Merging pattern-matching networks including retes

Abstract

A pattern-matching network is merged with an existing RETE network without recompiling the whole network, while still matching existing data. Permanent, temporary, full and partial merging operations are described, depending on the origination of match patterns among LHS, on demand (procedural programming), or top level. The merge process combines functionally identical nodes from the networks for reducing the number of nodes in the resulting network. The merger process produces so-called alpha and beta synapse nodes; alpha synapse nodes have pointers to top nodes, while beta synapse nodes consist pointers to a drain node, which is a node in the grafted RETE portions of the resulting network and to source memory nodes in the pre-existing RETE network. The algorithm for obtaining a set of synapse nodes without duplication of any logic connecting paths is described. The synapse nodes are used in updating a grafted network, also can be a suspended portion, upon completion of the merger. Two network updating algorithms are presented.