Next: Parameter Summary
Up: Parameter Selection
Previous: Mutation
Finally, I tested the inclusion of niching Goldberg and Richardson (1987) and elitism.
The top panels of Figure compare the convergence
rates with niching and elitism (left)i, only elitism (middle), and only niching
(right). The bottom panels show a similar comparison for the trace match.
All the other evolution parameters were chosen according to the previous
analysis. It seems that the lack of niching makes the convergence a
little slower in the first iterations (compare the left and middle panels)
but makes it faster after about the 40th generation. The lack of elitism,
on the other hand, makes convergence a little erratic (compare left and right
panels) since we are not guaranteed to go to a
better-fit best individual in any generation compared with the previous one.
Elitism was thus included and niching excluded in the evolution parameters of
the final optimum standard genetic algorithm.
SG_compare_other1
Figure 8 Comparison of results for
the inclusion of niching and elitism. Top panel convergence rates: left with
niching and elitism, middle with elitism only and right with niching only.
Bottom panels show similar comparison for the trace match with
continuous line representing the reference trace and dotted line the inverted
trace.
Next: Parameter Summary
Up: Parameter Selection
Previous: Mutation
Stanford Exploration Project
11/11/2002