6.1.1 InitializeRBFNet

RBF networks are initialized with InitializeRBFNet.

Initialize a RBF network.

The returned network is an object with head RBFNet, following the general format described in Section 3.2.3, Network Format. The FeedForwardNet and the RBFNet have one replacement rule more then the other network models. Its name is NumberOfInputs and it indicates how many inputs the network take.

The number of inputs and outputs in the network are extracted from the number of columns in the input and output data, so only the number of neurons needs to be specified.

InitializeRBFNet takes almost the same options as InitializationFeedforwardNet. However, some of the default values for the options are different, as indicated in the table that follows.

Options of InitializeRBFNet.

The parameters of the network can be initialized in three different ways depending on the option RandomInitialization: False, which is the default; True; and LinearParameters. The default initialization is usually the best. It gives a random distribution of the basis centers over the input data range. This make sense since only basis functions that cover the data range can be used in the function approximation. Also the widths of the basis functions are scaled using the input data range. The linear parameters are fitted with the least-squares algorithm to the output data. The meanings of the options are the same as for feedforward networks, and they are further described in Section 5.1.1, InitializeFeedForwardNet. You can also define your own initialization algorithm and insert the parameters in an RBFNet object as described in Section 13.1, Change the Parameter Values of an Existing Network.

The options Regularization and FixedParameters can be set at the initialization of the network, or when the network is trained with NeuralFit. You can learn how to use these options in Section 7.5, Regularization and Stopped Search and Section 13.2, Fixed Parameters.

The default value of the option Neuron is Exp. You then obtain an RBF network with the Gaussian bell function as the basis function, which is the most commonly used choice. Section 13.3, Select Your Own Neuron Function, describes how you can use other basis functions.