Network Cross-Validation for Nested Models by Edge-Sampling: Selection Consistency

Abstract: In the network literature, a wide range of statistical models have been proposed to exploit structural patterns in the data. Therefore, model selection between different models is a fundamental problem. Cross-validation is a powerful candidate to solve this problem, and Li et al. have already proposed an edge-sampling procedure to choose the number of communities in the block model framework. In this paper, we propose a penalized edge-sampling cross-validation framework for nested network model selection, adding a penalty term to deal with overfitting. We give a general framework applicable in various settings, giving a theoretical guarantee of consistency of the model selection procedure for distinguishing between several widely used models, including the stochastic block model (SBM), the degree-corrected stochastic block model (DCBM), and the graphon model. In summary, our work addresses the problem of model selection over a broad range of settings and fills a theoretical gap in the existing literature. Further numerical investigations will be reported in a subsequent version.