Influence of Experiment Design in GPA Investigating with Respect to PRNGs
Alternative metrics PlumXhttp://hdl.handle.net/11012/179251
MetadataShow full item record
This paper analyses the influence of experiment parameters onto the reliability of experiments with genetic programming algorithms. The paper is focused on the required number of experiments and especially on the influence of parallel execution which affect not only the order of thread execution but also behaviors of pseudo random number generators, which frequently do not respect recommendation of C++11 standard and are not implemented as thread safe. The observations and the effect of the suggested improvements are demonstrated on results of 720,000 experiments.
KeywordsGenetic Programming Algorithm, Efficiency, Pseudo Random Number Generator, Experiment Repeatability, Number of Experiments, Experiment Results Reliability
Document typePeer reviewed
SourceMendel. 2018 vol. 24, č. 2, s. 69–74. ISSN 1803-3814
- Vol. 24, No. 2