dougsMomentum - trains the network using bounded momentum descent


    dougsMomentum [<num-updates>] [-report <report-interval> | -setOnly]


This is a shortcut for training the network using "Doug's momentum descent". The arguments are similar to those for train. The momentum term is taken from the network's momentum parameter.

Doug's momentum descent is similar to standard momentum descent with the exception that the pre-momentum weight step vector is bounded so that its length cannot exceed 1.0. After the momentum is added, the length of the resulting weight change vector can grow as high as 1 / (1 - momentum). This change allows stable behavior with much higher initial learning rates, resulting in less need to adjust the learning rate as training progresses.

If the -setOnly flag is used, no training will occur. However, the network's numUpdates, reportInterval, and default algorithm will be set. This can be used to set the default training behavior in an initialization script prior to actually training.


To train for 1000 epochs, printing reports every 100:

    lens> dougsMomentum 1000 -r 100


train, steepest, momentum, deltaBarDelta

Last modified: Fri Nov 17 12:37:06 EST 2000