Testing Coverage Criteria for Deep Forests

In practice, many unknown errors have emerged in deep learning systems. One of the main reasons is that the behaviors of deep learning systems are unpredictable and difficult to test. Proper testing criteria are vitally important to evaluate the adequacy of testing deep learning systems. However, there is no testing criterion available for the deep forest, which is a deep learning model that has achieved good performance on small-scale data sets and low-computing-power platform projects. To address this problem, we propose a set of testing coverage criteria for deep forests in this paper. The set of testing coverage criteria is composed of multi-grained scanning node coverage (MGNC), multi-grained scanning leaf coverage (MGLC), cascade forest output coverage (CFOC) and cascade forest class coverage (CFCC).