Fig. 8

Training and testing accuracy using greedy advisor sets. The figure plots average advising accuracy on training and testing benchmarks, using greedy advisor sets learned for the Facet accuracy estimator augmented by predicted coreness. The horizontal axis is advisor set cardinality; the vertical axis is advising accuracy uniformly averaged across difficulty bins. The dashed and solid curves give accuracies for training and testing benchmarks respectively, averaged over cross-validation folds. Since each benchmark is in the testing set of exactly onefold, and the training set of all other folds, averaging over folds uniformly averages over benchmarks