Cost-based Feature Selection for Network Model Choice

Selecting a small set of informative features from large number possibly noisy candidates is challenging problem with many applications in machine learning and approximate Bayesian computation. In practice, the cost computing also needs to be considered. This particularly important for networks because computational costs individual can span several orders magnitude. We addressed this issue network model selection using two approaches. First, we adapted nine feature methods account features. show classes models that reduced by magnitude without considerably affecting classification accuracy (proportion correctly identified models). Second, selected pilot simulations smaller networks. approach factor 50 accuracy. To demonstrate utility our approach, applied it three different yeast protein interaction best-fitting duplication divergence model. Supplementary materials, including computer code reproduce results, are available online.