AUC keeps 0.5 when training on taobao datasets

[Woolseyyy]

When training DIEN on taobao datasets with the train.py script, I observed that AUC_TEST keeps 0.5 in many hyperparameters settings, e.g., 1e-3 learning rate & 1e-3 reg_lambdas, 1e-3 learning rate & 1e-4 reg_lambdas. The AUC_TEST became normal when the reg_lambda became 1e-5.
Is this normal? Is this consistent with your experiment?