5.1.2 NeuralFit

The initialized FF network is trained with NeuralFit. This command, which is also used for radial basis function networks and for dynamic networks, is described here with all its variants. Section 2.5.3, Training Feedforward and Radial Basis Function Networks, describes the algorithms in some detail.

Train a feedforward network.

NeuralFit returns a list of two variables. The first one is the trained FF network, and the second is a record containing information about the training.

An existing network can be submitted for more training by setting net equal to the network or its training record. The advantage of submitting the training record is that the information about the earlier training is combined with the additional training.

During the training, intermediate results are displayed in an automatically created notebook. After each training iteration the following information is displayed:

Training iteration number

The value of the root-mean-square error (RMSE)

For validation data submitted in the call, the RMSE value for this second data set is also displayed

The step length control parameter of the minimization algorithm ( or ), which is described in Section 2.5.3, Training Feedforward and Radial Basis Function Networks

At the end of the training, a plot is displayed with RMSE as a function of iteration.

Using the options of NeuralFit, as described in Section 7.7, Options Controlling Training Results Presentation, you can change the way the training results are presented.

At the end of the training you often receive different warning messages. These give you information on how the training performed. By looking at the performance plot, you can usually tell whether more training is required. If the plot has not flattened out toward the end of the training, then you should consider applying more training iterations. This can be done by resubmitting the trained network to NeuralFit so that you do not have to initiate training anew.

There are many options to NeuralFit that can be used to modify its behavior. They are described in Section 7.1, NeuralFit.

A derived FF network can be applied to new inputs using function evaluation. The result given by the network is its estimate of the output.

Function evaluation of a feedforward network.

The input argument x can be a vector containing one input sample, or a matrix containing one input sample on each row.

The function evaluation has one option.

Option of the network evaluation rule.