), c2_grid = c(0.1, 1), k = 3)
cv$best_params

Generate Concentric Circles Data

Description

Generate Concentric Circles Data

Usage

gen_circles(n, noise = 0.05)

Arguments

Value

A list with numeric matrix x and factor y.

Examples

dat <- gen_circles(20, noise = 0.05)
str(dat)

Generate Two-Moons Data

Description

Generate Two-Moons Data

Usage

gen_moons(n, noise = 0.2)

Arguments

Value

A list with numeric matrix x and factor y.

Examples

dat <- gen_moons(20, noise = 0.1)
str(dat)

Compatibility Alias for the RBF Kernel Lift Plot

Description

kernel_lift() is kept for existing code. New code should call lift_plot().

Usage

kernel_lift(x, y, gamma = 1, center = NULL)

Arguments

Value

A ggplot object.

See Also

Other visualization: compare_methods(), lift_plot(), lift_plotly()

Examples

set.seed(22)
dat <- gen_circles(60, noise = 0.04)
kernel_lift(dat$x, dat$y, gamma = 1)

Static RBF Kernel Lift Plot

Description

Maps two-dimensional data to a third RBF height and draws a static oblique projection. The height is ⁠exp(-gamma * ||x - center||^2)⁠, so points near the center rise while points farther away stay low. A translucent plane is drawn halfway between the two class mean heights.

Usage

lift_plot(x, y, gamma = 1, center = NULL)

Arguments

Value

A ggplot object.

See Also

Other visualization: compare_methods(), kernel_lift(), lift_plotly()

Examples

set.seed(20)
dat <- gen_circles(80, noise = 0.04)
lift_plot(dat$x, dat$y, gamma = 1)

Interactive RBF Kernel Lift Plot

Description

Draws the same RBF lift as lift_plot() as a rotatable plotly 3D chart. This function is optional; the package does not require plotly to build or check.

Usage

lift_plotly(x, y, gamma = 1, center = NULL)

Arguments

Value

A plotly object.

See Also

Other visualization: compare_methods(), kernel_lift(), lift_plot()

Examples

set.seed(21)
dat <- gen_circles(80, noise = 0.04)
if (requireNamespace("plotly", quietly = TRUE)) {
  lift_plotly(dat$x, dat$y, gamma = 1)
}

Animate a Decision Boundary Across a Hyperparameter Range

Description

Refits a model over a sequence of hyperparameter values and returns a gganimate object showing the boundary change.

Usage

morph_boundary(
  x,
  y,
  param = c("gamma", "cost", "c1"),
  range,
  model = c("tsvm", "svms"),
  n = 30,
  ...
)

Arguments

Value

A gganim object.

Examples

if (interactive()) {
  dat <- gen_moons(40)
  morph_boundary(dat$x, dat$y, param = "gamma", range = c(0.5, 2), n = 4)
}

Plot Twin-SVM Cross-Validation Results

Description

Plot Twin-SVM Cross-Validation Results

Usage

## S3 method for class 'cv_tsvm'
plot(x, ...)

Arguments

Value

A ggplot object.

Examples

set.seed(11)
dat <- gen_moons(40, noise = 0.1)
cv <- cv_tsvm(dat$x, dat$y, c1_grid = c(0.1, 1), c2_grid = c(0.1, 1), k = 3)
plot(cv)

Plot the Accuracy Spread from Repeated Splits

Description

Draws the per-split test accuracies as a boxplot, one box per model, so the dispersion is visible alongside the mean.

Usage

## S3 method for class 'resample_accuracy'
plot(x, ...)

Arguments

Value

A ggplot object.

Examples

set.seed(2)
dat <- gen_moons(120, noise = 0.15)
rs <- resample_accuracy(dat$x, dat$y, reps = 10, kernel = "rbf", gamma = 1)
plot(rs)

Plot a Standard SVM Decision Boundary

Description

Plot a Standard SVM Decision Boundary

Usage

## S3 method for class 'svms'
plot(x, ...)

Arguments

Value

A ggplot object.

