reproducing results from paper (or near)

[eejlny]

Could you please suggest the configuration settings to get close to the paper results?

For example dominant with inj_cora Table 3 in paper "BOND: Benchmarking..." shows a 82.7 ROC-AUC but I do not seem to be able to go over around 76% which is also the accuracy that shows in the notebook example.

Also dominant with the enron dataset shows auc 57% using the benchmark directory while paper reports 73.1%.

benchmark$ python main.py --model dominant --dataset enron

enron DOMINANT AUC: 0.5715±nan (0.5715) AP: 0.0009±nan (0.0009) Recall: 0.0000±nan (0.0000)

What could I be doing wrong ? This seems to be too low for randomness effects.

A short table with the right settings would be useful.

Thanks,