7.4 Troubleshooting

Sometimes, due to numerical problems, the training stops before reaching the minimum and without completing the iterations specified. Listed here are some measures you can take to possibly circumvent this situation.

Re-initialize the network model and repeat the training.

Try a different training algorithm by changing the option Method (See Section 7.1, NeuralFit.)

Exclude (or include if it excluded) a linear part in the network by setting the option LinearPartFalse. (See Section 5.1.1, InitializeFeedForwardNet, or Section 6.1.1, InitializeRBFNet.)

Decrease/increase the number of neurons in the network model.

Check that the data is reasonably scaled so that unnecessary numeric problems are avoided.