Cannot reproduce the performance in the paper

[Wang-Yu-Qing]

I run the code via python src/main.py --input data/amazon --features data/amazon/feature.txt

And I get the performance:

Optimizing
2021-04-19 11:05:57.456952: I tensorflow/core/platform/cpu_feature_guard.cc:140] Your CPU supports instructions that this TensorFlow binary was not compiled to use: AVX2 AVX512F FMA
Training
epoch 0:   0%|| 0/140091 [00:00<?, ?it/s]valid auc: 0.6256135767967163
valid pr: 0.6428760893673962
valid f1: 0.5847500584986922

valid auc: 0.625584291466907:00<?, ?it/s]
valid pr: 0.6428245314390619
valid f1: 0.5844698679691321

valid auc: 0.625586886284441400<?, ?it/s]
valid pr: 0.6428480902602651
valid f1: 0.5844698679691321

valid auc: 0.625579769911787700<?, ?it/s]
valid pr: 0.6428063702870027
valid f1: 0.5847412312832689

valid auc: 0.625559174025684800<?, ?it/s]
valid pr: 0.6427925591891217
valid f1: 0.584881326548049

valid auc: 0.625550895969689500<?, ?it/s]
valid pr: 0.6427865765095414
valid f1: 0.5850214218128291

valid auc: 0.625546650579006500<?, ?it/s]
valid pr: 0.642746090965665
valid f1: 0.5850214218128291
Early Stopping
Overall ROC-AUC: 0.6281
Overall PR-AUC: 0.6443
Overall F1: 0.5863

This is far away from what is list in the origin paper. Is this because the dataset is sampled from the whole dataset?

[Wang-Yu-Qing]

My config is wrong. I can produce the right performance now.