Neural Net Node Learning Rates

CAUTION:

This information relates to a deprecated version of the Neural Net modeling node, and is provided here for reference only. A newer version of the node, with enhanced features, is available in this release. See the topic Neural networks for more information. Although you can still build and score a model with the deprecated version, we strongly recommend using the newer version.

Neural net training is controlled by several parameters. These parameters can be set by using the Expert tab of the Neural Net node dialog box.

Alpha. A momentum term used in updating the weights during training. Momentum tends to keep the weight changes moving in a consistent direction. Specify a value between 0 and 1. Higher values of alpha increase momentum, decreasing the tendency to change direction based on local variations in the data.

Eta. The learning rate, which controls how much the weights are adjusted at each update. Eta changes as training proceeds for all training methods except RBFN, where eta remains constant. Initial Eta is the starting value of eta. During training, eta starts at Initial Eta, decreases to Low Eta, then is reset to High Eta and decreases to Low Eta again. The last two steps are repeated until training is complete.

Eta decay specifies the rate at which eta decreases, expressed as the number of cycles to go from High Eta to Low Eta. Specify values for each eta option.