In the study of Intrusion Detection System (IDS), based on the traditional machine learning methods, the premise is that the number of samples is enough, therefore, only when the sample size tends to infinity, can the performance be guaranteed in theory.