Initialization Method for Grammar-Guided Genetic Programming

New techniques for Grammar Guided Genetic Programming: dealing with large derivation trees and high cardinality terminal symbol sets

“…The maximum number of arguments for the root is set to the maximum tree size. The GP performance may be substantially decreased when using this approach since it is quite common that part of search space syntax trees cannot be generated by this method (García-Arnau, et al, 2007).…”

“…The lowest minimum depth over all the associated production rules is assigned to the nonterminal symbols. These approaches are later extended and formalized (García-Arnau, et al, 2007). Table 7 defines the recursive algorithm to obtain the minimum depth, denoted as *(), of all production rules and symbols of a CFG.…”

“…In GGGP, these operators use a CFG so that the resulting individuals are always feasible solutions belonging to the search space of the problem. The general procedure for the initialization of derivation trees is described in Table 6 (García- Arnau, et al, 2007). Starting from the axiom of the grammar and continuing with the produced nonterminal symbols, productions rules are iteratively applied until only terminals are produced.…”

“…The population initialization method must provide a representative sample of the search space (García-Arnau, et al, 2007), that is, a diverse set of individuals.…”

“…Several approaches that claim more uniform population initialization and reduce code bloat have been presented for GP (Poli, et al, 2008) (Luke & Panait, 2001) (Böhm & Geyer-Schulz, 1996) (Iba, 1996) (Langdon, 2000) (Luke, 2000) ( Chellapilla, 1997) (Trujillo, et al, 2016) (Alonso & Schott, 2013) and GGGP (Gruau, 1996) (García-Arnau, et al, 2007) (Ratle & Sebag, 2000) (Ványi & Zvada, 2003). Moreover, there are also some GP and GGGP variants that pursues a more uniform initial population (Ratle & Sebag, 2002) (Ratle & Sebag, 2001) (Tanev, 2004) (Murphy, et al, 2012) .…”