Examples

dat <- gen_moons(40)
fit <- svms(dat$x, dat$y, kernel = "linear")
plot(fit)

Plot a Twin-SVM Decision Boundary

Description

Plot a Twin-SVM Decision Boundary

Usage

## S3 method for class 'tsvm'
plot(x, ...)

Arguments

Value

A ggplot object.

Examples

dat <- gen_moons(40)
fit <- tsvm(dat$x, dat$y)
plot(fit)

Predict from a Standard SVM

Description

Predict from a Standard SVM

Usage

## S3 method for class 'svms'
predict(object, newdata, decision.values = FALSE, ...)

Arguments

Value

A factor of predicted classes, or a numeric vector when decision.values = TRUE.

Examples

set.seed(2)
dat <- gen_moons(30)
fit <- svms(dat$x, dat$y)
predict(fit, dat$x, decision.values = TRUE)

Predict from a Multiclass Standard SVM

Description

Predicts from a one-vs-one multiclass standard SVM. Each binary model votes for one class. Ties are resolved deterministically by choosing the class that appears first in the training factor level order.

Usage

## S3 method for class 'svms_multiclass'
predict(object, newdata, type = c("class", "votes"), ...)

Arguments

Value

A factor of predicted classes, or an integer vote matrix when type = "votes".

See Also

Other multiclass: confusion(), predict.tsvm_multiclass(), print.svms_multiclass(), print.tsvm_multiclass()

Examples

set.seed(45)
x <- rbind(
  matrix(rnorm(8, -2, 0.2), ncol = 2),
  matrix(rnorm(8, 0, 0.2), ncol = 2),
  matrix(rnorm(8, 2, 0.2), ncol = 2)
)
y <- factor(rep(c("a", "b", "c"), each = 4))
fit <- svms(x, y, kernel = "linear", max_passes = 2, max_iter = 100)
predict(fit, x[1:3, , drop = FALSE])

Predict from a Multiclass Twin SVM

Description

Predicts from a one-vs-one multiclass twin SVM. Each binary model votes for one class. Ties are resolved deterministically by choosing the class that appears first in the training factor level order.

Usage

## S3 method for class 'tsvm_multiclass'
predict(object, newdata, type = c("class", "votes"), ...)

Arguments

Value

A factor of predicted classes, or an integer vote matrix when type = "votes".

See Also

Other multiclass: confusion(), predict.svms_multiclass(), print.svms_multiclass(), print.tsvm_multiclass()

Examples

set.seed(40)
x <- rbind(
  matrix(rnorm(20, -2, 0.2), ncol = 2),
  matrix(rnorm(20, 0, 0.2), ncol = 2),
  matrix(rnorm(20, 2, 0.2), ncol = 2)
)
y <- factor(rep(c("a", "b", "c"), each = 10))
fit <- tsvm(x, y, kernel = "linear")
predict(fit, x[1:3, , drop = FALSE])

Print a Multiclass Standard SVM

Description

Print a Multiclass Standard SVM

Usage

## S3 method for class 'svms_multiclass'
print(x, ...)

Arguments

Value

The input object, invisibly.

See Also

Other multiclass: confusion(), predict.svms_multiclass(), predict.tsvm_multiclass(), print.tsvm_multiclass()

Examples

set.seed(46)
x <- rbind(
  matrix(rnorm(8, -2, 0.2), ncol = 2),
  matrix(rnorm(8, 0, 0.2), ncol = 2),
  matrix(rnorm(8, 2, 0.2), ncol = 2)
)
y <- factor(rep(c("a", "b", "c"), each = 4))
print(svms(x, y, kernel = "linear", max_passes = 2, max_iter = 100))

Print a Multiclass Twin SVM

Description

Print a Multiclass Twin SVM

Usage

## S3 method for class 'tsvm_multiclass'
print(x, ...)

Arguments

Value

The input object, invisibly.

