Genetic_Programming_Theory_and_Practice_XIII

192 N.F. McPhee et al.

4.1 Working Backwards

A natural place to start our analysis is at the end of the run, when the GP system
created one or more individuals that solved the problem. So we used Neo4j to find
all the ancestors of any “winning” individual, i.e., an individual with a total error of
zero represented than others. As we’ve already mentioned, individual 86:261 has 45
successful offspring, and both individuals 82:447 and 83:047 have five offspring
in the graph, i.e., five offspring that were ancestors of a winning individual in
generation 87. Each of these is marked in Fig. 1 with a shaded diamond.

Gen 79

Gen 80

Gen 81

Gen 82

Gen 83

Gen 84

Gen 85

Gen 86

Gen 87

80:220

82:447

83:124 83:619 83:047

84:319

85:086

86:261

87:941 42 Other Winners 87:719 87:947

Fig. 1 Ancestry of the 45 “winners” from a successful run of replace-space-with-newline using
lexicase.Diamond-shapednodes had an unusually large number of offspring (over 100 each).
Shaded nodeshad at least five offspring that were ancestors of winners