Authors:
Iris Pijning
1
;
Levi Koppenhol
2
;
Danny Dijkzeul
3
;
Nielis Brouwer
4
;
Sarah L. Thomson
5
and
Daan van den Berg
2
Affiliations:
1
Master Information Studies, UvA Amsterdam, The Netherlands
;
2
VU Amsterdam, The Netherlands
;
3
Cover Genius, The Netherlands
;
4
Rabobank, The Netherlands
;
5
Edinburgh Napier University, U.K.
Keyword(s):
Optimization, Frequency Fitness Assignment, Hillclimber, Job Shop Scheduling Problem, Neutrality.
Abstract:
The Frequency Fitness Assignment (FFA) method steers evolutionary algorithms by objective rareness instead of objective goodness. Does this mean the size of the combinatorial search space influences its performance when compared to more traditional evolutionary algorithms? Our results suggest it does. To address to which extent the search space size matters for the effectiveness of the FFA-principle, we compare the algorithms on 420 Job Shop Scheduling Problem (JSSP) instances systematically generated in gridwise sizes. The comparison of the FFA-hillclimber and the standard hillclimber is done in both EQ setting, accepting equally good (or fitness-frequent) solutions, and NOEQ setting, only accepting improvement. FFA-hillclimbers are more successful than standard hillclimbers on smaller problem instances, but not on larger ones. It seems that the ratio between jobs and machines, influences the success of the respective algorithms for fixed computational budgets.