holdGroups - stops groups from updating activity but not weights

USAGE

    holdGroups [<group> ...]

DESCRIPTION

With no arguments, this returns a list of unheld groups. Otherwise, this holds each of the groups listed.

When a group is held it will not update its inputs or outputs. However, it will continue to backpropagate error derivatives. This is useful for networks with two converging paths in which one may want to hold one path steady while changing the inputs on the other path.

With standard networks, error derivatives will accumulate and will only be backpropagated once the groups are released. This saves unnecessary propagation while the groups are held. In continuous networks, backpropagation continues normally. Holding has not been tested for continuous networks, so it is recommended that you study the code before using it.

EXAMPLES

To hold the "input" and "hidden1" groups:

    lens> holdGroups input hidden1

SEE ALSO

releaseGroups, freezeNetwork


Last modified: Mon Jun 8 01:12:12 EDT 1998