How many characters? How many extracted features? What kind of data normalization? What kind of input dimensionality reduction? Are the targets columns from a unit matrix? What kind of trn/val/tst data division? What was your training goal? What kind of performance is obtained in te literature for tis kind of problem?
When you used PATTERNNET how did you choose the number of hidden nodes, H?
[ I N ] = size(inputs) = ?
[ O N ] = size(targets) = ?
Post your patternnet code.
