NEURC

Automatic neural network classifier

    W = NEURC (A,UNITS)

Input
A	Dataset
UNITS	Number of units  Default: 0.2 x size smallest class in A.

Output
W	Trained feed-forward neural network mapping

Description

Automatically trained feed-forward neural network classifier with UNITS units in a single hidden layer. Training, by LMNC, is stopped when the  performance on an artificially generated tuning set of 1000 samples per  class (based on k-nearest neighbour interpolation) does not improve anymore.

NEURC always tries three random initialisations, with fixed random seeds,  and returns the best result according to the tuning set. This is done in  order to obtain a reproducable result.

If UNITS is NaN it is optimised by REGOPTC. This may take a long  computing time and is often not significantly better than the default.

Uses the Mathworks' neural network toolbox.

See also

mappings, datasets, lmnc, bpxnc, gendatk, regoptc,