The Complete SPRLIB & ANNLIB

join_net

- construct a new network out of a number of networks

SYNOPSIS

NET *join_net (n1, net1, n2, net2)

ARGUMENTS

int n1; The number of networks in the top part.
NET **net1; An array of top part network pointers.
int n2; The number of networks in the bottom part.
NET **net2; An array of bottom part network pointers.

RETURNS

A new network, constructed out of the networks given as arguments.

FUNCTION

This function can be used to build modular networks. Several networks can be trained and concatenated to one new network by this routine. Networks can be concatenated from left to right and from top to bottom. The user supplies two arrays of pointers to network structures. The argument net1 is an array of networks that should go in the top part of the new network, net2 contains networks that should go in the bottom part of the network. The total number of output units in networks net1 should equal the total number of input units in networks net2. The same holds for maps, in case of shared weights neural networks. All networks in net1 should have the same number of layers; this also holds for the networks in net2. Larger modular networks can be built by repeatedly calling this function.

NOTE

The original networks in net1 and net2 are no longer usable after a call to this function, since many pointers will have been diverted. This routine is still experimental and may lead to unwanted results. Please use with care!

SEE ALSO

join_adapt_ids, join_relayer

This document was generated using api2html on Thu Mar 5 09:00:00 MET DST 1998