Pattern Recongnition Network: Class balancing

Hi,

How do I consider class balancing in a patternnet similar to the following example?

Class balancing in a pixel classification layer

Answer 1

[Niraj Gadakari]

You can specify the performance function in patternnet as crossentropy, and specify the performance weights to perform class balancing in Pattern Recognition Network.

[Thanujan Mohanadasan]

As far as I understand this, performance weights are not specified before the training step. But class balancing will affect the training, right?

[Greg Heath]

performance weights should be specified for training. greg

[Nick Van Berckelaer]

Can you provide instructions on or an example of how performance weights are to be specified during training? The parameter perfWeights is not a performParam as far as I see.

net = feedforwardnet(10);
net.performFcn = 'crossentropy';
net.performParam.regularization = 0.1;
net.performParam.normalization = 'none';