resetNet - randomizes all weights and clears direction information

USAGE

    resetNet

DESCRIPTION

This randomizes all non-frozen weights in the network, clears weight update direction information, resets unit outputs and other values, and sets the network's totalUpdates, error, and several other values to zero. This is typically used to begin training the network from scratch.

Weight randomization is based on the randMean and randRange of the link's block. If one of these is NaN, it defaults to the parameters of the receiving group and then the network.

SEE ALSO

randWeights, train


Last modified: Wed Nov 15 11:10:34 EST 2000