Automatic neural network classifier with one hidden layer
[W,UNITS] = NEURC(A)
Automatically trained feed-forward neural network classifier. All parameters are automatically determined. The number of hidden units is the number of objects of A divided by the number of classes plus the feature size, but at least 2 and at most 100.
The network is initialised by RNNC and trained by LMNC. Training is stopped when the performance measured by an automatically generated tuning set of 1000 objects does not improve anymore or after the time set by PRTIME.
Uses the Mathworks' neural network toolbox. Consequently it is not available under Octave.
This routine escapes to KNNC if no classes have more than 2 objects.