RprobitB (version 1.2.0)

update_classes_wb: Weight-based class updates

Description

Weight-based class updates

Usage

update_classes_wb(
  s,
  b,
  Omega,
  epsmin = 0.01,
  epsmax = 0.7,
  deltamin = 0.1,
  deltashift = 0.5,
  identify_classes = FALSE,
  Cmax = 10L
)

Value

A list of updated values for s, b, and Omega and the indicator update_type which signals the type of class update:

• 0: no update

• 1: removed class

• 2: split class

• 3: merged classes

Arguments

s

[numeric(C)]
The vector of class weights.

b

[matrix(nrow = P_r, ncol = C)]
The matrix of class means as columns.

Omega

[matrix(nrow = P_r * P_r, ncol = C)]
The matrix of vectorized class covariance matrices as columns.

epsmin

[numeric(1)]
The threshold weight for removing a class.

epsmax

[numeric(1)]
The threshold weight for splitting a class.

deltamin

[numeric(1)]
The threshold difference in class means for joining two classes.

deltashift

[numeric(1)]
The scale for shifting the class means after a split.

identify_classes

[logical(1)]
Identify classes by decreasing class weights?

Cmax

[integer(1)]
The maximum number of classes, used to allocate space.

Details

The following updating rules apply:

• Class \(c\) is removed if \(s_c < \epsilon_{min}\).

• Class \(c\) is split into two classes, if \(s_c > \epsilon_{max}\).

• Two classes \(c_1\) and \(c_2\) are merged to one class, if \(||b_{c_1} - b_{c_2}|| < \delta_{min}\).

Examples

s <- c(0.7, 0.3)
b <- matrix(c(1, 1, 1, -1), ncol = 2)
Omega <- matrix(c(0.5, 0.3, 0.3, 0.5, 1, -0.1, -0.1, 0.8), ncol = 2)

### no update
update_classes_wb(s = s, b = b, Omega = Omega)

### remove class 2
update_classes_wb(s = s, b = b, Omega = Omega, epsmin = 0.31)

### split class 1
update_classes_wb(s = s, b = b, Omega = Omega, epsmax = 0.69)

### merge classes 1 and 2
update_classes_wb(s = s, b = b, Omega = Omega, deltamin = 3)