See Also

Other multiclass: confusion(), predict.svms_multiclass(), predict.tsvm_multiclass(), print.svms_multiclass()

Examples

set.seed(41)
x <- rbind(
  matrix(rnorm(20, -2, 0.2), ncol = 2),
  matrix(rnorm(20, 0, 0.2), ncol = 2),
  matrix(rnorm(20, 2, 0.2), ncol = 2)
)
y <- factor(rep(c("a", "b", "c"), each = 10))
print(tsvm(x, y, kernel = "linear"))

Accuracy and Its Spread over Repeated Random Splits

Description

Evaluates a classifier by repeated random holdout: the data are split into training and test sets reps times, the model is refitted on each training set, and test accuracy is recorded. Reporting the mean together with its between-split standard deviation and standard error turns a single accuracy number into an estimate with a stated uncertainty, so that differences between methods can be read against the noise rather than taken at face value.

Usage

resample_accuracy(
  x,
  y,
  model = c("both", "tsvm", "svms"),
  reps = 50L,
  prop = 0.7,
  seed = NULL,
  ...
)

Arguments

Details

The splits are drawn once, up front, and shared by every model, so when model = "both" the twin SVM and the standard SVM are compared on identical training and test sets (a paired comparison). Splits are stratified by class, so each class keeps roughly its overall proportion in both parts.

Value

An object of class resample_accuracy, a list with

summary

a data frame with one row per model giving mean, sd, se and n of the test accuracy;

accuracy

a long data frame of the per-split accuracies (model, rep, accuracy);

reps, prop, models, call

the settings used.

The plot() method draws the accuracy distributions as boxplots.

Examples

set.seed(1)
dat <- gen_moons(120, noise = 0.15)
rs <- resample_accuracy(dat$x, dat$y, model = "both",
                        reps = 10, kernel = "rbf", gamma = 1, seed = 1)
rs                 # mean, sd and se per model
plot(rs)           # accuracy spread as boxplots

Fit a Standard C-SVC Support Vector Machine

Description

Fits a C-SVC support vector machine with a Platt SMO solver. With two classes this uses the validated binary path. With three or more classes, the function fits one binary SVM for each class pair and predicts by majority vote. Multiclass ties are resolved by choosing the class that appears first in the factor level order.

Usage

svms(
  x,
  y,
  kernel = c("linear", "rbf", "poly"),
  cost = 1,
  gamma = NULL,
  degree = 3,
  coef0 = 1,
  tol = 0.001,
  max_passes = 10L,
  max_iter = 10000L
)

Arguments

Value

A fitted svms object for two classes, or a svms_multiclass object for three or more classes.

Examples

set.seed(1)
dat <- gen_moons(40, noise = 0.1)
fit <- svms(dat$x, dat$y, kernel = "linear", cost = 1)
predict(fit, dat$x[1:3, ])

Fit a Twin Support Vector Machine

Description

Fits a twin support vector machine. With two classes this uses the validated binary twin-SVM path: level 1 of y is class B, level 2 is class A, plane 1 is close to class A, and plane 2 is close to class B. With three or more classes, the function fits one binary twin SVM for each class pair and predicts by majority vote. Multiclass ties are resolved by choosing the class that appears first in the factor level order.

Usage

tsvm(
  x,
  y,
  method = c("ls", "twin"),
  kernel = c("linear", "rbf", "poly"),
  c1 = 1,
  c2 = 1,
  gamma = NULL,
  degree = 3,
  coef0 = 1,
  eps = 1e-06
)

Arguments

Value

A fitted tsvm object for two classes, or a tsvm_multiclass object for three or more classes.

References

Jayadeva, Khemchandani, R., and Chandra, S. (2007). Twin support vector machines for pattern classification. IEEE Transactions on Pattern Analysis and Machine Intelligence, 29(5), 905-910.

Kumar, M. A. and Gopal, M. (2009). Least squares twin support vector machines for pattern classification. Expert Systems with Applications, 36(4), 7535-7543.

Examples

set.seed(3)
dat <- gen_moons(50, noise = 0.05)
fit <- tsvm(dat$x, dat$y)
predict(fit, dat$x[1:4, ])

Example Two-Moons Data

Description

A small two-class two-moons data set for examples and tests.

Usage

data(twomoons)

Format

A data frame with 120 rows and 3 variables:

x1

First coordinate.

x2

Second coordinate.

y

Two-class factor with levels B and A.

Examples

data(twomoons)
str(twomoons)