Scaling Code Pattern Inference with Interactive What-If Analysis

Figure 1: To guide incremental pattern construction and refinement, SURF summarizes the global distribution of how individual features appear in the entire population in a single collated view. In SURF, users can directly provide code-line level feedback as features in addition to labeling positive and negative instances. SURF then re-generates a refined pattern. Users can contrast the impact of different feature choices using impact analysis and what-if-analysis. It visualizes how a specific feature choice is consistent with already labeled positive and negative instances and can match more instances in the unlabelled population.