MKLR: Multi-class Kernel Logistic Regression

In hyj12345/MKLR: Multi-class Kernel Logistic Regression

Description

This function fit a Multi-class Kernel Logistic Regression model to the data (y, x) using some pre-specified kernel. The return list contains the estimated kernel weights as well as the original data to perform predictions.There are two types of kernel, they are 'RBF' and 'polynomial'

Usage

MKLR(
  y,
  x,
  kernel = c("RBF", "polynomial")[1],
  lambda = 0.01,
  sigma2 = 1,
  d = 3,
  lr = 0.001,
  threshold = 1e-06,
  max_iter = 1000
)

Arguments

y	A n x 1 column vector containing multiclass responses.
x	A n x p matrix containing the covariates.
kernel	The kernel to use. Either RBF (default) or polynomial.
lambda	The regularization parameter.
sigma2	The scale in the RBF and polynomial kernel. See details.
d	The degree in the polynomial kernel.
threshold	The convergence threshold.
max_iter	The maximum number of iterations.

Details

The RBF kernel has the following form:

exp(-||x-y||^2/sigma2).

The polynomial kernel has the following form:

(1+x'y/sigma2)^d.

Value

A list containing:

x

The original x.

alpha

The vector of fitted weights.

kernel

The kernel.

sigma2

The scale parameter.

d

The polynomial degree.

See Also

predict.MKLR

hyj12345/MKLR documentation built on Dec. 20, 2021, 5:54 p.m.