qda-train {stochmod}R Documentation

Quadratic Discriminant Analysis

Description

Training method for QDA.

Usage

QDA.train( x, y, cov.reg = 0.0 )

Arguments

x N x p data matrix of N samples in p dimensions
y N x 1 vector of labels
cov.reg Covariance matrix regularization (towards identity), value must be in [0, 1]

Details

Models each class as a single (multivariate) Gaussian. Relaxes the common covariance matrix constraint of LDA. Computes the distribution parameters, the Bayesian class priors, and the discriminant functions. QDA is insensitive to temporal structure of the data and, therefore, only needs to work with a single observation sequence. This, in turn, requires a label for each sample.

Value

An QDA classifier defined by:

labels Vector of unique class labels
priors K x 1 vector of priors, estimated as fraction of points from each class
means K x p matrix of means approximated from the data
covmats K x p x p array of covariance matrices esimated from the data
icovmats K x p x p array of inverse covariance matrices
bias K x 1 vector of bias terms for discriminant function computations

Author(s)

Artem Sokolov Artem.Sokolov@gmail.com


[Package stochmod version 1.2 Index]