Although genetic programming (GP) is derived from genetic
algorithm (GA), there are issues that prevent GP being as efficient as
GA. The validity of a generated program, the design of ephemeral constants,
and the empirical considerations of GP parameters are part of the
issues affecting the convergence of a GP process. Based on many experimental
results, an improved GP scheme is proposed in this paper. The
design of the program representation and the storage of the computation
result(s) ensure that a program is both structurally and behaviorally
valid. Empirical considerations are also discussed, and a few guidelines of
setting GP parameters are derived from the experimental observations.
