PARZENC

Optimisation of the Parzen classifier

   [W,H] = PARZENC(A)
   W = PARZENC(A,H,FID)

Input
A	dataset
H	smoothing parameter (may be scalar, vector of per-class  parameters, or matrix with parameters for each class (rows) and  dimension (columns))
FID	File ID to write progress to (default [], see PRPROGRESS)

Output
W	trained mapping
H	estimated smoothing (scalar value)

Description

Computation of the optimum smoothing parameter H for the Parzen  classifier between the classes in the dataset A. The leave-one-out  Lissack & Fu estimate is used for the classification error E. The  final classifier is stored as a mapping in W. It may be converted  into a classifier by W*CLASSC. PARZENC cannot be used for density  estimation.

In case smoothing H is specified, no learning is performed, just the  discriminant W is produced for the given smoothing parameters H.  Smoothing parameters may be scalar, vector of per-class parameters, or  a matrix with individual smoothing for each class (rows) and feature  directions (columns)

Reference(s)

T. Lissack and K.S. Fu, Error estimation in pattern recognition via L-distance between posterior density functions, IEEE Trans. Inform. Theory, vol. 22, pp. 34-45, 1976.

See also

datasets, mappings, parzen_map, parzenml, parzendc, classc, prprogress,