Extended Version of Table 2

Due to space, we were not able to include complete information on the datasets and models learned within the paper. The table below extends Table 2 from the paper with additional information.

Explanation of column headings:

1. Name -- name of dataset
2. #Vars -- dimension of the dataset
3. Examples -- total number of training and test examples for dataset
4. Tuning -- number of tuning examples (all training examples except for the hold-out set
5. Holdout -- number of examples in the hold-out set
6. Test -- number of examples in the test set
7. NBE -- log likelihood of NBE model on the test data
8. WinMine -- log likelihood of WinMine model on the test data
9. Marginal -- log likelihood of a one-cluster NBE model on the test data
10. Clusters -- number of clusters chosen by NBE
11. Splits -- number of splits in all decision trees in the model learned by WinMine

For cross-validated datasets, numbers listed are the means across all 10 training/test splits. When computing the mean, log likelihoods were properly weighted to give every test example equal weight, even though some cross-validation splits have more test examples than others.