PERLC

PERLC - Trainable linear perceptron classifier

    W = PERLC(A,MAXITER,ETA,W_INI,TYPE)
    W = A*PERLC([],MAXITER,ETA,W_INI,TYPE)
    W = A*PERLC(MAXITER,ETA,W_INI,TYPE)

Input
A	Training dataset
MAXITER	Maximum number of iterations (default 1000)
ETA	Learning rate (default 0.1)
W_INI	Initial weights, as affine mapping, e.g W_INI = NMC(A) (default: random initialisation)
TYPE	'batch': update by batch processing (default)  'seq' : update sequentially

Output
W	Linear perceptron classifier mapping

Description

Outputs a perceptron W trained on dataset A using learning rate ETA for  a maximum of MAXITER iterations (or until convergence).

If ETA is NaN it is optimised by REGOPTC.

The resulting linear base-classifiers are combined by the maximum  confidence rule. A better combiner usually will be QDC, e.g.  W = A*(PERLC*QDC([],[],1e-6)).

See also

datasets, mappings, nmc, fisherc, bpxnc, lmnc, regoptc, fishercc,