Perfect agreement Clause Samples

Perfect agreement. We evaluated how well GAM would perform if the agreement model produced perfect predictions. This is done by letting the agreement model see the true labels, and always return 1 when nodes agree, and 0 oth- erwise. We ran this experiment for all 3 datasets with an MLP base classifier. The results in Ta- Table 2: Accuracy (%) of an MLP with 128 hidden units using GAM with a perfect agreement model. Model Datasets ▇▇▇▇ Citeseer Pubmed MLP128 + GAMp 90.5 76.5 91.6 ble 2 show that a perfect agreement model produces a huge boost, up to 38.8% over the baseline. For Citeseer, the smaller improvement is not surprising, given that only 49% nodes are connected by agreement (see Section 3.2).