LOGDENS

Force density based classifiers to use log-densities

    V = LOGDENS(W)
    V = W*LOGDENS

Input
W	Density based trained classifier

Output
V	Log-density based trained classifier

Description

Density based classifiers suffer from a low numeric accuracy in the tails  of the distributions. Especially for overtrained or high dimensional  classifiers this may cause zero-density estimates for many test samples,  resulting in a bad performance. This can be avoided by computing  log-densities offered by this routine. This works for all classifiers  based on normal distributions (e.g. LDC, QDC, MOGC) and Parzen estimates  (PARZENC, PARZENDC). The computation of log-densities is , in order to be

effective,	combined with a normalisation, resulting in posterior
distributions.	As a consequence, the possibility to output densities is
lost.

 SEE   MAPPINGS, LDC, UDC, QDC, MOGC, PARZENC, PARZENDC, NORMAL_MAP, PARZEN_MAP,  CLASSC