LOGMLC

Logistic Multi-Class Linear Classifier (Multinomial Logit)

    W = LOGMLC(A)
    W = LOGMLC(A, LABLIST_NAMES)

Input
A	Dataset
LABLISTNAMES	The character array of additional labelings which priors  (along with the current labeling prior) have to be taken  into account to compute object weights. It is useful if  there are object subgroups which abundances in the  (optional; default: '')

Output
W	Logistic Multi-Class Linear Classifier

Description

Computation of the linear classifier for the dataset A by maximizing the  likelihood of the assumed posterior probalility model

    p(c|x) = exp(w_c'x + b_c)/(exp(w_1'x + b_1) + exp(w_2'x + b_2) + ...)

Differences from the LOGLC
1) Genuine multi-class classification. All weights are optimized simultaneously.  2) More careful Newtonian optimization procedure.  3) The use of Cholesky factorization instead of pinv for Hessian inversion.

See also

mappings, loglc,