A Least Random Shuffled Frog-Leaping Algorithm

In the classic implementation of shuffled frog-leaping algorithm, in order to compute the evolution of frog groups (each of size n), the most common way is to first select q (q ≤ n) frogs with unequal probability from each group based on frogs’ individual fitness, and then, choosing the frog with the smallest fitness from selected ones and evolve it. However, the implementation has to generate random number many times and thus lead to remarkable time consumption. To solve this problem, in this paper, we first compute the probability of every frog to be chosen as the one with the lowest fitness and then choose the final frog with roulette algorithm. No random number is generated in the first step, and only one is generated in the second step. Empirical results show that the proposed method outperforms the classic one in time consumption in most common cases. The speedup is 3.2(q = 0.5n, LCG).