mlffNetworkInit¶
Initializes an Imsls_d_NN_Network data structure for use in training a neural network.
Synopsis¶
mlffNetworkInit (nInputs, nOutputs)
Required Arguments¶
- int
nInputs(Input) - The number of network inputs. If the network uses nominal input attributes, the number of inputs equals the number of encoded columns used to represent these attributes plus the number of continuous and ordinal input attributes, if any.
- int
nOutputs(Input) - The number of network outputs. For neural networks used for forecasting
continuous responses,
nOutputsis equal to the number of variables being forecasted. Networks used for binary classification have only one output. Other classification networks have one output for every possible target category.
Return Value¶
An Imsls_d_NN_Network data structure initialized with the number of inputs
and outputs specified by nInputs and nOutputs. To release this space
use mlffNetworkFree.
Description¶
The function mlffNetworkInit is used to initialize the network, the
function mlffNetwork is used to build up the
network in preparation for training, and the function
mlffNetworkFree is used to free the structure.
Function mlffNetworkInit initializes and returns an Imsls_d_NN_Network
data structure. This structure is required input to
mlffNetwork and the network trainers. This function
initializes the structure to accommodate a network with the number of inputs
and outputs specified by nInputs and nOutputs respectively. This
function must be called prior to building the complete network architecture
using mlffNetwork.
Function mlffNetwork modifies the structure initialized by this function and builds the network architecture consisting of hidden layers, perceptrons and links among these objects. This architecture is independent of the training data. Once the architecture is complete, the Imsls_d_NN_Network data structure can be used with mlffNetworkTrainer or mlffClassificationTrainer to train the network.
After the network is trained, not only does the Imsls_d_NN_Network data structure contains a description of the network architecture, it also contains the network weights needed for neural network forecasting or classification.