Automated discovery of test statistics using genetic programming.

摘要: The process of developing new test statistics is laborious, requiring the manual development and evaluation mathematical functions that satisfy several theoretical properties. Automating this process, hitherto not done, would greatly accelerate discovery much-needed, statistics. This automation a challenging problem because it requires method to know something about desirable properties good statistic in addition having an engine can develop explore candidate solutions with intuitive representation. In paper we describe genetic programming-based system for automated Specifically, our was able discover as powerful t-test comparing sample means from two distributions equal variances.