Model interpretation through lower-dimensional posterior summarization
share
Nonparametric regression models have recently surged in their power and popularity, accompanying the trend of increasing dataset size and complexity. While these models have proven their predictive ability in empirical settings, they are often difficult to interpret, and by themselves often do not address the underlying inferential goals of the analyst or decision maker. In this paper, we propose a modular two-stage approach for creating parsimonious, interpretable summaries of complex models which allow freedom in the choice of modeling technique and the inferential target. In the first stage, a flexible model is fit which is believed to be as accurate as possible. Then, in the second stage, a lower-dimensional summary model is fit which is suited to interpretably explain global or local predictive trends in the original model. The summary is refined and refitted as necessary to give adequate explanations of the original model, and we provide heuristics for this summary search. Our methodology is an example of posterior summarization, and so these summaries naturally come with valid Bayesian uncertainty estimates. We apply our technique and demonstrate its strengths on several real datasets.
READ FULL TEXT