Wilcoxon methods are meant to be used for repeated measurements: is the improvement using the same algorithm A still statistically significiant after n itterations? For two independant samples I think the Mann-Whithney test is more appropriated. For a few independant samples the Kruskal-Wallis test.

So I think in your case 50 runs of algorithm A would form a population, 50 runs on algorithm B the second population and afterwards you can use the MW test to check for statistically significiant differences.

signedrank  and ranksum returns the p-value of a two-sided Wilcoxon  test. ranksum test is equivalent to a Mann-Whitney U-test. You should use the ranksum test when the data are not paired. Where the signedrank is used when the the data are paired.

"The Wilcoxon signed-ranks test [Wilcoxon1945] is a nonparametric counterpart of

the paired t-test, which ranks the differences in performances of two algorithms over

each data set. It omits the signs, and compares the ranks for the positive and the

negative differences. The differences are ranked based on their absolute values; in

case of ties average ranks are computed"

http://www.kms.pwr.wroc.pl/staff/trawinski/wp-content/uploads/2013/09/2012-before-shortening-Nonparametric-Statistical-Analysis.pdf

I used one sample KS test and paired sample t- test to identify statistically significant differences among three different algorithms. You may check the following:

Ahmed, Ashik, and Md Shahid Ullah. "Optimal design of proportional–integral controllers for stand-alone solid oxide fuel cell power plant using differential evolution algorithm." SpringerPlus 5.1 (2016): 1. 

@Ibrahim Aljarah Could you tell me, how can I select null hypothesis?